Merge branch 'fix/restore-outcome-and-walk-gaps' into fix/sim-flakes-and-open-race

# Conflicts:
#	docs/BUGS.md

.github/ISSUE_TEMPLATE/config.yml

@@ -3,6 +3,9 @@ contact_links:
   - name: Security vulnerability
     url: https://github.com/narrrl/proton-drive-linux/security/advisories/new
     about: Report privately — never in a public issue.
+  - name: Troubleshooting and FAQ
+    url: https://github.com/narrrl/proton-drive-linux/wiki/Troubleshooting
+    about: Fixes for common problems, and what to include in a bug report.
   - name: Known issues
     url: https://github.com/narrrl/proton-drive-linux/blob/main/docs/BUGS.md
     about: The authoritative ledger of tracked bugs and their status.

.github/PULL_REQUEST_TEMPLATE.md

@@ -21,6 +21,7 @@ cargo test --workspace --locked
 - [ ] Touches recovery, staging, or anything that can be the only copy of user data
 - [ ] Changes packaging, the systemd unit, or desktop entries
 - [ ] Changes the control protocol (daemon and front ends updated together)
+- [ ] Adds or renames a page in `docs/` (listed in `SECTIONS` in `scripts/publish-wiki.sh`, which publishes it to the wiki)
 - [ ] None of the above
 
 <!-- Screenshots for GTK changes. -->

.github/workflows/aur.yml

@@ -0,0 +1,107 @@
+name: AUR
+
+on:
+  # The -git package computes its pkgver from git at build time, so a new commit
+  # upstream needs no AUR push. Only a changed -git recipe is published.
+  push:
+    branches: [main]
+    paths:
+      - 'packaging/aur/proton-drive-for-linux-git/**'
+  # Called by release.yml once the GitHub release and its assets exist.
+  workflow_call:
+    inputs:
+      version:
+        required: true
+        type: string
+    secrets:
+      AUR_SSH_PRIVATE_KEY:
+        required: true
+  workflow_dispatch:
+    inputs:
+      version:
+        description: 'Release to publish, such as 2.1.0. Leave empty to publish only the -git package.'
+        required: false
+        type: string
+
+permissions:
+  contents: write
+
+concurrency:
+  group: aur
+  cancel-in-progress: false
+
+jobs:
+  publish:
+    name: Publish AUR packages
+    runs-on: ubuntu-24.04
+    container: archlinux:base-devel
+    env:
+      # makepkg refuses to run as root, so the packaging steps run as `builder`.
+      HOME: /home/builder
+      VERSION: ${{ inputs.version }}
+
+    steps:
+      - name: Install packaging tools
+        run: pacman -Syu --noconfirm --needed git openssh pacman-contrib
+
+      - name: Checkout repository
+        uses: actions/checkout@v7
+        with:
+          # The updated recipes are committed back to main, not to the tag.
+          ref: main
+
+      - name: Create build user
+        run: |
+          useradd --create-home builder
+          chown -R builder: "$GITHUB_WORKSPACE"
+
+      - name: Configure AUR access
+        env:
+          AUR_SSH_PRIVATE_KEY: ${{ secrets.AUR_SSH_PRIVATE_KEY }}
+        run: |
+          install -d -m700 -o builder -g builder "$HOME/.ssh"
+          printf '%s\n' "$AUR_SSH_PRIVATE_KEY" > "$HOME/.ssh/aur"
+          cat > "$HOME/.ssh/config" <<'EOF'
+          Host aur.archlinux.org
+            IdentityFile ~/.ssh/aur
+            User aur
+          EOF
+          # Pin the host key to the fingerprint published on aur.archlinux.org.
+          # The scan sometimes returns nothing from the runner, so retry it.
+          for attempt in 1 2 3 4 5; do
+            ssh-keyscan -T 15 -t ed25519 aur.archlinux.org > "$HOME/.ssh/known_hosts" &&
+              [ -s "$HOME/.ssh/known_hosts" ] && break
+            echo "ssh-keyscan attempt $attempt returned no key; retrying"
+            sleep 10
+          done
+          ssh-keygen -lf "$HOME/.ssh/known_hosts" | grep -q 'SHA256:RFzBCUItH9LZS0cKB5UE6ceAYhBD5C8GeOBip8Z11+4'
+          chown -R builder: "$HOME/.ssh"
+          chmod 600 "$HOME/.ssh/aur"
+          runuser -u builder -- git config --global user.name 'Nils Pukropp'
+          runuser -u builder -- git config --global user.email 'contact@narl.io'
+
+      - name: Update release packages
+        if: env.VERSION != ''
+        run: runuser -u builder -- packaging/aur/update.sh "${VERSION#v}"
+
+      - name: Refresh the -git metadata
+        if: env.VERSION == ''
+        working-directory: packaging/aur/proton-drive-for-linux-git
+        run: runuser -u builder -- sh -c 'makepkg --printsrcinfo > .SRCINFO'
+
+      - name: Publish to the AUR
+        run: |
+          if [ -n "$VERSION" ]; then
+            runuser -u builder -- packaging/aur/push.sh
+          else
+            runuser -u builder -- packaging/aur/push.sh proton-drive-for-linux-git
+          fi
+
+      - name: Commit the updated recipes to main
+        if: env.VERSION != ''
+        run: |
+          runuser -u builder -- git add packaging/aur
+          if ! runuser -u builder -- git diff --cached --quiet; then
+            runuser -u builder -- git commit -m "chore(aur): update packages to ${VERSION#v}"
+            runuser -u builder -- git push origin HEAD:main
+          fi

.github/workflows/ci.yml

@@ -13,7 +13,7 @@ jobs:
   rust:
     runs-on: ubuntu-24.04
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v7
 
       - name: Install native dependencies
         run: |
@@ -32,14 +32,20 @@ jobs:
         with:
           toolchain: '1.96'
           components: rustfmt, clippy
+          cache: false
 
+      # Tags can only restore caches saved on main, so only main saves one. The
+      # release workflow restores this key for its quality gates.
       - uses: swatinem/rust-cache@v2
+        with:
+          shared-key: debug
+          save-if: ${{ github.ref == 'refs/heads/main' }}
 
       - name: Formatting
         run: cargo fmt --all -- --check
 
       - name: Lints
-        run: cargo clippy --workspace --all-targets -- -D warnings
+        run: cargo clippy --workspace --all-targets --locked -- -D warnings
 
       - name: Unit tests
         run: cargo test --workspace --locked
@@ -57,3 +63,85 @@ jobs:
 
   # Real Proton credentials and /dev/fuse deliberately do not run on pull
   # requests. Follow docs/TESTING.md on a dedicated test-account runner.
+
+  # The seeded simulation runs (docs/TESTING.md): real FUSE
+  # daemons against an in-memory Drive, so they need /dev/fuse but no account.
+  # They run on pull requests too: most data-loss fixes are only exercised
+  # here, and a change that breaks one should not reach main first.
+  simulation:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 30
+
+    steps:
+      - uses: actions/checkout@v7
+
+      - name: Install native dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y \
+            pkg-config \
+            fuse3 \
+            libfuse3-dev \
+            libgtk-4-dev \
+            libadwaita-1-dev \
+            libwebkitgtk-6.0-dev \
+            libsecret-1-dev \
+            libdbus-1-dev
+
+      - name: Install minimum supported Rust
+        uses: actions-rust-lang/setup-rust-toolchain@v1
+        with:
+          toolchain: '1.96'
+          cache: false
+
+      - uses: swatinem/rust-cache@v2
+        with:
+          shared-key: debug
+          save-if: false
+
+      - name: Simulation runs
+        run: 'cargo test -p pdfs-fuse --lib --locked sim:: -- --ignored --test-threads=1'
+
+      # A hung step writes every daemon thread's stack next to the run's
+      # state; without them a hang that does not replay cannot be told apart.
+      - name: Keep the stacks of a hung run
+        if: failure()
+        uses: actions/upload-artifact@v7
+        with:
+          name: simulation-stacks
+          path: /tmp/pdfs-sim-*/stacks.txt
+          if-no-files-found: ignore
+
+  # Keeps the release-profile dependencies cached on main, where the release
+  # workflow can restore them; a tag cannot reuse the caches of earlier tags.
+  release-cache:
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
+    runs-on: ubuntu-24.04
+
+    steps:
+      - uses: actions/checkout@v7
+
+      - name: Install native dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y \
+            pkg-config \
+            libfuse3-dev \
+            libgtk-4-dev \
+            libadwaita-1-dev \
+            libwebkitgtk-6.0-dev \
+            libsecret-1-dev \
+            libdbus-1-dev
+
+      - name: Install minimum supported Rust
+        uses: actions-rust-lang/setup-rust-toolchain@v1
+        with:
+          toolchain: '1.96'
+          cache: false
+
+      - uses: swatinem/rust-cache@v2
+        with:
+          shared-key: release
+
+      - name: Release build
+        run: cargo build --release --locked

.github/workflows/copr.yml

@@ -0,0 +1,66 @@
+name: COPR
+
+on:
+  # Called by release.yml once the GitHub release and its assets exist: COPR
+  # downloads the tag archive and the vendored crates from the release.
+  workflow_call:
+    inputs:
+      version:
+        required: true
+        type: string
+    secrets:
+      COPR_CONFIG:
+        required: true
+  workflow_dispatch:
+    inputs:
+      version:
+        description: 'Release to build, such as 3.5.0. It must already have its vendor tarball.'
+        required: true
+        type: string
+
+permissions:
+  contents: read
+
+concurrency:
+  group: copr
+  cancel-in-progress: false
+
+jobs:
+  build:
+    name: Build in COPR
+    runs-on: ubuntu-24.04
+    container: fedora:latest
+    env:
+      VERSION: ${{ inputs.version }}
+      COPR_PROJECT: narl-rs/proton-drive-linux
+
+    steps:
+      - name: Install packaging tools
+        run: dnf -y install copr-cli rpm-build rpmdevtools
+
+      - name: Checkout repository
+        uses: actions/checkout@v7
+
+      - name: Configure COPR access
+        env:
+          COPR_CONFIG: ${{ secrets.COPR_CONFIG }}
+        run: |
+          install -d -m700 ~/.config
+          printf '%s\n' "$COPR_CONFIG" > ~/.config/copr
+          chmod 600 ~/.config/copr
+
+      # COPR rebuilds the source package without our --define, so the version
+      # has to be written into the spec itself.
+      - name: Build the source package
+        run: |
+          version="${VERSION#v}"
+          spec=packaging/proton-drive-linux.spec
+          sed -i -E "s/^(%\{!\?version: %global version )[^}]*\}/\1${version}}/" "$spec"
+          grep -qF "%global version ${version}}" "$spec"
+          mkdir -p rpmbuild/SOURCES
+          spectool -g -C rpmbuild/SOURCES "$spec"
+          rpmbuild -bs --define "_topdir $PWD/rpmbuild" "$spec"
+
+      # Waits for the build, so a failed COPR build fails this job.
+      - name: Submit to COPR
+        run: copr-cli build "$COPR_PROJECT" rpmbuild/SRPMS/*.src.rpm

.github/workflows/nix.yml

@@ -0,0 +1,59 @@
+name: Nix
+
+# Builds the flake (docs/INSTALL.md#nixos-flake) whenever its recipe, the lock
+# files or the code change, so a NixOS user's `nix build` keeps working. Run it
+# from the Actions tab for any branch to get a build to try.
+on:
+  push:
+    paths:
+      - 'flake.nix'
+      - 'flake.lock'
+      - 'packaging/nix/**'
+      - 'Cargo.lock'
+      - 'Cargo.toml'
+      - 'crates/**'
+      - 'po/**'
+      - '.github/workflows/nix.yml'
+  pull_request:
+    paths:
+      - 'flake.nix'
+      - 'flake.lock'
+      - 'packaging/nix/**'
+      - 'Cargo.lock'
+      - '.github/workflows/nix.yml'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: nix-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  build:
+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v7
+
+      - uses: cachix/install-nix-action@v31
+
+      # Pulls from the project's binary cache, and pushes what this run built
+      # when the token is there; a fork's pull request has none and only pulls.
+      - uses: cachix/cachix-action@v17
+        with:
+          name: proton-drive-linux
+          authToken: ${{ secrets.CACHIX_AUTH_TOKEN }}
+
+      - name: Check the flake
+        run: nix flake check --print-build-logs --no-build
+
+      - name: Build the package
+        run: nix build --print-build-logs .#proton-drive-linux
+
+      - name: Smoke test
+        run: |
+          ./result/bin/pdfs --version
+          for bin in pdfs-app pdfs-tray pdfs-prompt; do test -x "result/bin/$bin"; done
+          grep -q "ExecStart=$(readlink -f result)/bin/pdfs" result/lib/systemd/user/proton-drive.service
+          ls result/share/locale/*/LC_MESSAGES/pdfs.mo

.github/workflows/release.yml

@@ -10,14 +10,68 @@ permissions:
   contents: write
 
 jobs:
+  checks:
+    name: Release quality gates
+    runs-on: ubuntu-24.04
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v7
+
+      - name: Install system dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y \
+            pkg-config \
+            libfuse3-dev \
+            libgtk-4-dev \
+            libadwaita-1-dev \
+            libwebkitgtk-6.0-dev \
+            libsecret-1-dev \
+            libdbus-1-dev
+
+      - name: Set up Rust
+        uses: actions-rust-lang/setup-rust-toolchain@v1
+        with:
+          toolchain: '1.96'
+          components: rustfmt, clippy
+          cache: false
+
+      # A tag can read the caches of main but its own caches are useless to the next
+      # tag, so this only restores what CI saved on main.
+      - name: Cache Cargo target
+        uses: swatinem/rust-cache@v2
+        with:
+          shared-key: debug
+          save-if: false
+
+      - name: Verify tag and manifest versions
+        if: startsWith(github.ref, 'refs/tags/')
+        shell: bash
+        run: |
+          tag_version="${GITHUB_REF_NAME#v}"
+          manifest_version="$(sed -n 's/^version = "\([^"]*\)"/\1/p' Cargo.toml | head -n1)"
+          pkgbuild_version="$(sed -n 's/^pkgver=//p' packaging/PKGBUILD | head -n1)"
+          spec_version="$(sed -n 's/^%{!?version: %global version \([^}]*\)}/\1/p' packaging/proton-drive-linux.spec)"
+          test "$tag_version" = "$manifest_version"
+          test "$tag_version" = "$pkgbuild_version"
+          test "$tag_version" = "$spec_version"
+
+      - name: Release quality gates
+        run: |
+          cargo fmt --all -- --check
+          cargo clippy --workspace --all-targets -- -D warnings
+          cargo test --workspace --locked
+          scripts/fuse-acceptance.sh --offline-only
+
   build-release:
-    name: Build and Release
+    name: Build packages
     runs-on: ubuntu-24.04
     environment: 'rpm signing'
 
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
 
       - name: Install system dependencies
         run: |
@@ -31,33 +85,21 @@ jobs:
             libsecret-1-dev \
             libdbus-1-dev \
             desktop-file-utils \
+            gettext \
             rpm
 
       - name: Set up Rust
         uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           toolchain: '1.96'
-          components: rustfmt, clippy
+          cache: false
 
+      # Restores the release-profile dependencies that CI builds on main.
       - name: Cache Cargo target
         uses: swatinem/rust-cache@v2
-
-      - name: Verify tag and manifest versions
-        if: startsWith(github.ref, 'refs/tags/')
-        shell: bash
-        run: |
-          tag_version="${GITHUB_REF_NAME#v}"
-          manifest_version="$(sed -n 's/^version = "\([^"]*\)"/\1/p' Cargo.toml | head -n1)"
-          pkgbuild_version="$(sed -n 's/^pkgver=//p' packaging/PKGBUILD | head -n1)"
-          test "$tag_version" = "$manifest_version"
-          test "$tag_version" = "$pkgbuild_version"
-
-      - name: Release quality gates
-        run: |
-          cargo fmt --all -- --check
-          cargo clippy --workspace --all-targets -- -D warnings
-          cargo test --workspace --locked
-          scripts/fuse-acceptance.sh --offline-only
+        with:
+          shared-key: release
+          save-if: false
 
       - name: Build release binaries
         run: |
@@ -84,10 +126,17 @@ jobs:
           cp target/release/pdfs-tray dist/binaries/
           cp target/release/pdfs-app dist/binaries/
           cp target/release/pdfs-prompt dist/binaries/
+          po/build.sh dist/binaries/locale
           
           # Create tarball
           tar -czf proton-drive-linux-${{ env.VERSION }}-x86_64.tar.gz -C dist/binaries .
 
+      # Every crate the build needs, so COPR can build the spec without network.
+      - name: Package vendored crates (.tar.xz)
+        run: |
+          cargo vendor --locked dist/vendor > /dev/null
+          tar -cJf proton-drive-linux-vendor-${{ env.VERSION }}.tar.xz -C dist vendor
+
       - name: Package Debian package (.deb)
         run: |
           mkdir -p deb/DEBIAN
@@ -110,6 +159,10 @@ jobs:
           cp packaging/io.narl.proton-drive-linux.svg deb/usr/share/icons/hicolor/scalable/apps/
           cp packaging/proton-drive.service deb/usr/lib/systemd/user/
           cp LICENSE deb/usr/share/licenses/proton-drive-linux/
+          po/build.sh deb/usr/share/locale
+
+          # Stops the running daemon before removal, so the mount stops mirroring deletes
+          install -m755 packaging/deb/prerm deb/DEBIAN/prerm
 
           # Calculate Installed-Size in KB for the control file
           INSTALLED_SIZE=$(du -sk deb | cut -f1)
@@ -124,7 +177,7 @@ jobs:
           Maintainer: Nils Pukropp <contact@narl.io>
           Depends: fuse3, libgtk-4-1 (>= 4.12.0), libadwaita-1-0 (>= 1.5.0), libsecret-1-0, libwebkitgtk-6.0-4, libimage-exiftool-perl
           Installed-Size: ${INSTALLED_SIZE}
-          Homepage: https://github.com/narrrl/proton-drive-linux
+          Homepage: https://proton-drive.narl.io
           Description: Unofficial Proton Drive client for Linux
            Files-on-demand Proton Drive for the Linux desktop, unaffiliated with Proton AG:
            - FUSE virtual mount with block-level caching
@@ -149,6 +202,7 @@ jobs:
             --define "_topdir $PWD/rpmbuild" \
             --define "_sourcedir $PWD" \
             --define "git_dir $PWD" \
+            --define "prebuilt 1" \
             --define "version ${RPM_VERSION}" \
             --define "_dbpath $PWD/rpmbuild/db" \
             packaging/proton-drive-linux.spec
@@ -178,24 +232,111 @@ jobs:
 
 
       - name: Upload Workflow Artifacts
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         with:
           name: proton-drive-linux-builds
           path: |
             *.deb
             *.tar.gz
+            *.tar.xz
             *.rpm
           if-no-files-found: error
 
-      - name: Create GitHub Release
-        if: startsWith(github.ref, 'refs/tags/')
-        uses: softprops/action-gh-release@v2
+  # The flake is the Nix release: a tag is what `github:narrrl/proton-drive-linux/vX.Y.Z`
+  # pins, so it has to build before the release is announced.
+  nix:
+    name: Build the Nix package
+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v7
+
+      - uses: cachix/install-nix-action@v31
+
+      # Pulls from the project's binary cache, and pushes what this run built
+      # when the token is there; a fork's pull request has none and only pulls.
+      - uses: cachix/cachix-action@v17
         with:
+          name: proton-drive-linux
+          authToken: ${{ secrets.CACHIX_AUTH_TOKEN }}
+
+      - name: Build the flake
+        run: nix build --print-build-logs .#proton-drive-linux
+
+      - name: Check the version
+        run: |
+          built="$(./result/bin/pdfs --version)"
+          echo "$built"
+          case "$GITHUB_REF" in
+            refs/tags/v*) grep -qF "${GITHUB_REF_NAME#v}" <<< "$built" ;;
+          esac
+
+  publish:
+    name: Create GitHub Release
+    needs: [checks, build-release, nix]
+    if: startsWith(github.ref, 'refs/tags/')
+    runs-on: ubuntu-24.04
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v7
+
+      - name: Download packages
+        uses: actions/download-artifact@v8
+        with:
+          name: proton-drive-linux-builds
+
+      # The release notes are this version's section of the changelog, from its
+      # heading up to the next one.
+      - name: Extract release notes
+        run: |
+          version="${GITHUB_REF_NAME#v}"
+          awk -v heading="## [$version]" \
+            'index($0, heading) == 1 { found = 1; print; next }
+             found && /^## \[/ { exit }
+             found' docs/CHANGELOG.md > release-notes.md
+          if [ ! -s release-notes.md ]; then
+            echo "::error::docs/CHANGELOG.md has no section for $version"
+            exit 1
+          fi
+
+      - name: Create GitHub Release
+        uses: softprops/action-gh-release@v3
+        with:
+          body_path: release-notes.md
           files: |
             *.deb
             *.tar.gz
+            *.tar.xz
             *.rpm
           draft: false
           prerelease: false
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+  aur:
+    name: Publish AUR packages
+    needs: publish
+    if: startsWith(github.ref, 'refs/tags/')
+    uses: ./.github/workflows/aur.yml
+    with:
+      version: ${{ github.ref_name }}
+    secrets:
+      AUR_SSH_PRIVATE_KEY: ${{ secrets.AUR_SSH_PRIVATE_KEY }}
+
+  copr:
+    name: Build COPR packages
+    needs: publish
+    if: startsWith(github.ref, 'refs/tags/')
+    uses: ./.github/workflows/copr.yml
+    with:
+      version: ${{ github.ref_name }}
+    secrets:
+      COPR_CONFIG: ${{ secrets.COPR_CONFIG }}
+
+  wiki:
+    name: Publish the wiki
+    needs: publish
+    if: startsWith(github.ref, 'refs/tags/')
+    uses: ./.github/workflows/wiki.yml
+    with:
+      ref: ${{ github.ref_name }}

.github/workflows/test-build.yml

@@ -0,0 +1,121 @@
+name: Test build
+
+# On-demand Fedora RPM of any branch, for someone to try a fix before it ships.
+# Run it from the Actions tab ("Run workflow", pick the branch). Built inside a
+# Fedora container so the binaries link against Fedora's own GTK, libadwaita and
+# WebKitGTK, not the Ubuntu ones the release RPM is built against.
+#
+# The RPM is unsigned and versioned `<Cargo version>-1.test.<sha>`, so it sorts
+# below the release it precedes and `dnf` replaces it when that ships.
+on:
+  workflow_dispatch:
+    inputs:
+      fedora:
+        description: 'Fedora release to build on'
+        required: false
+        default: '44'
+        type: string
+      prerelease:
+        description: 'Also publish a pre-release anyone can download without a GitHub login'
+        required: false
+        default: false
+        type: boolean
+
+permissions:
+  contents: read
+
+concurrency:
+  group: test-build-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  rpm:
+    name: Fedora ${{ inputs.fedora }} RPM
+    runs-on: ubuntu-24.04
+    container: fedora:${{ inputs.fedora }}
+    steps:
+      - name: Install build dependencies
+        run: |
+          dnf install -y --setopt=install_weak_deps=False \
+            git rpm-build gcc gcc-c++ make gettext curl \
+            pkgconf-pkg-config fuse3-devel gtk4-devel libadwaita-devel \
+            libsecret-devel dbus-devel glib2-devel webkitgtk6.0-devel
+
+      - name: Checkout repository
+        uses: actions/checkout@v7
+
+      # Fedora's packaged Rust can trail the workspace's minimum (1.96).
+      - name: Set up Rust
+        uses: actions-rust-lang/setup-rust-toolchain@v1
+        with:
+          toolchain: '1.96'
+          cache: false
+
+      - name: Cache Cargo target
+        uses: swatinem/rust-cache@v2
+        with:
+          shared-key: fedora-test-build
+          save-if: false
+
+      - name: Build release binaries
+        run: cargo build --release --locked
+
+      - name: Package RPM
+        run: |
+          git config --global --add safe.directory "$GITHUB_WORKSPACE"
+          VERSION=$(grep -m1 '^version' Cargo.toml | cut -d'"' -f2)
+          SHORT=${GITHUB_SHA::7}
+          mkdir -p rpmbuild/{BUILD,RPMS,SOURCES,SPECS,SRPMS,BUILDROOT,db}
+          rpmbuild -bb \
+            --nodeps \
+            --define "_topdir $PWD/rpmbuild" \
+            --define "_sourcedir $PWD" \
+            --define "git_dir $PWD" \
+            --define "prebuilt 1" \
+            --define "version ${VERSION}" \
+            --define "dist .test.${SHORT}" \
+            --define "_dbpath $PWD/rpmbuild/db" \
+            packaging/proton-drive-linux.spec
+          mkdir -p out
+          cp rpmbuild/RPMS/x86_64/*.rpm out/
+          # The bare binaries, for trying one without installing anything.
+          tar -czf "out/proton-drive-linux-${VERSION}-${SHORT}-fedora${{ inputs.fedora }}-x86_64.tar.gz" \
+            -C target/release pdfs pdfs-tray pdfs-app pdfs-prompt
+          ls -l out
+
+      - name: Upload artifact
+        uses: actions/upload-artifact@v7
+        with:
+          name: proton-drive-linux-fedora${{ inputs.fedora }}-${{ github.sha }}
+          path: out/*
+          retention-days: 14
+          if-no-files-found: error
+
+  prerelease:
+    name: Publish pre-release
+    needs: rpm
+    if: inputs.prerelease
+    runs-on: ubuntu-24.04
+    permissions:
+      contents: write
+    steps:
+      - name: Download build
+        uses: actions/download-artifact@v8
+        with:
+          path: out
+          merge-multiple: true
+
+      # One rolling pre-release per branch: a new run replaces the last one, so
+      # the link a tester was given keeps pointing at the newest build.
+      - name: Publish
+        env:
+          GH_TOKEN: ${{ github.token }}
+          GH_REPO: ${{ github.repository }}
+        run: |
+          TAG="test-$(echo "${GITHUB_REF_NAME}" | tr -c 'A-Za-z0-9._\n' '-')"
+          gh release delete "$TAG" --cleanup-tag --yes 2>/dev/null || true
+          gh release create "$TAG" out/* \
+            --target "$GITHUB_SHA" \
+            --prerelease \
+            --title "Test build of ${GITHUB_REF_NAME} (${GITHUB_SHA::7})" \
+            --notes "Unsigned test build of branch \`${GITHUB_REF_NAME}\` at ${GITHUB_SHA}. Not a release; replaced by the next test build of this branch."

.github/workflows/wiki.yml

@@ -0,0 +1,77 @@
+name: Wiki
+
+# docs/ is the source of the GitHub wiki (docs/DEVELOPMENT.md#common-changes).
+# Every change to the documentation checks its links; release.yml calls this
+# once the release is published, so the wiki documents the latest release. Run
+# it from the Actions tab to publish main in between.
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'docs/**'
+      - 'images/**'
+      - 'README.md'
+      - 'CONTRIBUTING.md'
+      - 'SECURITY.md'
+      - 'scripts/publish-wiki.sh'
+      - '.github/workflows/wiki.yml'
+  pull_request:
+    paths:
+      - 'docs/**'
+      - 'images/**'
+      - 'README.md'
+      - 'CONTRIBUTING.md'
+      - 'SECURITY.md'
+      - 'scripts/publish-wiki.sh'
+      - '.github/workflows/wiki.yml'
+  workflow_call:
+    inputs:
+      ref:
+        description: 'Tag or branch to publish'
+        required: true
+        type: string
+  workflow_dispatch:
+    inputs:
+      ref:
+        description: 'Tag or branch to publish'
+        required: false
+        default: 'main'
+        type: string
+
+permissions:
+  contents: read
+
+jobs:
+  check:
+    name: Check documentation links
+    if: inputs.ref == ''
+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v7
+
+      - name: Check links
+        run: scripts/publish-wiki.sh --check "$RUNNER_TEMP"
+
+  publish:
+    name: Publish to the wiki
+    # Set only by release.yml and the Actions tab, so pushes and pull requests
+    # never publish.
+    if: inputs.ref != ''
+    runs-on: ubuntu-24.04
+    concurrency:
+      group: wiki-publish
+      cancel-in-progress: false
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          ref: ${{ inputs.ref }}
+
+      - name: Publish
+        env:
+          WIKI_REMOTE: https://x-access-token:${{ github.token }}@github.com/${{ github.repository }}.wiki.git
+        run: |
+          git config --global user.name 'github-actions[bot]'
+          git config --global user.email '41898282+github-actions[bot]@users.noreply.github.com'
+          scripts/publish-wiki.sh --push "$RUNNER_TEMP"

.gitignore

@@ -3,3 +3,5 @@
 *.pdb
 .env
 /rpmbuild
+/result
+/result-*

CONTRIBUTING.md

@@ -6,9 +6,10 @@ deliberately high — everything else is ordinary Rust review.
 
 ## Before you start
 
-- Read [`AGENTS.md`](AGENTS.md) for repository conventions and
-  [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) for how the daemon, front ends, and SDK fit
-  together.
+- Read [`docs/DEVELOPMENT.md`](docs/DEVELOPMENT.md) for building, running and debugging from
+  source, and [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) for how the daemon, front ends, and
+  SDK fit together. [`docs/README.md`](docs/README.md) indexes the rest. The same pages are on
+  the [wiki](https://github.com/narrrl/proton-drive-linux/wiki).
 - Check [`docs/BUGS.md`](docs/BUGS.md) — it is the authoritative issue ledger, and your problem may
   already be tracked there with context.
 - For anything larger than a fix, open an issue first so the design can be discussed before you
@@ -23,8 +24,9 @@ cargo build --workspace
 ```
 
 GUI and FUSE targets need the GTK4, libadwaita, WebKitGTK 6, libsecret, D-Bus, and FUSE3
-development packages listed in the [README](README.md#prerequisites). Minimum supported Rust
-version is 1.96.
+development packages listed in [`docs/INSTALL.md`](docs/INSTALL.md#build-dependencies). Minimum
+supported Rust version is 1.96. [`docs/DEVELOPMENT.md`](docs/DEVELOPMENT.md#running-from-source)
+explains how to run your build next to, or instead of, the installed service.
 
 ## Quality gates
 
@@ -37,9 +39,9 @@ cargo test --workspace --locked
 scripts/fuse-acceptance.sh --offline-only
 ```
 
-Changes to the FUSE layer, sync, or the pending-operation queue should also be exercised against a
-real mount before review — see [`docs/TESTING.md`](docs/TESTING.md). Live runs are destructive:
-use a dedicated test account, never one holding data you cannot lose.
+Changes to the FUSE layer, sync, or the queue should also pass the simulation runs and be
+exercised against a real mount before review — see [`docs/TESTING.md`](docs/TESTING.md). Live
+runs are destructive: use a dedicated test account, never one holding data you cannot lose.
 
 ## Making changes
 
@@ -51,6 +53,15 @@ use a dedicated test account, never one holding data you cannot lose.
   control-socket work stays off GTK's main thread.
 - **`staging/` and `recovery/` may hold the only copy of user data.** Nothing may clear or purge
   them opportunistically.
+- **The GUI is fully translated.** Wrap every user-visible string in the helpers from
+  `crates/pdfs-gui/src/i18n.rs`, run `po/update.sh`, and translate the new entries in every
+  catalog in `po/LINGUAS` in the same change. See [`docs/TRANSLATING.md`](docs/TRANSLATING.md).
+  The CLI and daemon messages stay in English.
+- **Document what users see.** A new command, setting or behaviour belongs in the matching page
+  under `docs/` (a user guide topic, the CLI reference or configuration) and in the `[Unreleased]`
+  section of [`docs/CHANGELOG.md`](docs/CHANGELOG.md). Each release publishes `docs/` to the
+  wiki. A new page goes into `SECTIONS` in `scripts/publish-wiki.sh`, and
+  `scripts/publish-wiki.sh --check` must pass. CI runs it on every change to `docs/`.
 
 ## Commits and pull requests
 

Cargo.lock

@@ -467,6 +467,12 @@ dependencies = [
  "digest 0.10.7",
 ]
 
+[[package]]
+name = "block"
+version = "0.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0d8c1fef690941d3e7788d328517591fecc684c084084702d6ff1641e993699a"
+
 [[package]]
 name = "block-buffer"
 version = "0.10.4"
@@ -1814,6 +1820,26 @@ dependencies = [
  "wasm-bindgen",
 ]
 
+[[package]]
+name = "gettext-rs"
+version = "0.7.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5d5857dc1b7f0fee86961de833f434e29494d72af102ce5355738c0664222bdf"
+dependencies = [
+ "gettext-sys",
+ "locale_config",
+]
+
+[[package]]
+name = "gettext-sys"
+version = "0.26.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4ea859ab0dd7e70ff823032b3e077d03d39c965d68c6c10775add60e999d8ee9"
+dependencies = [
+ "cc",
+ "temp-dir",
+]
+
 [[package]]
 name = "ghash"
 version = "0.5.1"
@@ -2829,6 +2855,19 @@ version = "0.8.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "47d9d19d1d6efa0109d2f65ff4c85cddd50bd572e5a00127ab10987290bcefae"
 
+[[package]]
+name = "locale_config"
+version = "0.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "08d2c35b16f4483f6c26f0e4e9550717a2f6575bcd6f12a53ff0c490a94a6934"
+dependencies = [
+ "lazy_static",
+ "objc",
+ "objc-foundation",
+ "regex",
+ "winapi",
+]
+
 [[package]]
 name = "lock_api"
 version = "0.4.14"
@@ -2868,6 +2907,15 @@ version = "0.1.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4050469837a6ff301cd14c1f8f24f88549e6d548f24f64e2148eb0f72cebc51f"
 
+[[package]]
+name = "malloc_buf"
+version = "0.0.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "62bb907fe88d54d8d9ce32a3cceab4218ed2f6b7d35617cafe9adf84e43919cb"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "matchers"
 version = "0.2.0"
@@ -3189,6 +3237,35 @@ dependencies = [
  "syn 2.0.119",
 ]
 
+[[package]]
+name = "objc"
+version = "0.2.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "915b1b472bc21c53464d6c8461c9d3af805ba1ef837e1cac254428f4a77177b1"
+dependencies = [
+ "malloc_buf",
+]
+
+[[package]]
+name = "objc-foundation"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1add1b659e36c9607c7aab864a76c7a4c2760cd0cd2e120f3fb8b952c7e22bf9"
+dependencies = [
+ "block",
+ "objc",
+ "objc_id",
+]
+
+[[package]]
+name = "objc_id"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c92d4ddb4bd7b50d730c215ff871754d0da6b2178849f8a2a2ab69712d0c073b"
+dependencies = [
+ "objc",
+]
+
 [[package]]
 name = "ocb3"
 version = "0.1.0"
@@ -3312,6 +3389,31 @@ dependencies = [
  "system-deps 9.0.0",
 ]
 
+[[package]]
+name = "pangocairo"
+version = "0.22.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bd199a0fef965b3ebda91a929a1ab5e7729538882c68e8de38e1904d2e094757"
+dependencies = [
+ "cairo-rs",
+ "glib",
+ "pango",
+ "pangocairo-sys",
+]
+
+[[package]]
+name = "pangocairo-sys"
+version = "0.22.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a649a616d67ae68539bc85f00fb45d36fba742ed20fc20e6712a15d8ee52bd3f"
+dependencies = [
+ "cairo-sys-rs",
+ "glib-sys",
+ "libc",
+ "pango-sys",
+ "system-deps 9.0.0",
+]
+
 [[package]]
 name = "parking"
 version = "2.2.1"
@@ -3372,7 +3474,7 @@ checksum = "2ee67f1008b1ba2321834326597b8e186293b049a023cdef258527550b9935b4"
 
 [[package]]
 name = "pdfs-cli"
-version = "1.11.0"
+version = "3.6.0"
 dependencies = [
  "anyhow",
  "clap",
@@ -3383,16 +3485,19 @@ dependencies = [
  "tokio",
  "tracing",
  "tracing-subscriber",
+ "zeroize",
 ]
 
 [[package]]
 name = "pdfs-core"
-version = "1.11.0"
+version = "3.6.0"
 dependencies = [
  "async-trait",
+ "base64 0.23.1",
  "dbus-secret-service-keyring-store",
  "directories",
  "ignore",
+ "kamadak-exif",
  "keyring-core",
  "libc",
  "parking_lot",
@@ -3406,13 +3511,15 @@ dependencies = [
  "thiserror 2.0.20",
  "tokio",
  "tracing",
+ "zeroize",
  "zip",
 ]
 
 [[package]]
 name = "pdfs-fuse"
-version = "1.11.0"
+version = "3.6.0"
 dependencies = [
+ "async-trait",
  "base64 0.23.1",
  "fuser",
  "futures",
@@ -3426,20 +3533,24 @@ dependencies = [
  "sha1 0.11.0",
  "tokio",
  "tracing",
+ "tracing-subscriber",
  "zip",
 ]
 
 [[package]]
 name = "pdfs-gui"
-version = "1.11.0"
+version = "3.6.0"
 dependencies = [
  "anyhow",
  "async-channel",
+ "gettext-rs",
  "gtk4",
  "kamadak-exif",
  "ksni",
  "libadwaita",
+ "pangocairo",
  "pdfs-core",
+ "reqwest",
  "serde",
  "serde_json",
  "tokio",
@@ -3465,9 +3576,9 @@ checksum = "9b4f627cb1b25917193a259e49bdad08f671f8d9708acfd5fe0a8c1455d87220"
 
 [[package]]
 name = "pgp"
-version = "0.20.0"
+version = "0.21.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1cfa4743b28656065ff4c0ba09e46b357a65e8c00fc2341e89084b82f87cbdf1"
+checksum = "ae70f4d9325a391db30d115d6191a7bb67cec856aa0bba131290f0c4e09a532a"
 dependencies = [
  "aead 0.5.2",
  "aes 0.8.4",
@@ -3668,9 +3779,9 @@ dependencies = [
 
 [[package]]
 name = "proton-drive-rs"
-version = "0.6.5"
+version = "0.10.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "46cbe05623d9b73bd48b1473f42d43180142b2ee676b1cb9e9a06b562c60ba26"
+checksum = "8f70318394b2815b009c52f804d75747086db93a338ceb6eae62541f2b25d5bf"
 dependencies = [
  "async-trait",
  "base64 0.23.1",
@@ -3695,9 +3806,9 @@ dependencies = [
 
 [[package]]
 name = "proton-sdk"
-version = "0.6.5"
+version = "0.10.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7310d93bed440dfe4fd60306923f6a491901ee0d618b65ace4e5e22ff0387c1f"
+checksum = "4196392032f56b3bbaac0bb1f24eea62fa7e2a0af72004de8d08e518d14720fb"
 dependencies = [
  "aes-gcm 0.11.1",
  "async-trait",
@@ -3724,6 +3835,7 @@ dependencies = [
  "thiserror 2.0.20",
  "tokio",
  "tracing",
+ "zeroize",
 ]
 
 [[package]]
@@ -4055,6 +4167,18 @@ dependencies = [
  "syn 3.0.5",
 ]
 
+[[package]]
+name = "regex"
+version = "1.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f020237b6c8eed93db2e2cb53c00c60a8e1bc73da7d073199a1180401450218d"
+dependencies = [
+ "aho-corasick",
+ "memchr",
+ "regex-automata",
+ "regex-syntax",
+]
+
 [[package]]
 name = "regex-automata"
 version = "0.4.18"
@@ -4796,6 +4920,12 @@ version = "0.13.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "adb6935a6f5c20170eeceb1a3835a49e12e19d792f6dd344ccc76a985ca5a6ca"
 
+[[package]]
+name = "temp-dir"
+version = "0.1.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "83176759e9416cf81ee66cb6508dbfe9c96f20b8b56265a39917551c23c70964"
+
 [[package]]
 name = "tempfile"
 version = "3.27.0"

Cargo.toml

@@ -3,20 +3,20 @@ resolver = "2"
 members = ["crates/pdfs-core", "crates/pdfs-fuse", "crates/pdfs-cli", "crates/pdfs-gui"]
 
 [workspace.package]
-version = "1.11.0"
+version = "3.6.0"
 edition = "2024"
 license = "MIT"
 authors = ["Nils Pukropp <contact@narl.io>"]
 rust-version = "1.96"
 repository = "https://github.com/narrrl/proton-drive-linux"
-homepage = "https://github.com/narrrl/proton-drive-linux"
+homepage = "https://proton-drive.narl.io"
 readme = "README.md"
 keywords = ["proton", "drive", "fuse", "sync", "linux"]
 categories = ["filesystem", "command-line-utilities"]
 
 [workspace.dependencies]
-proton-sdk = { version = "0.6.5" }
-proton-drive-rs = { version = "0.6.5" }
+proton-sdk = { version = "0.10.0" }
+proton-drive-rs = { version = "0.10.0" }
 
 # Internal crates.
 pdfs-core = { path = "crates/pdfs-core" }
@@ -26,6 +26,7 @@ tokio = { version = "1", features = ["rt-multi-thread", "macros", "sync", "fs",
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 base64 = "0.23"
+zeroize = "1"
 thiserror = "2"
 anyhow = "1"
 # Non-poisoning mutexes. A handler that panics while holding a lock must not
@@ -60,3 +61,15 @@ async-trait = "0.1"
 futures = "0.3"
 fuser = "0.18"
 libc = "0.2"
+
+[profile.release]
+# No package ships a separate debug package (the PKGBUILD sets `!debug`, the
+# RPM spec sets `debug_package` to nil), and the unstripped binaries weighed
+# 43 MB (`pdfs`), 24 MB (`pdfs-app`) and 12 MB (`pdfs-tray`).
+strip = "symbols"
+# Off on purpose: LTO has broken GTK links for this workspace, which is why
+# the packages turn it off as well.
+lto = false
+# `Workers` and the thumbnail build catch panics to keep the daemon alive, so
+# unwinding must stay on.
+panic = "unwind"

README.md

@@ -4,488 +4,197 @@
 
 # Proton Drive for Linux
 
-**Files-on-demand Proton Drive for the Linux desktop** — a FUSE virtual mount with block-level
-caching, a scriptable CLI, and a native GTK4 app with tray and search launcher.
+**Proton Drive on the Linux desktop.** Your files as a folder that downloads on demand, synced
+local folders, Photos, sharing, and a native GTK4 app with tray and search launcher, plus a
+scriptable CLI.
+
+**[proton-drive.narl.io](https://proton-drive.narl.io)**
 
 [![CI](https://github.com/narrrl/proton-drive-linux/actions/workflows/ci.yml/badge.svg)](https://github.com/narrrl/proton-drive-linux/actions/workflows/ci.yml)
 [![Release](https://img.shields.io/github/v/release/narrrl/proton-drive-linux?sort=semver)](https://github.com/narrrl/proton-drive-linux/releases/latest)
+[![AUR](https://img.shields.io/aur/version/proton-drive-for-linux)](https://aur.archlinux.org/packages/proton-drive-for-linux)
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 [![Rust 1.96+](https://img.shields.io/badge/rust-1.96%2B-orange.svg)](https://www.rust-lang.org)
-[![Platform: Linux](https://img.shields.io/badge/platform-linux-lightgrey.svg)](#prerequisites)
+[![Buy me a coffee](https://img.shields.io/badge/Buy%20me%20a%20coffee-FFDD00?logo=buymeacoffee&logoColor=black)](https://buymeacoffee.com/narl)
 
-<img src="images/files.png" alt="The GTK4 file browser" width="820">
+<img src="images/files.png" alt="My files in the Proton Drive app" width="820">
 
 </div>
 
 > [!IMPORTANT]
 > This is an **unofficial**, community-built client. It is not affiliated with, endorsed by, or
 > supported by Proton AG. "Proton" and "Proton Drive" are trademarks of Proton AG, used here only
-> to describe what the software talks to. Use a test account first, and read
-> [`docs/RECOVERY.md`](docs/RECOVERY.md) before trusting it with data you cannot lose.
-
-## Install
-
-```bash
-# Debian / Ubuntu
-sudo apt install ./proton-drive-linux_*.deb
-
-# Fedora
-sudo dnf install ./proton-drive-linux-*.rpm
-
-# Arch Linux
-cd packaging && makepkg -fi
-```
-
-Packages for each release are attached to the [latest
-release](https://github.com/narrrl/proton-drive-linux/releases/latest); see
-[Installation & Packages](#installation--packages) for details and
-[Building from Source](#building-from-source) to compile it yourself.
-
-```bash
-pdfs login          # sign in (SRP + 2FA, credentials go to the system keyring)
-pdfs status         # mount state, queued writes, cache usage
-pdfs-app            # the desktop app
-```
-
-## Contents
-
-- [Features](#features) · [Quick search](#quick-search-and-media-opening) · [Opening files](#choosing-how-files-open)
-- [Selective sync](#selective-sync-pdfsignore) · [Diagnostics](#diagnostics--maintenance) · [Scripting](#scripting---json)
-- [Performance](#performance--caching) · [Filesystem safety](#filesystem-safety) · [Screenshots](#screenshots)
-- [Prerequisites](#prerequisites) · [Building](#building-from-source) · [Packages](#installation--packages) · [Releases](#automated-releases-cicd)
-
-**Documentation:** [Architecture](docs/ARCHITECTURE.md) · [Testing](docs/TESTING.md) ·
-[Recovery](docs/RECOVERY.md) · [Development](docs/DEVELOPMENT.md) ·
-[Known issues](docs/BUGS.md) · [Changelog](docs/CHANGELOG.md) ·
-[Contributing](CONTRIBUTING.md) · [Security policy](SECURITY.md)
+> to describe what the software talks to. Try it with data you have another copy of first, and
+> read [what survives a lost machine](docs/RECOVERY.md) before you rely on it.
 
 ## Features
 
-- **Files-on-Demand (FUSE)**: Mount your Proton Drive as a virtual filesystem where files are downloaded only when opened, utilizing smart block-level caching and disk-backed writes.
-- **Command-Line Interface (CLI)**: Manage your drive, authenticate, and monitor sync status directly from the terminal.
-- **Non-Blocking GTK4 Desktop App**: Browse files, manage pins, configure options, and monitor active transfers through a modern, native GUI with a fully non-blocking asynchronous main loop.
-- **File Thumbnails**: Browse locally cached image previews in Files, Shared, and Trash, including embedded previews from camera RAW files. Build a whole folder tree from the Files toolbar and resize the grid from the status bar.
-- **Shared Links & Invites**: Browse files shared with you by other users, and view/manage your own public shared links directly in the GUI.
-- **Local Backup (Computers)**: Sync and back up local directories (like Downloads, Documents, Pictures, etc.) directly to your Proton Drive account.
-- **Locations**: One page (and `pdfs locations`) listing every place Proton Drive occupies on this machine — the main mount plus each backed-up folder — with its mode, sync state, and whether it is read-only. Switching a folder between a full local copy and on-demand happens here.
-- **System Tray Integration**: Background indicator for status monitoring, quick actions, and fast sync controls.
-- **Unified Search Launcher (HUD)**: A resident Google Drive-style launcher (`pdfs-prompt`) that searches Proton Drive and local files together, ranks the best matches, tolerates abbreviations and typos, and is ideal for a system-wide hotkey. It can also present the same search in your own launcher (`--dmenu`) or, searching live as you type, in `fzf` (`--fzf`).
-- **Secure Credential Storage**: Integrates with the system Secret Service (GNOME Keyring, KWallet, etc.) with smart in-memory credential caching to avoid UI thread blockages.
-- **Proton Photos Support**: Access your Proton Photos timeline, view thumbnails, and download backed-up media natively (available in the GUI as a navigation tab and via the CLI).
-- **File Version History**: Every revision Proton Drive still holds for a file, from the browser's details pane (**Versions**) or `pdfs versions list|restore|save|rm`. Restoring is server-side — no re-upload — and an old version can be written out to a local file without touching the live one.
-- **Photo Favourites**: Star a photo in the lightbox, filter the gallery to favourites, or use `pdfs favorite <uid>` / `pdfs photos --favorites`.
-- **Photo Albums**: Browse your albums — including the ones shared with you — from the Albums view of the Photos page, or with `pdfs albums` / `pdfs album <uid>`. Album contents open in the same gallery as the timeline.
-- **Human Verification (CAPTCHA) Recovery**: Detects sign-in gates (VPN/new IP challenges) and launches an embedded `WebKitWebView` dialog to safely complete the challenge, transparently retrying authentication with the earned token.
-- **Selective Sync (`.pdfsignore`)**: Keep build trees, dependency directories, and editor leftovers out of synced folders using gitignore-style rules.
-- **Data-Safe Offline Writes**: Durable scratch/staging files and a transactional pending queue preserve acknowledged writes across network failures and restarts.
+**Files**
 
-## Quick Search and Media Opening
+- **Your Drive as a folder.** `~/ProtonDrive` works with every program. Files download when they
+  are opened, a block at a time, so a video starts playing and a large archive can be read
+  without fetching all of it.
+- **Available offline.** Keep chosen files and folders fully downloaded; everything else is
+  cached within a size limit you set.
+- **Synced folders.** Back up folders such as `~/Documents` under your computer in Proton Drive,
+  as a full two-way copy or online only. Gitignore-style `.pdfsignore` rules, conflict copies
+  instead of overwrites, pause and bandwidth limits.
+- **Works at local speed.** Making, renaming and deleting files and folders returns at once, online
+  or offline. Changes are staged on disk and queued before they are acknowledged, so they survive
+  a lost connection, a crash or a reboot and reach Drive when they can.
 
-`pdfs-prompt` searches Drive metadata and the local home-directory index in one request. Results are ranked together rather than split by source. Matching is case-insensitive and supports exact names, prefixes, parent paths, multiple terms, punctuation-separated words, ordered abbreviations, common typos, and adjacent transpositions.
+**Desktop app**
 
-The prompt stays resident after its first launch, so invoking the shortcut again reuses and resets the existing window. Drive folders open from the mount. Audio and video results also open through FUSE, allowing media players to seek and request ranges without downloading the entire file first; ordinary Drive files are materialized into the local cache before opening.
+- **Browse** My files, Shared with me, Shared by me, Computers and Trash, with thumbnails
+  (including camera RAW), drag and drop, details and **version history**.
+- **Photos.** Timeline with favorites, albums, and Places on a map; RAW+JPEG pairs shown as one
+  photo; duplicate finder; slideshow; **import from Google Photos** (Takeout).
+- **Sharing.** Invite people as viewer, editor or admin, create public links with password and
+  expiry, accept invitations.
+- **Sync overview.** What is uploading, what is waiting, what failed and why, and a history of
+  what happened. A change Drive refuses (storage full, access taken away) shows at once, on the
+  file too, and can be exported or discarded.
+- **Tray icon** with status, pause and quick actions, and a **search launcher** (`pdfs-prompt`)
+  that searches Proton Drive and your home folder together, as its own window, through fuzzel,
+  rofi or wofi, or live in `fzf`.
+- **Restore a computer.** Continue a backup on a new machine and bring its folders back in one
+  step.
+- **Sign-in** with two-factor authentication and CAPTCHA; the session is kept in the system
+  keyring.
+- Available in English, German, Spanish, French, Italian, Dutch, Polish and Brazilian
+  Portuguese.
 
-### Using your own launcher (fuzzel, rofi, wofi, …)
+**Command line**
 
-`pdfs-prompt --dmenu` runs the same search through an external dmenu-style
-launcher instead of the built-in window, so the HUD matches the rest of a
-tiling-WM desktop:
+- About 55 `pdfs` commands cover files, sync, sharing, Photos, Trash, versions and diagnostics.
+  Query commands print JSON with `--json`.
 
-```bash
-pdfs-prompt --dmenu                                   # fuzzel/rofi/wofi/tofi/bemenu/dmenu, whichever is installed
-pdfs-prompt --dmenu --menu 'fuzzel --dmenu --width 60'
-pdfs-prompt --dmenu --query invoice                   # skip straight to a search
-```
+## Install
 
-A launcher filters a fixed list — it cannot ask for a new one per keystroke — so
-searching takes two steps, and the prompt says which step you are on:
-
-- `Search Drive ›` lists your pinned files. Type anything that isn't one of them
-  and press Enter to search for it.
-- `Drive: invoice ›` lists the results. Enter opens one; typing something else
-  and pressing Enter searches again. Escape closes.
-
-`--query` skips the first step. fuzzel and rofi also get file-type icons.
-
-To make an existing keybinding use it without re-binding, and to pin the
-launcher command, set them in `config.json`:
-
-```json
-{
-  "prompt": {
-    "mode": "dmenu",
-    "menu": ["fuzzel", "--dmenu", "--width", "60"],
-    "menu_limit": 50
-  }
-}
-```
-
-`--gtk` overrides `"mode": "dmenu"` for one invocation. A `{prompt}` token
-anywhere in `menu` is replaced by the prompt text; without one, the launcher's
-own prompt flag is appended.
-
-### Live launcher search (fzf)
-
-The two-step above exists because no dmenu-style launcher can call back for new
-results per keystroke. `fzf` can, so `pdfs-prompt --fzf` searches as you type,
-exactly like the built-in window:
-
-```bash
-pdfs-prompt --fzf                    # pinned files, then live results as you type
-pdfs-prompt --fzf --query invoice    # start on a search
-```
-
-Every keystroke re-queries the daemon (debounced, like the GTK prompt) and fzf's
-own matcher is switched off, so the ordering you see is the daemon's ranking —
-abbreviations and typos included, which a launcher filtering a fixed list cannot
-do. Enter opens; Escape closes.
-
-fzf is a terminal program, so when it is started without one — from a WM
-keybinding — it spawns a terminal for itself: foot, ghostty, kitty, alacritty,
-wezterm or xterm, whichever is installed. Each is launched with the app-id/class
-`pdfs-prompt`, so one window rule floats it whichever terminal you have. In
-Hyprland:
-
-```
-windowrulev2 = float, class:^(pdfs-prompt)$
-windowrulev2 = size 900 500, class:^(pdfs-prompt)$
-windowrulev2 = center, class:^(pdfs-prompt)$
-bind = SUPER, space, exec, pdfs-prompt --fzf
-```
-
-Set it as the default front end, and pick the terminal explicitly, in
-`config.json`:
-
-```json
-{
-  "prompt": {
-    "mode": "fzf",
-    "terminal": ["foot", "--app-id=pdfs-prompt", "--"],
-    "menu_limit": 50
-  }
-}
-```
-
-A `{cmd}` token anywhere in `terminal` is replaced by the command to run;
-without one it is appended, which is what every detected terminal expects.
-
-## Choosing How Files Open
-
-By default every result is handed to `xdg-open`. An `open_with` block in
-`config.json` overrides that per file type — for instance opening text in
-Neovim inside Alacritty, while everything else still goes to the desktop:
-
-```json
-{
-  "open_with": {
-    "terminal": ["alacritty", "-e"],
-    "rules": [
-      { "match": ["@text"], "command": ["nvim"], "terminal": true },
-      { "match": ["*.png", "*.jpg"], "command": ["imv"] }
-    ]
-  }
-}
-```
-
-- `match` takes file-name globs (`*.md`, `notes-*.txt`) or the classes `@dir`,
-  `@text`, `@document`, `@image`, `@media`, `@any`. First matching rule wins.
-- `command` is argv, not a shell line. A `{}` token is replaced by the path;
-  without one the path is appended. `$VAR` tokens expand from the environment,
-  so `["$EDITOR"]` follows your editor.
-- `"terminal": true` wraps the command in `terminal`. When that is unset, the
-  terminal comes from `$PDFS_TERMINAL`, then `$TERMINAL`, then the first known
-  emulator on `PATH`; a bare name like `TERMINAL=alacritty` gains the right
-  "run this" flag automatically.
-- `"default": [...]` replaces `xdg-open` for everything unmatched.
-
-The rules apply to the prompt (both front ends) and to the GTK browser, so a
-file opens the same way wherever it was picked. Patterns are matched against the
-Drive name — a downloaded file is stored in the cache under its content hash,
-which says nothing about its type, so matching that would send everything to
-`xdg-open`.
-
-## Selective Sync (`.pdfsignore`)
-
-Two-way synced folders skip paths matched by ignore rules, so syncing a project
-directory does not upload `node_modules/`, `target/`, or `.git/`.
-
-Rules come from two places, and both apply:
-
-1. **Per folder** — a `.pdfsignore` file at the root of the synced folder
-   (`.protonignore` also works). Gitignore syntax, including negation:
-
-   ```gitignore
-   # everything build-related
-   build/
-   *.log
-
-   # ...except this one
-   !important.log
-   ```
-
-2. **Globally** — an `ignore_patterns` list in `config.json`, applied to every
-   synced folder. When unset, sensible defaults apply: `.git/`, `.hg/`, `.svn/`,
-   `node_modules/`, `target/`, `.venv/`, `__pycache__/`, `*~`, `*.swp`, `*.tmp`,
-   `.DS_Store`, and `Thumbs.db`.
-
-   ```json
-   {
-     "ignore_patterns": ["node_modules/", "target/", "*.iso"]
-   }
-   ```
-
-   Set it to `[]` to opt out of the defaults entirely.
-
-Rules are re-read at the start of every sync pass, so edits take effect on the
-next pass without restarting the daemon.
-
-**Ignoring is never destructive.** If a rule starts matching a file that was
-already synced, its copy on Drive is left untouched — the file simply stops
-being tracked. Removing the rule later picks the existing remote file back up
-rather than re-uploading it.
-
-## Diagnostics & Maintenance
-
-When something looks wrong, `pdfs diagnose` checks the installation and prints a
-report. It runs without a daemon on purpose — the state worth diagnosing is
-usually the state where the daemon will not start:
-
-```console
-$ pdfs diagnose
-Paths
-[ok  ]   state dir: /home/you/.local/state/proton-drive-linux
-[ok  ]   database: /home/you/.local/state/proton-drive-linux/cache.db (170.3 MiB)
-
-Account
-[ok  ]   keyring session: you@proton.me
-
-Daemon
-[ok  ]   daemon responding
-[ok  ]   mounted at: /home/you/ProtonDrive
-[ok  ]   queued writes: none
-
-No problems found.
-```
-
-It exits non-zero if any check fails, so it works in a health-check script.
-
-For the local metadata database and content cache:
-
-| Command | What it does |
+| Distribution | Package |
 |---|---|
-| `pdfs cache inspect` | Database size, reclaimable space, per-table row counts, cache usage against budget |
-| `pdfs cache inspect --deep` | Also runs SQLite's integrity check — reads every page, so it is slow on a large database |
-| `pdfs cache vacuum` | Checkpoints the write-ahead log and compacts the database |
-| `pdfs cache clear` | Deletes cached file content, keeping pinned files |
+| Fedora 44 or newer | [`narl-rs/proton-drive-linux`](https://copr.fedorainfracloud.org/coprs/narl-rs/proton-drive-linux/) in COPR, updated with `dnf upgrade` |
+| Arch Linux | [`proton-drive-for-linux`](https://aur.archlinux.org/packages/proton-drive-for-linux) (also `-bin` and `-git`) in the AUR |
+| Debian, Ubuntu 24.04 or newer | `.deb` from the [latest release](https://github.com/narrrl/proton-drive-linux/releases/latest) |
+| NixOS | The flake in this repository: see [Installation → NixOS](docs/INSTALL.md#nixos-flake) |
+| Anything else | [Build from source](docs/INSTALL.md#build-from-source) |
 
-`vacuum` takes a write lock for its duration and needs room for a second copy of
-the database while it runs, so it is a deliberate operation rather than
-something the daemon does on a timer.
+**Fedora**
 
-Before a release, run the real-kernel acceptance suite against a dedicated test
-account and mount. See [`docs/TESTING.md`](docs/TESTING.md); it includes optional
-two-mount convergence checks and confines destructive operations to a fresh
-test directory.
-
-## Scripting (`--json`)
-
-Query commands accept a global `--json` flag and emit machine-readable output:
-
-```console
-$ pdfs --json sync list | jq '.items[] | select(.state != "idle") | .local_path'
-$ pdfs --json cache inspect | jq '.db_reclaimable_bytes'
-$ pdfs --json status | jq -r '.mount.mountpoint'
+```bash
+sudo dnf copr enable narl-rs/proton-drive-linux
+sudo dnf install proton-drive-linux
 ```
 
-Supported on `status`, `ls`, `pins`, `sync list`, `devices list`, `transfers`,
-`activity`, and `cache inspect`. Commands that perform an action keep their
-human output — a script that needs to know whether one worked has the exit code.
+**Arch Linux**
 
-Two things worth relying on:
+```bash
+paru -S proton-drive-for-linux
+```
 
-- **The payload is unwrapped.** Output is `{"items": […]}`, not the daemon's
-  internal `{"SyncFolders": {"items": […]}}`, so scripts never name an internal
-  variant.
-- **Errors still fail.** A daemon-side error prints its JSON body (with a
-  machine-readable `kind`) on stdout *and* exits non-zero, so `set -e` and
-  `if pdfs …` behave as expected rather than treating a failed lookup as
-  success.
+**Debian and Ubuntu:** download the `.deb` from the
+[latest release](https://github.com/narrrl/proton-drive-linux/releases/latest), then:
 
-## Performance & Caching
+```bash
+sudo apt install ./proton-drive-linux_*_amd64.deb
+```
 
-The client includes several optimizations designed for high efficiency, a low memory footprint, and a responsive user experience:
+Then open **Proton Drive** from your applications menu and sign in. The app starts the background
+service, and your files appear in `~/ProtonDrive`.
 
-- **On-Demand Block Cache**: Files are read in fixed 4 MiB blocks. For unpinned files, the client fetches and caches only the blocks containing the requested byte ranges. This enables fast sequential and sparse reads (e.g. media streaming or metadata scanning) without downloading entire files.
-- **Disk-Backed Writes**: Large file writes are staged on disk in temporary scratch files (rather than fully buffered in RAM) and track modified byte intervals. Only the unedited remote segments are fetched at commit time, keeping memory usage minimal.
-- **Non-Blocking GTK4 Loop**: To prevent UI freezes, all synchronous D-Bus credential checks, control socket requests, and cache usage calculations are offloaded to background worker threads or fetched asynchronously.
-- **Flicker-Free UI Rendering**: The GTK4 application performs differential rendering of the pins list, only modifying list rows when the list content changes, preserving the user's scroll position and widget focus.
-- **Durable Staging**: Scratch metadata and staged writes are synced and atomically published before an upload is acknowledged locally. Rapid revisions supersede pending work transactionally.
+Signing in from a terminal instead? `pdfs login` stores the session but does not start the
+service. Enable it once:
 
-## Filesystem Safety
+```bash
+pdfs login
+systemctl --user enable --now proton-drive.service
+pdfs status
+```
 
-Version 1.0 strengthens the boundaries around local-only data and destructive reconciliation:
+New to the client? [Getting started](docs/GETTING_STARTED.md) walks through the first steps.
+[Installation](docs/INSTALL.md) has the optional extras (tray on GNOME, RAW and video
+thumbnails), updating, headless use and uninstalling.
 
-- Truncate composes with queued revisions, preserving the authoritative prefix and correct zero-filled growth.
-- Combined move-and-rename operations are queued durably and tolerate partially completed remote state.
-- Session tombstones prevent successfully unlinked entries from reappearing through eventually consistent listings.
-- Names are valid UTF-8 and no longer than Linux's 255-byte component limit.
-- Incomplete mirror scans are non-destructive, and total-wipe protection covers every non-empty baseline.
-- Failed conflict preservation, local deletion, or staging publication does not settle a new baseline or discard the only local copy.
-- State, cache, and config directories must be real, current-user-owned `0700` directories; control sockets are `0600`.
+## Documentation
 
+The documentation is in [`docs/`](docs/README.md) and on the
+[wiki](https://github.com/narrrl/proton-drive-linux/wiki).
+
+| | |
+|---|---|
+| [Getting started](docs/GETTING_STARTED.md) | From installing to your first synced folder |
+| [User guide](docs/USER_GUIDE.md) | Everything the app, tray and launcher do |
+| [FAQ](docs/FAQ.md) | Short answers to common questions |
+| [Command-line reference](docs/CLI.md) | Every `pdfs` command and scripting with `--json` |
+| [Configuration](docs/CONFIGURATION.md) | `config.json`, file locations, environment variables, the service |
+| [Troubleshooting](docs/TROUBLESHOOTING.md) | Diagnosing problems and fixing common ones |
+| [Recovering on a new computer](docs/RECOVERY.md) | Getting your synced folders back after a reinstall or a lost machine |
+| [Security model](docs/SECURITY_MODEL.md) | What protects your files, what stays on this computer, what the client connects to |
+| [Changelog](docs/CHANGELOG.md) | What changed in each release |
+
+For contributors: [development guide](docs/DEVELOPMENT.md), [architecture](docs/ARCHITECTURE.md),
+[testing](docs/TESTING.md), [translating](docs/TRANSLATING.md), [roadmap](docs/ROADMAP.md) and the
+[bug ledger](docs/BUGS.md). The [documentation index](docs/README.md) lists everything.
 
 ## Screenshots
 
-### GUI Application & Launcher
-
 <table>
   <tr>
-    <td align="center" width="50%"><img src="images/login.png" alt="Login Screen" width="100%"><br><sub><b>Login Screen</b></sub></td>
-    <td align="center" width="50%"><img src="images/files.png" alt="Files Browser" width="100%"><br><sub><b>Files Browser</b></sub></td>
+    <td align="center" width="50%"><img src="images/files.png" alt="My files" width="100%"><br><sub><b>My files</b></sub></td>
+    <td align="center" width="50%"><img src="images/photos.png" alt="Photos timeline" width="100%"><br><sub><b>Photos</b></sub></td>
   </tr>
   <tr>
-    <td align="center" width="50%"><img src="images/shared.png" alt="Shared Links" width="100%"><br><sub><b>Shared Links</b></sub></td>
-    <td align="center" width="50%"><img src="images/shared_with_me.png" alt="Shared with me" width="100%"><br><sub><b>Shared with me</b></sub></td>
+    <td align="center" width="50%"><img src="images/photos_map.png" alt="Photos places map" width="100%"><br><sub><b>Places</b></sub></td>
+    <td align="center" width="50%"><img src="images/computers.png" alt="Computers" width="100%"><br><sub><b>Computers</b></sub></td>
   </tr>
   <tr>
-    <td align="center" width="50%"><img src="images/computers.png" alt="Local Backups (Computers)" width="100%"><br><sub><b>Local Backups (Computers)</b></sub></td>
-    <td align="center" width="50%"><img src="images/photos.png" alt="Photos Timeline" width="100%"><br><sub><b>Photos Timeline</b></sub></td>
+    <td align="center" width="50%"><img src="images/sync.png" alt="Sync folders" width="100%"><br><sub><b>Sync</b></sub></td>
+    <td align="center" width="50%"><img src="images/conflicts.png" alt="Sync history showing resolved conflicts" width="100%"><br><sub><b>Sync history</b></sub></td>
   </tr>
   <tr>
-    <td align="center" width="50%"><img src="images/prompt.png" alt="Search Launcher Prompt" width="100%"><br><sub><b>Search Launcher Prompt</b></sub></td>
-    <td align="center" width="50%"><img src="images/settings.png" alt="Settings" width="100%"><br><sub><b>Settings</b></sub></td>
+    <td align="center" width="50%"><img src="images/prompt_gtk.png" alt="Search launcher" width="100%"><br><sub><b>Search launcher</b></sub></td>
+    <td align="center" width="50%"><img src="images/prompt_fzf.png" alt="Search launcher in fzf" width="100%"><br><sub><b>Search launcher in fzf</b></sub></td>
+  </tr>
+  <tr>
+    <td align="center" width="50%"><img src="images/login.png" alt="Sign in" width="100%"><br><sub><b>Sign in</b></sub></td>
+    <td></td>
   </tr>
 </table>
 
----
+## How it works
 
-## Prerequisites
+A background service (`proton-drive.service`, the `pdfs daemon` command) holds your session,
+mounts `~/ProtonDrive` through FUSE, runs the upload queue and the sync engine, and follows
+Proton's event stream for changes made elsewhere. The app, the tray, the launcher and the CLI are
+clients of that service over a local socket. Everything is end-to-end encrypted with your Proton
+keys before it leaves the machine, through a Rust implementation of Proton's Drive SDK.
 
-To compile the application from source or run the built binaries, ensure you have the following system libraries installed on your distribution:
-
-### Ubuntu / Debian (24.04+)
-```bash
-sudo apt-get update
-sudo apt-get install -y \
-  pkg-config \
-  libfuse3-dev \
-  libgtk-4-dev \
-  libadwaita-1-dev \
-  libwebkitgtk-6.0-dev \
-  libsecret-1-dev \
-  libdbus-1-dev \
-  libimage-exiftool-perl
-```
-
-### Arch Linux
-```bash
-sudo pacman -S --needed pkgconf fuse3 gtk4 libadwaita libsecret dbus webkitgtk-6.0 perl-image-exiftool
-```
-
-### Fedora (44+)
-```bash
-sudo dnf install -y \
-  pkgconf-pkg-config fuse3-devel gtk4-devel libadwaita-devel \
-  webkitgtk6.0-devel libsecret-devel dbus-devel glib2-devel \
-  perl-Image-ExifTool cargo rust
-```
-
-Runtime extras (pick your desktop):
-```bash
-# GNOME — keyring + tray (AppIndicator)
-sudo dnf install -y gnome-keyring gnome-shell-extension-appindicator xdg-utils
-
-# KDE Plasma — KWallet (Secret Service); tray works via built-in SNI
-sudo dnf install -y kwallet xdg-utils
-```
-
----
-
-## Building from Source
-
-Ensure you have Rust and Cargo installed (minimum supported Rust version is 1.96).
-
-1. Clone the repository and navigate into the project directory:
-   ```bash
-   git clone https://github.com/narrrl/proton-drive-linux.git
-   cd proton-drive-linux
-   ```
-2. Build the workspace in release mode:
-   ```bash
-   cargo build --release --locked
-   ```
-
-The compiled binaries will be available under `target/release/`:
-- `pdfs`: The CLI utility.
-- `pdfs-app`: The GTK4 application.
-- `pdfs-tray`: The tray status notifier.
-- `pdfs-prompt`: The launcher prompt for quick HUD search.
-
----
-
-## Installation & Packages
-
-### 1. Debian / Ubuntu (.deb)
-Install the debian package via `dpkg` or `apt`:
-```bash
-sudo apt install ./proton-drive-linux_*.deb
-```
-
-### 2. Arch Linux
-A local `PKGBUILD` is available under the `packaging/` directory. You can build and install it using:
-```bash
-cd packaging && makepkg -fi
-```
-
-### 3. Fedora (local RPM)
-A local `.spec` is available under `packaging/`. From the repository root:
-```bash
-sudo dnf install -y rpm-build
-rpmbuild -bb packaging/proton-drive-linux.spec \
-  --define "git_dir $PWD" \
-  --define "_rpmdir $PWD/packaging/out" \
-  --define "_builddir $PWD/packaging/build" \
-  --define "_sourcedir $PWD" \
-  --define "_specdir $PWD/packaging" \
-  --define "_srcrpmdir $PWD/packaging/out"
-sudo dnf install packaging/out/x86_64/proton-drive-linux-*.rpm
-```
-
----
-
-## Automated Releases (CI/CD)
-
-This project has a GitHub Actions CI workflow configured under `.github/workflows/release.yml`.
-
-### How it works:
-1. **Triggers**: 
-   - Pushing a git tag matching `v*` (e.g. `git tag v0.1.0 && git push origin v0.1.0`).
-   - Manual runs via the **Actions** tab in GitHub (**workflow_dispatch**).
-2. **Quality Gates and Build Process**:
-   - For tagged builds, verifies that the tag, workspace version, and `packaging/PKGBUILD` version agree.
-   - Runs `cargo fmt`, Clippy with warnings denied, the locked workspace test suite, and the account-free FUSE acceptance contract.
-   - Spawns an Ubuntu runner and installs GTK4, Libadwaita, FUSE3, and Secret Service packages.
-   - Sets up the Rust compiler and caches build targets to speed up runs.
-   - Compiles the workspace members in release mode.
-3. **Artifact Packaging**:
-   - Generates a `.tar.gz` containing the raw binaries (`pdfs`, `pdfs-app`, `pdfs-tray`, `pdfs-prompt`).
-   - Packs them into Debian (`.deb`) and Fedora (`.rpm`) packages.
-   - Includes the systemd user service and tray autostart entry in the Debian package.
-4. **Publishing**:
-   - Creates a GitHub Release matching the pushed tag and uploads the `.deb`, `.rpm`, and `.tar.gz` packages as release assets.
-   - For manual runs, compiles and exposes the packages as workflow run artifacts for testing.
-
----
+Files you open are decrypted into a local cache so programs can read them. What that means for
+the security of the machine is spelled out in the
+[security model](docs/SECURITY_MODEL.md#what-is-kept-on-this-computer), and in full detail in
+the [threat model](docs/ARCHITECTURE.md#10-threat-model-what-this-client-writes-to-disk-in-plaintext).
 
 ## Contributing
 
-Bug reports, packaging fixes, and patches are welcome — see [CONTRIBUTING.md](CONTRIBUTING.md) for
-the workflow and the quality gates CI enforces, and [SECURITY.md](SECURITY.md) for reporting a
-vulnerability privately.
+Bug reports, translations and patches are welcome. See [CONTRIBUTING.md](CONTRIBUTING.md) for the
+workflow and the checks CI runs, and [SECURITY.md](SECURITY.md) to report a vulnerability
+privately.
+
+## Support
+
+Proton Drive Linux is free and built in spare time. If it saves you a few headaches, you can
+[buy me a coffee](https://buymeacoffee.com/narl).
+
+<a href="https://buymeacoffee.com/narl"><img src="https://img.shields.io/badge/Buy%20me%20a%20coffee-FFDD00?logo=buymeacoffee&logoColor=black" alt="Buy me a coffee"></a>
 
 ## License
 
 Released under the [MIT License](LICENSE).
 
+The Places map uses [GeoNames](https://www.geonames.org/) data (CC BY 4.0) and
+[Natural Earth](https://www.naturalearthdata.com/) outlines (public domain). The optional street
+map comes from [OpenFreeMap](https://openfreemap.org/), © [OpenStreetMap](https://www.openstreetmap.org/copyright)
+contributors.
+
 Proton Drive is a service of Proton AG. This project is an independent client and carries no
 affiliation with or endorsement from Proton AG.

SECURITY.md

@@ -4,6 +4,9 @@
 
 Fixes land on the latest release. Please reproduce on the most recent version before reporting.
 
+For what the client protects and what it does not, written for users, see the
+[security model](docs/SECURITY_MODEL.md).
+
 ## Reporting a vulnerability
 
 **Do not open a public issue.** Report privately through GitHub's
@@ -30,5 +33,10 @@ the interesting boundaries are:
 - **Cryptography.** Key derivation, signature handling, and anything that could publish plaintext
   or the wrong key material to Proton.
 
+What the client deliberately keeps in plaintext on disk, and why, is documented in the
+[threat model](docs/ARCHITECTURE.md#10-threat-model-what-this-client-writes-to-disk-in-plaintext).
+Behaviour described there is known and not a vulnerability by itself; a way to make it worse
+than described is.
+
 Out of scope: vulnerabilities in Proton's own service (report those to Proton AG), and issues that
 require an attacker who already has your user account on the machine.

crates/pdfs-cli/Cargo.toml

@@ -23,3 +23,4 @@ tracing-subscriber.workspace = true
 clap.workspace = true
 rpassword.workspace = true
 serde_json.workspace = true
+zeroize.workspace = true

crates/pdfs-cli/src/journal.rs

@@ -0,0 +1,122 @@
+//! Log priorities for journald.
+//!
+//! journald files each line a service writes at the unit's default priority,
+//! info, unless the line opens with a syslog `<N>` prefix. Without one every
+//! WARN and ERROR the daemon logs is stored as info, and the
+//! `journalctl -p warning` that the troubleshooting guide asks for shows
+//! nothing (`docs/BUGS.md` B258).
+
+use std::fmt;
+use std::os::unix::fs::MetadataExt;
+
+use tracing::{Event, Level, Subscriber};
+use tracing_subscriber::fmt::format::Writer;
+use tracing_subscriber::fmt::{FmtContext, FormatEvent, FormatFields};
+use tracing_subscriber::registry::LookupSpan;
+
+/// Whether standard output is the journal stream systemd connected this
+/// process to.
+///
+/// `JOURNAL_STREAM` alone does not say so: a terminal started from a user unit
+/// inherits it, and every `pdfs` run in that terminal would print the
+/// prefixes. systemd sets it to the stream's `device:inode`, so it is compared
+/// against what fd 1 actually is.
+pub(crate) fn stdout_is_journal() -> bool {
+    let Some(stream) = std::env::var_os("JOURNAL_STREAM") else {
+        return false;
+    };
+    let Ok(stdout) = std::fs::metadata("/proc/self/fd/1") else {
+        return false;
+    };
+    stream
+        .to_str()
+        .is_some_and(|value| names_stream(value, stdout.dev(), stdout.ino()))
+}
+
+/// Whether a `JOURNAL_STREAM` value names the file `dev`:`ino`.
+fn names_stream(value: &str, dev: u64, ino: u64) -> bool {
+    value
+        .split_once(':')
+        .is_some_and(|(d, i)| d.parse() == Ok(dev) && i.parse() == Ok(ino))
+}
+
+/// The syslog priority journald should file a `level` line at.
+fn priority(level: &Level) -> u8 {
+    match *level {
+        Level::ERROR => 3,
+        Level::WARN => 4,
+        Level::INFO => 6,
+        Level::DEBUG | Level::TRACE => 7,
+    }
+}
+
+/// An event format that opens each line with its `<N>` priority, then writes
+/// the line as `F` would.
+pub(crate) struct Prioritized<F>(pub(crate) F);
+
+impl<S, N, F> FormatEvent<S, N> for Prioritized<F>
+where
+    S: Subscriber + for<'a> LookupSpan<'a>,
+    N: for<'a> FormatFields<'a> + 'static,
+    F: FormatEvent<S, N>,
+{
+    fn format_event(
+        &self,
+        ctx: &FmtContext<'_, S, N>,
+        mut writer: Writer<'_>,
+        event: &Event<'_>,
+    ) -> fmt::Result {
+        write!(writer, "<{}>", priority(event.metadata().level()))?;
+        self.0.format_event(ctx, writer, event)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::io;
+    use std::sync::{Arc, Mutex};
+
+    #[derive(Clone)]
+    struct Buf(Arc<Mutex<Vec<u8>>>);
+
+    impl io::Write for Buf {
+        fn write(&mut self, bytes: &[u8]) -> io::Result<usize> {
+            self.0.lock().unwrap().extend_from_slice(bytes);
+            Ok(bytes.len())
+        }
+        fn flush(&mut self) -> io::Result<()> {
+            Ok(())
+        }
+    }
+
+    #[test]
+    fn each_line_opens_with_the_priority_of_its_level() {
+        let buf = Buf(Arc::new(Mutex::new(Vec::new())));
+        let writer = buf.clone();
+        let subscriber = tracing_subscriber::fmt()
+            .with_max_level(Level::DEBUG)
+            .with_ansi(false)
+            .with_writer(move || writer.clone())
+            .event_format(Prioritized(tracing_subscriber::fmt::format()))
+            .finish();
+        tracing::subscriber::with_default(subscriber, || {
+            tracing::error!("e");
+            tracing::warn!("w");
+            tracing::info!("i");
+            tracing::debug!("d");
+        });
+        let out = String::from_utf8(buf.0.lock().unwrap().clone()).unwrap();
+        let prefixes: Vec<&str> = out.lines().map(|line| &line[..3]).collect();
+        assert_eq!(prefixes, ["<3>", "<4>", "<6>", "<7>"]);
+    }
+
+    #[test]
+    fn journal_stream_must_name_the_same_file() {
+        assert!(names_stream("8:12345", 8, 12345));
+        assert!(!names_stream("8:12345", 8, 54321));
+        assert!(!names_stream("9:12345", 8, 12345));
+        assert!(!names_stream("8", 8, 12345));
+        assert!(!names_stream("", 8, 12345));
+    }
+}

crates/pdfs-core/Cargo.toml

@@ -28,3 +28,6 @@ dbus-secret-service-keyring-store.workspace = true
 parking_lot.workspace = true
 async-trait.workspace = true
 libc.workspace = true
+zeroize.workspace = true
+base64.workspace = true
+exif = { version = "0.6", package = "kamadak-exif" }

crates/pdfs-core/data/countries.tsv

@@ -0,0 +1,250 @@
+AD	Andorra
+AE	United Arab Emirates
+AF	Afghanistan
+AG	Antigua and Barbuda
+AI	Anguilla
+AL	Albania
+AM	Armenia
+AO	Angola
+AQ	Antarctica
+AR	Argentina
+AS	American Samoa
+AT	Austria
+AU	Australia
+AW	Aruba
+AX	Åland Islands
+AZ	Azerbaijan
+BA	Bosnia and Herzegovina
+BB	Barbados
+BD	Bangladesh
+BE	Belgium
+BF	Burkina Faso
+BG	Bulgaria
+BH	Bahrain
+BI	Burundi
+BJ	Benin
+BL	Saint Barthélemy
+BM	Bermuda
+BN	Brunei Darussalam
+BO	Bolivia, Plurinational State of
+BQ	Bonaire, Sint Eustatius and Saba
+BR	Brazil
+BS	Bahamas
+BT	Bhutan
+BV	Bouvet Island
+BW	Botswana
+BY	Belarus
+BZ	Belize
+CA	Canada
+CC	Cocos (Keeling) Islands
+CD	Congo, The Democratic Republic of the
+CF	Central African Republic
+CG	Congo
+CH	Switzerland
+CI	Côte d'Ivoire
+CK	Cook Islands
+CL	Chile
+CM	Cameroon
+CN	China
+CO	Colombia
+CR	Costa Rica
+CU	Cuba
+CV	Cabo Verde
+CW	Curaçao
+CX	Christmas Island
+CY	Cyprus
+CZ	Czechia
+DE	Germany
+DJ	Djibouti
+DK	Denmark
+DM	Dominica
+DO	Dominican Republic
+DZ	Algeria
+EC	Ecuador
+EE	Estonia
+EG	Egypt
+EH	Western Sahara
+ER	Eritrea
+ES	Spain
+ET	Ethiopia
+FI	Finland
+FJ	Fiji
+FK	Falkland Islands (Malvinas)
+FM	Micronesia, Federated States of
+FO	Faroe Islands
+FR	France
+GA	Gabon
+GB	United Kingdom
+GD	Grenada
+GE	Georgia
+GF	French Guiana
+GG	Guernsey
+GH	Ghana
+GI	Gibraltar
+GL	Greenland
+GM	Gambia
+GN	Guinea
+GP	Guadeloupe
+GQ	Equatorial Guinea
+GR	Greece
+GS	South Georgia and the South Sandwich Islands
+GT	Guatemala
+GU	Guam
+GW	Guinea-Bissau
+GY	Guyana
+HK	Hong Kong
+HM	Heard Island and McDonald Islands
+HN	Honduras
+HR	Croatia
+HT	Haiti
+HU	Hungary
+ID	Indonesia
+IE	Ireland
+IL	Israel
+IM	Isle of Man
+IN	India
+IO	British Indian Ocean Territory
+IQ	Iraq
+IR	Iran, Islamic Republic of
+IS	Iceland
+IT	Italy
+JE	Jersey
+JM	Jamaica
+JO	Jordan
+JP	Japan
+KE	Kenya
+KG	Kyrgyzstan
+KH	Cambodia
+KI	Kiribati
+KM	Comoros
+KN	Saint Kitts and Nevis
+KP	Korea, Democratic People's Republic of
+KR	Korea, Republic of
+KW	Kuwait
+KY	Cayman Islands
+KZ	Kazakhstan
+LA	Lao People's Democratic Republic
+LB	Lebanon
+LC	Saint Lucia
+LI	Liechtenstein
+LK	Sri Lanka
+LR	Liberia
+LS	Lesotho
+LT	Lithuania
+LU	Luxembourg
+LV	Latvia
+LY	Libya
+MA	Morocco
+MC	Monaco
+MD	Moldova, Republic of
+ME	Montenegro
+MF	Saint Martin (French part)
+MG	Madagascar
+MH	Marshall Islands
+MK	North Macedonia
+ML	Mali
+MM	Myanmar
+MN	Mongolia
+MO	Macao
+MP	Northern Mariana Islands
+MQ	Martinique
+MR	Mauritania
+MS	Montserrat
+MT	Malta
+MU	Mauritius
+MV	Maldives
+MW	Malawi
+MX	Mexico
+MY	Malaysia
+MZ	Mozambique
+NA	Namibia
+NC	New Caledonia
+NE	Niger
+NF	Norfolk Island
+NG	Nigeria
+NI	Nicaragua
+NL	Netherlands
+NO	Norway
+NP	Nepal
+NR	Nauru
+NU	Niue
+NZ	New Zealand
+OM	Oman
+PA	Panama
+PE	Peru
+PF	French Polynesia
+PG	Papua New Guinea
+PH	Philippines
+PK	Pakistan
+PL	Poland
+PM	Saint Pierre and Miquelon
+PN	Pitcairn
+PR	Puerto Rico
+PS	Palestine, State of
+PT	Portugal
+PW	Palau
+PY	Paraguay
+QA	Qatar
+RE	Réunion
+RO	Romania
+RS	Serbia
+RU	Russian Federation
+RW	Rwanda
+SA	Saudi Arabia
+SB	Solomon Islands
+SC	Seychelles
+SD	Sudan
+SE	Sweden
+SG	Singapore
+SH	Saint Helena, Ascension and Tristan da Cunha
+SI	Slovenia
+SJ	Svalbard and Jan Mayen
+SK	Slovakia
+SL	Sierra Leone
+SM	San Marino
+SN	Senegal
+SO	Somalia
+SR	Suriname
+SS	South Sudan
+ST	Sao Tome and Principe
+SV	El Salvador
+SX	Sint Maarten (Dutch part)
+SY	Syrian Arab Republic
+SZ	Eswatini
+TC	Turks and Caicos Islands
+TD	Chad
+TF	French Southern Territories
+TG	Togo
+TH	Thailand
+TJ	Tajikistan
+TK	Tokelau
+TL	Timor-Leste
+TM	Turkmenistan
+TN	Tunisia
+TO	Tonga
+TR	Türkiye
+TT	Trinidad and Tobago
+TV	Tuvalu
+TW	Taiwan, Province of China
+TZ	Tanzania, United Republic of
+UA	Ukraine
+UG	Uganda
+UM	United States Minor Outlying Islands
+US	United States
+UY	Uruguay
+UZ	Uzbekistan
+VA	Holy See (Vatican City State)
+VC	Saint Vincent and the Grenadines
+VE	Venezuela, Bolivarian Republic of
+VG	Virgin Islands, British
+VI	Virgin Islands, U.S.
+VN	Viet Nam
+VU	Vanuatu
+WF	Wallis and Futuna
+WS	Samoa
+XK	Kosovo
+YE	Yemen
+YT	Mayotte
+ZA	South Africa
+ZM	Zambia
+ZW	Zimbabwe

crates/pdfs-core/src/cache.rs

@@ -91,7 +91,7 @@ const TOUCH_FLUSH_AT: usize = 256;
 ///
 /// Public because it is the *contract* a block has to satisfy, not a detail of
 /// this cache: the network read path validates against it too, so a block that
-/// comes back short is refused before it can truncate a file (bugs.md B84).
+/// comes back short is refused before it can truncate a file (`docs/BUGS.md` B84).
 pub fn block_len(size: u64, idx: u64) -> u64 {
     let start = idx.saturating_mul(BLOCK_SIZE);
     size.saturating_sub(start).min(BLOCK_SIZE)
@@ -298,6 +298,10 @@ pub struct Pin {
     /// which is why it is defaulted.
     #[serde(default)]
     pub is_dir: Option<bool>,
+    /// A pinned file whose content is on disk right now. The daemon fills it
+    /// in when it answers `ListPins`; folders and older daemons say `false`.
+    #[serde(default)]
+    pub cached: bool,
 }
 
 /// The legacy JSON pin registry, kept only to import a pre-P5 `pins.json` into
@@ -307,8 +311,7 @@ struct PinFile {
     pins: BTreeMap<String, Pin>,
 }
 
-/// What a staged file actually contains, written beside it as `<name>.json`
-/// (offline.md Phase 2).
+/// What a staged file actually contains, written beside it as `<name>.json`.
 ///
 /// A staged file is **not** necessarily valid whole-file content, and that is
 /// the whole reason this exists. A partial overwrite commits by filling the
@@ -361,6 +364,14 @@ pub struct StagedWrite {
     pub based_on: Option<Baseline>,
 }
 
+impl StagedWrite {
+    /// Whether `authored` ranges cover a file of `len` bytes, so the staged
+    /// bytes are its complete new content. An empty file has nothing to cover.
+    pub fn covers(authored: &[(u64, u64)], len: u64) -> bool {
+        len == 0 || authored == [(0, len)]
+    }
+}
+
 /// The identity of a remote revision, as far as we can observe it.
 ///
 /// `revision_id` is the authoritative identity when present: the server assigns a
@@ -427,9 +438,9 @@ pub struct ContentCache {
     /// Subdirectory for write-handle scratch files (disk-backed write buffers).
     /// Emptied on open so a crashed run leaves no orphans.
     scratch_dir: PathBuf,
-    /// Subdirectory holding the bytes of writes that have not been uploaded yet
-    /// (offline.md Phase 2/3) — every released write passes through here, since
-    /// the upload is a queued op performed later. Unlike `scratch_dir` this is
+    /// Subdirectory holding the bytes of writes that have not been uploaded yet.
+    /// Every released write passes through here, since the upload is a queued
+    /// op performed later. Unlike `scratch_dir` this is
     /// **never** emptied on open: these are the only copy of content the user
     /// authored, and the whole point is that they outlive the daemon.
     staging_dir: PathBuf,
@@ -464,7 +475,7 @@ pub struct ContentCache {
     thumb_bytes: AtomicU64,
     /// Unified metadata DB. Its `cache_entries` table is the LRU index: every
     /// store/read/evict updates it, and the budget enforcers query it instead of
-    /// scanning the cache directories (plan.md P4).
+    /// scanning the cache directories.
     db: Arc<Db>,
     /// LRU touches not yet written to `cache_entries`, keyed by cache key.
     ///
@@ -726,6 +737,60 @@ impl ContentCache {
         Some(buf)
     }
 
+    /// Write the whole content of `uid` at `(mtime, size)` to `dst` from the
+    /// cache alone, and report whether that worked.
+    ///
+    /// A whole-file blob is cloned. Otherwise every block of the revision has to
+    /// be cached, under its recorded geometry or, failing that, the uniform one
+    /// the read path would use. Anything short of the full content leaves
+    /// nothing at `dst`, so a caller falls back to a download with a clean slate.
+    ///
+    /// This is what lets a mirror folder take a file an on-demand mount has
+    /// already read, after a move between them, without fetching it again. The
+    /// trust is the read path's: the same `(mtime, size)` key, and blocks that
+    /// were validated when they were fetched.
+    pub fn copy_cached_to(&self, uid: &NodeUid, mtime: i64, size: u64, dst: &Path) -> bool {
+        if size == 0 {
+            return false;
+        }
+        if let Some(blob) = self.valid_blob(uid, mtime, size) {
+            if Self::clone_or_copy(&blob, dst).is_ok_and(|len| len == size) {
+                self.touch(Self::key(uid));
+                return true;
+            }
+            let _ = std::fs::remove_file(dst);
+            return false;
+        }
+        let geometry = self
+            .block_geometry(uid, mtime, size)
+            .unwrap_or_else(|| BlockGeometry::uniform(size));
+        if geometry.size() != size {
+            return false;
+        }
+        let spans = geometry.spans(0, size);
+        // Checked up front so a file with one block missing costs stats, not a
+        // copy of every block before it.
+        if !spans
+            .iter()
+            .all(|&span| self.has_block(uid, mtime, size, span))
+        {
+            return false;
+        }
+        let written = (|| -> Option<()> {
+            use std::io::Write;
+            let mut out = std::fs::File::create(dst).ok()?;
+            for span in spans {
+                out.write_all(&self.cached_block(uid, mtime, size, span)?)
+                    .ok()?;
+            }
+            (out.metadata().ok()?.len() == size).then_some(())
+        })();
+        if written.is_none() {
+            let _ = std::fs::remove_file(dst);
+        }
+        written.is_some()
+    }
+
     /// Store `bytes` as the cached content for `uid`, tagged with `(mtime,
     /// size)`. The blob is written to a temp file then renamed so a concurrent
     /// reader never sees a partial blob; the meta tag is written last so a
@@ -779,6 +844,9 @@ impl ContentCache {
         }
         let bytes = std::fs::metadata(&tmp)?.len();
         std::fs::rename(&tmp, &blob)?;
+        // Linked from the blob already cached, the temp name was a second name
+        // of it, and the rename did nothing (`docs/BUGS.md` B187).
+        let _ = std::fs::remove_file(&tmp);
         let meta = serde_json::to_vec(&Meta { mtime, size })?;
         std::fs::write(self.meta_path(uid), meta)?;
         self.db
@@ -920,7 +988,7 @@ impl ContentCache {
     /// a miss ([`cached_block`](Self::cached_block) checks), so a bad block
     /// cannot escape this cache even though this does not check. Validating a
     /// block is the *reader's* job, at the point it is fetched — see
-    /// `Core::read_block` (bugs.md B84).
+    /// `Core::read_block` (`docs/BUGS.md` B84).
     pub fn store_block(
         &self,
         uid: &NodeUid,
@@ -1422,9 +1490,8 @@ impl ContentCache {
     /// This is what makes a write survive its upload: the caller is releasing a
     /// write handle and would otherwise delete the file, so until the bytes are
     /// on the remote, staging holds the only copy. Every dirty handle goes
-    /// through here — the upload is a queued op performed later (offline.md
-    /// Phase 3), and a staged file is also what a human can recover from if the
-    /// queue never drains.
+    /// through here — the upload is a queued op performed later, and a staged
+    /// file is also what a human can recover from if the queue never drains.
     ///
     /// Falls back to a copy when the rename crosses a filesystem boundary; a
     /// failure here means we could not save the bytes at all, so it is reported
@@ -1854,6 +1921,7 @@ impl ContentCache {
                 path: row.path,
                 recursive: row.recursive,
                 is_dir: row.is_dir,
+                cached: false,
             })
             .collect()
     }
@@ -1997,6 +2065,17 @@ mod tests {
         NodeUid::new(VolumeId::from("vol"), LinkId::from(link))
     }
 
+    #[test]
+    fn an_empty_write_covers_the_file_and_a_gap_does_not() {
+        assert!(
+            StagedWrite::covers(&[], 0),
+            "an empty file has nothing to cover"
+        );
+        assert!(StagedWrite::covers(&[(0, 7)], 7));
+        assert!(!StagedWrite::covers(&[(0, 3), (4, 7)], 7));
+        assert!(!StagedWrite::covers(&[], 7));
+    }
+
     /// A unique temp directory removed on drop; avoids a dev-dependency.
     struct TempDir(PathBuf);
     impl TempDir {
@@ -2119,6 +2198,24 @@ mod tests {
         assert_eq!(c.read_range(&u, 200, 5, 0, 5).unwrap(), b"newer");
     }
 
+    /// A landed create and the revision after it can adopt the same staged
+    /// blob. The link to the temp name is then a second name of the cached
+    /// blob, and a rename between two names of one file does nothing: the
+    /// temp name stayed until the next start swept it (`docs/BUGS.md` B187).
+    #[test]
+    fn store_file_twice_from_the_same_blob_leaves_no_temp_file() {
+        let (c, d) = cache();
+        let u = uid("a");
+        let src = d.path().join("staged-blob");
+        std::fs::write(&src, b"staged").unwrap();
+
+        c.store_file(&u, 100, 6, &src).unwrap();
+        c.store_file(&u, 100, 6, &src).unwrap();
+
+        assert!(!c.blob_path(&u).with_extension("tmp").exists());
+        assert_eq!(c.read_range(&u, 100, 6, 0, 6).unwrap(), b"staged");
+    }
+
     #[test]
     fn reconcile_rebuilds_index_from_disk() {
         let dir = TempDir::new();
@@ -2212,42 +2309,26 @@ mod tests {
     /// call also needs.
     #[test]
     fn budget_check_is_free_when_under_budget() {
-        use std::time::Instant;
-
         // Cap far above what we store, so every store stays under budget.
         let (c, _d) = cache_capped(1 << 30);
         let payload = vec![0u8; 4096];
-
-        // Cost of a store against a near-empty index...
-        let t = Instant::now();
-        for i in 0..500u64 {
+        for i in 0..8u64 {
             c.store_block(&uid("a"), 1, 1 << 30, blk(1 << 30, i), &payload)
                 .unwrap();
         }
-        let empty = t.elapsed() / 500;
 
-        // ...and against one with a few thousand entries in it.
-        for i in 500..2500u64 {
-            c.store_block(&uid("a"), 1, 1 << 30, blk(1 << 30, i), &payload)
-                .unwrap();
-        }
-        let t = Instant::now();
-        for i in 2500..3000u64 {
-            c.store_block(&uid("a"), 1, 1 << 30, blk(1 << 30, i), &payload)
-                .unwrap();
-        }
-        let full = t.elapsed() / 500;
+        // Plant an index row far over the cap behind the running total's back,
+        // and older than every real block. Any budget check that consults the
+        // index would see the pool over budget, evict, and reseed the total.
+        c.db.cache_touch("planted", KIND_BLOCK, 1 << 40, 0).unwrap();
 
-        println!("B4: store_block under budget — {empty:?} at ~0 entries, {full:?} at 2500");
-        // A ratio, not a wall-clock bound: the claim is that the under-budget
-        // path does not grow with the index, and only a ratio tests that. An
-        // absolute threshold measures the machine's load instead, and this test
-        // runs alongside the rest of the suite.
-        assert!(
-            full < empty * 3,
-            "under-budget store should not scale with the index: \
-             {empty:?} at ~0 entries vs {full:?} at 2500"
-        );
+        c.store_block(&uid("a"), 1, 1 << 30, blk(1 << 30, 8), &payload)
+            .unwrap();
+
+        // Not a timing ratio: that measured the CI runner's load and flaked.
+        assert_eq!(c.block_bytes.load(Ordering::Relaxed), 9 * 4096);
+        let entries = c.db.cache_entries_by_kind(KIND_BLOCK).unwrap();
+        assert_eq!(entries.len(), 10, "under-budget store evicted: {entries:?}");
     }
 
     /// The running totals are a fast path, not a second source of truth: after
@@ -2556,6 +2637,46 @@ mod tests {
         assert!(c.cached_block(&u, 100, 5000, blk(5000, 0)).is_none());
     }
 
+    #[test]
+    fn a_cached_blob_is_copied_out_whole() {
+        let (c, d) = cache();
+        let u = uid("blob");
+        c.store(&u, 7, 5, b"hello").unwrap();
+        let dst = d.path().join("out");
+        assert!(c.copy_cached_to(&u, 7, 5, &dst));
+        assert_eq!(std::fs::read(&dst).unwrap(), b"hello");
+        // Another revision is not this one.
+        assert!(!c.copy_cached_to(&u, 8, 5, &d.path().join("stale")));
+        assert!(!d.path().join("stale").exists());
+    }
+
+    #[test]
+    fn cached_blocks_are_copied_out_in_order() {
+        let (c, d) = cache();
+        let u = uid("blocks");
+        let size = BLOCK_SIZE + 3;
+        let first = vec![1_u8; BLOCK_SIZE as usize];
+        c.store_block(&u, 7, size, blk(size, 0), &first).unwrap();
+        c.store_block(&u, 7, size, blk(size, 1), b"end").unwrap();
+        let dst = d.path().join("out");
+        assert!(c.copy_cached_to(&u, 7, size, &dst));
+        let bytes = std::fs::read(&dst).unwrap();
+        assert_eq!(bytes.len() as u64, size);
+        assert_eq!(&bytes[..BLOCK_SIZE as usize], first.as_slice());
+        assert_eq!(&bytes[BLOCK_SIZE as usize..], b"end");
+    }
+
+    #[test]
+    fn a_file_with_a_block_missing_is_not_copied_out() {
+        let (c, d) = cache();
+        let u = uid("partial");
+        let size = BLOCK_SIZE + 3;
+        c.store_block(&u, 7, size, blk(size, 1), b"end").unwrap();
+        let dst = d.path().join("out");
+        assert!(!c.copy_cached_to(&u, 7, size, &dst));
+        assert!(!dst.exists(), "a partial copy must not be left behind");
+    }
+
     /// The [`BlockSpan`] the uniform geometry gives block `idx` of a file of
     /// `size` bytes — the shape these tests were written against, and the one
     /// this client's own uploads still have.
@@ -2568,7 +2689,7 @@ mod tests {
     }
 
     /// The geometry the whole read path validates against, here and over the
-    /// network (bugs.md B84): full blocks, then whatever is left.
+    /// network (`docs/BUGS.md` B84): full blocks, then whatever is left.
     #[test]
     fn block_len_is_full_blocks_then_the_remainder() {
         let size = BLOCK_SIZE * 2 + 17;
@@ -2648,7 +2769,7 @@ mod tests {
     /// A cached block is addressed by the range it covers, not by its index
     /// alone. Re-planning a file onto its real geometry must therefore miss the
     /// blocks the uniform fallback cached, rather than serve their bytes at the
-    /// wrong offset (bugs.md B85).
+    /// wrong offset (`docs/BUGS.md` B85).
     #[test]
     fn a_cached_block_is_not_served_to_a_different_geometry() {
         let (c, _d) = cache();
@@ -2715,7 +2836,7 @@ mod tests {
 
     /// A block of the wrong length cannot escape the cache even though the
     /// store does not check: the read validates it and reports a miss, so the
-    /// next read refetches rather than serving a hole (bugs.md B84).
+    /// next read refetches rather than serving a hole (`docs/BUGS.md` B84).
     #[test]
     fn a_block_of_the_wrong_length_reads_as_a_miss() {
         let (c, _d) = cache();
@@ -2780,7 +2901,7 @@ mod tests {
         );
     }
 
-    /// A failed upload must never cost the user their bytes (offline.md Phase 2):
+    /// A failed upload must never cost the user their bytes:
     /// the scratch file moves to staging intact, and staging survives a reopen
     /// the way scratch deliberately does not.
     #[test]

crates/pdfs-core/src/config.rs

@@ -3,7 +3,7 @@
 use std::path::{Path, PathBuf};
 
 use directories::ProjectDirs;
-use serde::{Deserialize, Serialize};
+use serde::{Deserialize, Deserializer, Serialize};
 
 use crate::error::{Error, Result};
 use crate::menu::PromptConfig;
@@ -12,8 +12,19 @@ use crate::syncignore::DEFAULT_IGNORE_PATTERNS;
 
 /// Proton requires `external-drive-{name}@{semver}-{channel}` (channel ∈
 /// stable/beta/alpha); a malformed value trips the 422 anti-abuse path.
-pub const APP_VERSION: &str = "external-drive-linux@0.1.0-alpha";
-pub const USER_AGENT: &str = "proton-drive-linux/0.1.0";
+/// Follows the release so Proton can tell releases apart and retire old ones
+/// without retiring every one.
+pub const APP_VERSION: &str = concat!(
+    "external-drive-linux@",
+    env!("CARGO_PKG_VERSION"),
+    "-stable"
+);
+pub const USER_AGENT: &str = concat!("proton-drive-linux/", env!("CARGO_PKG_VERSION"));
+
+/// The fixed identity every release up to 3.1.0 wrote into `config.json`. Read
+/// back as "unset" so those configs follow the release instead of pinning it.
+const LEGACY_APP_VERSION: &str = "external-drive-linux@0.1.0-alpha";
+const LEGACY_USER_AGENT: &str = "proton-drive-linux/0.1.0";
 
 /// Keyring service name; one entry per credential kind keyed by username.
 pub const KEYRING_SERVICE: &str = "proton-drive-linux";
@@ -76,15 +87,38 @@ fn resolve_sweep_mode(env: Option<&str>, stored: Option<SweepMode>) -> SweepMode
 }
 
 /// Configuration structure allowing the user to customize client identification headers.
-#[derive(Serialize, Deserialize, Debug, Clone)]
+#[derive(Serialize, Deserialize, Debug, Clone, Default)]
 pub struct AppConfig {
-    pub app_version: String,
-    pub user_agent: String,
+    /// Override for [`APP_VERSION`]. `None` means "use the release's own";
+    /// the old fixed default reads as `None`, so it leaves the file on the
+    /// next save.
+    #[serde(
+        default,
+        deserialize_with = "identity_override",
+        skip_serializing_if = "Option::is_none"
+    )]
+    pub app_version: Option<String>,
+    /// Override for [`USER_AGENT`], handled like `app_version`.
+    #[serde(
+        default,
+        deserialize_with = "identity_override",
+        skip_serializing_if = "Option::is_none"
+    )]
+    pub user_agent: Option<String>,
     /// Soft cap on the on-disk content cache, in bytes (`0` = unlimited).
     /// `None` means "use [`DEFAULT_CACHE_BUDGET_BYTES`]"; the Settings page
     /// writes an explicit value here. Defaulted for configs predating the field.
     #[serde(default)]
     pub cache_budget: Option<u64>,
+    /// Upload cap in bytes per second shared by every upload. `None` or `0`
+    /// means no cap. Defaulted for configs predating the field.
+    #[serde(default)]
+    pub upload_limit: Option<u64>,
+    /// Download cap in bytes per second shared by every download, FUSE reads
+    /// included. `None` or `0` means no cap. Defaulted for configs predating
+    /// the field.
+    #[serde(default)]
+    pub download_limit: Option<u64>,
     /// Mountpoint the daemon mounts at. `None` means
     /// [`AppDirs::default_mountpoint`]; the Settings page writes an explicit
     /// path here. Defaulted for configs predating the field.
@@ -119,25 +153,176 @@ pub struct AppConfig {
     /// field.
     #[serde(default)]
     pub prompt: Option<PromptConfig>,
+    /// Paint the desktop app in Proton's colours instead of the system theme
+    /// and accent colour. `None` means "follow the system". Defaulted for
+    /// configs predating the field; also read from `proton_accent`, its name
+    /// while it only changed the accent.
+    #[serde(default, alias = "proton_accent")]
+    pub proton_theme: Option<bool>,
+    /// How the desktop app's My Files page lays out and orders a folder.
+    /// Defaulted for configs predating the field.
+    #[serde(default)]
+    pub files_view: FilesView,
+    /// The desktop app's main window size, remembered between runs. Defaulted
+    /// for configs predating the field.
+    #[serde(default)]
+    pub window: WindowState,
+    /// The user hid the tray icon from its own menu. The tray exits at start
+    /// while this is set, and the app does not spawn it; Preferences turns it
+    /// back on. Defaulted for configs predating the field.
+    #[serde(default)]
+    pub tray_hidden: bool,
+    /// Language for the desktop app, tray and search prompt, as a gettext code
+    /// such as `de` or `pt_BR`. `None` means "follow the system locale".
+    /// Defaulted for configs predating the field.
+    #[serde(default)]
+    pub language: Option<String>,
+    /// Draw the Places map from OpenFreeMap's street map instead of the
+    /// built-in outline. Off by default: the tiles come from a third party,
+    /// which then learns which parts of the world are looked at. Defaulted for
+    /// configs predating the field.
+    #[serde(default)]
+    pub online_map: bool,
+    /// Record a new folder, a new file, a rename or a delete through the mount
+    /// locally and send it from the queue, instead of waiting for Drive in the
+    /// syscall. `None` means on. `false` brings back the old path for one
+    /// release (docs/ROADMAP.md). Defaulted for configs predating
+    /// the field.
+    #[serde(default)]
+    pub local_first: Option<bool>,
+    /// Ask GitHub once a day whether a newer release is out, and say so in the
+    /// desktop app. Off by default: GitHub then sees the address and version
+    /// asking. Never used where a package manager installs updates (see
+    /// [`crate::update::updates_managed`]). Defaulted for configs predating the
+    /// field.
+    #[serde(default)]
+    pub update_check: bool,
+    /// When the update check last asked GitHub, in Unix seconds, so a restart
+    /// does not ask again within [`crate::update::CHECK_INTERVAL`]. Defaulted
+    /// for configs predating the field.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub update_last_checked: Option<i64>,
+    /// The latest release that check found, such as `3.6.0`, so the app can
+    /// say so again after a restart without asking. Older than or equal to the
+    /// running version once the update is installed. Defaulted for configs
+    /// predating the field.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub update_latest: Option<String>,
 }
 
-impl Default for AppConfig {
+/// The My Files page's view choices, remembered between runs.
+#[derive(Serialize, Deserialize, Debug, Clone, Copy, PartialEq, Eq)]
+#[serde(default)]
+pub struct FilesView {
+    /// Columns instead of the icon grid.
+    pub list: bool,
+    pub sort: FileSort,
+    pub descending: bool,
+    /// Folders ahead of files, whatever the sort.
+    pub folders_first: bool,
+}
+
+impl Default for FilesView {
     fn default() -> Self {
         Self {
-            app_version: APP_VERSION.to_string(),
-            user_agent: USER_AGENT.to_string(),
-            cache_budget: None,
-            mountpoint: None,
-            ignore_patterns: None,
-            device_uid: None,
-            conflict_sweep: None,
-            open_with: None,
-            prompt: None,
+            list: false,
+            sort: FileSort::Name,
+            descending: false,
+            folders_first: true,
         }
     }
 }
 
+/// The desktop app's main window, as it was when it last closed.
+#[derive(Serialize, Deserialize, Debug, Clone, Copy, PartialEq, Eq)]
+#[serde(default)]
+pub struct WindowState {
+    /// The unmaximised size, in logical pixels.
+    pub width: i32,
+    pub height: i32,
+    pub maximized: bool,
+}
+
+impl WindowState {
+    /// Below this the saved size is taken for damage and the default is used.
+    pub const MIN_WIDTH: i32 = 480;
+    pub const MIN_HEIGHT: i32 = 400;
+
+    /// This state, with a size too small to use replaced by the default.
+    pub fn usable(self) -> Self {
+        if self.width < Self::MIN_WIDTH || self.height < Self::MIN_HEIGHT {
+            Self {
+                maximized: self.maximized,
+                ..Self::default()
+            }
+        } else {
+            self
+        }
+    }
+}
+
+impl Default for WindowState {
+    fn default() -> Self {
+        Self {
+            width: 980,
+            height: 680,
+            maximized: false,
+        }
+    }
+}
+
+/// The key a My Files listing is ordered by.
+#[derive(Serialize, Deserialize, Debug, Clone, Copy, PartialEq, Eq, Default)]
+#[serde(rename_all = "lowercase")]
+pub enum FileSort {
+    #[default]
+    Name,
+    Size,
+    Modified,
+}
+
+impl FileSort {
+    /// The name used in the desktop app's actions and the config file.
+    pub fn as_str(self) -> &'static str {
+        match self {
+            FileSort::Name => "name",
+            FileSort::Size => "size",
+            FileSort::Modified => "modified",
+        }
+    }
+
+    pub fn parse(value: &str) -> Option<Self> {
+        match value {
+            "name" => Some(FileSort::Name),
+            "size" => Some(FileSort::Size),
+            "modified" => Some(FileSort::Modified),
+            _ => None,
+        }
+    }
+}
+
+/// Reads an `app_version` or `user_agent` override, dropping the fixed
+/// identity older releases stored as the default.
+fn identity_override<'de, D: Deserializer<'de>>(
+    deserializer: D,
+) -> std::result::Result<Option<String>, D::Error> {
+    let value = Option::<String>::deserialize(deserializer)?;
+    Ok(value.filter(|v| v != LEGACY_APP_VERSION && v != LEGACY_USER_AGENT))
+}
+
 impl AppConfig {
+    /// The `x-pm-appversion` sent to Proton: the user's override, or
+    /// [`APP_VERSION`].
+    pub fn resolved_app_version(&self) -> &str {
+        self.app_version.as_deref().unwrap_or(APP_VERSION)
+    }
+
+    /// The `User-Agent` sent to Proton: the user's override, or
+    /// [`USER_AGENT`].
+    pub fn resolved_user_agent(&self) -> &str {
+        self.user_agent.as_deref().unwrap_or(USER_AGENT)
+    }
+
     /// The effective cache budget in bytes: the user's explicit choice, or
     /// [`DEFAULT_CACHE_BUDGET_BYTES`] when unset.
     pub fn resolved_cache_budget(&self) -> u64 {
@@ -185,6 +370,22 @@ impl AppConfig {
     }
 }
 
+/// Fail when the config file at `path` exists but does not parse, so that
+/// [`AppDirs::save_config`] never writes over settings it could not read.
+fn refuse_to_replace_unreadable(path: &Path) -> Result<()> {
+    match std::fs::read_to_string(path) {
+        Ok(content) => match serde_json::from_str::<AppConfig>(&content) {
+            Ok(_) => Ok(()),
+            Err(e) => Err(Error::Other(format!(
+                "{} cannot be read ({e}); fix it or move it aside before changing settings",
+                path.display()
+            ))),
+        },
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(()),
+        Err(e) => Err(e.into()),
+    }
+}
+
 /// Resolved XDG locations for state, cache, and the default mountpoint.
 pub struct AppDirs {
     dirs: ProjectDirs,
@@ -216,6 +417,7 @@ impl AppDirs {
                     // Preserve the original bytes for diagnosis/recovery.  In
                     // particular, never replace a truncated config with defaults
                     // and silently lose mount, device, or ignore settings.
+                    // `save_config` refuses to write over it for the same reason.
                     tracing::error!(path = %path.display(), error = %e, "config is unreadable; using in-memory defaults without overwriting it");
                     return AppConfig::default();
                 }
@@ -223,9 +425,8 @@ impl AppDirs {
         }
 
         let default_config = AppConfig::default();
-        if let Ok(content) = serde_json::to_string_pretty(&default_config) {
-            let _ = std::fs::create_dir_all(self.dirs.config_dir());
-            let _ = std::fs::write(&path, content);
+        if let Err(e) = self.save_config(&default_config) {
+            tracing::warn!(path = %path.display(), error = %e, "could not write the default config");
         }
         default_config
     }
@@ -233,8 +434,15 @@ impl AppDirs {
     /// Persist `config` to [`config_path`](Self::config_path), creating the
     /// config dir if missing. The Settings page calls this after editing the
     /// cache budget or mountpoint; the daemon re-reads the file on its next mount.
+    ///
+    /// Refuses while the file on disk does not parse. Every caller loads,
+    /// changes one field and saves, and [`load_config`](Self::load_config)
+    /// gives defaults for a file it cannot read, so the first change after
+    /// the file was damaged wrote defaults over the mountpoint, the device and
+    /// the ignore rules the user had (`docs/BUGS.md` B223).
     pub fn save_config(&self, config: &AppConfig) -> Result<()> {
         let path = self.config_path();
+        refuse_to_replace_unreadable(&path)?;
         let parent = path
             .parent()
             .ok_or_else(|| Error::Other("configuration path has no parent".into()))?;
@@ -258,6 +466,46 @@ impl AppDirs {
         Ok(())
     }
 
+    /// This machine's id (e.g. ~/.config/proton-drive-linux/machine-id). See
+    /// [`machine_id`](Self::machine_id).
+    pub fn machine_id_path(&self) -> PathBuf {
+        self.dirs.config_dir().join("machine-id")
+    }
+
+    /// A random id for this machine, made on first use and kept in
+    /// [`machine_id_path`](Self::machine_id_path).
+    ///
+    /// It names this machine's profile in a device it may share with other
+    /// computers. Random rather than `/etc/machine-id`, so nothing that
+    /// identifies the host goes to Drive. In a file of its own rather than
+    /// `config.json`, because the app saves the config from a copy it loaded
+    /// earlier, and that would drop an id the daemon had just added.
+    pub fn machine_id(&self) -> Result<String> {
+        let path = self.machine_id_path();
+        match std::fs::read_to_string(&path) {
+            Ok(id) if is_machine_id(id.trim()) => return Ok(id.trim().to_string()),
+            Ok(_) => {
+                tracing::warn!(path = %path.display(), "machine id is unusable; making a new one");
+            }
+            Err(e) if e.kind() == std::io::ErrorKind::NotFound => {}
+            Err(e) => return Err(e.into()),
+        }
+        let mut bytes = [0u8; 16];
+        use std::io::Read as _;
+        std::fs::File::open("/dev/urandom")?.read_exact(&mut bytes)?;
+        let id: String = bytes.iter().map(|b| format!("{b:02x}")).collect();
+        let parent = path
+            .parent()
+            .ok_or_else(|| Error::Other("machine id path has no parent".into()))?;
+        std::fs::create_dir_all(parent)?;
+        let tmp = path.with_extension(format!("tmp-{}", std::process::id()));
+        std::fs::write(&tmp, format!("{id}\n"))?;
+        std::fs::File::open(&tmp)?.sync_all()?;
+        std::fs::rename(&tmp, &path)?;
+        std::fs::File::open(parent)?.sync_all()?;
+        Ok(id)
+    }
+
     /// Persistent state: inode map / cache index DB lives here.
     pub fn state_dir(&self) -> PathBuf {
         // `state_dir` is Linux-only in `directories`; fall back to data dir.
@@ -329,7 +577,7 @@ impl AppDirs {
     /// typical umask makes them `0755`. Until this was fixed, the only thing
     /// keeping another local user out was `~/.cache` and `~/.local/state`
     /// happening to be `0700`, which is a convention of the user's system and
-    /// not something this client established (bugs.md B6).
+    /// not something this client established (`docs/BUGS.md` B6).
     ///
     /// Applied on every start, not just at creation: a directory that already
     /// exists with a permissive mode — restored from a backup that flattened
@@ -346,6 +594,12 @@ impl AppDirs {
     }
 }
 
+/// Whether `id` is what [`AppDirs::machine_id`] writes: 32 lowercase hex
+/// digits.
+fn is_machine_id(id: &str) -> bool {
+    id.len() == 32 && id.bytes().all(|b| matches!(b, b'0'..=b'9' | b'a'..=b'f'))
+}
+
 /// Make `dir` owner-only (`0700`). These directories contain decrypted bytes,
 /// plaintext names, and an authenticated control socket, so failure is fatal.
 fn restrict_dir(dir: &Path) -> Result<()> {
@@ -380,7 +634,7 @@ fn restrict_dir(dir: &Path) -> Result<()> {
 /// session — enumerate, read, upload, trash, share — without presenting any
 /// credential, since the daemon already holds one. It is an authority boundary,
 /// not merely private data, and `UnixListener::bind` applies the umask like any
-/// other file (bugs.md B6).
+/// other file (`docs/BUGS.md` B6).
 ///
 /// There is a window between `bind` and this call during which the socket
 /// carries the umask's mode. It is closed in practice by the containing
@@ -401,7 +655,29 @@ mod tests {
         std::fs::metadata(p).unwrap().permissions().mode() & 0o777
     }
 
-    /// bugs.md B6. A permissive mode here is not a cosmetic problem: these
+    /// A damaged config must not be replaced by defaults the next time a
+    /// setting changes (`docs/BUGS.md` B223).
+    #[test]
+    fn an_unreadable_config_is_never_written_over() {
+        let dir = std::env::temp_dir().join(format!(
+            "pdfs-cfg-guard-{}-{:?}",
+            std::process::id(),
+            std::thread::current().id()
+        ));
+        let _ = std::fs::remove_dir_all(&dir);
+        std::fs::create_dir_all(&dir).unwrap();
+        let path = dir.join("config.json");
+
+        assert!(refuse_to_replace_unreadable(&path).is_ok(), "no file yet");
+        std::fs::write(&path, serde_json::to_vec(&AppConfig::default()).unwrap()).unwrap();
+        assert!(refuse_to_replace_unreadable(&path).is_ok());
+        std::fs::write(&path, br#"{"mountpoint": "/home/me/Pro"#).unwrap();
+        assert!(refuse_to_replace_unreadable(&path).is_err(), "truncated");
+
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+
+    /// `docs/BUGS.md` B6. A permissive mode here is not a cosmetic problem: these
     /// directories hold decrypted content, a plaintext index of every node name,
     /// and the control socket that commands the daemon's session.
     #[test]
@@ -468,27 +744,159 @@ mod tests {
     #[test]
     fn test_app_config_serialization() {
         let config = AppConfig {
-            app_version: "external-drive-test-client@1.0.0".to_string(),
-            user_agent: "test-agent/1.0".to_string(),
+            app_version: Some("external-drive-test-client@1.0.0".to_string()),
+            user_agent: Some("test-agent/1.0".to_string()),
             cache_budget: Some(1234),
+            upload_limit: Some(500_000),
+            download_limit: None,
             mountpoint: Some("/tmp/x".to_string()),
             ignore_patterns: Some(vec!["build/".to_string()]),
             device_uid: Some("dev-uid".to_string()),
             conflict_sweep: Some(SweepMode::Enforce),
             open_with: None,
             prompt: None,
+            proton_theme: Some(true),
+            files_view: FilesView {
+                list: true,
+                sort: FileSort::Modified,
+                descending: true,
+                folders_first: false,
+            },
+            window: WindowState {
+                width: 1200,
+                height: 800,
+                maximized: true,
+            },
+            tray_hidden: true,
+            language: Some("de".to_string()),
+            online_map: true,
+            local_first: Some(false),
+            update_check: true,
+            update_last_checked: Some(1_760_000_000),
+            update_latest: Some("3.6.0".to_string()),
         };
         let json = serde_json::to_string(&config).unwrap();
         let decoded: AppConfig = serde_json::from_str(&json).unwrap();
-        assert_eq!(decoded.app_version, "external-drive-test-client@1.0.0");
-        assert_eq!(decoded.user_agent, "test-agent/1.0");
+        assert_eq!(
+            decoded.resolved_app_version(),
+            "external-drive-test-client@1.0.0"
+        );
+        assert_eq!(decoded.resolved_user_agent(), "test-agent/1.0");
+        assert_eq!(decoded.proton_theme, Some(true));
+        assert!(decoded.tray_hidden);
+        assert!(decoded.online_map);
+        assert!(decoded.update_check);
+        assert_eq!(decoded.update_last_checked, Some(1_760_000_000));
+        assert_eq!(decoded.update_latest.as_deref(), Some("3.6.0"));
+        assert_eq!(decoded.language.as_deref(), Some("de"));
+        assert_eq!(decoded.upload_limit, Some(500_000));
+        assert_eq!(decoded.download_limit, None);
+        assert_eq!(decoded.files_view, config.files_view);
+        assert_eq!(decoded.window, config.window);
+    }
+
+    #[test]
+    fn a_config_without_a_window_opens_at_the_default_size() {
+        let decoded: AppConfig =
+            serde_json::from_str(r#"{"app_version":"a","user_agent":"b"}"#).unwrap();
+        assert_eq!(decoded.window, WindowState::default());
+        let partial: WindowState = serde_json::from_str(r#"{"maximized":true}"#).unwrap();
+        assert_eq!((partial.width, partial.height), (980, 680));
+        assert!(partial.maximized);
+    }
+
+    #[test]
+    fn a_window_saved_too_small_opens_at_the_default_size() {
+        let tiny = WindowState {
+            width: 0,
+            height: 500,
+            maximized: true,
+        };
+        assert_eq!(
+            tiny.usable(),
+            WindowState {
+                maximized: true,
+                ..WindowState::default()
+            }
+        );
+        let fine = WindowState {
+            width: 640,
+            height: 480,
+            maximized: false,
+        };
+        assert_eq!(fine.usable(), fine);
+    }
+
+    #[test]
+    fn a_config_without_a_files_view_gets_the_default() {
+        let decoded: AppConfig =
+            serde_json::from_str(r#"{"app_version":"a","user_agent":"b"}"#).unwrap();
+        assert_eq!(decoded.files_view, FilesView::default());
+        let partial: FilesView = serde_json::from_str(r#"{"sort":"size"}"#).unwrap();
+        assert_eq!(partial.sort, FileSort::Size);
+        assert!(partial.folders_first);
+    }
+
+    #[test]
+    fn the_old_proton_accent_setting_turns_on_the_proton_theme() {
+        let decoded: AppConfig =
+            serde_json::from_str(r#"{"app_version":"a","user_agent":"b","proton_accent":true}"#)
+                .unwrap();
+        assert_eq!(decoded.proton_theme, Some(true));
     }
 
     #[test]
     fn test_default_app_config() {
         let default_config = AppConfig::default();
-        assert_eq!(default_config.app_version, APP_VERSION);
-        assert_eq!(default_config.user_agent, USER_AGENT);
+        assert_eq!(default_config.resolved_app_version(), APP_VERSION);
+        assert_eq!(default_config.resolved_user_agent(), USER_AGENT);
+    }
+
+    #[test]
+    fn the_client_identity_follows_the_release() {
+        let version = env!("CARGO_PKG_VERSION");
+        assert_eq!(
+            APP_VERSION,
+            format!("external-drive-linux@{version}-stable")
+        );
+        assert_eq!(USER_AGENT, format!("proton-drive-linux/{version}"));
+    }
+
+    #[test]
+    fn an_old_config_with_the_fixed_identity_follows_the_release() {
+        // Every config written up to 3.1.0 holds these two values, so they must
+        // not pin the identity once Proton retires 0.1.0.
+        let json = format!(
+            r#"{{"app_version":"{LEGACY_APP_VERSION}","user_agent":"{LEGACY_USER_AGENT}","tray_hidden":true}}"#
+        );
+        let decoded: AppConfig = serde_json::from_str(&json).unwrap();
+        assert_eq!(decoded.resolved_app_version(), APP_VERSION);
+        assert_eq!(decoded.resolved_user_agent(), USER_AGENT);
+        assert!(decoded.tray_hidden, "the rest of the config survives");
+
+        let saved = serde_json::to_string(&decoded).unwrap();
+        assert!(!saved.contains("app_version"), "{saved}");
+        assert!(!saved.contains("user_agent"), "{saved}");
+    }
+
+    #[test]
+    fn a_hand_set_client_identity_is_kept() {
+        let json = r#"{"app_version":"external-drive-linux@9.9.9-beta","user_agent":"custom/1"}"#;
+        let decoded: AppConfig = serde_json::from_str(json).unwrap();
+        assert_eq!(
+            decoded.resolved_app_version(),
+            "external-drive-linux@9.9.9-beta"
+        );
+        assert_eq!(decoded.resolved_user_agent(), "custom/1");
+        let saved = serde_json::to_string(&decoded).unwrap();
+        assert!(saved.contains("external-drive-linux@9.9.9-beta"), "{saved}");
+    }
+
+    #[test]
+    fn a_config_without_a_client_identity_loads() {
+        let decoded: AppConfig = serde_json::from_str("{}").unwrap();
+        assert_eq!(decoded.resolved_app_version(), APP_VERSION);
+        assert_eq!(decoded.resolved_user_agent(), USER_AGENT);
     }
 
     #[test]
@@ -547,6 +955,15 @@ mod tests {
         assert_eq!(decoded.conflict_sweep, Some(SweepMode::Enforce));
     }
 
+    #[test]
+    fn a_machine_id_is_32_lowercase_hex_digits() {
+        assert!(is_machine_id("0123456789abcdef0123456789abcdef"));
+        assert!(!is_machine_id("0123456789ABCDEF0123456789ABCDEF"));
+        assert!(!is_machine_id("0123456789abcdef"));
+        assert!(!is_machine_id("../../../../etc/passwd/0123456789"));
+        assert!(!is_machine_id(""));
+    }
+
     #[test]
     fn a_config_predating_ignore_patterns_still_loads() {
         // Every installed config on disk looks like this; the field must not be
@@ -554,6 +971,11 @@ mod tests {
         let json = r#"{"app_version":"v","user_agent":"u"}"#;
         let decoded: AppConfig = serde_json::from_str(json).unwrap();
         assert!(decoded.ignore_patterns.is_none());
+        assert!(!decoded.tray_hidden, "an old config keeps the tray icon");
+        assert!(
+            decoded.language.is_none(),
+            "an old config follows the system"
+        );
         assert_eq!(
             decoded.resolved_ignore_patterns().len(),
             DEFAULT_IGNORE_PATTERNS.len(),

crates/pdfs-core/src/db/activity.rs

@@ -29,6 +29,17 @@ impl Db {
         Ok(())
     }
 
+    /// Whether the kept log already holds an entry of `kind` for `target`. Lets
+    /// the daemon report something once across restarts rather than once per
+    /// run, without keeping a second table in step with the log.
+    pub fn activity_has(&self, kind: ActivityKind, target: &str) -> Result<bool> {
+        let kind = serde_json::to_string(&kind)?;
+        let conn = self.read();
+        let mut stmt =
+            conn.prepare("SELECT 1 FROM activity WHERE kind = ?1 AND target = ?2 LIMIT 1")?;
+        Ok(stmt.exists(params![kind, target])?)
+    }
+
     /// The most recent activity, newest first, capped at `limit` entries. Rows
     /// whose stored `kind` no longer parses (written by an older build) are
     /// skipped rather than failing the whole read.

crates/pdfs-core/src/db/albums.rs

@@ -178,7 +178,7 @@ impl Db {
                     ratio: r.get(3)?,
                     thumb_state: r.get(4)?,
                     kind: crate::control::PhotoKind::from_i64(r.get(5)?),
-                    // Album rows have no favourite state of their own: an album
+                    // Album rows have no favorite state of their own: an album
                     // photo on someone else's volume is not in our timeline, and
                     // favouriting one is not supported yet anyway.
                     favorite: false,
@@ -239,7 +239,7 @@ impl Db {
                     ratio: r.get(3)?,
                     thumb_state: r.get(4)?,
                     kind: crate::control::PhotoKind::from_i64(r.get(5)?),
-                    // Album rows have no favourite state of their own: an album
+                    // Album rows have no favorite state of their own: an album
                     // photo on someone else's volume is not in our timeline, and
                     // favouriting one is not supported yet anyway.
                     favorite: false,

crates/pdfs-core/src/db/fixtures/v2.8.2.sql

@@ -0,0 +1,249 @@
+/* WARNING: Script requires that SQLITE_DBCONFIG_DEFENSIVE be disabled */
+PRAGMA foreign_keys=OFF;
+BEGIN TRANSACTION;
+CREATE TABLE sync_state (key TEXT PRIMARY KEY, value TEXT);
+INSERT INTO sync_state VALUES('schema_version','35');
+CREATE TABLE nodes (
+  uid           TEXT PRIMARY KEY,
+  parent_uid    TEXT,
+  name          TEXT NOT NULL,
+  is_dir        INTEGER NOT NULL,
+  size          INTEGER,
+  mtime         INTEGER NOT NULL,
+  revision_hash TEXT,
+  trashed       INTEGER NOT NULL DEFAULT 0
+, node_json TEXT, listed INTEGER NOT NULL DEFAULT 0, path TEXT);
+INSERT INTO nodes VALUES('vol~root',NULL,'My Files',1,0,200,NULL,0,'{"uid":{"volume_id":"vol","link_id":"root"},"parent_uid":null,"kind":"Folder","name":"My Files","creation_time":100,"modification_time":200,"trashed":false,"is_shared":false,"is_shared_publicly":false,"signature_email":null,"membership":null,"direct_role":null,"share_id":null,"photo":null,"album":null,"verification":{"name":"NotSigned","passphrase":"NotSigned","content_key":null,"extended_attributes":null}}',0,'');
+INSERT INTO nodes VALUES('vol~docs','vol~root','Docs',1,0,200,NULL,0,'{"uid":{"volume_id":"vol","link_id":"docs"},"parent_uid":{"volume_id":"vol","link_id":"root"},"kind":"Folder","name":"Docs","creation_time":100,"modification_time":200,"trashed":false,"is_shared":false,"is_shared_publicly":false,"signature_email":null,"membership":null,"direct_role":null,"share_id":null,"photo":null,"album":null,"verification":{"name":"NotSigned","passphrase":"NotSigned","content_key":null,"extended_attributes":null}}',0,'Docs');
+INSERT INTO nodes VALUES('vol~report','vol~docs','report.txt',0,10,200,NULL,0,'{"uid":{"volume_id":"vol","link_id":"report"},"parent_uid":{"volume_id":"vol","link_id":"docs"},"kind":{"File":{"media_type":"text/plain","total_size_on_storage":20,"active_revision_state":null,"active_revision_id":null,"claimed_size":10,"claimed_modification_time":null,"content_sha1":null}},"name":"report.txt","creation_time":100,"modification_time":200,"trashed":false,"is_shared":false,"is_shared_publicly":false,"signature_email":null,"membership":null,"direct_role":null,"share_id":null,"photo":null,"album":null,"verification":{"name":"NotSigned","passphrase":"NotSigned","content_key":null,"extended_attributes":null}}',0,'Docs/report.txt');
+INSERT INTO nodes VALUES('vol~old','vol~docs','old.txt',0,3,200,NULL,0,'{"uid":{"volume_id":"vol","link_id":"old"},"parent_uid":{"volume_id":"vol","link_id":"docs"},"kind":{"File":{"media_type":"text/plain","total_size_on_storage":13,"active_revision_state":null,"active_revision_id":null,"claimed_size":3,"claimed_modification_time":null,"content_sha1":null}},"name":"old.txt","creation_time":100,"modification_time":200,"trashed":false,"is_shared":false,"is_shared_publicly":false,"signature_email":null,"membership":null,"direct_role":null,"share_id":null,"photo":null,"album":null,"verification":{"name":"NotSigned","passphrase":"NotSigned","content_key":null,"extended_attributes":null}}',0,'Docs/old.txt');
+INSERT INTO nodes VALUES('vol~shared','vol~docs','shared.txt',0,4,200,NULL,0,'{"uid":{"volume_id":"vol","link_id":"shared"},"parent_uid":{"volume_id":"vol","link_id":"docs"},"kind":{"File":{"media_type":"text/plain","total_size_on_storage":14,"active_revision_state":null,"active_revision_id":null,"claimed_size":4,"claimed_modification_time":null,"content_sha1":null}},"name":"shared.txt","creation_time":100,"modification_time":200,"trashed":false,"is_shared":false,"is_shared_publicly":false,"signature_email":null,"membership":null,"direct_role":null,"share_id":null,"photo":null,"album":null,"verification":{"name":"NotSigned","passphrase":"NotSigned","content_key":null,"extended_attributes":null}}',0,'Docs/shared.txt');
+INSERT INTO nodes VALUES('local~dir','vol~root','Offline folder',1,0,200,NULL,0,'{"uid":{"volume_id":"local","link_id":"dir"},"parent_uid":{"volume_id":"vol","link_id":"root"},"kind":"Folder","name":"Offline folder","creation_time":100,"modification_time":200,"trashed":false,"is_shared":false,"is_shared_publicly":false,"signature_email":null,"membership":null,"direct_role":null,"share_id":null,"photo":null,"album":null,"verification":{"name":"NotSigned","passphrase":"NotSigned","content_key":null,"extended_attributes":null}}',1,'Offline folder');
+INSERT INTO nodes VALUES('local~inner','local~dir','inner.txt',0,5,200,NULL,0,'{"uid":{"volume_id":"local","link_id":"inner"},"parent_uid":{"volume_id":"local","link_id":"dir"},"kind":{"File":{"media_type":"text/plain","total_size_on_storage":15,"active_revision_state":null,"active_revision_id":null,"claimed_size":5,"claimed_modification_time":null,"content_sha1":null}},"name":"inner.txt","creation_time":100,"modification_time":200,"trashed":false,"is_shared":false,"is_shared_publicly":false,"signature_email":null,"membership":null,"direct_role":null,"share_id":null,"photo":null,"album":null,"verification":{"name":"NotSigned","passphrase":"NotSigned","content_key":null,"extended_attributes":null}}',0,'Offline folder/inner.txt');
+INSERT INTO nodes VALUES('local~part','vol~root','movie.part',0,7,200,NULL,0,'{"uid":{"volume_id":"local","link_id":"part"},"parent_uid":{"volume_id":"vol","link_id":"root"},"kind":{"File":{"media_type":"text/plain","total_size_on_storage":17,"active_revision_state":null,"active_revision_id":null,"claimed_size":7,"claimed_modification_time":null,"content_sha1":null}},"name":"movie.part","creation_time":100,"modification_time":200,"trashed":false,"is_shared":false,"is_shared_publicly":false,"signature_email":null,"membership":null,"direct_role":null,"share_id":null,"photo":null,"album":null,"verification":{"name":"NotSigned","passphrase":"NotSigned","content_key":null,"extended_attributes":null}}',0,'movie.part');
+CREATE TABLE cache_entries (
+  cache_key     TEXT PRIMARY KEY,
+  size_bytes    INTEGER,
+  last_accessed INTEGER,
+  is_pinned     INTEGER NOT NULL DEFAULT 0
+, kind TEXT NOT NULL DEFAULT 'blob');
+CREATE TABLE pins (
+  uid       TEXT PRIMARY KEY,
+  path      TEXT NOT NULL,
+  recursive INTEGER NOT NULL DEFAULT 0
+);
+CREATE TABLE local_files (
+  id       INTEGER PRIMARY KEY,
+  path     TEXT NOT NULL UNIQUE,
+  name     TEXT NOT NULL,
+  is_dir   INTEGER NOT NULL,
+  size     INTEGER NOT NULL DEFAULT 0,
+  mtime    INTEGER NOT NULL DEFAULT 0,
+  scan_gen INTEGER NOT NULL DEFAULT 0
+);
+CREATE TABLE photos (
+  uid          TEXT PRIMARY KEY,
+  capture_time INTEGER NOT NULL,
+  name         TEXT,
+  ratio        REAL,
+  thumb_state  INTEGER NOT NULL DEFAULT 0,
+  seq          INTEGER NOT NULL
+, media_type TEXT, kind INTEGER NOT NULL DEFAULT 0, favorite INTEGER NOT NULL DEFAULT 0, content_hash TEXT, main_uid TEXT, group_key TEXT, resolved_at INTEGER, latitude REAL, longitude REAL, place_id INTEGER, similar_hash INTEGER);
+CREATE TABLE trash (
+  uid    TEXT PRIMARY KEY,
+  name   TEXT NOT NULL,
+  is_dir INTEGER NOT NULL,
+  size   INTEGER NOT NULL DEFAULT 0,
+  mtime  INTEGER NOT NULL DEFAULT 0
+, parent_uid TEXT, trashed_at INTEGER);
+CREATE TABLE device (
+  uid      TEXT PRIMARY KEY,
+  share_id TEXT NOT NULL,
+  root_uid TEXT NOT NULL,
+  name     TEXT NOT NULL,
+  created  INTEGER NOT NULL DEFAULT 0
+);
+CREATE TABLE sync_folder (
+  id              INTEGER PRIMARY KEY,
+  local_path      TEXT NOT NULL UNIQUE,
+  remote_uid      TEXT NOT NULL,
+  remote_share_id TEXT NOT NULL,
+  mode            TEXT NOT NULL DEFAULT 'mirror',
+  state           TEXT NOT NULL DEFAULT 'idle',
+  last_sync       INTEGER NOT NULL DEFAULT 0
+, pending_mode TEXT);
+CREATE TABLE sync_entry (
+  folder_id   INTEGER NOT NULL,
+  rel_path    TEXT NOT NULL,
+  remote_uid  TEXT,
+  local_mtime INTEGER NOT NULL DEFAULT 0,
+  local_size  INTEGER NOT NULL DEFAULT 0,
+  remote_hash TEXT,
+  remote_rev  TEXT, local_mtime_ns INTEGER,
+  PRIMARY KEY (folder_id, rel_path)
+);
+CREATE TABLE activity (
+  id     INTEGER PRIMARY KEY,
+  time   INTEGER NOT NULL,
+  kind   TEXT NOT NULL,
+  target TEXT NOT NULL,
+  detail TEXT NOT NULL DEFAULT '',
+  ok     INTEGER NOT NULL DEFAULT 1
+);
+CREATE TABLE pending_op (
+  id              INTEGER PRIMARY KEY,
+  kind            TEXT NOT NULL,
+  uid             TEXT NOT NULL,
+  blob_path       TEXT,
+  meta_json       TEXT,
+  created_at      INTEGER NOT NULL,
+  attempts        INTEGER NOT NULL DEFAULT 0,
+  last_error      TEXT,
+  next_attempt_at INTEGER NOT NULL DEFAULT 0
+, parent_uid TEXT, name TEXT, claimed_at INTEGER NOT NULL DEFAULT 0, access_deferred_since INTEGER NOT NULL DEFAULT 0);
+INSERT INTO pending_op VALUES(1,'mkdir','local~dir',NULL,NULL,1700000000000,0,NULL,0,'vol~root','Offline folder',4611686018427387903,0);
+INSERT INTO pending_op VALUES(2,'create','local~inner','/blobs/inner','{"len":5,"mtime":1700000000,"based_on":null}',1700000000000,0,NULL,0,'local~dir','inner.txt',0,0);
+INSERT INTO pending_op VALUES(3,'create','local~part','/blobs/part','{"len":5,"mtime":1700000000,"based_on":null}',1700000000000,0,NULL,8000000000000,'vol~root','movie.part',0,0);
+INSERT INTO pending_op VALUES(4,'revision','vol~report','/blobs/report','{"len":5,"mtime":1700000000,"based_on":null}',1700000000000,2,'network unreachable',1700000900000,NULL,NULL,0,0);
+INSERT INTO pending_op VALUES(5,'rename','vol~shared',NULL,'{"original_parent_uid":"vol~docs","original_name":"shared.txt"}',1700000000000,0,NULL,1700000700000,'vol~root','shared-moved.txt',0,1700000100000);
+INSERT INTO pending_op VALUES(6,'trash','vol~old',NULL,NULL,1700000000000,0,NULL,0,NULL,NULL,0,0);
+CREATE TABLE IF NOT EXISTS 'local_fts_data'(id INTEGER PRIMARY KEY, block BLOB);
+INSERT INTO local_fts_data VALUES(1,x'000000');
+INSERT INTO local_fts_data VALUES(10,x'00000000000000');
+CREATE TABLE IF NOT EXISTS 'local_fts_idx'(segid, term, pgno, PRIMARY KEY(segid, term)) WITHOUT ROWID;
+CREATE TABLE IF NOT EXISTS 'local_fts_docsize'(id INTEGER PRIMARY KEY, sz BLOB);
+CREATE TABLE IF NOT EXISTS 'local_fts_config'(k PRIMARY KEY, v) WITHOUT ROWID;
+INSERT INTO local_fts_config VALUES('version',4);
+CREATE TABLE share_access (
+  root_uid TEXT PRIMARY KEY,
+  access   TEXT NOT NULL
+           CHECK (access IN ('owner', 'editor', 'viewer', 'unknown'))
+);
+CREATE TABLE mount (
+  id              INTEGER PRIMARY KEY,
+  kind            TEXT NOT NULL
+                  CHECK (kind IN ('myfiles', 'device', 'shared')),
+  sync_folder_id  INTEGER REFERENCES sync_folder(id) ON DELETE CASCADE,
+  share_root_uid  TEXT,
+  local_path      TEXT NOT NULL DEFAULT '',
+  root_uid        TEXT NOT NULL DEFAULT '',
+  root_share_id   TEXT NOT NULL DEFAULT '',
+  mode            TEXT NOT NULL DEFAULT 'unknown',
+  access          TEXT NOT NULL DEFAULT 'rw'
+                  CHECK (access IN ('rw', 'ro')),
+  CHECK (
+    (kind = 'myfiles' AND sync_folder_id IS NULL AND share_root_uid IS NULL)
+    OR
+    (kind = 'device' AND sync_folder_id IS NOT NULL AND share_root_uid IS NULL)
+    OR
+    (kind = 'shared' AND sync_folder_id IS NULL AND share_root_uid IS NOT NULL)
+  )
+);
+CREATE TABLE albums (
+  uid           TEXT PRIMARY KEY,
+  name          TEXT NOT NULL,
+  photo_count   INTEGER NOT NULL DEFAULT 0,
+  cover_uid     TEXT,
+  last_activity INTEGER,
+  shared        INTEGER NOT NULL DEFAULT 0,
+  seq           INTEGER NOT NULL
+);
+CREATE TABLE album_photos (
+  album_uid    TEXT NOT NULL,
+  uid          TEXT NOT NULL,
+  capture_time INTEGER NOT NULL,
+  name         TEXT,
+  media_type   TEXT,
+  kind         INTEGER NOT NULL DEFAULT 0,
+  ratio        REAL,
+  thumb_state  INTEGER NOT NULL DEFAULT 0,
+  seq          INTEGER NOT NULL,
+  PRIMARY KEY (album_uid, uid)
+);
+CREATE TABLE IF NOT EXISTS 'nodes_fts_data'(id INTEGER PRIMARY KEY, block BLOB);
+INSERT INTO nodes_fts_data VALUES(1,x'083850');
+INSERT INTO nodes_fts_data VALUES(10,x'000000000108080008010101020101030101040101050101060101070101080101');
+INSERT INTO nodes_fts_data VALUES(137438953473,x'000000340430206669010204010366696c0102050103696c6501020601036c657301020701036d79200102020103792066010203040808080808');
+INSERT INTO nodes_fts_data VALUES(274877906945,x'0000001a0430646f6302080201010201036f6373020803010103040b');
+INSERT INTO nodes_fts_data VALUES(412316860417,x'0000008b04302e747803080801010d01032f72650306010106010363732f03060101040103646f630306010102010365706f03080301010801036f637303060101030202727403080501010a0103706f7203080401010901037265700308020101070202742e03080601010b0103732f7203060101050103742e7403080701010c0202787403080901010e040b0a0a0a0b0a0a0b0b0a0a0b');
+INSERT INTO nodes_fts_data VALUES(549755813889,x'0000006b04302e747804080501010a01032f6f6c0406010106010363732f04060101040103642e7404080401010902026f63040601010201036c642e04080301010801036f6373040601010302026c640408020101070103732f6f0406010105010374787404080601010b040b0a0a0b090b0a0a0a');
+INSERT INTO nodes_fts_data VALUES(687194767361,x'0000008c04302e747805080801010d01032f736805060101060103617265050804010109010363732f05060101040103642e7405080701010c02026f630506010102010365642e05080601010b010368617205080301010801036f63730506010103010372656405080501010a0103732f73050601010502026861050802010107010374787405080901010e040b0a0b0a0b090b0b0a0b0a0a');
+INSERT INTO nodes_fts_data VALUES(824633720833,x'00000084043020666f060809010109010364657206080d01010d0103652066060808010108010366666c06080301010302026c6906080401010402026f6c06080a01010a0103696e6506080601010601036c646506080c01010c0202696e06080501010501036e652006080701010701036f666606080201010202026c6406080b01010b040b0b0b0b0a0a0b0b0a0b0b');
+INSERT INTO nodes_fts_data VALUES(962072674305,x'000000da043020666f070601010901032e747807080701011601032f696e07060101100103646572070601010d010365206607060101080202722e07080501011403012f070601010e010366666c070601010302026c69070601010402026f6c070601010a0103696e65070601010603016e07080201011101036c6465070601010c0202696e070601010501036e6520070601010703017207080401011302026e6507080301011201036f6666070601010202026c64070601010b0103722e7407080601011502022f69070601010f0103747874070808010117040a0b0a0a0a0a080a09090a090a090a090a0a090b09');
+INSERT INTO nodes_fts_data VALUES(1099511627777,x'0000005c04302e706108080701010701036172740808090101090103652e70080806010106010369652e08080501010501036d6f7608080201010201036f766908080301010301037061720808080101080103766965080804010104040b0b0b0b0b0b0b');
+CREATE TABLE IF NOT EXISTS 'nodes_fts_idx'(segid, term, pgno, PRIMARY KEY(segid, term)) WITHOUT ROWID;
+INSERT INTO nodes_fts_idx VALUES(1,x'',2);
+INSERT INTO nodes_fts_idx VALUES(2,x'',2);
+INSERT INTO nodes_fts_idx VALUES(3,x'',2);
+INSERT INTO nodes_fts_idx VALUES(4,x'',2);
+INSERT INTO nodes_fts_idx VALUES(5,x'',2);
+INSERT INTO nodes_fts_idx VALUES(6,x'',2);
+INSERT INTO nodes_fts_idx VALUES(7,x'',2);
+INSERT INTO nodes_fts_idx VALUES(8,x'',2);
+CREATE TABLE IF NOT EXISTS 'nodes_fts_content'(id INTEGER PRIMARY KEY, c0, c1);
+INSERT INTO nodes_fts_content VALUES(1,'My Files','');
+INSERT INTO nodes_fts_content VALUES(2,'Docs','Docs');
+INSERT INTO nodes_fts_content VALUES(3,'report.txt','Docs/report.txt');
+INSERT INTO nodes_fts_content VALUES(4,'old.txt','Docs/old.txt');
+INSERT INTO nodes_fts_content VALUES(5,'shared.txt','Docs/shared.txt');
+INSERT INTO nodes_fts_content VALUES(6,'Offline folder','Offline folder');
+INSERT INTO nodes_fts_content VALUES(7,'inner.txt','Offline folder/inner.txt');
+INSERT INTO nodes_fts_content VALUES(8,'movie.part','movie.part');
+CREATE TABLE IF NOT EXISTS 'nodes_fts_docsize'(id INTEGER PRIMARY KEY, sz BLOB);
+INSERT INTO nodes_fts_docsize VALUES(1,x'0600');
+INSERT INTO nodes_fts_docsize VALUES(2,x'0202');
+INSERT INTO nodes_fts_docsize VALUES(3,x'080d');
+INSERT INTO nodes_fts_docsize VALUES(4,x'050a');
+INSERT INTO nodes_fts_docsize VALUES(5,x'080d');
+INSERT INTO nodes_fts_docsize VALUES(6,x'0c0c');
+INSERT INTO nodes_fts_docsize VALUES(7,x'0716');
+INSERT INTO nodes_fts_docsize VALUES(8,x'0808');
+CREATE TABLE IF NOT EXISTS 'nodes_fts_config'(k PRIMARY KEY, v) WITHOUT ROWID;
+INSERT INTO nodes_fts_config VALUES('version',4);
+CREATE TABLE own_sealed_rev (
+  uid         TEXT PRIMARY KEY,
+  revision_id TEXT NOT NULL,
+  sealed_at   INTEGER NOT NULL
+);
+INSERT INTO own_sealed_rev VALUES('vol~report','rev-1',1700000000000);
+PRAGMA writable_schema=ON;
+INSERT INTO sqlite_schema(type,name,tbl_name,rootpage,sql)VALUES('table','local_fts','local_fts',0,'CREATE VIRTUAL TABLE local_fts USING fts5(
+  name, path, content=''local_files'', content_rowid=''id'', tokenize=''trigram''
+)');
+INSERT INTO sqlite_schema(type,name,tbl_name,rootpage,sql)VALUES('table','nodes_fts','nodes_fts',0,'CREATE VIRTUAL TABLE nodes_fts USING fts5(
+  name, path, tokenize=''trigram''
+)');
+CREATE TRIGGER mount_sync_folder_insert
+AFTER INSERT ON sync_folder
+BEGIN
+  INSERT OR IGNORE INTO mount (kind, sync_folder_id)
+  VALUES ('device', NEW.id);
+END;
+CREATE INDEX idx_nodes_parent ON nodes(parent_uid);
+CREATE INDEX idx_photos_seq ON photos(seq);
+CREATE INDEX pending_op_uid ON pending_op(uid);
+CREATE INDEX pending_op_parent ON pending_op(parent_uid);
+CREATE INDEX cache_entries_lru ON cache_entries(kind, last_accessed);
+CREATE INDEX idx_nodes_name_nocase ON nodes(name COLLATE NOCASE);
+CREATE INDEX idx_local_files_name_nocase
+  ON local_files(name COLLATE NOCASE);
+CREATE UNIQUE INDEX mount_myfiles
+  ON mount(kind) WHERE kind = 'myfiles';
+CREATE UNIQUE INDEX mount_device
+  ON mount(sync_folder_id) WHERE kind = 'device';
+CREATE UNIQUE INDEX mount_shared
+  ON mount(share_root_uid) WHERE kind = 'shared';
+CREATE INDEX idx_album_photos_seq ON album_photos(album_uid, seq);
+CREATE INDEX idx_album_photos_uid ON album_photos(uid);
+CREATE INDEX idx_photos_favorite ON photos(favorite, seq);
+CREATE INDEX idx_sync_entry_remote_uid ON sync_entry(remote_uid);
+CREATE INDEX idx_local_files_scan_gen ON local_files(scan_gen);
+CREATE INDEX idx_nodes_parent_live ON nodes(parent_uid) WHERE trashed = 0;
+CREATE INDEX idx_pending_op_claimed
+    ON pending_op(uid) WHERE claimed_at <> 0;
+CREATE INDEX idx_trash_parent ON trash(parent_uid);
+CREATE INDEX idx_photos_group ON photos(group_key);
+CREATE INDEX idx_photos_unresolved ON photos(resolved_at);
+CREATE INDEX idx_photos_place ON photos(place_id);
+PRAGMA writable_schema=OFF;
+COMMIT;

crates/pdfs-core/src/db/local.rs

@@ -111,9 +111,80 @@ impl Db {
         )?;
         let count: i64 = tx.query_row("SELECT COUNT(*) FROM local_files", [], |r| r.get(0))?;
         tx.commit()?;
+        drop(conn);
+        self.compact_search_indexes_if_due(finished_at)?;
         Ok(count)
     }
 
+    /// Close a scan that did not see everything: it stopped at the entry cap,
+    /// or a batch failed to write. Nothing is pruned, since a path it missed is
+    /// not a path that is gone; the next complete scan prunes. Returns the
+    /// number of indexed entries.
+    ///
+    /// Pruning after such a scan dropped every entry it had not reached, and
+    /// the next scan put them back. That churn is most of what grew
+    /// `local_fts` to 16 times its size (`docs/BUGS.md` B217).
+    pub fn local_finish_partial_scan(&self, finished_at: i64) -> Result<i64> {
+        let conn = self.conn.lock();
+        conn.execute(
+            "INSERT INTO sync_state (key, value) VALUES ('local_indexed_at', ?1)
+             ON CONFLICT(key) DO UPDATE SET value = excluded.value",
+            [finished_at.to_string()],
+        )?;
+        let count: i64 = conn.query_row("SELECT COUNT(*) FROM local_files", [], |r| r.get(0))?;
+        Ok(count)
+    }
+
+    /// Compact both full-text indexes, at most once a day.
+    ///
+    /// FTS5 keeps a deleted entry's data until the segments holding it are
+    /// merged, and with the few large segments these indexes settle into that
+    /// is close to never. Every file the local scan sees disappear or appear
+    /// leaves some behind, and every node row the tree replaces does too: on
+    /// one machine `local_fts` held 317 MiB for 40,000 paths, against 20 MiB
+    /// rebuilt, and a search took four times as long (`docs/BUGS.md` B217).
+    ///
+    /// `local_fts` is rebuilt from `local_files`, which also drops anything
+    /// left over, and `nodes_fts`, which has no content table to rebuild from,
+    /// is merged into one segment. Each runs in its own transaction on the
+    /// write connection, about half a second on an index that size; once a
+    /// day keeps the leftovers small without paying that at every scan.
+    pub fn compact_search_indexes_if_due(&self, now: i64) -> Result<()> {
+        const EVERY_SECS: i64 = 24 * 60 * 60;
+        let last: i64 = self
+            .read()
+            .query_row(
+                "SELECT value FROM sync_state WHERE key = 'fts_compacted_at'",
+                [],
+                |row| row.get::<_, String>(0),
+            )
+            .optional()?
+            .and_then(|v| v.parse().ok())
+            .unwrap_or(0);
+        if now - last < EVERY_SECS {
+            return Ok(());
+        }
+        let started = std::time::Instant::now();
+        {
+            let conn = self.conn.lock();
+            conn.execute("INSERT INTO local_fts(local_fts) VALUES('rebuild')", [])?;
+        }
+        {
+            let conn = self.conn.lock();
+            conn.execute("INSERT INTO nodes_fts(nodes_fts) VALUES('optimize')", [])?;
+            conn.execute(
+                "INSERT INTO sync_state (key, value) VALUES ('fts_compacted_at', ?1)
+                 ON CONFLICT(key) DO UPDATE SET value = excluded.value",
+                [now.to_string()],
+            )?;
+        }
+        tracing::info!(
+            elapsed_ms = started.elapsed().as_millis() as u64,
+            "compacted the search indexes"
+        );
+        Ok(())
+    }
+
     /// When the last local scan completed (epoch seconds), or `None` if the index
     /// has never been built. The daemon uses this to decide whether a fresh mount
     /// needs an immediate rescan or can serve the existing index.

crates/pdfs-core/src/db/maintenance.rs

@@ -1,4 +1,4 @@
-//! Database health reporting and repair (features.md 4.3).
+//! Database health reporting and repair.
 //!
 //! Everything here exists for the case where something has already gone wrong:
 //! a crash mid-write, a full disk, a cache that has grown out of proportion to
@@ -91,8 +91,13 @@ impl Db {
     /// Returns an empty vector when the database is sound; otherwise each entry
     /// is one problem SQLite found. Read-only and safe against a live daemon,
     /// though it reads every page, so it is not free on a large file.
+    ///
+    /// Runs on a read connection: it reads every page, which took 34 s on a
+    /// 567 MiB file, and on the write connection every write in the daemon
+    /// waited that long behind it (`docs/BUGS.md` B218). WAL lets it read a
+    /// consistent snapshot while the writer goes on.
     pub fn integrity_check(&self) -> Result<Vec<String>> {
-        let conn = self.conn.lock();
+        let conn = self.read();
         let mut stmt = conn.prepare("PRAGMA integrity_check")?;
         let rows = stmt.query_map([], |r| r.get::<_, String>(0))?;
         let mut problems = Vec::new();
@@ -107,9 +112,11 @@ impl Db {
         Ok(problems)
     }
 
-    /// Size and row-count snapshot for `pdfs cache inspect`.
+    /// Size and row-count snapshot for `pdfs cache inspect`. On a read
+    /// connection, for the reason [`integrity_check`](Self::integrity_check)
+    /// gives: counting the rows of every table reads most of the file.
     pub fn stats(&self) -> Result<DbStats> {
-        let conn = self.conn.lock();
+        let conn = self.read();
         let page_count: i64 = conn.query_row("PRAGMA page_count", [], |r| r.get(0))?;
         let free_pages: i64 = conn.query_row("PRAGMA freelist_count", [], |r| r.get(0))?;
         let page_size: i64 = conn.query_row("PRAGMA page_size", [], |r| r.get(0))?;
@@ -218,6 +225,40 @@ mod tests {
         assert!(db.integrity_check().unwrap().is_empty());
     }
 
+    /// `pdfs diagnose` and `pdfs cache inspect` read the whole file. Behind the
+    /// write lock they froze every write in the daemon for as long as that
+    /// took (`docs/BUGS.md` B218).
+    #[test]
+    fn the_integrity_check_and_stats_do_not_wait_for_the_writer() {
+        let path = std::env::temp_dir().join(format!(
+            "pdfs-check-{}-{}.db",
+            std::process::id(),
+            std::time::SystemTime::now()
+                .duration_since(std::time::UNIX_EPOCH)
+                .unwrap()
+                .as_nanos()
+        ));
+        let db = std::sync::Arc::new(Db::open(&path).unwrap());
+        let writer = db.conn.lock();
+        let (tx, rx) = std::sync::mpsc::channel();
+        let reader = std::sync::Arc::clone(&db);
+        std::thread::spawn(move || {
+            let checked = reader.integrity_check().map(|p| p.is_empty());
+            let counted = reader.stats().map(|s| s.page_count > 0);
+            let _ = tx.send((checked.ok(), counted.ok()));
+        });
+        let answer = rx.recv_timeout(std::time::Duration::from_secs(10));
+        drop(writer);
+        assert_eq!(answer.ok(), Some((Some(true), Some(true))));
+        drop(db);
+        for suffix in ["", "-wal", "-shm"] {
+            let mut file = path.as_os_str().to_owned();
+            file.push(suffix);
+            let _ = std::fs::remove_file(std::path::PathBuf::from(file));
+        }
+        let _ = std::fs::remove_file(path.with_extension("db.lock"));
+    }
+
     #[test]
     fn stats_report_the_schema_and_the_known_tables() {
         let db = Db::open_in_memory().unwrap();

crates/pdfs-core/src/db/migrations.rs

@@ -5,11 +5,13 @@
 //! is already on disk somewhere, so a change here would silently diverge from
 //! what an existing install actually has. Add a new one instead.
 
+use rusqlite::OptionalExtension;
+
 use super::Db;
 use crate::Result;
 
 /// Current schema version. Bump on every forward migration added below.
-pub(super) const SCHEMA_VERSION: i64 = 31;
+pub(super) const SCHEMA_VERSION: i64 = 44;
 
 impl Db {
     pub(super) fn migrate(&self) -> Result<()> {
@@ -270,6 +272,201 @@ impl Db {
                 tx.execute_batch(MIGRATION_V31)?;
             }
         }
+        if current < 32 {
+            // Data only, so safe to repeat; each table is guarded for a very
+            // old fixture that never had it.
+            for table in ["photos", "album_photos"] {
+                let has_table: bool = tx.query_row(
+                    "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = ?1",
+                    [table],
+                    |row| row.get::<_, i64>(0),
+                )? > 0;
+                if has_table {
+                    tx.execute(&MIGRATION_V32.replace("{table}", table), [])?;
+                }
+            }
+        }
+        if current < 33 {
+            // Same guards as V26-V31.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('photos') WHERE name = 'place_id'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_photos: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'photos'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_photos && !has_column {
+                tx.execute_batch(MIGRATION_V33)?;
+            }
+        }
+        if current < 34 {
+            // Same guards as V26-V33.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('photos') WHERE name = 'similar_hash'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_photos: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'photos'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_photos && !has_column {
+                tx.execute_batch(MIGRATION_V34)?;
+            }
+        }
+        if current < 35 {
+            // Same guards as V26-V34.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('trash') WHERE name = 'trashed_at'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_trash: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'trash'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_trash && !has_column {
+                tx.execute_batch(MIGRATION_V35)?;
+            }
+        }
+        if current < 36 {
+            // Same guards as V26-V35.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('nodes') WHERE name = 'lid'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_nodes: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'nodes'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_nodes && !has_column {
+                tx.execute_batch(MIGRATION_V36)?;
+            }
+        }
+        if current < 37 {
+            // Same guards as V26-V36.
+            let sql: Option<String> = tx
+                .query_row(
+                    "SELECT sql FROM sqlite_master WHERE type = 'table' AND name = 'pending_op'",
+                    [],
+                    |row| row.get(0),
+                )
+                .optional()?;
+            if sql.is_some_and(|sql| !sql.contains("AUTOINCREMENT")) {
+                tx.execute_batch(MIGRATION_V37)?;
+            }
+        }
+        if current < 38 {
+            // Same guards as V26-V37.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('pending_op') WHERE name = 'issue'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_ops: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'pending_op'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_ops && !has_column {
+                tx.execute_batch(MIGRATION_V38)?;
+            }
+        }
+        if current < 39 {
+            // Same guards as V26-V38.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('pending_op') WHERE name = 'lid'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_ops: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'pending_op'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_ops && !has_column {
+                tx.execute_batch(MIGRATION_V39)?;
+            }
+        }
+        if current < 40 {
+            // Same guards as V26-V39.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('nodes') WHERE name = 'parent_lid'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_nodes: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'nodes'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_nodes && !has_column {
+                tx.execute_batch(MIGRATION_V40)?;
+            }
+        }
+        if current < 41 {
+            // Needs both tables; on a database that never had them there is
+            // nothing to rewrite.
+            let tables: i64 = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master
+                 WHERE type = 'table' AND name IN ('nodes', 'pending_op', 'pins')",
+                [],
+                |row| row.get(0),
+            )?;
+            if tables == 3 {
+                tx.execute_batch(MIGRATION_V41)?;
+            }
+        }
+        if current < 42 {
+            // Same guards as V26-V40.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('pending_op') WHERE name = 'sent_sha1'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_ops: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'pending_op'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_ops && !has_column {
+                tx.execute_batch(MIGRATION_V42)?;
+            }
+        }
+        if current < 43 {
+            let has_ops: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'pending_op'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_ops {
+                tx.execute_batch(MIGRATION_V43)?;
+            }
+        }
+        if current < 44 {
+            // Same guards as V26-V42.
+            let has_column: bool = tx.query_row(
+                "SELECT COUNT(*) FROM pragma_table_info('sync_entry') \
+                 WHERE name = 'remote_revision_id'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            let has_entries: bool = tx.query_row(
+                "SELECT COUNT(*) FROM sqlite_master WHERE type = 'table' AND name = 'sync_entry'",
+                [],
+                |row| row.get::<_, i64>(0),
+            )? > 0;
+            if has_entries && !has_column {
+                tx.execute_batch(MIGRATION_V44)?;
+            }
+        }
         tx.execute(
             "INSERT INTO sync_state (key, value) VALUES ('schema_version', ?1)
              ON CONFLICT(key) DO UPDATE SET value = excluded.value",
@@ -407,7 +604,7 @@ CREATE TABLE trash (
 );
 ";
 
-/// Schema v8: device sync (devices.md). This machine registers as one Proton Drive
+/// Schema v8: device sync. This machine registers as one Proton Drive
 /// **Device** (a share + root folder on the main volume); `device` is a singleton
 /// row cached so we reuse the same device across restarts instead of creating a
 /// new one each run. `sync_folder` is one row per local folder the user added,
@@ -468,8 +665,8 @@ ALTER TABLE sync_folder ADD COLUMN pending_mode TEXT;
 ";
 
 /// Schema v11: writes no longer upload inside the FUSE handler. `release` stages
-/// the bytes and records the intended upload here, and a drain worker performs it
-/// (offline.md Phase 3). The row outlives the process, so a write survives both a
+/// the bytes and records the intended upload here, and a drain worker performs it.
+/// The row outlives the process, so a write survives both a
 /// dead network and a restart.
 ///
 /// `blob_path` points into the content cache's `staging/` dir and `meta_json` is
@@ -496,7 +693,7 @@ CREATE INDEX pending_op_uid ON pending_op(uid);
 
 /// Schema v12: `pending_op` also carries mutations that *create* a node, which a
 /// revision op never had to describe — it always addressed a node the server had
-/// already minted a uid for (offline.md Phase 3b).
+/// already minted a uid for.
 ///
 /// An offline `create`/`mkdir` cannot get a real uid, so the node is invented
 /// locally under a `local~<uuid>` placeholder and the op records where it goes
@@ -684,7 +881,7 @@ CREATE INDEX idx_album_photos_seq ON album_photos(album_uid, seq);
 CREATE INDEX idx_album_photos_uid ON album_photos(uid);
 ";
 
-/// V20: whether a photo is a favourite.
+/// V20: whether a photo is a favorite.
 ///
 /// A column rather than a tag table: `Favorite` is the only Proton photo tag a
 /// user sets by hand (the rest — Video, Screenshot, Selfie, Raw… — are the
@@ -886,7 +1083,7 @@ ALTER TABLE pending_op ADD COLUMN access_deferred_since INTEGER NOT NULL DEFAULT
 /// as the last sync and left the file the same length — a fixed-size record
 /// rewritten, an image re-exported, a database page updated in place — read as
 /// *unchanged* and was never uploaded. The remote copy silently stayed at the
-/// older content until something else about the file moved (bugs.md B25).
+/// older content until something else about the file moved (`docs/BUGS.md` B25).
 ///
 /// Nullable rather than a converted `local_mtime * 1_000_000_000`: the
 /// nanosecond part of an existing baseline is genuinely unknown, and inventing
@@ -938,7 +1135,7 @@ CREATE INDEX IF NOT EXISTS idx_photos_group ON photos(group_key);
 /// Schema v31: which photos a refresh still has to read from the server.
 ///
 /// A refresh used to resolve every photo's node to learn its name, media type,
-/// favourite tag, content hash and photo relation — minutes of round-trips for a
+/// favorite tag, content hash and photo relation — minutes of round-trips for a
 /// library of any size, repeated every time the timeline went stale. `resolved_at`
 /// records that the work was already done, so a refresh only reads the photos it
 /// has never read and the ones a remote event marked stale (`resolved_at = NULL`).
@@ -952,3 +1149,236 @@ ALTER TABLE photos ADD COLUMN resolved_at INTEGER;
 CREATE INDEX IF NOT EXISTS idx_photos_unresolved ON photos(resolved_at);
 UPDATE photos SET resolved_at = strftime('%s','now') WHERE name IS NOT NULL;
 ";
+
+/// V32: give videos written off as un-thumbnailable another chance.
+///
+/// Before local video thumbnails, the fallback fed a video's bytes to the image
+/// decoder, which could only fail, and that failure was recorded as the
+/// permanent "no thumbnail" verdict. Run once per table in `{table}`.
+const MIGRATION_V32: &str = "UPDATE {table} SET thumb_state = 0 WHERE kind = 1 AND thumb_state = 2";
+
+/// Schema v33: where each photo was taken.
+///
+/// `latitude` and `longitude` come from the location other clients write into
+/// a photo's extended attributes; `place_id` is the GeoNames id of the nearest
+/// town (see `crate::places`), so the Places view is a plain `GROUP BY`.
+///
+/// Every photo is marked unresolved, so the next refresh reads the location of
+/// the whole library once. That is the backfill: the metadata is on the node
+/// the refresh reads anyway, and nothing has to be downloaded.
+const MIGRATION_V33: &str = "
+ALTER TABLE photos ADD COLUMN latitude REAL;
+ALTER TABLE photos ADD COLUMN longitude REAL;
+ALTER TABLE photos ADD COLUMN place_id INTEGER;
+CREATE INDEX IF NOT EXISTS idx_photos_place ON photos(place_id);
+UPDATE photos SET resolved_at = NULL;
+";
+
+/// Schema v34: what each photo looks like, for finding similar ones.
+///
+/// `similar_hash` is a 64-bit difference hash of the photo's thumbnail (see
+/// `crate::similar`): resized and re-encoded copies of one picture land a few
+/// bits apart, where the server's `content_hash` only matches byte-identical
+/// files. `NULL` until the daemon has hashed the thumbnail, which it does as
+/// thumbnails arrive and when the similar-photo finder asks for the rest.
+const MIGRATION_V34: &str = "
+ALTER TABLE photos ADD COLUMN similar_hash INTEGER;
+";
+
+/// Schema v35: when each trashed node went to the trash.
+///
+/// The API does not tell this client, so `trashed_at` is when the daemon first
+/// saw the node in the trash, in epoch seconds. `NULL` when it cannot say: rows
+/// from before this version, and nodes found by a refresh that did not follow a
+/// complete one, which may have been in the trash for a long time already.
+const MIGRATION_V35: &str = "
+ALTER TABLE trash ADD COLUMN trashed_at INTEGER;
+";
+
+/// Schema v36: every node row has a local id (`lid`); see
+/// docs/ARCHITECTURE.md §2.
+///
+/// The `uid` was the row's key, so a node had no identity until Drive minted
+/// one, and a create that landed moved its row to a new key. The `lid` is given
+/// when the row is first written and never changes or comes back, which is what
+/// `AUTOINCREMENT` adds over a plain rowid. It takes the old rowid, so the
+/// search index, keyed by rowid since v15, stays valid. The `uid` becomes
+/// nullable for a node that is not on Drive yet.
+const MIGRATION_V36: &str = "
+CREATE TABLE nodes_v36 (
+  lid           INTEGER PRIMARY KEY AUTOINCREMENT,
+  uid           TEXT UNIQUE,
+  parent_uid    TEXT,
+  name          TEXT NOT NULL,
+  is_dir        INTEGER NOT NULL,
+  size          INTEGER,
+  mtime         INTEGER NOT NULL,
+  revision_hash TEXT,
+  trashed       INTEGER NOT NULL DEFAULT 0,
+  node_json     TEXT,
+  listed        INTEGER NOT NULL DEFAULT 0,
+  path          TEXT
+);
+INSERT INTO nodes_v36
+  (lid, uid, parent_uid, name, is_dir, size, mtime, revision_hash, trashed, node_json, listed,
+   path)
+  SELECT rowid, uid, parent_uid, name, is_dir, size, mtime, revision_hash, trashed, node_json,
+         listed, path
+    FROM nodes;
+DROP TABLE nodes;
+ALTER TABLE nodes_v36 RENAME TO nodes;
+CREATE INDEX idx_nodes_parent ON nodes(parent_uid);
+CREATE INDEX idx_nodes_name_nocase ON nodes(name COLLATE NOCASE);
+CREATE INDEX idx_nodes_parent_live ON nodes(parent_uid) WHERE trashed = 0;
+";
+
+/// Schema v37: a `pending_op` id is never given out twice.
+///
+/// A plain rowid is the largest one plus one, so an op queued right after the
+/// newest one was dropped took its id. The drain retires the op it is
+/// working on by id, and so deleted the new op: a trash queued over a rename
+/// on the wire was lost when the rename landed. `AUTOINCREMENT` never reuses
+/// an id. Ids are kept, as is every index.
+const MIGRATION_V37: &str = "
+CREATE TABLE pending_op_v37 (
+  id                    INTEGER PRIMARY KEY AUTOINCREMENT,
+  kind                  TEXT NOT NULL,
+  uid                   TEXT NOT NULL,
+  blob_path             TEXT,
+  meta_json             TEXT,
+  created_at            INTEGER NOT NULL,
+  attempts              INTEGER NOT NULL DEFAULT 0,
+  last_error            TEXT,
+  next_attempt_at       INTEGER NOT NULL DEFAULT 0,
+  parent_uid            TEXT,
+  name                  TEXT,
+  claimed_at            INTEGER NOT NULL DEFAULT 0,
+  access_deferred_since INTEGER NOT NULL DEFAULT 0
+);
+INSERT INTO pending_op_v37
+  (id, kind, uid, blob_path, meta_json, created_at, attempts, last_error, next_attempt_at,
+   parent_uid, name, claimed_at, access_deferred_since)
+  SELECT id, kind, uid, blob_path, meta_json, created_at, attempts, last_error,
+         next_attempt_at, parent_uid, name, claimed_at, access_deferred_since
+    FROM pending_op;
+DROP TABLE pending_op;
+ALTER TABLE pending_op_v37 RENAME TO pending_op;
+CREATE INDEX pending_op_uid ON pending_op(uid);
+CREATE INDEX pending_op_parent ON pending_op(parent_uid);
+CREATE INDEX idx_pending_op_claimed
+    ON pending_op(uid) WHERE claimed_at <> 0;
+";
+
+/// Schema v38: why Drive refused a queued op, when it did (a sync issue).
+/// `NULL` for an op that has not been refused, which is every op migrated.
+const MIGRATION_V38: &str = "
+ALTER TABLE pending_op ADD COLUMN issue TEXT;
+";
+
+/// Schema v39: a queued op names its node and its parent by local id too.
+///
+/// A create made inside a folder that had not landed named the folder by its
+/// placeholder uid, and every landing had to find and rewrite those strings.
+/// The local id stays the same when the folder lands. An op whose node has no
+/// row keeps `NULL` and is matched by uid, as before.
+const MIGRATION_V39: &str = "
+ALTER TABLE pending_op ADD COLUMN lid INTEGER;
+ALTER TABLE pending_op ADD COLUMN parent_lid INTEGER;
+UPDATE pending_op
+   SET lid = (SELECT lid FROM nodes WHERE nodes.uid = pending_op.uid),
+       parent_lid = (SELECT lid FROM nodes WHERE nodes.uid = pending_op.parent_uid);
+CREATE INDEX pending_op_lid ON pending_op(lid);
+CREATE INDEX pending_op_parent_lid ON pending_op(parent_lid);
+";
+
+/// Schema v40: a node row names its parent by local id too.
+///
+/// The local id stays when a folder made here lands and its row takes the uid
+/// Drive gave it, so a child linked by it needs no rewrite. `NULL` while the
+/// parent has no row, as `parent_uid` then names nothing either.
+const MIGRATION_V40: &str = "
+ALTER TABLE nodes ADD COLUMN parent_lid INTEGER;
+UPDATE nodes SET parent_lid = (SELECT p.lid FROM nodes p WHERE p.uid = nodes.parent_uid);
+CREATE INDEX idx_nodes_parent_lid ON nodes(parent_lid);
+";
+
+/// Schema v41: a node made on this machine goes by `local~<lid>` until it
+/// lands, a function of its row, where it used to get a `local~<ms>-<n>`
+/// placeholder. Every placeholder still stored is rewritten to its row's
+/// stand-in: the row itself and its `node_json`, its children, the ops on it
+/// or into it, and its pins. A placeholder op with no row keeps its uid and
+/// drains as before.
+const MIGRATION_V41: &str = "
+CREATE TEMP TABLE local_v41 AS
+  SELECT uid AS old, 'local~' || lid AS new, lid FROM nodes
+   WHERE uid LIKE 'local~%' AND uid <> 'local~' || lid;
+UPDATE pending_op SET uid = (SELECT new FROM local_v41 WHERE old = pending_op.uid)
+ WHERE uid IN (SELECT old FROM local_v41);
+UPDATE pending_op SET parent_uid = (SELECT new FROM local_v41 WHERE old = pending_op.parent_uid)
+ WHERE parent_uid IN (SELECT old FROM local_v41);
+UPDATE pending_op
+   SET meta_json = json_set(meta_json, '$.uid',
+         (SELECT new FROM local_v41 WHERE old = json_extract(pending_op.meta_json, '$.uid')))
+ WHERE json_valid(meta_json)
+   AND json_extract(meta_json, '$.uid') IN (SELECT old FROM local_v41);
+UPDATE pending_op
+   SET meta_json = json_set(meta_json, '$.original_parent_uid',
+         (SELECT new FROM local_v41
+           WHERE old = json_extract(pending_op.meta_json, '$.original_parent_uid')))
+ WHERE json_valid(meta_json)
+   AND json_extract(meta_json, '$.original_parent_uid') IN (SELECT old FROM local_v41);
+UPDATE pins SET uid = (SELECT new FROM local_v41 WHERE old = pins.uid)
+ WHERE uid IN (SELECT old FROM local_v41);
+UPDATE nodes
+   SET node_json = json_set(node_json, '$.parent_uid.link_id',
+         (SELECT CAST(lid AS TEXT) FROM local_v41 WHERE old = nodes.parent_uid))
+ WHERE parent_uid IN (SELECT old FROM local_v41)
+   AND json_valid(node_json)
+   AND json_extract(node_json, '$.parent_uid.volume_id') = 'local';
+UPDATE nodes SET parent_uid = (SELECT new FROM local_v41 WHERE old = nodes.parent_uid)
+ WHERE parent_uid IN (SELECT old FROM local_v41);
+UPDATE nodes
+   SET node_json = json_set(node_json, '$.uid.link_id', CAST(lid AS TEXT))
+ WHERE lid IN (SELECT lid FROM local_v41)
+   AND json_valid(node_json)
+   AND json_extract(node_json, '$.uid.volume_id') = 'local';
+UPDATE nodes SET uid = 'local~' || lid WHERE lid IN (SELECT lid FROM local_v41);
+DROP TABLE local_v41;
+";
+
+/// Schema v42: the SHA-1 of each blob a queued create sent, space-separated.
+/// A create whose answer was lost may have made its file with any of them,
+/// and the op that finds that file again may run after a restart. `NULL` for
+/// an op that has sent nothing, which is every op migrated: what the daemon
+/// that wrote it sent is not known. Likewise the parent and name each attempt
+/// was sent under, as a JSON list of pairs: a rename since then moves the op
+/// but not the file Drive may have made.
+const MIGRATION_V42: &str = "
+ALTER TABLE pending_op ADD COLUMN sent_sha1 TEXT;
+ALTER TABLE pending_op ADD COLUMN sent_to TEXT;
+";
+
+/// Schema v43: queued ops by kind and name.
+///
+/// Before each create, rename and trash the drain asks whether a queued trash
+/// or rename holds the name it wants, and a released name wakes the ops
+/// waiting for it. Each of those scanned the whole queue, so a large offline
+/// batch made draining it quadratic. The kind leads, so a lookup of every
+/// queued rename uses the index too.
+const MIGRATION_V43: &str = "
+CREATE INDEX IF NOT EXISTS pending_op_kind_name ON pending_op(kind, name);
+";
+
+/// Schema v44: a synced file's baseline remembers the revision it was synced
+/// at.
+///
+/// A mirrored file edited here is uploaded as a new revision of whatever is
+/// current on Drive, and another computer syncing the same folder may have
+/// committed one since this pass looked. The revision id is what the upload
+/// checks first, so that edit becomes a conflict copy instead of being
+/// replaced (B249). `NULL` means "this row predates the column", and the check
+/// falls back to the modification time and size; each row gains a real value
+/// the next time its path is synced.
+const MIGRATION_V44: &str = "
+ALTER TABLE sync_entry ADD COLUMN remote_revision_id TEXT;
+";

crates/pdfs-core/src/db/mod.rs

@@ -56,10 +56,11 @@ pub use local::LocalFileHit;
 pub use maintenance::{DbStats, VacuumOutcome};
 pub use nodes::{PublishedSharedRoot, SearchHit, StoredNode};
 pub use ops::{
-    AttachedBlob, LOCAL_VOLUME, OP_CREATE, OP_MKDIR, OP_RENAME, OP_REVISION, OP_TRASH, PARK_UNTIL,
-    PendingCounts, PendingOp, RenameMeta, op_supersedes,
+    AttachedBlob, CreateLanding, CreateRetired, FAILING_ATTEMPTS, LOCAL_VOLUME, OP_CREATE,
+    OP_MKDIR, OP_RENAME, OP_REVISION, OP_TRASH, PARK_EXPIRY_MS, PARK_UNTIL, PendingCounts,
+    PendingOp, RenameMeta, local_lid, local_uid, op_supersedes,
 };
-pub use photos::{StoredPhoto, THUMB_HAVE, THUMB_NONE, THUMB_UNKNOWN, TimelineRow};
+pub use photos::{StoredPhoto, StoredPlace, THUMB_HAVE, THUMB_NONE, THUMB_UNKNOWN, TimelineRow};
 pub use pins::PinRow;
 pub use sync::{StoredSyncEntry, StoredSyncFolder};
 pub use trash::StoredTrash;
@@ -178,6 +179,45 @@ impl Drop for Reader<'_> {
     }
 }
 
+/// The write connection with `synchronous=FULL`, back to NORMAL on drop.
+///
+/// NORMAL in WAL mode keeps a commit across a crash of the process but not
+/// across a power loss: the WAL is synced only at checkpoints. That is fine for
+/// metadata the next listing re-fetches, and not for the queue. A staged blob
+/// is synced before its op is queued, and the op row is what makes it
+/// reachable; losing that row loses a write `close()` already reported as
+/// done. A FULL commit syncs the WAL, which covers every NORMAL commit before
+/// it as well.
+pub(super) struct Durable<'a>(parking_lot::MutexGuard<'a, Connection>);
+
+impl<'a> Durable<'a> {
+    pub(super) fn new(conn: parking_lot::MutexGuard<'a, Connection>) -> Result<Self> {
+        conn.pragma_update(None, "synchronous", "FULL")?;
+        Ok(Self(conn))
+    }
+}
+
+impl std::ops::Deref for Durable<'_> {
+    type Target = Connection;
+
+    fn deref(&self) -> &Connection {
+        &self.0
+    }
+}
+
+impl std::ops::DerefMut for Durable<'_> {
+    fn deref_mut(&mut self) -> &mut Connection {
+        &mut self.0
+    }
+}
+
+impl Drop for Durable<'_> {
+    fn drop(&mut self) {
+        // A failure leaves the writer at FULL: slower, never less safe.
+        let _ = self.0.pragma_update(None, "synchronous", "NORMAL");
+    }
+}
+
 /// Open a second connection to the same file for reads only.
 ///
 /// `SQLITE_OPEN_READ_ONLY` plus `query_only` is belt and braces: the flag stops
@@ -215,18 +255,34 @@ impl Db {
         let conn = match Self::open_configured(path) {
             Ok(conn) => conn,
             Err(e) if is_corrupt(&e) => {
-                // Every byte in here is derived from the account or from the
-                // content cache directory, so the recovery is to start over —
-                // loudly, and keeping the damaged file for a post-mortem rather
-                // than deleting evidence.
+                // Most of what is in here is derived from the account and is
+                // fetched again. The queue is not: it holds writes accepted
+                // here that Drive does not have yet, and the mirror baselines
+                // and pins are this machine's alone. So the damaged file is
+                // moved aside, kept as evidence, and those tables are copied
+                // out of it into the new one as far as SQLite can still read
+                // them (`docs/BUGS.md` B216).
                 tracing::error!(
                     db = %path.display(),
                     error = %e,
-                    "cache database is corrupt — moving it aside and rebuilding \
-                     (metadata will be re-fetched; no user data is stored here)"
+                    "cache database is corrupt; moving it aside, rebuilding, and \
+                     copying the upload queue, sync baselines and pins out of it"
                 );
-                quarantine_corrupt_db(path)?;
-                Self::open_configured(path)?
+                let damaged = quarantine_corrupt_db(path)?;
+                let conn = Self::open_configured(path)?;
+                for (table, result) in salvage(&conn, &damaged) {
+                    match result {
+                        Ok(rows) => {
+                            tracing::warn!(table, rows, "salvaged from the damaged database")
+                        }
+                        Err(error) => tracing::error!(
+                            table,
+                            %error,
+                            "could not salvage from the damaged database"
+                        ),
+                    }
+                }
+                conn
             }
             Err(e) => return Err(e),
         };
@@ -253,6 +309,12 @@ impl Db {
         Ok(db)
     }
 
+    /// The write connection for a commit that accepts a user's change into
+    /// the queue. See [`Durable`].
+    pub(super) fn durable(&self) -> Result<Durable<'_>> {
+        Durable::new(self.conn.lock())
+    }
+
     /// A connection for a `SELECT`-only method.
     ///
     /// Reads see the last committed state, which is the same thing they saw
@@ -280,7 +342,8 @@ impl Db {
     fn open_configured(path: &Path) -> Result<Connection> {
         let conn = Connection::open(path)?;
         // WAL: readers never block the single writer. NORMAL sync is the
-        // standard durability/throughput tradeoff for WAL.
+        // standard durability/throughput tradeoff for WAL; the queue commits
+        // that need more raise it to FULL for themselves (`Durable`).
         conn.pragma_update(None, "journal_mode", "WAL")?;
         conn.pragma_update(None, "synchronous", "NORMAL")?;
         conn.pragma_update(None, "foreign_keys", "ON")?;
@@ -427,12 +490,96 @@ fn is_corrupt(e: &Error) -> bool {
     )
 }
 
-/// Move a damaged database (and its WAL sidecars) out of the way so the next
-/// open starts from empty.
+/// The tables [`salvage`] copies out of a damaged database, in order, each
+/// with the rows it keeps.
 ///
-/// Renamed rather than deleted: this is a cache, but it is also the only
-/// evidence of what went wrong, and it costs one file the user can remove.
-fn quarantine_corrupt_db(path: &Path) -> Result<()> {
+/// The queue goes with the node rows it names by local id: a create made here
+/// has no other record of its node, and a real node keeps its local id when its
+/// folder is listed again. Those rows come back unlisted, so their folders are
+/// fetched again rather than trusted. Claims are cleared by [`Db::open`] as
+/// after any other crash.
+const SALVAGE: &[(&str, &str)] = &[
+    (
+        "nodes",
+        "lid IN (SELECT lid FROM old.pending_op WHERE lid IS NOT NULL
+                 UNION SELECT parent_lid FROM old.pending_op WHERE parent_lid IS NOT NULL)
+         OR uid LIKE 'local~%'",
+    ),
+    ("pending_op", "1"),
+    ("sync_folder", "1"),
+    ("sync_entry", "1"),
+    ("pins", "1"),
+    ("own_sealed_rev", "1"),
+];
+
+/// Copy what only this machine has out of the damaged database at `damaged`
+/// into the fresh one `conn` just created. Returns how each table went.
+///
+/// Each table is copied in its own transaction, so a table SQLite cannot read
+/// to the end is left out whole rather than half copied, and does not stop
+/// the others. Only the columns both files have are copied, which covers a
+/// file that broke in the middle of an upgrade. The damaged file is opened read
+/// only and left as it was.
+fn salvage(conn: &Connection, damaged: &Path) -> Vec<(&'static str, Result<usize>)> {
+    let uri = format!(
+        "file:{}?mode=ro",
+        damaged
+            .to_string_lossy()
+            .replace('%', "%25")
+            .replace('?', "%3f")
+            .replace('#', "%23")
+    );
+    if let Err(e) = conn.execute("ATTACH DATABASE ?1 AS old", [&uri]) {
+        let e = Error::Db(e);
+        tracing::error!(error = %e, "could not open the damaged database to salvage it");
+        return Vec::new();
+    }
+    let results = SALVAGE
+        .iter()
+        .map(|&(table, rows)| (table, salvage_table(conn, table, rows)))
+        .collect();
+    let _ = conn.execute("UPDATE nodes SET listed = 0", []);
+    if let Err(e) = conn.execute("DETACH DATABASE old", []) {
+        tracing::warn!(error = %e, "could not detach the damaged database");
+    }
+    results
+}
+
+fn salvage_table(conn: &Connection, table: &str, rows: &str) -> Result<usize> {
+    let columns = |schema: &str| -> Result<Vec<String>> {
+        let mut stmt = conn.prepare(&format!("PRAGMA {schema}.table_info({table})"))?;
+        let names = stmt
+            .query_map([], |r| r.get::<_, String>(1))?
+            .collect::<rusqlite::Result<Vec<_>>>()?;
+        Ok(names)
+    };
+    let new = columns("main")?;
+    let shared: Vec<String> = columns("old")?
+        .into_iter()
+        .filter(|c| new.contains(c))
+        .map(|c| format!("\"{c}\""))
+        .collect();
+    if shared.is_empty() {
+        return Ok(0);
+    }
+    let shared = shared.join(", ");
+    let tx = conn.unchecked_transaction()?;
+    let copied = tx.execute(
+        &format!(
+            "INSERT OR IGNORE INTO main.{table} ({shared}) SELECT {shared} FROM old.{table} WHERE {rows}"
+        ),
+        [],
+    )?;
+    tx.commit()?;
+    Ok(copied)
+}
+
+/// Move a damaged database (and its WAL sidecars) out of the way so the next
+/// open starts from empty. Returns where the database itself went.
+///
+/// Renamed rather than deleted: it is the only evidence of what went wrong,
+/// and [`salvage`] reads what only this machine has back out of it.
+fn quarantine_corrupt_db(path: &Path) -> Result<PathBuf> {
     let stamp = std::time::SystemTime::now()
         .duration_since(std::time::UNIX_EPOCH)
         .map(|d| d.as_secs())
@@ -456,7 +603,9 @@ fn quarantine_corrupt_db(path: &Path) -> Result<()> {
             Err(e) => tracing::warn!(file = %from.display(), error = %e, "could not quarantine"),
         }
     }
-    Ok(())
+    let mut kept = path.as_os_str().to_owned();
+    kept.push(format!(".corrupt-{stamp}"));
+    Ok(PathBuf::from(kept))
 }
 
 #[cfg(test)]

crates/pdfs-core/src/db/mounts.rs

@@ -171,6 +171,7 @@ impl Db {
                 pending_mode: pending_mode.as_deref().map(MountMode::from),
                 mounted: false,
                 progress: None,
+                paused: false,
             });
         }
         Ok(mounts)

crates/pdfs-core/src/db/nodes.rs

@@ -3,9 +3,11 @@
 
 use std::collections::HashSet;
 
+use rusqlite::types::Value;
 use rusqlite::{OptionalExtension, Transaction, params};
 
 use super::Db;
+use super::ops::{OP_CREATE, OP_MKDIR, OP_RENAME, local_lid, local_uid, wake_ops_waiting_for_tx};
 use crate::{Access, Result};
 use proton_drive_rs::proton_sdk::ids::{LinkId, NodeUid, VolumeId};
 use proton_drive_rs::{Node, NodeKind};
@@ -17,6 +19,8 @@ use super::utils::{
 pub struct StoredNode {
     pub node: Node,
     pub listed: bool,
+    /// The row's local id ([`Db::lid_of`]).
+    pub lid: i64,
 }
 
 /// One full-text search match: the stored [`Node`] plus its mountpoint-relative
@@ -62,12 +66,13 @@ impl Db {
     /// snapshot while offline, including after a live event expired its TTL.
     pub fn visible_children(&self, parent: &NodeUid) -> Result<Vec<Node>> {
         let conn = self.read();
-        let mut stmt = conn.prepare(
-            "SELECT node_json FROM nodes
-              WHERE parent_uid = ?1 AND trashed = 0 AND node_json IS NOT NULL
-              ORDER BY name, uid",
-        )?;
-        let rows = stmt.query_map([parent.to_string()], |row| row.get::<_, String>(0))?;
+        let (key, parent) = parent_key(&conn, &parent.to_string())?;
+        let mut stmt = conn.prepare(&format!(
+            "SELECT {NODE_NOW} FROM nodes n LEFT JOIN nodes p ON p.lid = n.parent_lid
+              WHERE n.{key} = ?1 AND n.trashed = 0 AND n.node_json IS NOT NULL
+              ORDER BY n.name, n.uid"
+        ))?;
+        let rows = stmt.query_map([parent], |row| row.get::<_, String>(0))?;
         let mut nodes = Vec::new();
         for row in rows {
             nodes.push(serde_json::from_str(&row?)?);
@@ -264,10 +269,25 @@ impl Db {
     /// resolve its shared-tree access. Removing the op first could resurrect
     /// that row after a crash; removing the row first could strand a now
     /// unauthorizable op. One transaction closes both windows.
-    pub fn complete_trash_op(&self, op_id: i64, uid: &NodeUid) -> Result<()> {
+    ///
+    /// An op that found the name still held by this trash failed with a
+    /// backoff, and the name is free now. It is made due at once
+    /// ([`Db::wake_ops_waiting_for`]).
+    ///
+    /// Returns the staged blob of the create the trash withdrew, if it kept
+    /// one, for the caller to discard.
+    pub fn complete_trash_op(&self, op_id: i64, uid: &NodeUid) -> Result<Option<String>> {
         let uid = uid.to_string();
         let mut conn = self.conn.lock();
         let tx = conn.transaction()?;
+        let (name, blob): (Option<String>, Option<String>) = tx
+            .query_row(
+                "SELECT name, blob_path FROM pending_op WHERE id = ?1",
+                params![op_id],
+                |row| Ok((row.get(0)?, row.get(1)?)),
+            )
+            .optional()?
+            .unwrap_or_default();
         let rowid: Option<i64> = tx
             .query_row(
                 "SELECT rowid FROM nodes WHERE uid = ?1",
@@ -280,17 +300,23 @@ impl Db {
             tx.execute("DELETE FROM nodes_fts WHERE rowid = ?1", params![rowid])?;
         }
         tx.execute("DELETE FROM pending_op WHERE id = ?1", params![op_id])?;
+        if let Some(name) = name {
+            wake_ops_waiting_for_tx(&tx, &name)?;
+        }
         tx.commit()?;
-        Ok(())
+        Ok(blob)
     }
 
     /// Check if a folder node has any non-trashed children in the database.
     pub fn has_children(&self, parent_uid: &NodeUid) -> Result<bool> {
-        let uid_str = parent_uid.to_string();
         let conn = self.read();
+        let (key, parent) = parent_key(&conn, &parent_uid.to_string())?;
         let count: i64 = conn.query_row(
-            "SELECT COUNT(*) FROM nodes WHERE parent_uid = ?1 AND trashed = 0",
-            params![uid_str],
+            &format!(
+                "SELECT COUNT(*) FROM nodes
+                 WHERE {key} = ?1 AND trashed = 0 AND node_json IS NOT NULL"
+            ),
+            params![parent],
             |row| row.get(0),
         )?;
         Ok(count > 0)
@@ -305,9 +331,11 @@ impl Db {
     pub fn node_path(&self, uid: &str) -> Result<Option<String>> {
         let conn = self.read();
         let exists: Option<i64> = conn
-            .query_row("SELECT 1 FROM nodes WHERE uid = ?1", params![uid], |r| {
-                r.get(0)
-            })
+            .query_row(
+                "SELECT 1 FROM nodes WHERE uid = ?1 AND node_json IS NOT NULL",
+                params![uid],
+                |r| r.get(0),
+            )
             .optional()?;
         if exists.is_none() {
             return Ok(None);
@@ -325,11 +353,11 @@ impl Db {
     pub fn path_relative_to(&self, ancestor_uid: &str, uid: &str) -> Result<Option<String>> {
         let conn = self.read();
         let mut stmt = conn.prepare(
-            "WITH RECURSIVE anc(uid, parent_uid, name, depth) AS (
-               SELECT uid, parent_uid, name, 0 FROM nodes WHERE uid = ?1
+            "WITH RECURSIVE anc(uid, parent_lid, name, depth) AS (
+               SELECT uid, parent_lid, name, 0 FROM nodes WHERE uid = ?1
                UNION ALL
-               SELECT n.uid, n.parent_uid, n.name, anc.depth + 1
-               FROM nodes n JOIN anc ON n.uid = anc.parent_uid
+               SELECT n.uid, n.parent_lid, n.name, anc.depth + 1
+               FROM nodes n JOIN anc ON n.lid = anc.parent_lid
                WHERE anc.depth < 1024
              )
              SELECT uid, name FROM anc ORDER BY depth",
@@ -355,6 +383,22 @@ impl Db {
     /// than [`TRIGRAM_MIN`] fall back to a `LIKE` scan since trigram indexes
     /// nothing below 3 chars.
     pub fn search(&self, query: &str, limit: usize) -> Result<Vec<SearchHit>> {
+        self.search_in(query, limit, None)
+    }
+
+    /// [`search`](Self::search) restricted to the subtree under the
+    /// mountpoint-relative folder `scope`. `None` or an empty scope searches
+    /// everything.
+    pub fn search_in(
+        &self,
+        query: &str,
+        limit: usize,
+        scope: Option<&str>,
+    ) -> Result<Vec<SearchHit>> {
+        let scope = scope.map(|s| s.trim_matches('/')).filter(|s| !s.is_empty());
+        // Rows whose path is not persisted yet pass the SQL filter and are
+        // checked once their path is walked below.
+        let scope_pat = scope.map(|s| format!("{}/%", like_escape(s)));
         let query = query.trim();
         if query.is_empty() {
             return Ok(Vec::new());
@@ -362,12 +406,14 @@ impl Db {
         let conn = self.read();
         let rows: Vec<HitRow> = if query.chars().count() < TRIGRAM_MIN {
             let pat = format!("%{}%", like_escape(query));
-            let mut stmt = conn.prepare(
-                "SELECT node_json, uid, path FROM nodes
-                 WHERE name LIKE ?1 ESCAPE '\\' AND trashed = 0 AND node_json IS NOT NULL
-                 ORDER BY name LIMIT ?2",
-            )?;
-            collect_hits(stmt.query_map(params![pat, limit as i64], hit_row)?)?
+            let mut stmt = conn.prepare(&format!(
+                "SELECT {NODE_NOW}, n.uid, n.path
+                 FROM nodes n LEFT JOIN nodes p ON p.lid = n.parent_lid
+                 WHERE n.name LIKE ?1 ESCAPE '\\' AND n.trashed = 0 AND n.node_json IS NOT NULL
+                   AND (?3 IS NULL OR n.path IS NULL OR n.path LIKE ?3 ESCAPE '\\')
+                 ORDER BY n.name LIMIT ?2"
+            ))?;
+            collect_hits(stmt.query_map(params![pat, limit as i64, scope_pat], hit_row)?)?
         } else {
             // Escape double quotes and quote each term, then combine with AND so
             // all terms must match but can appear in any order or position.
@@ -376,13 +422,15 @@ impl Db {
                 .map(|word| format!("\"{}\"", word.replace('"', "\"\"")))
                 .collect::<Vec<_>>()
                 .join(" AND ");
-            let mut stmt = conn.prepare(
-                "SELECT n.node_json, n.uid, n.path
+            let mut stmt = conn.prepare(&format!(
+                "SELECT {NODE_NOW}, n.uid, n.path
                  FROM nodes_fts f JOIN nodes n ON n.rowid = f.rowid
+                 LEFT JOIN nodes p ON p.lid = n.parent_lid
                  WHERE f.name MATCH ?1 AND n.trashed = 0 AND n.node_json IS NOT NULL
-                 ORDER BY f.rank LIMIT ?2",
-            )?;
-            collect_hits(stmt.query_map(params![phrase, limit as i64], hit_row)?)?
+                   AND (?3 IS NULL OR n.path IS NULL OR n.path LIKE ?3 ESCAPE '\\')
+                 ORDER BY f.rank LIMIT ?2"
+            ))?;
+            collect_hits(stmt.query_map(params![phrase, limit as i64, scope_pat], hit_row)?)?
         };
 
         let mut hits = Vec::with_capacity(rows.len());
@@ -392,6 +440,13 @@ impl Db {
                 Some(path) => path,
                 None => walk_path_of(&conn, &uid)?,
             };
+            if let Some(scope) = scope
+                && !path
+                    .strip_prefix(scope)
+                    .is_some_and(|rest| rest.starts_with('/'))
+            {
+                continue;
+            }
             hits.push(SearchHit { node, path });
         }
         Ok(hits)
@@ -411,20 +466,22 @@ impl Db {
         let lane_limit = limit.div_ceil(2);
         let mut rows: Vec<HitRow> = if terms.is_empty() {
             let pat = format!("%{}%", like_escape(query));
-            let mut stmt = conn.prepare(
-                "SELECT node_json, uid, path FROM nodes
-                 WHERE name LIKE ?1 ESCAPE '\\' AND trashed = 0 AND node_json IS NOT NULL
-                 ORDER BY name LIMIT ?2",
-            )?;
+            let mut stmt = conn.prepare(&format!(
+                "SELECT {NODE_NOW}, n.uid, n.path
+                 FROM nodes n LEFT JOIN nodes p ON p.lid = n.parent_lid
+                 WHERE n.name LIKE ?1 ESCAPE '\\' AND n.trashed = 0 AND n.node_json IS NOT NULL
+                 ORDER BY n.name LIMIT ?2"
+            ))?;
             collect_hits(stmt.query_map(params![pat, limit as i64], hit_row)?)?
         } else {
             let expression = terms.join(" OR ");
-            let mut stmt = conn.prepare(
-                "SELECT n.node_json, n.uid, n.path
+            let mut stmt = conn.prepare(&format!(
+                "SELECT {NODE_NOW}, n.uid, n.path
                    FROM nodes_fts f JOIN nodes n ON n.rowid = f.rowid
+                   LEFT JOIN nodes p ON p.lid = n.parent_lid
                   WHERE nodes_fts MATCH ?1 AND n.trashed = 0 AND n.node_json IS NOT NULL
-                  ORDER BY f.rank LIMIT ?2",
-            )?;
+                  ORDER BY f.rank LIMIT ?2"
+            ))?;
             collect_hits(stmt.query_map(params![expression, lane_limit as i64], hit_row)?)?
         };
         // Trigram rank does not know that a name *starts* with the query, and a
@@ -434,12 +491,13 @@ impl Db {
         if !terms.is_empty()
             && let Some(first) = query.chars().next()
         {
-            let mut stmt = conn.prepare(
-                "SELECT node_json, uid, path FROM nodes
-                 WHERE name COLLATE NOCASE LIKE ?1 ESCAPE '\\'
-                   AND trashed = 0 AND node_json IS NOT NULL
-                 ORDER BY name COLLATE NOCASE LIMIT ?2",
-            )?;
+            let mut stmt = conn.prepare(&format!(
+                "SELECT {NODE_NOW}, n.uid, n.path
+                 FROM nodes n LEFT JOIN nodes p ON p.lid = n.parent_lid
+                 WHERE n.name COLLATE NOCASE LIKE ?1 ESCAPE '\\'
+                   AND n.trashed = 0 AND n.node_json IS NOT NULL
+                 ORDER BY n.name COLLATE NOCASE LIMIT ?2"
+            ))?;
             let mut extra = Vec::new();
             for pattern in [
                 format!("{}%", like_escape(query)),
@@ -475,40 +533,50 @@ impl Db {
     /// re-enumerates from the remote on next access.
     pub fn set_listed(&self, uid: &NodeUid, listed: bool) -> Result<()> {
         let conn = self.conn.lock();
+        let (key, uid) = row_key(&uid.to_string());
         conn.execute(
-            "UPDATE nodes SET listed = ?2 WHERE uid = ?1",
-            params![uid.to_string(), listed as i64],
+            &format!("UPDATE nodes SET listed = ?2 WHERE {key} = ?1"),
+            params![uid, listed as i64],
         )?;
         Ok(())
     }
 
     /// Load every persisted node for cold-start hydration of the `State` maps.
+    ///
+    /// Each node names its parent as the parent's row has it now, found by
+    /// local id: a folder that landed has its new uid there, while the node
+    /// stored below it may still hold the placeholder.
     pub fn load_all(&self) -> Result<Vec<StoredNode>> {
         let conn = self.read();
-        let mut stmt =
-            conn.prepare("SELECT node_json, listed FROM nodes WHERE node_json IS NOT NULL")?;
+        let mut stmt = conn.prepare(&format!(
+            "SELECT {NODE_NOW}, n.listed, n.lid
+             FROM nodes n LEFT JOIN nodes p ON p.lid = n.parent_lid
+             WHERE n.node_json IS NOT NULL"
+        ))?;
         let rows = stmt.query_map([], |row| {
             let json: String = row.get(0)?;
             let listed: i64 = row.get(1)?;
-            Ok((json, listed != 0))
+            Ok((json, listed != 0, row.get(2)?))
         })?;
         let mut out = Vec::new();
         for row in rows {
-            let (json, listed) = row?;
+            let (json, listed, lid) = row?;
             let node: Node = serde_json::from_str(&json)?;
-            out.push(StoredNode { node, listed });
+            out.push(StoredNode { node, listed, lid });
         }
         Ok(out)
     }
 
     /// Load one persisted node back by uid. Used to recover the My Files root
-    /// when the API is unreachable, so the mount can still serve the cached tree
-    /// (offline.md Phase 1).
+    /// when the API is unreachable, so the mount can still serve the cached tree.
     pub fn node_by_uid(&self, uid: &str) -> Result<Option<Node>> {
         let conn = self.read();
         let json: Option<String> = conn
             .query_row(
-                "SELECT node_json FROM nodes WHERE uid = ?1 AND node_json IS NOT NULL",
+                &format!(
+                    "SELECT {NODE_NOW} FROM nodes n LEFT JOIN nodes p ON p.lid = n.parent_lid
+                     WHERE n.uid = ?1 AND n.node_json IS NOT NULL"
+                ),
                 params![uid],
                 |r| r.get(0),
             )
@@ -519,31 +587,178 @@ impl Db {
         }
     }
 
+    /// The uid of the node with local id `lid`: `None` when it has no row, or
+    /// is not on Drive yet.
+    pub fn uid_of_lid(&self, lid: i64) -> Result<Option<String>> {
+        let conn = self.read();
+        Ok(conn
+            .query_row("SELECT uid FROM nodes WHERE lid = ?1", params![lid], |r| {
+                r.get(0)
+            })
+            .optional()?
+            .flatten())
+    }
+
+    /// The local id of the node stored under `uid`, or `None` when it has no
+    /// row. It stays the same for as long as the row exists, including when a
+    /// create lands and the row takes the uid Drive gave it: a node made on
+    /// this machine is found by its stand-in ([`local_uid`]) before and after.
+    pub fn lid_of(&self, uid: &str) -> Result<Option<i64>> {
+        node_lid(&self.read(), uid)
+    }
+
+    /// The local id of each node, in order, giving a row to each that has none.
+    ///
+    /// A new row is only a stub: it has no `node_json`, so no listing or search
+    /// returns it until the node is upserted, which the caller still owes. That
+    /// keeps this cheap enough to call with the inode lock held, which is
+    /// where an inode number is first needed.
+    pub fn lids_for(&self, nodes: &[&Node]) -> Result<Vec<i64>> {
+        let mut conn = self.conn.lock();
+        let tx = conn.transaction()?;
+        let mut lids = Vec::with_capacity(nodes.len());
+        {
+            let mut known = tx.prepare_cached("SELECT lid FROM nodes WHERE uid = ?1")?;
+            let mut stub = tx.prepare_cached(
+                "INSERT INTO nodes (uid, parent_uid, name, is_dir, mtime, trashed, parent_lid)
+                 VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7)",
+            )?;
+            for node in nodes {
+                let uid = node.uid.to_string();
+                let lid = match known.query_row(params![uid], |r| r.get(0)).optional()? {
+                    Some(lid) => lid,
+                    None => {
+                        let parent = node.parent_uid.as_ref().map(ToString::to_string);
+                        let parent_lid = match &parent {
+                            Some(parent) => node_lid(&tx, parent)?,
+                            None => None,
+                        };
+                        stub.execute(params![
+                            uid,
+                            parent,
+                            node.name,
+                            node.is_folder() as i64,
+                            node.modification_time,
+                            node.trashed as i64,
+                            parent_lid,
+                        ])?;
+                        let lid = tx.last_insert_rowid();
+                        adopt_children_tx(&tx, &uid, lid)?;
+                        lid
+                    }
+                };
+                lids.push(lid);
+            }
+        }
+        tx.commit()?;
+        Ok(lids)
+    }
+
+    /// Give a node being made on this machine its row under `parent_uid`, and
+    /// return its local id. The row goes by [`local_uid`] of that id until
+    /// the node lands.
+    ///
+    /// Like [`Db::lids_for`], only a stub: the caller upserts the node.
+    pub fn add_local_row(
+        &self,
+        parent_uid: &str,
+        name: &str,
+        is_dir: bool,
+        mtime: i64,
+    ) -> Result<i64> {
+        let mut conn = self.conn.lock();
+        let tx = conn.transaction()?;
+        tx.execute(
+            "INSERT INTO nodes (uid, parent_uid, name, is_dir, mtime, trashed, parent_lid)
+             VALUES (NULL, ?1, ?2, ?3, ?4, 0,
+                     COALESCE(?5, (SELECT lid FROM nodes WHERE uid = ?1)))",
+            params![
+                parent_uid,
+                name,
+                is_dir as i64,
+                mtime,
+                local_lid(parent_uid)
+            ],
+        )?;
+        let lid = tx.last_insert_rowid();
+        tx.execute(
+            "UPDATE nodes SET uid = ?2 WHERE lid = ?1",
+            params![lid, local_uid(lid)],
+        )?;
+        tx.commit()?;
+        Ok(lid)
+    }
+
     /// Read the persisted incremental-sync cursor (a `DriveEventId`), if any.
     /// The daemon resumes from this on restart instead of reseeding to the
     /// server head, so changes made while unmounted are still applied (P2).
     pub fn children_if_listed(&self, parent: &NodeUid) -> Result<Option<Vec<Node>>> {
         let conn = self.read();
+        let (key, folder) = row_key(&parent.to_string());
         let listed: Option<i64> = conn
             .query_row(
-                "SELECT listed FROM nodes WHERE uid = ?1",
-                params![parent.to_string()],
+                &format!("SELECT listed FROM nodes WHERE {key} = ?1"),
+                params![folder],
                 |r| r.get(0),
             )
             .optional()?;
         if listed != Some(1) {
             return Ok(None);
         }
-        let mut stmt = conn.prepare(
-            "SELECT node_json FROM nodes
-             WHERE parent_uid = ?1 AND node_json IS NOT NULL AND trashed = 0",
-        )?;
-        let rows = stmt.query_map(params![parent.to_string()], |r| r.get::<_, String>(0))?;
+        drop(conn);
+        self.known_children(parent).map(Some)
+    }
+
+    /// The children the DB knows under `parent`, whether or not the folder's
+    /// listing is still marked complete. Only for when nothing better can be
+    /// had: a listing dropped as stale may miss or keep a child.
+    pub fn known_children(&self, parent: &NodeUid) -> Result<Vec<Node>> {
+        let conn = self.read();
+        let (key, parent) = parent_key(&conn, &parent.to_string())?;
+        let mut stmt = conn.prepare(&format!(
+            "SELECT {NODE_NOW} FROM nodes n LEFT JOIN nodes p ON p.lid = n.parent_lid
+             WHERE n.{key} = ?1 AND n.node_json IS NOT NULL AND n.trashed = 0"
+        ))?;
+        let rows = stmt.query_map(params![parent], |r| r.get::<_, String>(0))?;
         let mut out = Vec::new();
         for json in rows {
             out.push(serde_json::from_str(&json?)?);
         }
-        Ok(Some(out))
+        Ok(out)
+    }
+
+    /// What a queued change does to `parent`'s listing, which Drive does not
+    /// show until it lands: the children made or moved in by a queued create,
+    /// mkdir or rename, as the DB holds them, and the nodes a queued rename
+    /// moves elsewhere.
+    ///
+    /// A listing read from Drive is laid over with these, or a folder listed
+    /// while a create in it is queued, after a restart say, loses the file
+    /// until the create lands (`docs/BUGS.md` B132).
+    pub fn queued_children(&self, parent: &NodeUid) -> Result<(Vec<Node>, HashSet<String>)> {
+        let conn = self.read();
+        let (key, parent) = parent_key(&conn, &parent.to_string())?;
+        let mut stmt = conn.prepare(&format!(
+            "SELECT {NODE_NOW} FROM nodes n LEFT JOIN nodes p ON p.lid = n.parent_lid
+             WHERE n.{key} = ?1 AND n.node_json IS NOT NULL AND n.trashed = 0
+               AND EXISTS (SELECT 1 FROM pending_op o
+                           WHERE (o.lid = n.lid OR o.uid = n.uid) AND o.kind IN (?2, ?3, ?4))"
+        ))?;
+        let rows = stmt.query_map(params![parent, OP_CREATE, OP_MKDIR, OP_RENAME], |r| {
+            r.get::<_, String>(0)
+        })?;
+        let mut here = Vec::new();
+        for json in rows {
+            here.push(serde_json::from_str(&json?)?);
+        }
+        let mut stmt = conn.prepare(&format!(
+            "SELECT DISTINCT n.uid FROM pending_op p JOIN nodes n ON n.lid = p.lid OR n.uid = p.uid
+             WHERE p.kind = ?2 AND n.{key} IS NOT ?1"
+        ))?;
+        let gone = stmt
+            .query_map(params![parent, OP_RENAME], |r| r.get::<_, String>(0))?
+            .collect::<std::result::Result<HashSet<_>, _>>()?;
+        Ok((here, gone))
     }
 
     // --- Content-cache LRU index (P4) -------------------------------------
@@ -565,9 +780,157 @@ struct PriorRow {
     trashed: bool,
 }
 
+/// [`Db::lid_of`] on `conn`, which may be a transaction.
+pub(super) fn node_lid(conn: &rusqlite::Connection, uid: &str) -> Result<Option<i64>> {
+    let row = match local_lid(uid) {
+        Some(lid) => conn.query_row("SELECT lid FROM nodes WHERE lid = ?1", params![lid], |r| {
+            r.get(0)
+        }),
+        None => conn.query_row("SELECT lid FROM nodes WHERE uid = ?1", params![uid], |r| {
+            r.get(0)
+        }),
+    };
+    Ok(row.optional()?)
+}
+
+/// The column and value that find `uid`'s own row. A stand-in is found by its
+/// local id: a create that lands gives the row the uid Drive made before the
+/// tree has it, and the tree asks by the stand-in until then (`docs/BUGS.md`
+/// B177).
+fn row_key(uid: &str) -> (&'static str, Value) {
+    match local_lid(uid) {
+        Some(lid) => ("lid", Value::Integer(lid)),
+        None => ("uid", Value::Text(uid.to_owned())),
+    }
+}
+
+/// The column and value a row names `parent` by: its local id when it has a
+/// row, which the children keep when it lands, or its uid when it has none,
+/// as a device folder's root does.
+fn parent_key(conn: &rusqlite::Connection, parent: &str) -> Result<(&'static str, Value)> {
+    Ok(match node_lid(conn, parent)? {
+        Some(lid) => ("parent_lid", Value::Integer(lid)),
+        None => ("parent_uid", Value::Text(parent.to_owned())),
+    })
+}
+
+/// `n.node_json` naming its parent as the parent's row `p`, joined by
+/// `n.parent_lid`, has it now. A folder that lands takes its real uid in its
+/// own row only, so a node stored below it still names the stand-in.
+const NODE_NOW: &str = "CASE WHEN p.uid IS NULL OR p.uid = n.parent_uid THEN n.node_json
+     ELSE json_set(n.node_json, '$.parent_uid', json_object(
+            'volume_id', substr(p.uid, 1, instr(p.uid, '~') - 1),
+            'link_id', substr(p.uid, instr(p.uid, '~') + 1))) END";
+
+/// Hand a drained placeholder's row to the real uid it landed as.
+///
+/// Nothing else is rewritten. The rows below it and the ops made inside it
+/// name it by its local id, which stays; what reads them for a uid takes the
+/// one the row has now ([`NODE_NOW`], `PARENT_NOW` in `ops.rs`).
+///
+/// Dropping the placeholder row and waiting for the server's copy of the real
+/// node left a window with no row for either uid. A queued child drained in
+/// that window found no authority for its parent, re-read it over the network
+/// and was deferred five seconds behind a folder that had just landed
+/// (`docs/BUGS.md` B120). The placeholder's own node, readdressed, stands in
+/// until the server's copy is upserted over it. A row the real uid already
+/// has is newer than the placeholder, so its node is kept, under the
+/// placeholder's `lid`: the node was known by that one first.
+pub(super) fn adopt_placeholder_row_tx(
+    tx: &Transaction<'_>,
+    local: &str,
+    real: &str,
+) -> Result<()> {
+    // A folder pinned before it landed stays pinned: the pin keyed by the
+    // placeholder was dropped with it (`docs/BUGS.md` B148).
+    tx.execute(
+        "UPDATE OR IGNORE pins SET uid = ?2 WHERE uid = ?1",
+        params![local, real],
+    )?;
+    tx.execute("DELETE FROM pins WHERE uid = ?1", params![local])?;
+    let row: Option<(i64, Option<String>)> = tx
+        .query_row(
+            "SELECT lid, node_json FROM nodes WHERE uid = ?1",
+            params![local],
+            |row| Ok((row.get(0)?, row.get(1)?)),
+        )
+        .optional()?;
+    // Ops queued before schema 39 have no `parent_lid`, nor do ops made
+    // under a placeholder that never had a row.
+    tx.execute(
+        "UPDATE pending_op SET parent_uid = ?2 WHERE parent_lid IS NULL AND parent_uid = ?1",
+        params![local, real],
+    )?;
+    let Some((lid, json)) = row else {
+        return Ok(());
+    };
+    let known: Option<(i64, Option<String>)> = tx
+        .query_row(
+            "SELECT lid, node_json FROM nodes WHERE uid = ?1",
+            params![real],
+            |row| Ok((row.get(0)?, row.get(1)?)),
+        )
+        .optional()?;
+    if let Some((known_lid, _)) = known {
+        // Ops queued against the row Drive listed now name the landing one.
+        tx.execute(
+            "UPDATE pending_op SET lid = ?2 WHERE lid = ?1",
+            params![known_lid, lid],
+        )?;
+        tx.execute(
+            "UPDATE pending_op SET parent_lid = ?2 WHERE parent_lid = ?1",
+            params![known_lid, lid],
+        )?;
+        tx.execute(
+            "UPDATE nodes SET parent_lid = ?2 WHERE parent_lid = ?1",
+            params![known_lid, lid],
+        )?;
+        tx.execute("DELETE FROM nodes WHERE lid = ?1", params![known_lid])?;
+        tx.execute("DELETE FROM nodes_fts WHERE rowid = ?1", params![known_lid])?;
+    }
+    tx.execute(
+        "UPDATE nodes SET uid = ?2 WHERE lid = ?1",
+        params![lid, real],
+    )?;
+    let json = known.and_then(|(_, known_json)| known_json).or(json);
+    let (Some(json), Some(real)) = (json, parse_node_uid(real)) else {
+        return Ok(());
+    };
+    let mut node: Node = serde_json::from_str(&json)?;
+    node.uid = real;
+    upsert_node_tx(tx, &node)
+}
+
+/// Link the rows that name `uid` as their parent to its new row `lid`: they
+/// were stored before it, or under a row of the same uid since dropped.
+fn adopt_children_tx(tx: &Transaction<'_>, uid: &str, lid: i64) -> Result<()> {
+    tx.execute(
+        "UPDATE nodes SET parent_lid = ?2 WHERE parent_uid = ?1 AND parent_lid IS NOT ?2",
+        params![uid, lid],
+    )?;
+    Ok(())
+}
+
 fn upsert_node_tx(tx: &Transaction<'_>, node: &Node) -> Result<()> {
-    let json = serde_json::to_string(node)?;
     let uid = node.uid.to_string();
+    // A node written under its stand-in after it landed is its row, which has
+    // the uid Drive gave it by now: written under the stand-in, it would bring
+    // the placeholder row back beside it.
+    if let Some(lid) = local_lid(&uid) {
+        let landed: Option<String> = tx
+            .query_row(
+                "SELECT uid FROM nodes WHERE lid = ?1 AND uid IS NOT ?2",
+                params![lid, uid],
+                |row| row.get(0),
+            )
+            .optional()?;
+        if let Some(real) = landed.as_deref().and_then(parse_node_uid) {
+            let mut node = node.clone();
+            node.uid = real;
+            return upsert_node_tx(tx, &node);
+        }
+    }
+    let json = serde_json::to_string(node)?;
     let parent_uid = node.parent_uid.as_ref().map(|u| u.to_string());
 
     let prior: Option<PriorRow> = tx
@@ -589,26 +952,34 @@ fn upsert_node_tx(tx: &Transaction<'_>, node: &Node) -> Result<()> {
     // where resolving it from scratch is a recursive walk to the root. A node
     // whose parent is not cached (a device folder's root lives in `device`, never
     // in `nodes`) starts a path of its own, which is what the walk did too.
-    let path = match &parent_uid {
-        None => String::new(),
+    let (parent_lid, path) = match &parent_uid {
+        None => (None, String::new()),
         Some(parent) => {
-            let parent_path: Option<Option<String>> = tx
-                .query_row(
-                    "SELECT path FROM nodes WHERE uid = ?1",
-                    params![parent],
-                    |row| row.get(0),
-                )
-                .optional()?;
-            join_path(parent_path.flatten().as_deref().unwrap_or(""), &node.name)
+            let parent_row: Option<(i64, Option<String>)> = match node_lid(tx, parent)? {
+                Some(lid) => tx
+                    .query_row(
+                        "SELECT lid, path FROM nodes WHERE lid = ?1",
+                        params![lid],
+                        |row| Ok((row.get(0)?, row.get(1)?)),
+                    )
+                    .optional()?,
+                None => None,
+            };
+            let (lid, parent_path) = parent_row.unzip();
+            (
+                lid,
+                join_path(parent_path.flatten().as_deref().unwrap_or(""), &node.name),
+            )
         }
     };
 
     tx.execute(
         "INSERT INTO nodes
-           (uid, parent_uid, name, is_dir, size, mtime, trashed, node_json, path)
-         VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)
+           (uid, parent_uid, name, is_dir, size, mtime, trashed, node_json, path, parent_lid)
+         VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10)
          ON CONFLICT(uid) DO UPDATE SET
            parent_uid = excluded.parent_uid,
+           parent_lid = excluded.parent_lid,
            name       = excluded.name,
            is_dir     = excluded.is_dir,
            size       = excluded.size,
@@ -626,6 +997,7 @@ fn upsert_node_tx(tx: &Transaction<'_>, node: &Node) -> Result<()> {
             node.trashed as i64,
             json,
             path,
+            parent_lid,
         ],
     )?;
     let rowid: i64 = tx.query_row(
@@ -633,6 +1005,9 @@ fn upsert_node_tx(tx: &Transaction<'_>, node: &Node) -> Result<()> {
         params![uid],
         |row| row.get(0),
     )?;
+    if prior.is_none() {
+        adopt_children_tx(tx, &uid, rowid)?;
+    }
 
     // FTS5 has no UPSERT, so the node's own row is always replaced.
     let indexable = node_is_indexable_tx(tx, &uid)?;
@@ -655,7 +1030,7 @@ fn upsert_node_tx(tx: &Transaction<'_>, node: &Node) -> Result<()> {
             || prior.trashed != node.trashed
     });
     if node.is_folder() && subtree_moved {
-        reindex_subtree_tx(tx, &uid, &path, indexable)?;
+        reindex_subtree_tx(tx, rowid, &path, indexable)?;
     }
     Ok(())
 }
@@ -676,22 +1051,22 @@ fn upsert_node_tx(tx: &Transaction<'_>, node: &Node) -> Result<()> {
 /// this deep.
 fn reindex_subtree_tx(
     tx: &Transaction<'_>,
-    folder: &str,
+    folder: i64,
     folder_path: &str,
     folder_indexable: bool,
 ) -> Result<()> {
     let descendants: Vec<(i64, String, bool)> = {
         let mut stmt = tx.prepare(
-            "WITH RECURSIVE sub(rowid, uid, name, path, blocked, depth) AS (
-               SELECT n.rowid, n.uid, n.name,
+            "WITH RECURSIVE sub(rowid, name, path, blocked, depth) AS (
+               SELECT n.rowid, n.name,
                       CASE WHEN ?2 = '' THEN n.name ELSE ?2 || '/' || n.name END,
                       n.trashed, 0
-                 FROM nodes n WHERE n.parent_uid = ?1
+                 FROM nodes n WHERE n.parent_lid = ?1
                UNION ALL
-               SELECT n.rowid, n.uid, n.name,
+               SELECT n.rowid, n.name,
                       sub.path || '/' || n.name,
                       MAX(sub.blocked, n.trashed), sub.depth + 1
-                 FROM nodes n JOIN sub ON n.parent_uid = sub.uid
+                 FROM nodes n JOIN sub ON n.parent_lid = sub.rowid
                 WHERE sub.depth < 256
              )
              SELECT rowid, path, blocked FROM sub",
@@ -716,8 +1091,10 @@ fn reindex_subtree_tx(
     let mut set_path = tx.prepare("UPDATE nodes SET path = ?2 WHERE rowid = ?1")?;
     let mut unindex = tx.prepare("DELETE FROM nodes_fts WHERE rowid = ?1")?;
     let mut index = tx.prepare(
+        // Not a stub row (`Db::lids_for`): it is indexed when its node is
+        // written.
         "INSERT INTO nodes_fts (rowid, name, path)
-         SELECT rowid, name, ?2 FROM nodes WHERE rowid = ?1",
+         SELECT rowid, name, ?2 FROM nodes WHERE rowid = ?1 AND node_json IS NOT NULL",
     )?;
     for (rowid, path, blocked) in descendants {
         set_path.execute(params![rowid, path])?;
@@ -730,7 +1107,8 @@ fn reindex_subtree_tx(
 }
 
 fn direct_child_uids_tx(tx: &Transaction<'_>, parent: &str) -> Result<Vec<String>> {
-    let mut stmt = tx.prepare("SELECT uid FROM nodes WHERE parent_uid = ?1")?;
+    let (key, parent) = parent_key(tx, parent)?;
+    let mut stmt = tx.prepare(&format!("SELECT uid FROM nodes WHERE {key} = ?1"))?;
     let rows = stmt.query_map([parent], |row| row.get::<_, String>(0))?;
     let mut uids = Vec::new();
     for row in rows {
@@ -755,14 +1133,14 @@ fn direct_child_uids_tx(tx: &Transaction<'_>, parent: &str) -> Result<Vec<String
 /// ends at one whose parent is a row we have already visited.
 fn node_is_indexable_tx(tx: &Transaction<'_>, uid: &str) -> Result<bool> {
     let indexable: i64 = tx.query_row(
-        "WITH RECURSIVE ancestors(uid, parent_uid, trashed, depth, path) AS (
-           SELECT uid, parent_uid, trashed, 0, char(31) || uid || char(31)
+        "WITH RECURSIVE ancestors(lid, parent_lid, trashed, depth, path) AS (
+           SELECT lid, parent_lid, trashed, 0, ',' || lid || ','
              FROM nodes WHERE uid = ?1
            UNION ALL
-           SELECT n.uid, n.parent_uid, n.trashed, a.depth + 1,
-                  a.path || n.uid || char(31)
-             FROM ancestors a JOIN nodes n ON n.uid = a.parent_uid
-            WHERE instr(a.path, char(31) || n.uid || char(31)) = 0
+           SELECT n.lid, n.parent_lid, n.trashed, a.depth + 1,
+                  a.path || n.lid || ','
+             FROM ancestors a JOIN nodes n ON n.lid = a.parent_lid
+            WHERE instr(a.path, ',' || n.lid || ',') = 0
          )
          SELECT CASE
            WHEN COUNT(*) = 0 THEN 0
@@ -770,8 +1148,7 @@ fn node_is_indexable_tx(tx: &Transaction<'_>, uid: &str) -> Result<bool> {
            WHEN EXISTS (
              SELECT 1 FROM ancestors a
               WHERE a.depth = (SELECT MAX(depth) FROM ancestors)
-                AND a.parent_uid IS NOT NULL
-                AND EXISTS (SELECT 1 FROM nodes p WHERE p.uid = a.parent_uid)
+                AND EXISTS (SELECT 1 FROM nodes p WHERE p.lid = a.parent_lid)
            ) THEN 0
            ELSE 1
          END
@@ -789,7 +1166,7 @@ fn tombstone_subtree_tx(tx: &Transaction<'_>, root_uid: &str) -> Result<()> {
                SELECT rowid, uid, node_json FROM nodes WHERE uid = ?1
                UNION
                SELECT n.rowid, n.uid, n.node_json FROM nodes n
-                 JOIN subtree s ON n.parent_uid = s.uid
+                 JOIN subtree s ON n.parent_lid = s.rowid
              )
              SELECT rowid, uid, node_json FROM subtree",
         )?;

crates/pdfs-core/src/db/photos.rs

@@ -3,7 +3,7 @@
 
 use rusqlite::params;
 
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 
 use super::Db;
 use crate::Result;
@@ -42,6 +42,15 @@ pub struct StoredPhoto {
     pub has_raw: bool,
 }
 
+/// A town photos were taken in: its GeoNames id (see [`crate::places`]), how
+/// many photos it holds, and the newest of them.
+#[derive(Clone, Debug, PartialEq)]
+pub struct StoredPlace {
+    pub id: u32,
+    pub count: usize,
+    pub cover: StoredPhoto,
+}
+
 /// One row of a timeline replacement. Everything but the uid and the capture
 /// time is "what this refresh learned": a `None` keeps whatever is already
 /// stored rather than clearing it, because the timeline DTO carries only the uid
@@ -58,6 +67,8 @@ pub struct TimelineRow {
     /// The main photo this one is *related* to (a live-photo video, a burst
     /// sibling), when the server says it belongs to one.
     pub main_uid: Option<String>,
+    /// Where the photo was taken, as `(latitude, longitude)` in decimal degrees.
+    pub location: Option<(f64, f64)>,
     /// When this refresh resolved the photo's node, if it did at all. `Some`
     /// makes every other field authoritative — including a `None` that means
     /// "the server no longer reports one" — while `None` means the refresh
@@ -86,6 +97,9 @@ struct Learned {
     favorite: bool,
     content_hash: Option<String>,
     main_uid: Option<String>,
+    location: Option<(f64, f64)>,
+    /// What its thumbnail looks like (see [`crate::similar`]).
+    similar_hash: Option<i64>,
     /// When this photo's node was last resolved, kept so a refresh that skips it
     /// does not make it look unresolved again.
     resolved_at: Option<i64>,
@@ -281,7 +295,7 @@ impl Db {
         let learned: HashMap<String, Learned> = {
             let mut stmt = tx.prepare(
                 "SELECT uid, name, ratio, thumb_state, media_type, favorite, content_hash, \
-                 main_uid, resolved_at FROM photos",
+                 main_uid, resolved_at, latitude, longitude, similar_hash FROM photos",
             )?;
             let rows = stmt.query_map([], |r| {
                 Ok((
@@ -295,6 +309,10 @@ impl Db {
                         content_hash: r.get(6)?,
                         main_uid: r.get(7)?,
                         resolved_at: r.get(8)?,
+                        location: r
+                            .get::<_, Option<f64>>(9)?
+                            .zip(r.get::<_, Option<f64>>(10)?),
+                        similar_hash: r.get(11)?,
                     },
                 ))
             })?;
@@ -317,14 +335,20 @@ impl Db {
                     carried.name = row.name.clone();
                     carried.media_type = row.media_type.clone();
                     carried.favorite = row.favorite.unwrap_or(false);
+                    // New bytes may show another picture: hash it again.
+                    if row.content_hash != carried.content_hash {
+                        carried.similar_hash = None;
+                    }
                     carried.content_hash = row.content_hash.clone();
                     carried.main_uid = row.main_uid.clone();
+                    carried.location = row.location;
                 } else {
                     carried.name = row.name.clone().or(carried.name);
                     carried.media_type = row.media_type.clone().or(carried.media_type);
                     carried.favorite = row.favorite.unwrap_or(carried.favorite);
                     carried.content_hash = row.content_hash.clone().or(carried.content_hash);
                     carried.main_uid = row.main_uid.clone().or(carried.main_uid);
+                    carried.location = row.location.or(carried.location);
                 }
                 carried.resolved_at = row.resolved_at.or(carried.resolved_at);
                 // The tab this photo lands in is derived here, once, so a page or
@@ -355,8 +379,10 @@ impl Db {
             let mut stmt = tx.prepare(
                 "INSERT INTO photos
                    (uid, capture_time, name, ratio, thumb_state, seq, media_type, kind, favorite,
-                    content_hash, main_uid, group_key, resolved_at)
-                 VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13)",
+                    content_hash, main_uid, group_key, resolved_at, latitude, longitude,
+                    place_id, similar_hash)
+                 VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14, ?15,
+                         ?16, ?17)",
             )?;
             for (seq, ((row, carried, kind), group_key)) in
                 resolved.iter().zip(group_keys.iter()).enumerate()
@@ -375,6 +401,13 @@ impl Db {
                     carried.main_uid,
                     group_key,
                     carried.resolved_at,
+                    carried.location.map(|(lat, _)| lat),
+                    carried.location.map(|(_, lon)| lon),
+                    carried
+                        .location
+                        .and_then(|(lat, lon)| crate::places::nearest(lat, lon))
+                        .map(|city| city.id),
+                    carried.similar_hash,
                 ])?;
             }
         }
@@ -447,6 +480,14 @@ impl Db {
                 albums.execute([uid])?;
             }
         }
+        // A group whose representative went is left pointing at nothing, which
+        // drops its other files off every page. Make them stand alone until the
+        // next refresh groups them again.
+        tx.execute(
+            "UPDATE photos SET group_key = NULL WHERE group_key IS NOT NULL \
+             AND group_key NOT IN (SELECT uid FROM photos)",
+            [],
+        )?;
         tx.commit()?;
         Ok(removed)
     }
@@ -455,7 +496,8 @@ impl Db {
     /// restricts the page to one tab (Photos / Videos / Raw); `range`, when set,
     /// restricts it to a `[from, to)` capture-time window (epoch seconds) — the
     /// date scrubber's jump. `offset` is relative to whatever the filters leave.
-    /// `favorites` restricts the page to favourited photos.
+    /// `favorites` restricts the page to favorited photos, and `not_in_album`
+    /// to the shots no album holds any file of.
     ///
     /// The page holds one row per *group*: the RAW and the JPEG of one shot are
     /// one tile. The Raw tab is the exception and lists every raw file, because
@@ -467,6 +509,7 @@ impl Db {
         kind: Option<crate::control::PhotoKind>,
         range: Option<(i64, i64)>,
         favorites: bool,
+        not_in_album: bool,
     ) -> Result<Vec<StoredPhoto>> {
         let conn = self.read();
         // Built up so any combination of the optional filters is one indexed
@@ -490,6 +533,15 @@ impl Db {
         if favorites {
             conds.push("p.favorite = 1".to_string());
         }
+        // By group, not by file: a shot filed as its JPEG is filed, even though
+        // its RAW was never added anywhere.
+        if not_in_album {
+            conds.push(
+                "NOT EXISTS (SELECT 1 FROM album_photos a JOIN photos m ON m.uid = a.uid \
+                 WHERE COALESCE(m.group_key, m.uid) = COALESCE(p.group_key, p.uid))"
+                    .to_string(),
+            );
+        }
         if !conds.is_empty() {
             sql.push_str(" WHERE ");
             sql.push_str(&conds.join(" AND "));
@@ -577,6 +629,256 @@ impl Db {
         Ok(rows)
     }
 
+    /// Photos taken on `today`'s calendar day in earlier years, newest first —
+    /// the "On this day" strip. `today` is epoch seconds: its local month and day
+    /// pick the date, and its own year is left out, because today's photos are
+    /// what the timeline already opens on. One row per group, like the timeline.
+    pub fn photos_on_this_day(&self, today: i64, limit: usize) -> Result<Vec<StoredPhoto>> {
+        let conn = self.read();
+        let sql = format!(
+            "{PHOTO_SELECT} WHERE p.uid = COALESCE(p.group_key, p.uid) \
+             AND strftime('%m-%d', p.capture_time, 'unixepoch', 'localtime') \
+                 = strftime('%m-%d', ?1, 'unixepoch', 'localtime') \
+             AND strftime('%Y', p.capture_time, 'unixepoch', 'localtime') \
+                 < strftime('%Y', ?1, 'unixepoch', 'localtime') \
+             ORDER BY p.seq LIMIT ?2"
+        );
+        let mut stmt = conn.prepare(&sql)?;
+        let rows = stmt
+            .query_map(params![today, limit as i64], stored_photo)?
+            .collect::<rusqlite::Result<_>>()?;
+        Ok(rows)
+    }
+
+    /// Every town photos were taken in, with how many and the newest one as its
+    /// cover, most photographed first. Counted per group, like the timeline;
+    /// photos without a location, or taken far from any town, are in none.
+    pub fn photo_places(&self) -> Result<Vec<StoredPlace>> {
+        let rows: Vec<(u32, usize, String)> = {
+            let conn = self.read();
+            // SQLite fills the bare `uid` from the row `MIN(seq)` picked, which
+            // is the newest photo of the place.
+            let mut stmt = conn.prepare(
+                "SELECT place_id, COUNT(*), uid, MIN(seq) FROM photos \
+                 WHERE uid = COALESCE(group_key, uid) AND place_id IS NOT NULL \
+                 GROUP BY place_id ORDER BY COUNT(*) DESC, MIN(seq)",
+            )?;
+            stmt.query_map([], |r| {
+                Ok((r.get(0)?, r.get::<_, i64>(1)? as usize, r.get(2)?))
+            })?
+            .collect::<rusqlite::Result<_>>()?
+        };
+        let uids: Vec<String> = rows.iter().map(|(_, _, uid)| uid.clone()).collect();
+        let mut covers: HashMap<String, StoredPhoto> = self
+            .photos_by_uid(&uids)?
+            .into_iter()
+            .map(|photo| (photo.uid.clone(), photo))
+            .collect();
+        Ok(rows
+            .into_iter()
+            .filter_map(|(id, count, uid)| {
+                Some(StoredPlace {
+                    id,
+                    count,
+                    cover: covers.remove(&uid)?,
+                })
+            })
+            .collect())
+    }
+
+    /// One page of the photos taken in a town, newest first, one per group.
+    pub fn photos_at_place(
+        &self,
+        id: u32,
+        offset: usize,
+        limit: usize,
+    ) -> Result<Vec<StoredPhoto>> {
+        let conn = self.read();
+        let sql = format!(
+            "{PHOTO_SELECT} WHERE p.uid = COALESCE(p.group_key, p.uid) AND p.place_id = ?1 \
+             ORDER BY p.seq LIMIT ?2 OFFSET ?3"
+        );
+        let mut stmt = conn.prepare(&sql)?;
+        let rows = stmt
+            .query_map(params![id, limit as i64, offset as i64], stored_photo)?
+            .collect::<rusqlite::Result<_>>()?;
+        Ok(rows)
+    }
+
+    /// Every photo taken in a town, one per group as in [`Self::photos_at_place`],
+    /// with where exactly it was taken. A group's own location is that of the
+    /// photo standing for it.
+    pub fn photo_locations_at_place(&self, id: u32) -> Result<Vec<(StoredPhoto, (f64, f64))>> {
+        let photos = self.photos_at_place(id, 0, usize::MAX >> 1)?;
+        let located: HashMap<String, (f64, f64)> = {
+            let conn = self.read();
+            let mut stmt = conn.prepare(
+                "SELECT uid, latitude, longitude FROM photos \
+                 WHERE place_id = ?1 AND latitude IS NOT NULL AND longitude IS NOT NULL",
+            )?;
+            stmt.query_map(params![id], |r| Ok((r.get(0)?, (r.get(1)?, r.get(2)?))))?
+                .collect::<rusqlite::Result<_>>()?
+        };
+        Ok(photos
+            .into_iter()
+            .filter_map(|photo| {
+                let at = *located.get(&photo.uid)?;
+                Some((photo, at))
+            })
+            .collect())
+    }
+
+    /// Sets of photos with the same content hash, each set ordered with the
+    /// copy worth keeping first, and the sets newest first.
+    ///
+    /// The copy to keep is the one in the most albums, then the one that is part
+    /// of a larger group (the JPEG's RAW is filed with it, a stray re-upload is
+    /// not), then the earliest in server order. At most one file per group is
+    /// listed: the files of one RAW+JPEG shot are not copies of each other, and a
+    /// set left with one member is no set.
+    pub fn photo_duplicates(&self) -> Result<Vec<Vec<StoredPhoto>>> {
+        let rows: Vec<(String, String, String)> = {
+            let conn = self.read();
+            let mut stmt = conn.prepare(
+                "SELECT p.uid, p.content_hash, COALESCE(p.group_key, p.uid) AS k FROM photos p \
+                 WHERE p.content_hash IN (SELECT content_hash FROM photos \
+                     WHERE content_hash IS NOT NULL GROUP BY content_hash HAVING COUNT(*) > 1) \
+                 ORDER BY p.content_hash, \
+                     (SELECT COUNT(*) FROM album_photos a WHERE a.uid = p.uid) DESC, \
+                     (SELECT COUNT(*) FROM photos m WHERE COALESCE(m.group_key, m.uid) = COALESCE(p.group_key, p.uid)) DESC, \
+                     p.seq",
+            )?;
+            stmt.query_map([], |r| Ok((r.get(0)?, r.get(1)?, r.get(2)?)))?
+                .collect::<rusqlite::Result<_>>()?
+        };
+        let uids: Vec<String> = rows.iter().map(|(uid, _, _)| uid.clone()).collect();
+        let mut stored: HashMap<String, StoredPhoto> = self
+            .photos_by_uid(&uids)?
+            .into_iter()
+            .map(|photo| (photo.uid.clone(), photo))
+            .collect();
+
+        let mut sets: Vec<Vec<StoredPhoto>> = Vec::new();
+        let mut hash_of_set: Option<&str> = None;
+        let mut groups_in_set: HashSet<&str> = HashSet::new();
+        for (uid, hash, group) in &rows {
+            if hash_of_set != Some(hash.as_str()) {
+                hash_of_set = Some(hash);
+                groups_in_set.clear();
+                sets.push(Vec::new());
+            }
+            if !groups_in_set.insert(group) {
+                continue;
+            }
+            if let (Some(set), Some(photo)) = (sets.last_mut(), stored.remove(uid)) {
+                set.push(photo);
+            }
+        }
+        sets.retain(|set| set.len() > 1);
+        sets.sort_by_key(|set| std::cmp::Reverse(set.iter().map(|p| p.capture_time).max()));
+        Ok(sets)
+    }
+
+    /// Shots whose look hasn't been hashed yet, by the uid of the photo the
+    /// grid shows for each, newest first, at most `limit`. Videos, and photos
+    /// with no thumbnail to hash, are left out. A shot counts as hashed once
+    /// any of its files is, since they all show one picture.
+    pub fn photos_unhashed(&self, limit: usize) -> Result<Vec<StoredPhoto>> {
+        let uids: Vec<String> = {
+            let conn = self.read();
+            let mut stmt = conn.prepare(
+                "SELECT COALESCE(group_key, uid) AS k FROM photos \
+                 GROUP BY k \
+                 HAVING MAX(similar_hash) IS NULL \
+                    AND MAX(kind != ?1 AND thumb_state != ?2) \
+                 ORDER BY MAX(capture_time) DESC LIMIT ?3",
+            )?;
+            stmt.query_map(
+                params![
+                    crate::control::PhotoKind::Video.as_i64(),
+                    THUMB_NONE,
+                    limit as i64
+                ],
+                |r| r.get(0),
+            )?
+            .collect::<rusqlite::Result<_>>()?
+        };
+        self.photos_by_uid(&uids)
+    }
+
+    /// Record the look of photo `uid`: its difference hash, or
+    /// [`crate::similar::NO_HASH`] when its thumbnail gave none.
+    pub fn photo_set_similar_hash(&self, uid: &str, hash: u64) -> Result<()> {
+        let conn = self.conn.lock();
+        conn.execute(
+            "UPDATE photos SET similar_hash = ?2 WHERE uid = ?1",
+            params![uid, hash as i64],
+        )?;
+        Ok(())
+    }
+
+    /// How many shots have been hashed, out of how many could be.
+    pub fn similar_progress(&self) -> Result<(usize, usize)> {
+        let conn = self.read();
+        let (hashed, total): (i64, i64) = conn.query_row(
+            "SELECT COUNT(*) FILTER (WHERE hashed), COUNT(*) FROM \
+               (SELECT MAX(similar_hash) IS NOT NULL AS hashed FROM photos \
+                GROUP BY COALESCE(group_key, uid) \
+                HAVING MAX(similar_hash) IS NOT NULL \
+                    OR MAX(kind != ?1 AND thumb_state != ?2))",
+            params![crate::control::PhotoKind::Video.as_i64(), THUMB_NONE],
+            |r| Ok((r.get(0)?, r.get(1)?)),
+        )?;
+        Ok((hashed.max(0) as usize, total.max(0) as usize))
+    }
+
+    /// Sets of shots that look alike (see [`crate::similar`]), one photo per
+    /// shot in capture order, and the sets newest first. A set whose photos
+    /// are all byte-identical copies is left to [`Self::photo_duplicates`].
+    pub fn photo_similar(&self) -> Result<Vec<Vec<StoredPhoto>>> {
+        let rows: Vec<(String, u64, Option<String>)> = {
+            let conn = self.read();
+            let mut stmt = conn.prepare(
+                "SELECT COALESCE(group_key, uid) AS k, MAX(similar_hash), \
+                        MAX(content_hash) FILTER (WHERE uid = COALESCE(group_key, uid)) \
+                 FROM photos GROUP BY k \
+                 HAVING MAX(similar_hash) IS NOT NULL AND MAX(similar_hash) != ?1",
+            )?;
+            stmt.query_map(params![crate::similar::NO_HASH as i64], |r| {
+                Ok((r.get(0)?, r.get::<_, i64>(1)? as u64, r.get(2)?))
+            })?
+            .collect::<rusqlite::Result<_>>()?
+        };
+        let hashes: Vec<u64> = rows.iter().map(|(_, hash, _)| *hash).collect();
+        let found: Vec<Vec<usize>> = crate::similar::sets(&hashes)
+            .into_iter()
+            .filter(|set| {
+                let first = &rows[set[0]].2;
+                first.is_none() || set.iter().any(|&i| rows[i].2 != *first)
+            })
+            .collect();
+        let uids: Vec<String> = found.iter().flatten().map(|&i| rows[i].0.clone()).collect();
+        let mut stored: HashMap<String, StoredPhoto> = self
+            .photos_by_uid(&uids)?
+            .into_iter()
+            .map(|photo| (photo.uid.clone(), photo))
+            .collect();
+        let mut sets: Vec<Vec<StoredPhoto>> = found
+            .iter()
+            .map(|set| {
+                let mut photos: Vec<StoredPhoto> = set
+                    .iter()
+                    .filter_map(|&i| stored.remove(&rows[i].0))
+                    .collect();
+                photos.sort_by_key(|p| p.capture_time);
+                photos
+            })
+            .filter(|set| set.len() > 1)
+            .collect();
+        sets.sort_by_key(|set| std::cmp::Reverse(set.iter().map(|p| p.capture_time).max()));
+        Ok(sets)
+    }
+
     /// Per-tab counts for the Photos page subtitle: `(photos, videos, raw)`.
     pub fn photos_counts(&self) -> Result<(usize, usize, usize)> {
         use crate::control::PhotoKind;
@@ -620,7 +922,7 @@ impl Db {
         Ok(photos)
     }
 
-    /// Record a photo's favourite flag locally, after the server accepted the
+    /// Record a photo's favorite flag locally, after the server accepted the
     /// change. A uid the timeline does not hold is a no-op — an album photo on
     /// someone else's volume is never in our own `photos` table.
     pub fn photos_set_favorite(&self, uid: &str, favorite: bool) -> Result<()> {

crates/pdfs-core/src/db/pins.rs

@@ -64,11 +64,11 @@ impl Db {
     }
 
     /// Whether `uid` is pinned — directly, or because a strict ancestor folder
-    /// is pinned recursively. The ancestor check walks `parent_uid` to the root
+    /// is pinned recursively. The ancestor check walks `parent_lid` to the root
     /// via a CTE; a direct pin is honoured even when the node has no `nodes` row
     /// yet (e.g. a CLI that never hydrates the node cache).
     ///
-    /// The walk is depth-capped like [`super::utils::path_of`]: a `parent_uid`
+    /// The walk is depth-capped like [`super::utils::path_of`]: a parent
     /// cycle is corrupt data the API can hand us, and `UNION ALL` over one never
     /// terminates — while holding the daemon's only SQLite connection, on a path
     /// that runs per cached read.
@@ -85,11 +85,11 @@ impl Db {
         let anc: Option<i64> = conn
             .query_row(
                 &format!(
-                    "WITH RECURSIVE anc(uid, parent_uid, depth) AS (
-                       SELECT uid, parent_uid, 0 FROM nodes WHERE uid = ?1
+                    "WITH RECURSIVE anc(uid, parent_lid, depth) AS (
+                       SELECT uid, parent_lid, 0 FROM nodes WHERE uid = ?1
                        UNION ALL
-                       SELECT n.uid, n.parent_uid, anc.depth + 1
-                       FROM nodes n JOIN anc ON n.uid = anc.parent_uid
+                       SELECT n.uid, n.parent_lid, anc.depth + 1
+                       FROM nodes n JOIN anc ON n.lid = anc.parent_lid
                        WHERE anc.depth < {MAX_PATH_DEPTH}
                      )
                      SELECT 1 FROM anc a JOIN pins p ON p.uid = a.uid
@@ -109,10 +109,11 @@ impl Db {
     pub fn pinned_uids(&self) -> Result<Vec<String>> {
         let conn = self.read();
         let mut stmt = conn.prepare(
-            "WITH RECURSIVE sub(uid) AS (
-               SELECT uid FROM pins WHERE recursive = 1
+            "WITH RECURSIVE sub(uid, lid) AS (
+               SELECT p.uid, n.lid FROM pins p LEFT JOIN nodes n ON n.uid = p.uid
+                WHERE p.recursive = 1
                UNION
-               SELECT n.uid FROM nodes n JOIN sub ON n.parent_uid = sub.uid
+               SELECT n.uid, n.lid FROM nodes n JOIN sub ON n.parent_lid = sub.lid
              )
              SELECT uid FROM sub
              UNION
@@ -126,15 +127,16 @@ impl Db {
         Ok(out)
     }
 
-    /// Every descendant uid of `folder` (all depths), via a `parent_uid` CTE.
+    /// Every descendant uid of `folder` (all depths), via a `parent_lid` CTE.
     /// Used to evict a recursively-pinned subtree's cached blobs on unpin.
     pub fn descendants(&self, folder: &str) -> Result<Vec<String>> {
         let conn = self.read();
         let mut stmt = conn.prepare(
-            "WITH RECURSIVE sub(uid) AS (
-               SELECT uid FROM nodes WHERE parent_uid = ?1
+            "WITH RECURSIVE sub(uid, lid) AS (
+               SELECT c.uid, c.lid FROM nodes c JOIN nodes f ON c.parent_lid = f.lid
+                WHERE f.uid = ?1
                UNION
-               SELECT n.uid FROM nodes n JOIN sub ON n.parent_uid = sub.uid
+               SELECT n.uid, n.lid FROM nodes n JOIN sub ON n.parent_lid = sub.lid
              )
              SELECT uid FROM sub",
         )?;

crates/pdfs-core/src/db/share_access.rs

@@ -95,11 +95,27 @@ impl Db {
     /// Resolve a persisted node's effective access through its nearest recorded
     /// share root. `None` means the node itself is not persisted, which callers
     /// must not treat as owned: stale handles need to fail closed.
+    ///
+    /// A stand-in ([`local_uid`](super::local_uid)) finds its row by local id,
+    /// so it still does once its create has landed and the row has the uid
+    /// Drive gave it. The tree learns that uid a moment later, and a rename in
+    /// between was refused (`docs/BUGS.md` B174).
     pub fn effective_node_access(&self, uid: &NodeUid) -> Result<Option<Access>> {
         let conn = self.read();
+        let key = uid.to_string();
+        let key = match super::local_lid(&key) {
+            Some(lid) => conn
+                .query_row("SELECT uid FROM nodes WHERE lid = ?1", [lid], |row| {
+                    row.get::<_, Option<String>>(0)
+                })
+                .optional()?
+                .flatten()
+                .unwrap_or(key),
+            None => key,
+        };
         let exists: bool = conn.query_row(
-            "SELECT EXISTS(SELECT 1 FROM nodes WHERE uid = ?1)",
-            [uid.to_string()],
+            "SELECT EXISTS(SELECT 1 FROM nodes WHERE uid = ?1 AND node_json IS NOT NULL)",
+            [&key],
             |row| row.get(0),
         )?;
         if !exists {
@@ -107,25 +123,26 @@ impl Db {
         }
         let value = conn
             .query_row(
-                "WITH RECURSIVE ancestors(uid, parent_uid, depth, path) AS (
-                   SELECT uid, parent_uid, 0, char(31) || uid || char(31)
+                "WITH RECURSIVE ancestors(uid, parent_uid, parent_lid, depth, path) AS (
+                   SELECT uid, parent_uid, parent_lid, 0, char(31) || uid || char(31)
                      FROM nodes WHERE uid = ?1
                    UNION ALL
-                   SELECT a.parent_uid, n.parent_uid, a.depth + 1,
-                          a.path || a.parent_uid || char(31)
+                   SELECT COALESCE(n.uid, a.parent_uid), n.parent_uid, n.parent_lid,
+                          a.depth + 1,
+                          a.path || COALESCE(n.uid, a.parent_uid) || char(31)
                      FROM ancestors a
-                     LEFT JOIN nodes n ON n.uid = a.parent_uid
+                     LEFT JOIN nodes n ON n.lid = a.parent_lid
                     WHERE a.parent_uid IS NOT NULL
                       AND instr(
                             a.path,
-                            char(31) || a.parent_uid || char(31)
+                            char(31) || COALESCE(n.uid, a.parent_uid) || char(31)
                           ) = 0
                  )
                  SELECT sa.access
                    FROM ancestors a JOIN share_access sa ON sa.root_uid = a.uid
                   ORDER BY a.depth ASC
                   LIMIT 1",
-                [uid.to_string()],
+                [&key],
                 |row| row.get::<_, String>(0),
             )
             .optional()?;

crates/pdfs-core/src/db/sync.rs

@@ -25,8 +25,8 @@ pub struct StoredSyncFolder {
 }
 
 /// One per-file sync baseline row: what a path looked like on both sides at the
-/// last successful sync, so the next reconcile can tell which side changed
-/// (devices.md Phase 2). `remote_rev`/`remote_hash` hold the remote signature —
+/// last successful sync, so the next reconcile can tell which side changed.
+/// `remote_rev`/`remote_hash` hold the remote signature —
 /// its modification time and size as strings — since no cheap content hash is
 /// exposed; change detection is `(mtime, size)` on each side.
 #[derive(Clone, Debug, PartialEq)]
@@ -39,12 +39,18 @@ pub struct StoredSyncEntry {
     ///
     /// `local_mtime` alone cannot tell an edit that landed inside the same
     /// second as the last sync — and left the file the same length — from no
-    /// edit at all, so those never uploaded (bugs.md B25). `None` on a row
+    /// edit at all, so those never uploaded (`docs/BUGS.md` B25). `None` on a row
     /// written before schema 28, which the comparison reads as "compare whole
     /// seconds", so upgrading does not make an entire mirror look changed.
     pub local_mtime_ns: Option<i64>,
     pub remote_rev: Option<String>,
     pub remote_hash: Option<String>,
+    /// The id of the remote revision this row was synced at. A mirror upload
+    /// checks it before it supersedes what Drive holds, because another client
+    /// may have committed a revision since (`docs/BUGS.md` B249). `None` on a
+    /// row written before schema 44, or by a pass that did not learn it; the
+    /// check then falls back to `remote_rev` and `remote_hash`.
+    pub remote_revision_id: Option<String>,
 }
 
 impl Db {
@@ -139,6 +145,17 @@ impl Db {
         Ok(())
     }
 
+    /// Re-point a synced folder at the share that now holds its remote folder,
+    /// after the folder moved to another device without changing its uid.
+    pub fn sync_folder_set_share(&self, id: i64, share_id: &str) -> Result<()> {
+        let conn = self.conn.lock();
+        conn.execute(
+            "UPDATE sync_folder SET remote_share_id = ?2 WHERE id = ?1",
+            params![id, share_id],
+        )?;
+        Ok(())
+    }
+
     /// Queue (or, with `None`, withdraw) a mode the folder should move to once it
     /// is safe to switch.
     pub fn sync_folder_set_pending_mode(&self, id: i64, mode: Option<&str>) -> Result<()> {
@@ -177,7 +194,7 @@ impl Db {
         let conn = self.read();
         let mut stmt = conn.prepare(
             "SELECT rel_path, remote_uid, local_mtime, local_size, remote_rev, remote_hash,
-                    local_mtime_ns
+                    local_mtime_ns, remote_revision_id
              FROM sync_entry WHERE folder_id = ?1",
         )?;
         let rows = stmt.query_map(params![folder_id], |r| {
@@ -189,6 +206,7 @@ impl Db {
                 remote_rev: r.get(4)?,
                 remote_hash: r.get(5)?,
                 local_mtime_ns: r.get(6)?,
+                remote_revision_id: r.get(7)?,
             })
         })?;
         let mut map = HashMap::new();
@@ -227,15 +245,16 @@ impl Db {
         conn.execute(
             "INSERT INTO sync_entry
                (folder_id, rel_path, remote_uid, local_mtime, local_size, remote_rev,
-                remote_hash, local_mtime_ns)
-             VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8)
+                remote_hash, local_mtime_ns, remote_revision_id)
+             VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)
              ON CONFLICT(folder_id, rel_path) DO UPDATE SET
-               remote_uid     = excluded.remote_uid,
-               local_mtime    = excluded.local_mtime,
-               local_size     = excluded.local_size,
-               remote_rev     = excluded.remote_rev,
-               remote_hash    = excluded.remote_hash,
-               local_mtime_ns = excluded.local_mtime_ns",
+               remote_uid         = excluded.remote_uid,
+               local_mtime        = excluded.local_mtime,
+               local_size         = excluded.local_size,
+               remote_rev         = excluded.remote_rev,
+               remote_hash        = excluded.remote_hash,
+               local_mtime_ns     = excluded.local_mtime_ns,
+               remote_revision_id = excluded.remote_revision_id",
             params![
                 folder_id,
                 e.rel_path,
@@ -245,6 +264,7 @@ impl Db {
                 e.remote_rev,
                 e.remote_hash,
                 e.local_mtime_ns,
+                e.remote_revision_id,
             ],
         )?;
         Ok(())
@@ -260,10 +280,94 @@ impl Db {
         Ok(())
     }
 
+    /// Drop the baseline rows of `rel_path` and everything under it.
+    ///
+    /// Used when a subtree leaves a mirror folder through a move on Drive: its
+    /// local copy is about to be removed, and rows left behind would read as
+    /// "deleted locally" and trash the moved nodes at their new place.
+    pub fn sync_entries_remove_subtree(&self, folder_id: i64, rel_path: &str) -> Result<()> {
+        let conn = self.conn.lock();
+        conn.execute(
+            "DELETE FROM sync_entry
+              WHERE folder_id = ?1
+                AND (rel_path = ?2 OR substr(rel_path, 1, length(?2) + 1) = ?2 || '/')",
+            params![folder_id, rel_path],
+        )?;
+        Ok(())
+    }
+
+    /// Move the baseline rows of `from_rel` and everything under it to `to_rel`
+    /// in folder `to_folder`, in one transaction.
+    ///
+    /// Used when a move on Drive took a mirror subtree along with its local
+    /// copy: the rows still describe both sides correctly, only under a new
+    /// path, so the next pass on either folder has nothing to do. The two
+    /// folders may be the same one.
+    pub fn sync_entries_move(
+        &self,
+        from_folder: i64,
+        from_rel: &str,
+        to_folder: i64,
+        to_rel: &str,
+    ) -> Result<()> {
+        let mut conn = self.conn.lock();
+        let tx = conn.transaction()?;
+        {
+            let mut select = tx.prepare(
+                "SELECT rel_path, remote_uid, local_mtime, local_size, remote_rev, remote_hash,
+                        local_mtime_ns, remote_revision_id
+                   FROM sync_entry
+                  WHERE folder_id = ?1
+                    AND (rel_path = ?2 OR substr(rel_path, 1, length(?2) + 1) = ?2 || '/')",
+            )?;
+            let rows = select
+                .query_map(params![from_folder, from_rel], |r| {
+                    Ok(StoredSyncEntry {
+                        rel_path: r.get(0)?,
+                        remote_uid: r.get(1)?,
+                        local_mtime: r.get(2)?,
+                        local_size: r.get(3)?,
+                        remote_rev: r.get(4)?,
+                        remote_hash: r.get(5)?,
+                        local_mtime_ns: r.get(6)?,
+                        remote_revision_id: r.get(7)?,
+                    })
+                })?
+                .collect::<rusqlite::Result<Vec<_>>>()?;
+            let mut delete =
+                tx.prepare("DELETE FROM sync_entry WHERE folder_id = ?1 AND rel_path = ?2")?;
+            for row in &rows {
+                delete.execute(params![from_folder, row.rel_path])?;
+            }
+            let mut insert = tx.prepare(
+                "INSERT OR REPLACE INTO sync_entry
+                   (folder_id, rel_path, remote_uid, local_mtime, local_size, remote_rev,
+                    remote_hash, local_mtime_ns, remote_revision_id)
+                 VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)",
+            )?;
+            for row in rows {
+                let rel = format!("{to_rel}{}", &row.rel_path[from_rel.len()..]);
+                insert.execute(params![
+                    to_folder,
+                    rel,
+                    row.remote_uid,
+                    row.local_mtime,
+                    row.local_size,
+                    row.remote_rev,
+                    row.remote_hash,
+                    row.local_mtime_ns,
+                    row.remote_revision_id,
+                ])?;
+            }
+        }
+        tx.commit()?;
+        Ok(())
+    }
+
     /// Drop the entire baseline for a folder. Used when flipping ondemand→mirror:
     /// the local tree was evicted, so the old baseline is stale and would make the
     /// next reconcile mistake "locally deleted" for "must re-download". Clearing it
-    /// leaves an empty baseline + full remote = pure download (devices.md P3).
+    /// leaves an empty baseline + full remote = pure download.
     pub fn sync_entries_clear(&self, folder_id: i64) -> Result<()> {
         let conn = self.conn.lock();
         conn.execute(

crates/pdfs-core/src/db/trash.rs

@@ -1,5 +1,7 @@
 //! Trashed nodes, persisted so the trash view opens without a round-trip.
 
+use std::collections::HashMap;
+
 use rusqlite::params;
 
 use super::Db;
@@ -15,6 +17,9 @@ pub struct StoredTrash {
     /// row written before schema v29, and for a node whose parent the server
     /// did not report.
     pub parent_uid: Option<String>,
+    /// When the daemon first saw the node in the trash, epoch seconds. `None`
+    /// when it cannot say (see schema v35).
+    pub trashed_at: Option<i64>,
 }
 
 impl Db {
@@ -24,8 +29,8 @@ impl Db {
         tx.execute("DELETE FROM trash", [])?;
         {
             let mut stmt = tx.prepare(
-                "INSERT INTO trash (uid, name, is_dir, size, mtime, parent_uid) \
-                 VALUES (?1, ?2, ?3, ?4, ?5, ?6)",
+                "INSERT INTO trash (uid, name, is_dir, size, mtime, parent_uid, trashed_at) \
+                 VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7)",
             )?;
             for item in items {
                 stmt.execute(params![
@@ -34,7 +39,8 @@ impl Db {
                     item.is_dir as i64,
                     item.size,
                     item.mtime,
-                    item.parent_uid
+                    item.parent_uid,
+                    item.trashed_at
                 ])?;
             }
         }
@@ -47,7 +53,7 @@ impl Db {
     pub fn trash_list(&self) -> Result<Vec<StoredTrash>> {
         let conn = self.read();
         let mut stmt = conn.prepare(
-            "SELECT uid, name, is_dir, size, mtime, parent_uid FROM trash
+            "SELECT uid, name, is_dir, size, mtime, parent_uid, trashed_at FROM trash
              ORDER BY is_dir DESC, name COLLATE NOCASE",
         )?;
         let rows = stmt.query_map([], |r| {
@@ -58,6 +64,7 @@ impl Db {
                 size: r.get(3)?,
                 mtime: r.get(4)?,
                 parent_uid: r.get(5)?,
+                trashed_at: r.get(6)?,
             })
         })?;
         let mut items = Vec::new();
@@ -81,5 +88,19 @@ impl Db {
         Ok(pairs)
     }
 
-    // ---- device sync (devices.md) -----------------------------------------
+    /// When each node now in the trash was first seen there, keyed by uid, so
+    /// a refresh can carry the times over into the listing it writes.
+    pub fn trash_seen(&self) -> Result<HashMap<String, Option<i64>>> {
+        let conn = self.read();
+        let mut stmt = conn.prepare("SELECT uid, trashed_at FROM trash")?;
+        let rows = stmt.query_map([], |r| Ok((r.get(0)?, r.get(1)?)))?;
+        let mut seen = HashMap::new();
+        for row in rows {
+            let (uid, at) = row?;
+            seen.insert(uid, at);
+        }
+        Ok(seen)
+    }
+
+    // ---- device sync -----------------------------------------
 }

crates/pdfs-core/src/db/utils.rs

@@ -72,18 +72,18 @@ pub(super) fn path_of(conn: &Connection, uid: &str) -> Result<String> {
     }
 }
 
-/// Resolve a path by walking `parent_uid` to the root via a recursive CTE.
+/// Resolve a path by walking `parent_lid` to the root via a recursive CTE.
 ///
-/// A cycle in `parent_uid` is corrupt data the API can still hand us, and
+/// A cycle in the parents is corrupt data the API can still hand us, and
 /// `UNION ALL` over one never terminates — the walk is therefore depth-capped
 /// and returns the truncated path rather than hanging the caller.
 pub(super) fn walk_path_of(conn: &Connection, uid: &str) -> Result<String> {
     let mut stmt = conn.prepare(&format!(
-        "WITH RECURSIVE anc(uid, parent_uid, name, depth) AS (
-           SELECT uid, parent_uid, name, 0 FROM nodes WHERE uid = ?1
+        "WITH RECURSIVE anc(parent_lid, parent_uid, name, depth) AS (
+           SELECT parent_lid, parent_uid, name, 0 FROM nodes WHERE uid = ?1
            UNION ALL
-           SELECT n.uid, n.parent_uid, n.name, anc.depth + 1
-           FROM nodes n JOIN anc ON n.uid = anc.parent_uid
+           SELECT n.parent_lid, n.parent_uid, n.name, anc.depth + 1
+           FROM nodes n JOIN anc ON n.lid = anc.parent_lid
            WHERE anc.depth < {MAX_PATH_DEPTH}
          )
          SELECT name, parent_uid FROM anc ORDER BY depth DESC"

crates/pdfs-core/src/error.rs

@@ -32,6 +32,16 @@ pub enum Error {
     #[error("saved session can no longer unlock your keys — run `pdfs login` again")]
     ReloginRequired,
 
+    /// The account has a separate mailbox password that has not been entered
+    /// since login, so its keys cannot be unlocked. The fix is
+    /// [`auth::unlock`](crate::auth::unlock), not a new login.
+    #[error("account is locked — enter the mailbox password with `pdfs unlock`")]
+    Locked,
+
+    /// The mailbox password was checked against the account and is wrong.
+    #[error("that mailbox password does not unlock your account")]
+    WrongMailboxPassword,
+
     /// The login was gated behind human verification. Not a failure the user can
     /// fix by retrying: the challenge has to be solved and the login restarted
     /// with the resulting token (see [`auth::login_verified`](crate::auth::login_verified)).
@@ -66,6 +76,20 @@ impl Error {
     }
 }
 
+impl Error {
+    /// Whether a control-socket round-trip gave up waiting for the reply: the
+    /// daemon is running but stuck, as opposed to not running at all.
+    pub fn is_timeout(&self) -> bool {
+        matches!(
+            self,
+            Error::Io(e) if matches!(
+                e.kind(),
+                std::io::ErrorKind::WouldBlock | std::io::ErrorKind::TimedOut
+            )
+        )
+    }
+}
+
 pub type Result<T> = std::result::Result<T, Error>;
 
 /// A failure on its way to a front-end: prose for the user, plus the
@@ -130,19 +154,20 @@ impl CoreError {
     /// through the boxes the drain deals in, where the concrete type survives
     /// but the static one does not.
     ///
-    /// A failure that is not an API response at all — a transport error, a
-    /// timeout, a TLS failure — is what being offline looks like from here, so
-    /// it classifies as [`ErrorKind::Offline`] rather than
-    /// [`ErrorKind::Remote`]: the request never reached Proton, and telling the
-    /// user their connection is down beats telling them the server refused.
+    /// A transport error, a timeout or a TLS failure is what being offline
+    /// looks like from here, so it classifies as [`ErrorKind::Offline`] rather
+    /// than [`ErrorKind::Remote`]: the request never reached Proton, and
+    /// telling the user their connection is down beats telling them the
+    /// server refused. The SDK's other failures say what went wrong instead
+    /// (see [`proton_kind`]).
+    ///
+    /// Callers also hand it this crate's [`Error`] from the cache and the
+    /// database, which is not a transport error at all. Taking those for
+    /// "offline" told the user their connection was down when a pin failed on
+    /// the database (`docs/BUGS.md` B266), so the source chain is searched for
+    /// either type before falling back.
     pub fn from_api(e: &(dyn std::error::Error + 'static), what: &str) -> Self {
-        let kind = match e.downcast_ref::<proton_sdk::ProtonError>() {
-            Some(proton_sdk::ProtonError::Api(api)) => api_kind(api.code),
-            // A `ProtonError` that is not an `Api` variant never reached the
-            // API; so does anything that is not a `ProtonError` at all.
-            _ => ErrorKind::Offline,
-        };
-        Self::new(kind, format!("{what}: {e}"))
+        Self::new(api_kind_of(e), format!("{what}: {e}"))
     }
 
     /// Prefix the message while keeping the classification.
@@ -184,16 +209,57 @@ impl From<serde_json::Error> for CoreError {
 
 impl From<Error> for CoreError {
     fn from(e: Error) -> Self {
-        let kind = match &e {
-            Error::NotLoggedIn | Error::ReloginRequired | Error::HumanVerificationRequired(_) => {
-                ErrorKind::Denied
-            }
-            Error::Proton(proton_sdk::ProtonError::Api(api)) => api_kind(api.code),
-            // A `ProtonError` that never reached the API is a transport failure.
-            Error::Proton(_) => ErrorKind::Offline,
-            _ => ErrorKind::Internal,
-        };
-        Self::new(kind, e.to_string())
+        Self::new(core_kind(&e), e.to_string())
+    }
+}
+
+/// What this crate's [`Error`] means to a front-end.
+fn core_kind(e: &Error) -> ErrorKind {
+    match e {
+        Error::NotLoggedIn
+        | Error::ReloginRequired
+        | Error::Locked
+        | Error::WrongMailboxPassword
+        | Error::HumanVerificationRequired(_) => ErrorKind::Denied,
+        Error::Proton(proton) => proton_kind(proton),
+        _ => ErrorKind::Internal,
+    }
+}
+
+/// The kind for [`CoreError::from_api`]: the first SDK or crate error in `e`'s
+/// source chain decides it.
+fn api_kind_of(e: &(dyn std::error::Error + 'static)) -> ErrorKind {
+    let mut cause = Some(e);
+    while let Some(err) = cause {
+        if let Some(proton) = err.downcast_ref::<proton_sdk::ProtonError>() {
+            return proton_kind(proton);
+        }
+        if let Some(core) = err.downcast_ref::<Error>() {
+            return core_kind(core);
+        }
+        cause = err.source();
+    }
+    // Not the SDK's at all: a timeout around the call, or the socket.
+    ErrorKind::Offline
+}
+
+/// What a failed SDK call means to a front-end.
+///
+/// Only a transport failure is the connection. The rest happened on this
+/// machine, and calling them "offline" sent the user to check a network that
+/// worked (issue #38: a photo upload on an account without a photo library
+/// read "You're offline").
+fn proton_kind(e: &proton_sdk::ProtonError) -> ErrorKind {
+    use proton_sdk::ProtonError as P;
+    match e {
+        P::Api(api) => api_kind(api.code),
+        P::Transport(_) => ErrorKind::Offline,
+        // Proton answered, but not with anything this client can read.
+        P::Serialization(_) => ErrorKind::Remote,
+        // The SDK refused the call before sending it, such as an upload to a
+        // photo library the account does not have.
+        P::InvalidOperation(_) => ErrorKind::Invalid,
+        P::Crypto(_) | P::KeysLocked => ErrorKind::Internal,
     }
 }
 
@@ -252,6 +318,32 @@ fn api_kind(code: proton_sdk::api::ResponseCode) -> ErrorKind {
     }
 }
 
+/// Whether `e` says the session itself is gone: its refresh token was refused,
+/// or the account was deleted or disabled.
+///
+/// Nothing but a new login helps then, and no retry will ever succeed, so a
+/// caller must not take it for a lost connection and wait for it to come back
+/// (issue #35: a revoked session was shown as "offline" for good).
+pub fn session_revoked(e: &proton_sdk::ProtonError) -> bool {
+    use proton_sdk::api::ResponseCode as C;
+    match e {
+        proton_sdk::ProtonError::Api(api) => {
+            api.is_invalid_refresh_token()
+                || matches!(api.code, C::AccountDeleted | C::AccountDisabled)
+        }
+        _ => false,
+    }
+}
+
+/// Whether `e` is Proton's temporary block on a sign-in (code 2028).
+///
+/// Its message says the app version is no longer supported, which is not what
+/// happened: the release is current and only waiting helps (issues #28, #39).
+/// A front-end must say that instead of passing the message through.
+pub fn sign_in_blocked(e: &proton_sdk::ProtonError) -> bool {
+    matches!(e, proton_sdk::ProtonError::Api(api) if api.code == proton_sdk::api::ResponseCode::Blocked)
+}
+
 /// The result every request-serving method in the daemon returns.
 pub type CoreResult<T> = std::result::Result<T, CoreError>;
 
@@ -283,6 +375,25 @@ mod tests {
         assert_eq!(api_kind(C::ProtonDriveUnknown), ErrorKind::Remote);
     }
 
+    #[test]
+    fn only_a_refused_refresh_or_a_closed_account_revokes_the_session() {
+        assert!(session_revoked(&api(C::InvalidRefreshToken)));
+        assert!(session_revoked(&api(C::AccountDeleted)));
+        assert!(session_revoked(&api(C::AccountDisabled)));
+        // One file the account may not touch says nothing about the session.
+        assert!(!session_revoked(&api(C::Forbidden)));
+        assert!(!session_revoked(&api(C::Timeout)));
+        assert!(!session_revoked(&ProtonError::KeysLocked));
+    }
+
+    #[test]
+    fn only_code_2028_is_a_blocked_sign_in() {
+        assert!(sign_in_blocked(&api(C::Blocked)));
+        assert!(!sign_in_blocked(&api(C::OutdatedApp)));
+        assert!(!sign_in_blocked(&api(C::TooManyRequests)));
+        assert!(!sign_in_blocked(&ProtonError::KeysLocked));
+    }
+
     /// The regression this whole pass exists to prevent: before it, every SDK
     /// failure was hardcoded to `Remote`, so a 403 asked the user to retry a
     /// request that could never succeed.
@@ -313,6 +424,32 @@ mod tests {
         assert_eq!(CoreError::from_api(&io, "list").kind, ErrorKind::Offline);
     }
 
+    /// Issue #38: the SDK refused a photo upload before sending it, and the
+    /// app said "You're offline" while every other call went through.
+    #[test]
+    fn a_call_the_sdk_refuses_to_send_is_not_offline() {
+        let refused = ProtonError::invalid_operation("account has no photos volume");
+        let e = CoreError::from_api(&refused, "upload photo failed");
+        assert_eq!(e.kind, ErrorKind::Invalid);
+        assert!(e.message.ends_with("account has no photos volume"));
+        let e: CoreError = Error::Proton(ProtonError::KeysLocked).into();
+        assert_ne!(e.kind, ErrorKind::Offline);
+    }
+
+    /// A pin or a search that fails on the database is not the connection, so
+    /// it must not say "You're offline" (`docs/BUGS.md` B266). An SDK error
+    /// wrapped in this crate's type classifies as the SDK error does.
+    #[test]
+    fn a_database_error_handed_to_from_api_is_not_offline() {
+        let db = Error::Other("database is locked".into());
+        assert_eq!(CoreError::from_api(&db, "pin").kind, ErrorKind::Internal);
+        let wrapped = Error::Proton(api(C::DoesNotExist));
+        assert_eq!(
+            CoreError::from_api(&wrapped, "load node").kind,
+            ErrorKind::NotFound
+        );
+    }
+
     /// `Error` is the plumbing type; converting it must classify as sharply as
     /// the direct path, or the two disagree about the same failure.
     #[test]

crates/pdfs-core/src/exiftime.rs

@@ -0,0 +1,133 @@
+//! When a photo was taken, as its own EXIF tags say.
+//!
+//! The fallback for a photo nothing else dates: a Takeout photo without a
+//! sidecar or a date in its name, and a photo uploaded from the app. Without
+//! it the first was filed under the moment of the import and the second under
+//! the file's modification time, which a copy resets.
+
+use std::io::{BufRead, Seek};
+use std::path::Path;
+
+use crate::takeout::days_from_civil;
+
+/// The capture time in the EXIF of the file at `path`, in epoch seconds.
+pub fn capture_time_of(path: &Path) -> Option<i64> {
+    let file = std::fs::File::open(path).ok()?;
+    capture_time(&mut std::io::BufReader::new(file))
+}
+
+/// The capture time in the EXIF `reader` holds: `DateTimeOriginal`, then
+/// `DateTimeDigitized`, each adjusted by its `OffsetTime…` tag when the camera
+/// wrote one. Without an offset the time is read as UTC, as a date in a file
+/// name is (`takeout::capture_time_from_name`).
+pub fn capture_time(reader: &mut (impl BufRead + Seek)) -> Option<i64> {
+    let exif = exif::Reader::new().read_from_container(reader).ok()?;
+    let ascii = |tag: exif::Tag| -> Option<String> {
+        let field = exif.get_field(tag, exif::In::PRIMARY)?;
+        match &field.value {
+            exif::Value::Ascii(parts) => parts.first().map(|part| {
+                String::from_utf8_lossy(part)
+                    .trim_matches('\0')
+                    .trim()
+                    .to_string()
+            }),
+            _ => None,
+        }
+    };
+    [
+        (exif::Tag::DateTimeOriginal, exif::Tag::OffsetTimeOriginal),
+        (exif::Tag::DateTimeDigitized, exif::Tag::OffsetTimeDigitized),
+    ]
+    .into_iter()
+    .find_map(|(time, offset)| {
+        let local = parse_exif_time(&ascii(time)?)?;
+        let offset = ascii(offset).and_then(|o| parse_offset(&o)).unwrap_or(0);
+        Some(local - offset)
+    })
+}
+
+/// `YYYY:MM:DD HH:MM:SS` as epoch seconds, read as UTC. `None` for the
+/// all-zero or blank value cameras write when they do not know.
+fn parse_exif_time(text: &str) -> Option<i64> {
+    let bytes = text.as_bytes();
+    if bytes.len() < 19 || bytes[4] != b':' || bytes[7] != b':' || bytes[13] != b':' {
+        return None;
+    }
+    let number = |range: std::ops::Range<usize>| text.get(range)?.parse::<i64>().ok();
+    let (year, month, day) = (number(0..4)?, number(5..7)?, number(8..10)?);
+    let (hour, minute, second) = (number(11..13)?, number(14..16)?, number(17..19)?);
+    if !(1800..=9999).contains(&year)
+        || !(1..=12).contains(&month)
+        || !(1..=31).contains(&day)
+        || hour > 23
+        || minute > 59
+        || second > 60
+    {
+        return None;
+    }
+    Some(days_from_civil(year, month, day) * 86_400 + hour * 3600 + minute * 60 + second)
+}
+
+/// `+HH:MM` or `-HH:MM` as seconds east of UTC.
+fn parse_offset(text: &str) -> Option<i64> {
+    let sign = match text.as_bytes().first()? {
+        b'+' => 1,
+        b'-' => -1,
+        _ => return None,
+    };
+    let (hours, minutes) = text.get(1..)?.split_once(':')?;
+    let (hours, minutes) = (hours.parse::<i64>().ok()?, minutes.parse::<i64>().ok()?);
+    (hours <= 14 && minutes < 60).then_some(sign * (hours * 3600 + minutes * 60))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn an_exif_time_reads_as_utc_and_its_offset_moves_it() {
+        // 2023-06-01 12:00:00 UTC.
+        let noon = 1_685_620_800;
+        assert_eq!(parse_exif_time("2023:06:01 12:00:00"), Some(noon));
+        assert_eq!(parse_offset("+02:00"), Some(7200));
+        assert_eq!(parse_offset("-05:30"), Some(-19_800));
+        assert_eq!(
+            parse_exif_time("2023:06:01 14:00:00").map(|t| t - parse_offset("+02:00").unwrap()),
+            Some(noon)
+        );
+    }
+
+    #[test]
+    fn an_unknown_exif_time_is_no_time() {
+        assert_eq!(parse_exif_time("0000:00:00 00:00:00"), None);
+        assert_eq!(parse_exif_time("    :  :     :  :  "), None);
+        assert_eq!(parse_exif_time(""), None);
+        assert_eq!(parse_offset("02:00"), None);
+    }
+
+    #[test]
+    fn a_photo_is_dated_by_its_original_time_and_offset() {
+        let ascii = |tag, text: &str| exif::Field {
+            tag,
+            ifd_num: exif::In::PRIMARY,
+            value: exif::Value::Ascii(vec![text.as_bytes().to_vec()]),
+        };
+        let taken = ascii(exif::Tag::DateTimeOriginal, "2023:06:01 14:00:00");
+        let offset = ascii(exif::Tag::OffsetTimeOriginal, "+02:00");
+        let digitized = ascii(exif::Tag::DateTimeDigitized, "2024:01:01 00:00:00");
+        let mut writer = exif::experimental::Writer::new();
+        writer.push_field(&taken);
+        writer.push_field(&offset);
+        writer.push_field(&digitized);
+        let mut tiff = std::io::Cursor::new(Vec::new());
+        writer.write(&mut tiff, false).unwrap();
+        tiff.set_position(0);
+        assert_eq!(capture_time(&mut tiff), Some(1_685_620_800));
+    }
+
+    #[test]
+    fn a_file_without_exif_has_no_capture_time() {
+        let mut png = std::io::Cursor::new(b"\x89PNG\r\n\x1a\n".to_vec());
+        assert_eq!(capture_time(&mut png), None);
+    }
+}

crates/pdfs-core/src/kernelkey.rs

@@ -0,0 +1,233 @@
+//! A secret held in the Linux kernel keyring, for as long as the user is logged in.
+//!
+//! This is where the key passphrases of a two-password account live by default:
+//! the mailbox password is never stored, and the secrets derived from it are
+//! kept in kernel memory — not on disk, not in the Secret Service — and vanish
+//! when the user's last session ends or the machine reboots. Both the daemon and
+//! `pdfs unlock` reach the same entry because it hangs off the *user* keyring.
+//! It is also where session tokens wait while the Secret Service refuses them
+//! ([`UNSAVED_TOKENS`]), so they outlive the process that holds them.
+//!
+//! Kernel keys are not paged to swap, are invisible to other users, and are
+//! wiped by [`clear`] (`KEYCTL_REVOKE`), which makes them a better home than a
+//! process's heap. Processes of the same user can still read them, exactly as
+//! they could ptrace the daemon: this protects against disk, backup and other
+//! users, not against code already running as you.
+//!
+//! The keyring may be unavailable (a container, a kernel built without
+//! `CONFIG_KEYS`); every operation then reports it as an error and the caller
+//! falls back to asking again, never to storing the secret somewhere weaker.
+
+use std::io;
+
+use zeroize::Zeroizing;
+
+/// One secret in the user keyring, by its description. Descriptions are
+/// namespaced so they cannot collide with another program's keys.
+pub struct Key {
+    description: &'static [u8],
+}
+
+/// The account's key passphrases (see the module docs).
+#[cfg(not(test))]
+pub const PASSPHRASES: Key = Key {
+    description: b"pdfs:key-passphrases\0",
+};
+/// Tests must never touch a real unlock.
+#[cfg(test)]
+pub const PASSPHRASES: Key = Key {
+    description: b"pdfs:key-passphrases:test\0",
+};
+
+/// Session tokens the Secret Service refused to store, until it takes them.
+/// A refresh token held only in a process's memory dies with the process.
+#[cfg(not(test))]
+pub const UNSAVED_TOKENS: Key = Key {
+    description: b"pdfs:unsaved-tokens\0",
+};
+#[cfg(test)]
+pub const UNSAVED_TOKENS: Key = Key {
+    description: b"pdfs:unsaved-tokens:test\0",
+};
+
+const KEY_TYPE: &[u8] = b"user\0";
+
+const KEY_SPEC_USER_KEYRING: i32 = -4;
+const KEYCTL_REVOKE: libc::c_long = 3;
+const KEYCTL_SETPERM: libc::c_long = 5;
+const KEYCTL_SEARCH: libc::c_long = 10;
+const KEYCTL_READ: libc::c_long = 11;
+
+/// Possessor and owner may view, read, write, search, link and set attributes;
+/// group and others get nothing.
+///
+/// A new key's default grants the owner only *view*, and read access to the
+/// possessor. A process possesses the user keyring only when its session
+/// keyring links to it, which a systemd user service or a container does not
+/// guarantee — so with the default, the daemon could store a key it then
+/// cannot read back. The owner is the same user that could ptrace the daemon
+/// anyway, so granting it read access gives up nothing.
+const PERM_OWNER_AND_POSSESSOR: u32 = 0x3f3f_0000;
+
+fn last_error<T>() -> io::Result<T> {
+    Err(io::Error::last_os_error())
+}
+
+fn find(description: &[u8]) -> io::Result<Option<libc::c_long>> {
+    // SAFETY: both strings are NUL-terminated and outlive the call.
+    let id = unsafe {
+        libc::syscall(
+            libc::SYS_keyctl,
+            KEYCTL_SEARCH,
+            KEY_SPEC_USER_KEYRING,
+            KEY_TYPE.as_ptr(),
+            description.as_ptr(),
+            0,
+        )
+    };
+    if id >= 0 {
+        return Ok(Some(id));
+    }
+    let err = io::Error::last_os_error();
+    match err.raw_os_error() {
+        // ENOKEY: not stored. EKEYREVOKED/EKEYEXPIRED: stored once, now gone.
+        Some(libc::ENOKEY | libc::EKEYREVOKED | libc::EKEYEXPIRED) => Ok(None),
+        _ => Err(err),
+    }
+}
+
+/// Stop this process from being core-dumped or ptraced by an unprivileged
+/// process of the same user, so the secrets in its memory stay out of crash
+/// reports. Call once at start of a long-lived process that holds them.
+pub fn forbid_dumps() {
+    // SAFETY: `PR_SET_DUMPABLE` takes plain integers and only affects this process.
+    let rc = unsafe { libc::prctl(libc::PR_SET_DUMPABLE, 0, 0, 0, 0) };
+    if rc != 0 {
+        tracing::warn!(error = %io::Error::last_os_error(), "could not mark the process non-dumpable");
+    }
+}
+
+impl Key {
+    /// Store `secret`, replacing any earlier one.
+    pub fn store(&self, secret: &[u8]) -> io::Result<()> {
+        // SAFETY: strings are NUL-terminated; the payload pointer/length pair
+        // describes `secret`, which outlives the call.
+        let id = unsafe {
+            libc::syscall(
+                libc::SYS_add_key,
+                KEY_TYPE.as_ptr(),
+                self.description.as_ptr(),
+                secret.as_ptr(),
+                secret.len(),
+                KEY_SPEC_USER_KEYRING,
+            )
+        };
+        if id < 0 {
+            return last_error();
+        }
+        // SAFETY: plain integer arguments.
+        let rc = unsafe {
+            libc::syscall(
+                libc::SYS_keyctl,
+                KEYCTL_SETPERM,
+                id,
+                PERM_OWNER_AND_POSSESSOR,
+            )
+        };
+        if rc < 0 {
+            // A key whose permissions are unknown is not one to leave behind.
+            let err = io::Error::last_os_error();
+            let _ = self.clear();
+            return Err(err);
+        }
+        Ok(())
+    }
+
+    /// Read the stored secret, or `None` when there is none.
+    pub fn load(&self) -> io::Result<Option<Zeroizing<Vec<u8>>>> {
+        let Some(id) = find(self.description)? else {
+            return Ok(None);
+        };
+        // Size first, then read; retry if the key was replaced in between.
+        for _ in 0..3 {
+            // SAFETY: a null buffer asks the kernel only for the payload length.
+            let len = unsafe {
+                libc::syscall(
+                    libc::SYS_keyctl,
+                    KEYCTL_READ,
+                    id,
+                    std::ptr::null_mut::<u8>(),
+                    0usize,
+                )
+            };
+            if len < 0 {
+                return match io::Error::last_os_error().raw_os_error() {
+                    Some(libc::ENOKEY | libc::EKEYREVOKED | libc::EKEYEXPIRED) => Ok(None),
+                    _ => last_error(),
+                };
+            }
+            let mut buf = Zeroizing::new(vec![0u8; len as usize]);
+            // SAFETY: `buf` is `len` writable bytes.
+            let read = unsafe {
+                libc::syscall(
+                    libc::SYS_keyctl,
+                    KEYCTL_READ,
+                    id,
+                    buf.as_mut_ptr(),
+                    buf.len(),
+                )
+            };
+            if read < 0 {
+                return last_error();
+            }
+            if read as usize <= buf.len() {
+                buf.truncate(read as usize);
+                return Ok(Some(buf));
+            }
+        }
+        Err(io::Error::other("the key changed while it was being read"))
+    }
+
+    /// Remove the stored secret. Succeeds when there is none.
+    pub fn clear(&self) -> io::Result<()> {
+        let Some(id) = find(self.description)? else {
+            return Ok(());
+        };
+        // SAFETY: plain integer arguments.
+        let rc = unsafe { libc::syscall(libc::SYS_keyctl, KEYCTL_REVOKE, id) };
+        if rc < 0 {
+            return match io::Error::last_os_error().raw_os_error() {
+                Some(libc::ENOKEY | libc::EKEYREVOKED) => Ok(()),
+                _ => last_error(),
+            };
+        }
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    // The kernel keyring is process-wide state and a sandbox may not offer it,
+    // so one test covers the whole round trip and skips when it is unavailable.
+    #[test]
+    fn stored_secret_round_trips_and_clears() {
+        if PASSPHRASES.store(b"probe").is_err() {
+            eprintln!("kernel keyring unavailable; skipping");
+            return;
+        }
+        let saved = PASSPHRASES
+            .load()
+            .unwrap()
+            .expect("stored secret is readable");
+        assert_eq!(saved.as_slice(), b"probe");
+
+        PASSPHRASES.store(b"replaced").unwrap();
+        assert_eq!(PASSPHRASES.load().unwrap().unwrap().as_slice(), b"replaced");
+
+        PASSPHRASES.clear().unwrap();
+        assert!(PASSPHRASES.load().unwrap().is_none());
+        PASSPHRASES.clear().unwrap();
+    }
+}

crates/pdfs-core/src/keycheck.rs

@@ -0,0 +1,294 @@
+//! Trace the key chain that opens My Files, for `pdfs diagnose`.
+//!
+//! The daemon only says "missing key" when the My Files share does not open
+//! (#40). This walks the chain the other way and names the link that breaks:
+//! the share's passphrase is encrypted to an address key, whose token is
+//! encrypted to an account key, which opens with the stored passphrase. Every
+//! finding holds key IDs and flags only, never key material, so the output is
+//! safe to post in an issue.
+
+use proton_sdk::account::{AddressKeyReport, KeyReport, UserKeyReport};
+use proton_sdk::ids::AddressId;
+
+use crate::auth;
+use crate::error::Result;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum Level {
+    Ok,
+    Warn,
+    Fail,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct Finding {
+    pub level: Level,
+    pub label: String,
+    pub detail: String,
+}
+
+/// The address and recipients of the My Files share passphrase, or why they
+/// could not be read.
+pub type ShareRecipients = std::result::Result<(AddressId, Vec<Option<String>>), String>;
+
+/// Resume the stored session and ask Proton for the account's keys and the
+/// My Files share. Fails when there is no session, the keys are locked, or
+/// Proton cannot be reached; reading the share is allowed to fail on its own.
+///
+/// The session's tokens are the daemon's too. A refresh here spends the
+/// refresh token the daemon holds, which it recovers from by resuming the
+/// stored session (see [`auth::stored_session_is_newer`]).
+pub async fn run() -> Result<(KeyReport, ShareRecipients)> {
+    let (client, _session) = auth::resume_client().await?;
+    let report = client.account().key_report().await?;
+    let share = client
+        .my_files_share_recipients()
+        .await
+        .map_err(|e| e.to_string());
+    Ok((report, share))
+}
+
+/// One finding per key, then one for the My Files share.
+pub fn findings(report: &KeyReport, share: &ShareRecipients) -> Vec<Finding> {
+    let mut findings = Vec::new();
+    for key in &report.user_keys {
+        findings.push(user_key_finding(key));
+    }
+    for address in &report.addresses {
+        for key in &address.keys {
+            findings.push(address_key_finding(report, &address.email, key));
+        }
+    }
+    findings.push(share_finding(report, share));
+    findings
+}
+
+fn user_key_finding(key: &UserKeyReport) -> Finding {
+    let label = format!("account key {}", short(key.id.as_str()));
+    let (level, detail) = match (key.active, key.unlocks) {
+        (true, true) => (Level::Ok, "opens".to_string()),
+        (true, false) => (
+            Level::Fail,
+            format!(
+                "does not open: {}",
+                key.error.as_deref().unwrap_or("unknown")
+            ),
+        ),
+        (false, true) => (Level::Ok, "inactive, opens with your password".to_string()),
+        (false, false) => (Level::Ok, "inactive, does not open".to_string()),
+    };
+    Finding {
+        level,
+        label,
+        detail: with_primary(key.primary, detail),
+    }
+}
+
+fn address_key_finding(report: &KeyReport, email: &str, key: &AddressKeyReport) -> Finding {
+    let label = format!("{email} key {}", short(key.id.as_str()));
+    let (level, detail) = if !key.active {
+        (Level::Ok, "inactive, not used".to_string())
+    } else if key.unlocks {
+        (Level::Ok, "opens".to_string())
+    } else {
+        // A non-primary key matters only for what was encrypted to it; the
+        // share finding says whether My Files is.
+        let level = if key.primary {
+            Level::Fail
+        } else {
+            Level::Warn
+        };
+        (level, format!("does not open: {}", why_closed(report, key)))
+    };
+    Finding {
+        level,
+        label,
+        detail: with_primary(key.primary, detail),
+    }
+}
+
+/// Why an active address key does not open, named by the account key its
+/// token was meant for when the token has one.
+fn why_closed(report: &KeyReport, key: &AddressKeyReport) -> String {
+    let error = key.error.as_deref().unwrap_or("unknown");
+    let Some(recipients) = &key.token_recipients else {
+        return error.to_string();
+    };
+    let named: Vec<&str> = recipients.iter().flatten().map(String::as_str).collect();
+    if let Some(user_key) = named.iter().find_map(|r| report.user_key_named(r)) {
+        let id = short(user_key.id.as_str());
+        return match (user_key.active, user_key.unlocks) {
+            (false, true) => {
+                format!(
+                    "its token is for inactive account key {id}, which opens with your password"
+                )
+            }
+            (false, false) => {
+                format!("its token is for inactive account key {id}, which does not open either")
+            }
+            (true, false) => format!("its token is for account key {id}, which does not open"),
+            // The token opened; the key itself refused its passphrase.
+            (true, true) => error.to_string(),
+        };
+    }
+    match named.first() {
+        Some(recipient) => {
+            format!("its token is for key {recipient}, which is none of your account keys")
+        }
+        None if recipients.is_empty() => error.to_string(),
+        None => "its token does not name the key it is for".to_string(),
+    }
+}
+
+fn share_finding(report: &KeyReport, share: &ShareRecipients) -> Finding {
+    let label = "My Files share".to_string();
+    let (level, detail) = match share {
+        Err(e) => (Level::Warn, format!("could not be read: {e}")),
+        Ok((address_id, recipients)) => {
+            let named: Vec<&str> = recipients.iter().flatten().map(String::as_str).collect();
+            match named.iter().find_map(|r| report.address_key_named(r)) {
+                Some((address, key)) => {
+                    let name = format!("{} key {}", address.email, short(key.id.as_str()));
+                    if key.unlocks {
+                        (Level::Ok, format!("encrypted to {name}, which opens"))
+                    } else {
+                        (
+                            Level::Fail,
+                            format!("encrypted to {name}, which does not open"),
+                        )
+                    }
+                }
+                None => (
+                    Level::Fail,
+                    match named.first() {
+                        Some(recipient) => format!(
+                            "encrypted to key {recipient}, which is none of the keys of address {address_id}"
+                        ),
+                        None => "its passphrase does not name the key it is for".to_string(),
+                    },
+                ),
+            }
+        }
+    };
+    Finding {
+        level,
+        label,
+        detail,
+    }
+}
+
+fn with_primary(primary: bool, detail: String) -> String {
+    if primary {
+        format!("primary, {detail}")
+    } else {
+        detail
+    }
+}
+
+/// The start of a Proton key ID: enough to tell keys apart and to find them
+/// in the daemon log, which prints the whole ID.
+fn short(id: &str) -> String {
+    match id.char_indices().nth(8) {
+        Some((end, _)) => format!("{}…", &id[..end]),
+        None => id.to_string(),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use proton_sdk::account::AddressReport;
+    use proton_sdk::ids::{AddressKeyId, UserKeyId};
+
+    fn user_key(id: &str, active: bool, unlocks: bool, pgp: &str) -> UserKeyReport {
+        UserKeyReport {
+            id: UserKeyId::new(id),
+            primary: active,
+            active,
+            unlocks,
+            error: (!unlocks).then(|| "unlock: wrong passphrase".to_string()),
+            key_ids: vec![pgp.to_string()],
+        }
+    }
+
+    fn address_key(id: &str, primary: bool, token_for: &str, pgp: &str) -> AddressKeyReport {
+        AddressKeyReport {
+            id: AddressKeyId::new(id),
+            primary,
+            active: true,
+            token_recipients: Some(vec![Some(token_for.to_string())]),
+            unlocks: token_for == "u1",
+            error: (token_for != "u1").then(|| "token: failed to decrypt: missing key".to_string()),
+            key_ids: vec![pgp.to_string()],
+        }
+    }
+
+    /// The account in #40, as far as its log shows: the primary address key
+    /// opens, an older one does not, and My Files is encrypted to the older one.
+    fn report(stranded_token_for: &str) -> KeyReport {
+        KeyReport {
+            user_keys: vec![
+                user_key("UserKeyOne-123", true, true, "u1"),
+                user_key("UserKeyTwo-456", false, true, "u2"),
+            ],
+            addresses: vec![AddressReport {
+                id: AddressId::new("address-1"),
+                email: "user@proton.me".to_string(),
+                keys: vec![
+                    address_key("AddrKeyNew-789", true, "u1", "a1"),
+                    address_key("AddrKeyOld-000", false, stranded_token_for, "a2"),
+                ],
+            }],
+        }
+    }
+
+    #[test]
+    fn a_stranded_address_key_is_traced_to_the_inactive_account_key() {
+        let report = report("u2");
+        let share = Ok((AddressId::new("address-1"), vec![Some("a2".to_string())]));
+        let findings = findings(&report, &share);
+
+        assert_eq!(findings[0].detail, "primary, opens");
+        assert_eq!(findings[1].detail, "inactive, opens with your password");
+        assert_eq!(findings[2].level, Level::Ok);
+        assert_eq!(findings[3].level, Level::Warn);
+        assert_eq!(findings[3].label, "user@proton.me key AddrKeyO…");
+        assert_eq!(
+            findings[3].detail,
+            "does not open: its token is for inactive account key UserKeyT…, which opens with your password"
+        );
+        assert_eq!(findings[4].level, Level::Fail);
+        assert_eq!(
+            findings[4].detail,
+            "encrypted to user@proton.me key AddrKeyO…, which does not open"
+        );
+    }
+
+    #[test]
+    fn a_token_for_a_key_the_account_does_not_list_says_so() {
+        let report = report("ffff");
+        let detail = &findings(&report, &Err("offline".to_string()))[3].detail;
+        assert_eq!(
+            detail,
+            "does not open: its token is for key ffff, which is none of your account keys"
+        );
+    }
+
+    #[test]
+    fn a_share_that_opens_and_one_that_cannot_be_read() {
+        let report = report("u2");
+        let open = Ok((AddressId::new("address-1"), vec![Some("a1".to_string())]));
+        let last = findings(&report, &open).pop().unwrap();
+        assert_eq!(last.level, Level::Ok);
+        assert_eq!(
+            last.detail,
+            "encrypted to user@proton.me key AddrKeyN…, which opens"
+        );
+
+        let last = findings(&report, &Err("offline".to_string()))
+            .pop()
+            .unwrap();
+        assert_eq!(last.level, Level::Warn);
+        assert_eq!(last.detail, "could not be read: offline");
+    }
+}

crates/pdfs-core/src/lib.rs

@@ -10,17 +10,26 @@ pub mod config;
 pub mod control;
 pub mod db;
 pub mod error;
+pub mod exiftime;
+pub mod kernelkey;
+pub mod keycheck;
 pub mod localindex;
 pub mod menu;
+pub mod mountfail;
 pub mod mounts;
 pub mod opener;
+pub mod places;
 pub mod profile;
 pub mod sdkcache;
 pub mod search;
 pub mod service;
 pub mod shell;
+pub mod similar;
 pub mod syncignore;
+pub mod syncroot;
 pub mod takeout;
+pub mod tray;
+pub mod update;
 
 pub use access::{Access, access_for, perm_bits};
 pub use error::{CoreError, CoreResult, Error, Result};

crates/pdfs-core/src/localindex.rs

@@ -42,7 +42,7 @@ const SKIP_PATH_SUFFIXES: &[&str] = &["go/pkg/mod"];
 
 /// Hard cap on indexed entries, so a pathological home directory cannot grow the
 /// database without bound. Reaching it stops the walk early.
-const MAX_ENTRIES: usize = 400_000;
+pub const MAX_ENTRIES: usize = 400_000;
 
 /// Rows handed to the sink at a time. Large enough that the per-transaction cost
 /// amortises, small enough that the writer lock is released often (FUSE

crates/pdfs-core/src/mountfail.rs

@@ -0,0 +1,75 @@
+//! Why the daemon's mount keeps failing, for the front ends to read.
+//!
+//! The control socket opens only once a mount has succeeded, so a daemon whose
+//! mount fails on every try had no way to say why. The app and the tray showed
+//! "Connecting…" for days while the reason sat in the journal (`docs/BUGS.md`
+//! B263). The daemon now writes its last failure to a small file in the state
+//! directory, next to the control socket, and removes it once a mount is up.
+
+use std::path::{Path, PathBuf};
+
+use serde::{Deserialize, Serialize};
+
+/// The file's name inside the state directory.
+const FILE_NAME: &str = "mount-failure.json";
+
+/// The last failed mount attempt.
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
+pub struct MountFailure {
+    /// The error, in English, as the journal has it.
+    pub error: String,
+    /// Unix seconds of the first failure in this run of failures.
+    pub since: i64,
+    /// Failed attempts in a row.
+    pub attempts: u32,
+    /// Unix seconds of the next attempt.
+    pub next_attempt: i64,
+}
+
+/// Where the failure is kept for the state directory `state_dir`.
+pub fn path(state_dir: &Path) -> PathBuf {
+    state_dir.join(FILE_NAME)
+}
+
+/// Record `failure`, replacing the one before it. Written to a temporary file
+/// and renamed, so a reader never sees half of it.
+pub fn record(state_dir: &Path, failure: &MountFailure) -> std::io::Result<()> {
+    let json = serde_json::to_vec(failure).map_err(std::io::Error::other)?;
+    let tmp = state_dir.join(format!("{FILE_NAME}.tmp"));
+    std::fs::write(&tmp, json)?;
+    std::fs::rename(&tmp, path(state_dir))
+}
+
+/// Forget the failure: a mount is up, or a new daemon has started.
+pub fn clear(state_dir: &Path) {
+    let _ = std::fs::remove_file(path(state_dir));
+}
+
+/// The recorded failure, if there is one and it can be read.
+pub fn read(state_dir: &Path) -> Option<MountFailure> {
+    let bytes = std::fs::read(path(state_dir)).ok()?;
+    serde_json::from_slice(&bytes).ok()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn a_recorded_failure_reads_back_until_cleared() {
+        let dir = std::env::temp_dir().join(format!("pdfs-mountfail-{}", std::process::id()));
+        std::fs::create_dir_all(&dir).unwrap();
+        assert_eq!(read(&dir), None);
+        let failure = MountFailure {
+            error: "fetch My Files root: missing key".into(),
+            since: 1_700_000_000,
+            attempts: 3,
+            next_attempt: 1_700_000_040,
+        };
+        record(&dir, &failure).unwrap();
+        assert_eq!(read(&dir), Some(failure));
+        clear(&dir);
+        assert_eq!(read(&dir), None);
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+}

crates/pdfs-core/src/mounts.rs

@@ -110,6 +110,9 @@ pub struct MountSpec {
     /// Live device sync progress, absent when no pass is running.
     #[serde(default)]
     pub progress: Option<SyncProgress>,
+    /// A synced folder paused on its own. Live daemon state, not a stored column.
+    #[serde(default)]
+    pub paused: bool,
 }
 
 #[cfg(test)]

crates/pdfs-core/src/opener.rs

@@ -189,11 +189,21 @@ fn spawn(argv: &[String], path: &Path) {
         .stderr(Stdio::null())
         .process_group(0)
         .spawn();
-    if let Err(e) = spawned {
-        tracing::error!(command = ?argv, "open failed: {e}");
+    match spawned {
+        Ok(child) => reap(child),
+        Err(e) => tracing::error!(command = ?argv, "open failed: {e}"),
     }
 }
 
+/// Wait for `child` on a thread of its own. A child nobody waits for stays a
+/// zombie for as long as its parent lives, which for the app and the tray is
+/// the whole session.
+pub fn reap(mut child: std::process::Child) {
+    std::thread::spawn(move || {
+        let _ = child.wait();
+    });
+}
+
 /// [`open`] with the policy from the user's `config.json`, read once per
 /// process. For callers that open the occasional path and have no reason to
 /// carry a policy around; long-lived windows should hold their own copy.
@@ -472,6 +482,21 @@ fn glob_match(pattern: &str, name: &str) -> bool {
 mod tests {
     use super::*;
 
+    #[test]
+    fn a_reaped_child_does_not_stay_a_zombie() {
+        let child = std::process::Command::new("true").spawn().unwrap();
+        let proc_entry = std::path::PathBuf::from(format!("/proc/{}", child.id()));
+        reap(child);
+        let deadline = std::time::Instant::now() + std::time::Duration::from_secs(5);
+        while proc_entry.exists() && std::time::Instant::now() < deadline {
+            std::thread::sleep(std::time::Duration::from_millis(10));
+        }
+        assert!(
+            !proc_entry.exists(),
+            "the child is still in the process table"
+        );
+    }
+
     fn rule(patterns: &[&str], command: &[&str], terminal: bool) -> OpenRule {
         OpenRule {
             patterns: patterns.iter().map(|p| (*p).to_string()).collect(),

crates/pdfs-core/src/places.rs

@@ -0,0 +1,185 @@
+//! Offline reverse geocoding: the town a photo was taken in, from its GPS
+//! position, without asking any service.
+//!
+//! The towns are GeoNames' `cities15000` dump (every place with more than
+//! 15,000 people), trimmed to id, name, position and country in
+//! `data/cities15000.tsv` by `scripts/geonames-cities.py`. The script drops the
+//! parts of a city, such as Kreuzberg or "Paris 15 Vaugirard", so that a photo
+//! lands in the town people would name. GeoNames publishes the dump under
+//! CC BY 4.0, which is why the About dialog credits it.
+//!
+//! `data/countries.tsv` maps each country code to its ISO 3166-1 English name,
+//! as the `iso-codes` project spells it, so a front end can translate it
+//! through that project's `iso_3166-1` gettext domain.
+
+use std::collections::HashMap;
+use std::sync::OnceLock;
+
+/// How far from the nearest town a photo may be and still count as taken
+/// there, in km. Farther out (at sea, in the wilderness) it has no place.
+pub const MAX_DISTANCE_KM: f64 = 50.0;
+
+/// Mean Earth radius in km, for the haversine distance.
+const EARTH_RADIUS_KM: f64 = 6371.0;
+
+const CITIES: &str = include_str!("../data/cities15000.tsv");
+const COUNTRIES: &str = include_str!("../data/countries.tsv");
+
+/// One town of the table.
+#[derive(Debug, Clone, PartialEq)]
+pub struct City {
+    /// The GeoNames id, stable across dumps.
+    pub id: u32,
+    pub name: &'static str,
+    pub latitude: f64,
+    pub longitude: f64,
+    /// ISO 3166-1 alpha-2 country code.
+    pub country: &'static str,
+}
+
+/// The towns, and an index of them by whole-degree cell.
+struct Table {
+    cities: Vec<City>,
+    cells: HashMap<(i32, i32), Vec<usize>>,
+}
+
+fn table() -> &'static Table {
+    static TABLE: OnceLock<Table> = OnceLock::new();
+    TABLE.get_or_init(|| {
+        let cities: Vec<City> = CITIES.lines().filter_map(parse_city).collect();
+        let mut cells: HashMap<(i32, i32), Vec<usize>> = HashMap::new();
+        for (index, city) in cities.iter().enumerate() {
+            cells
+                .entry(cell(city.latitude, city.longitude))
+                .or_default()
+                .push(index);
+        }
+        Table { cities, cells }
+    })
+}
+
+fn parse_city(line: &'static str) -> Option<City> {
+    let mut fields = line.split('\t');
+    Some(City {
+        id: fields.next()?.parse().ok()?,
+        name: fields.next()?,
+        latitude: fields.next()?.parse().ok()?,
+        longitude: fields.next()?.parse().ok()?,
+        country: fields.next()?,
+    })
+}
+
+fn cell(latitude: f64, longitude: f64) -> (i32, i32) {
+    (latitude.floor() as i32, longitude.floor() as i32)
+}
+
+/// Great-circle distance between two positions, in km.
+pub fn distance_km(a: (f64, f64), b: (f64, f64)) -> f64 {
+    let (lat1, lat2) = (a.0.to_radians(), b.0.to_radians());
+    let dlat = lat2 - lat1;
+    let dlon = (b.1 - a.1).to_radians();
+    let h = (dlat / 2.0).sin().powi(2) + lat1.cos() * lat2.cos() * (dlon / 2.0).sin().powi(2);
+    2.0 * EARTH_RADIUS_KM * h.sqrt().min(1.0).asin()
+}
+
+/// The town nearest to a position, if one lies within [`MAX_DISTANCE_KM`].
+pub fn nearest(latitude: f64, longitude: f64) -> Option<&'static City> {
+    if !latitude.is_finite() || !longitude.is_finite() {
+        return None;
+    }
+    let table = table();
+    // A degree of latitude is about 111 km everywhere; a degree of longitude
+    // shrinks towards the poles, so more cells are needed to cover the radius.
+    let lat_span = (MAX_DISTANCE_KM / 111.0).ceil() as i32;
+    let lon_span = {
+        let scale = latitude.to_radians().cos().abs().max(0.01);
+        ((MAX_DISTANCE_KM / (111.0 * scale)).ceil() as i32).min(180)
+    };
+    let (row, column) = cell(latitude, longitude);
+    let mut best: Option<(f64, &City)> = None;
+    for dlat in -lat_span..=lat_span {
+        for dlon in -lon_span..=lon_span {
+            // Longitude wraps at the antimeridian: -181 is 179.
+            let lon_cell = (column + dlon + 180).rem_euclid(360) - 180;
+            let Some(indices) = table.cells.get(&(row + dlat, lon_cell)) else {
+                continue;
+            };
+            for &index in indices {
+                let city = &table.cities[index];
+                let d = distance_km((latitude, longitude), (city.latitude, city.longitude));
+                if d <= MAX_DISTANCE_KM && best.is_none_or(|(bd, _)| d < bd) {
+                    best = Some((d, city));
+                }
+            }
+        }
+    }
+    best.map(|(_, city)| city)
+}
+
+/// The town with this GeoNames id.
+pub fn city(id: u32) -> Option<&'static City> {
+    let cities = &table().cities;
+    // The table is sorted by id.
+    cities
+        .binary_search_by_key(&id, |city| city.id)
+        .ok()
+        .map(|index| &cities[index])
+}
+
+/// The English ISO 3166-1 name of a country code, such as "Germany" for "DE".
+pub fn country_name(code: &str) -> Option<&'static str> {
+    COUNTRIES.lines().find_map(|line| {
+        let (c, name) = line.split_once('\t')?;
+        (c == code).then_some(name)
+    })
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn a_position_in_a_town_resolves_to_that_town() {
+        let berlin = nearest(52.52, 13.405).expect("Berlin is a town");
+        assert_eq!(berlin.name, "Berlin");
+        assert_eq!(berlin.country, "DE");
+        assert_eq!(city(berlin.id), Some(berlin));
+    }
+
+    #[test]
+    fn the_open_sea_has_no_town() {
+        assert_eq!(nearest(0.0, -30.0), None);
+        assert_eq!(nearest(f64::NAN, 0.0), None);
+    }
+
+    #[test]
+    fn a_town_across_the_antimeridian_is_still_found() {
+        // Just east of 180°, the nearest towns in Fiji are west of it.
+        let town = nearest(-16.8, 179.99);
+        assert!(town.is_none_or(|t| t.country == "FJ"));
+        let suva = nearest(-18.14, 178.44).expect("Suva is a town");
+        assert_eq!(suva.country, "FJ");
+    }
+
+    #[test]
+    fn every_town_has_a_named_country() {
+        for city in &table().cities {
+            assert!(
+                country_name(city.country).is_some(),
+                "{} has no country name for {}",
+                city.name,
+                city.country
+            );
+        }
+        assert_eq!(country_name("DE"), Some("Germany"));
+    }
+
+    #[test]
+    fn distances_are_great_circle_km() {
+        let d = distance_km((52.52, 13.405), (48.137, 11.575));
+        assert!(
+            (d - 504.0).abs() < 5.0,
+            "Berlin to Munich is about 504 km, got {d}"
+        );
+    }
+}

crates/pdfs-core/src/profile.rs

@@ -1,4 +1,4 @@
-//! The machine profile that travels with a device (features.md 5.3).
+//! The machine profile that travels with a device.
 //!
 //! Sync folder mappings, pins, and the settings that shape them live in
 //! `cache.db` and `config.json` — both on the local disk. When that disk dies,
@@ -19,9 +19,31 @@ use serde::{Deserialize, Serialize};
 
 use crate::mounts::MountMode;
 
-/// File name of the profile.
+/// File name of the profile clients before 3.5.0 wrote, one per device.
+///
+/// Read, never written: two computers sharing a device each overwrote the
+/// other's. See [`profile_file_name`].
 pub const PROFILE_FILE_NAME: &str = "profile.json";
 
+/// File name of one machine's profile: `profile-<machine id>.json`.
+///
+/// One per machine, because several computers may sync into one device, and
+/// each describes its own arrangement. A single shared file went to whichever
+/// computer saved last.
+pub fn profile_file_name(machine_id: &str) -> String {
+    format!("profile-{machine_id}.json")
+}
+
+/// Whether `name` is a profile: [`PROFILE_FILE_NAME`] or one machine's
+/// [`profile_file_name`].
+pub fn is_profile_file_name(name: &str) -> bool {
+    name == PROFILE_FILE_NAME
+        || name
+            .strip_prefix("profile-")
+            .and_then(|rest| rest.strip_suffix(".json"))
+            .is_some_and(|id| !id.is_empty() && id.bytes().all(|b| b.is_ascii_alphanumeric()))
+}
+
 /// Folder, directly under the device root, that holds the profile.
 ///
 /// The API refuses `Cannot create file at the root of a device` (HTTP 422), so
@@ -74,6 +96,10 @@ pub struct Profile {
     /// Hostname of the machine that wrote it, so a restore prompt can say
     /// *"restore settings from laptop-xyz?"* rather than quoting a uid.
     pub hostname: String,
+    /// The writing machine's id, as in its [`profile_file_name`]. `None` in a
+    /// profile written before 3.5.0.
+    #[serde(default)]
+    pub machine_id: Option<String>,
     /// When it was written, epoch seconds.
     pub saved_at: i64,
     #[serde(default)]
@@ -92,6 +118,25 @@ pub struct Profile {
     pub mountpoint: Option<String>,
 }
 
+/// The saved arrangement of the folder with uid `remote_uid`, from the
+/// profiles of every computer that syncs the device.
+///
+/// The profile of the machine named `hostname` wins, since it describes this
+/// computer; otherwise the most recently saved one that has the folder.
+pub fn saved_folder<'a>(
+    profiles: &'a [Profile],
+    remote_uid: &str,
+    hostname: &str,
+) -> Option<&'a ProfileFolder> {
+    let holding = |p: &&'a Profile| p.folders.iter().any(|f| f.remote_uid == remote_uid);
+    let profile = profiles
+        .iter()
+        .filter(holding)
+        .find(|p| p.hostname == hostname)
+        .or_else(|| profiles.iter().filter(holding).max_by_key(|p| p.saved_at))?;
+    profile.folders.iter().find(|f| f.remote_uid == remote_uid)
+}
+
 impl Profile {
     /// Parse a profile document, refusing one written by a newer client.
     pub fn parse(bytes: &[u8]) -> Result<Self, String> {
@@ -122,6 +167,7 @@ mod tests {
             version: PROFILE_VERSION,
             device_uid: "dev-1".to_string(),
             hostname: "laptop-xyz".to_string(),
+            machine_id: Some("0123456789abcdef0123456789abcdef".to_string()),
             saved_at: 1_700_000_000,
             folders: vec![ProfileFolder {
                 remote_uid: "vol~link".to_string(),
@@ -219,6 +265,52 @@ mod tests {
         assert!(p.ignore_patterns.is_none());
     }
 
+    #[test]
+    fn a_profile_name_is_the_legacy_one_or_one_per_machine() {
+        assert!(is_profile_file_name(PROFILE_FILE_NAME));
+        let name = profile_file_name("0123456789abcdef0123456789abcdef");
+        assert_eq!(name, "profile-0123456789abcdef0123456789abcdef.json");
+        assert!(is_profile_file_name(&name));
+        assert!(!is_profile_file_name("profile-.json"));
+        assert!(!is_profile_file_name("profile-a/b.json"));
+        assert!(!is_profile_file_name("profile (1).json"));
+        assert!(!is_profile_file_name("notes.json"));
+    }
+
+    /// A profile written before 3.5.0 has no machine id and still parses.
+    #[test]
+    fn a_profile_without_a_machine_id_parses() {
+        let json = br#"{"version":1,"device_uid":"d","hostname":"h","saved_at":0}"#;
+        assert_eq!(Profile::parse(json).unwrap().machine_id, None);
+    }
+
+    /// Several computers sync one device, each with its own profile. This
+    /// computer's own wins, then the newest that has the folder.
+    #[test]
+    fn the_saved_folder_comes_from_this_computer_then_the_newest_profile() {
+        let profile = |hostname: &str, saved_at: i64, path: &str| Profile {
+            hostname: hostname.to_string(),
+            saved_at,
+            folders: vec![ProfileFolder {
+                remote_uid: "vol~docs".to_string(),
+                local_path: path.to_string(),
+                mode: MountMode::Mirror,
+            }],
+            ..sample()
+        };
+        let profiles = vec![
+            profile("desktop", 10, "/home/a/Docs"),
+            profile("laptop", 5, "/home/b/Docs"),
+            profile("old", 1, "/home/c/Docs"),
+        ];
+        let path =
+            |host: &str| saved_folder(&profiles, "vol~docs", host).map(|f| f.local_path.as_str());
+        assert_eq!(path("laptop"), Some("/home/b/Docs"));
+        assert_eq!(path("new-machine"), Some("/home/a/Docs"));
+        assert_eq!(saved_folder(&profiles, "vol~other", "laptop"), None);
+        assert_eq!(saved_folder(&[], "vol~docs", "laptop"), None);
+    }
+
     #[test]
     fn future_version_is_refused() {
         let mut p = sample();

crates/pdfs-core/src/service.rs

@@ -55,9 +55,10 @@ fn systemctl_query(args: &[&str]) -> bool {
 /// keyring checks). `--now` is a no-op on an active unit, so the daemon would
 /// keep sleeping and the mount would not appear until the next poll fired.
 /// Restarting makes the new session take effect immediately, and starts the
-/// unit when it is not running at all.
+/// unit when it is not running at all. `reenable` for the reason given on
+/// [`enable`].
 pub fn enable_start() -> bool {
-    let enabled = systemctl(&["enable"]);
+    let enabled = systemctl(&["reenable"]);
     systemctl(&["restart"]) && enabled
 }
 
@@ -97,8 +98,14 @@ pub fn is_enabled() -> bool {
 /// Enable the service so it auto-starts on future logins, without necessarily
 /// starting it now beyond what `enable` does. Used by the Settings "Start on
 /// login" switch; pairs with [`disable`](disable).
+///
+/// `reenable` rather than `enable`: older installs were enabled
+/// into `default.target.wants`, and a plain `enable` would add the new
+/// `graphical-session.target.wants` link while leaving the old one to start the
+/// unit too early. `reenable` drops every existing link and recreates them from
+/// the current `[Install]` section.
 pub fn enable() -> bool {
-    systemctl(&["enable"])
+    systemctl(&["reenable"])
 }
 
 /// Disable auto-start without stopping the running mount. Used by the Settings

crates/pdfs-core/src/shell.rs

@@ -23,6 +23,11 @@ const SCRIPTS: [(&str, &str); 2] = [
 /// `NAUTILUS_SCRIPT_SELECTED_FILE_PATHS`; the positional args carry the same
 /// selection when the script runs on a local (non-URI) folder, so we prefer the
 /// environment and fall back to `"$@"` for the Nemo/Caja variants.
+///
+/// The `--` keeps a file named like an option, such as `--json`, a path. A
+/// name with a newline in it still arrives split in two from Nautilus, whose
+/// list is newline-separated; `pdfs` then fails on both halves, which touches
+/// nothing.
 fn script_body(subcommand: &str) -> String {
     format!(
         "#!/bin/sh\n\
@@ -41,7 +46,7 @@ fn script_body(subcommand: &str) -> String {
          \n\
          printf '%s' \"$paths\" | while IFS= read -r path; do\n\
          \x20   [ -n \"$path\" ] || continue\n\
-         \x20   pdfs {subcommand} \"$path\" || exit 1\n\
+         \x20   pdfs {subcommand} -- \"$path\" || exit 1\n\
          done\n"
     )
 }
@@ -95,3 +100,13 @@ fn write_scripts(dir: &PathBuf) -> io::Result<()> {
     tracing::debug!("installed file-manager scripts in {dir:?}");
     Ok(())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::script_body;
+
+    #[test]
+    fn the_script_ends_options_before_each_path() {
+        assert!(script_body("pin").contains(r#"pdfs pin -- "$path""#));
+    }
+}

crates/pdfs-core/src/similar.rs

@@ -0,0 +1,193 @@
+//! Telling photos apart by what they show rather than by their bytes.
+//!
+//! The server's content hash only matches byte-identical files. A photo that
+//! was resized, re-encoded or sent through a messenger is another file, so the
+//! similar-photo finder hashes each thumbnail instead: a 64-bit difference
+//! hash ("dHash"), which compares neighbouring brightness in a 9×8 grid. Two
+//! copies of one picture land a few bits apart, and different pictures about
+//! half of the bits apart.
+
+/// Hashes this many bits apart or fewer show the same picture.
+pub const SIMILAR_BITS: u32 = 6;
+
+/// Stored for a photo whose thumbnail was looked at and gave no usable hash
+/// (it couldn't be decoded, or is one flat colour), so it isn't looked at
+/// again. A real hash is never 0: a picture that would hash to 0 has no
+/// brightness changes, which [`difference_hash`] turns down as flat.
+pub const NO_HASH: u64 = 0;
+
+/// Samples of the grid that differ by less than this in brightness, all of
+/// them, make a flat picture — a black frame, a blank page — whose hash is
+/// noise and would match every other flat picture.
+const FLAT: u8 = 8;
+
+/// The difference hash of an 8-bit greyscale image `width` px wide, rows
+/// stored one after another in `luma`. `None` for an empty or flat image.
+pub fn difference_hash(width: usize, height: usize, luma: &[u8]) -> Option<u64> {
+    if width == 0 || height == 0 || luma.len() < width * height {
+        return None;
+    }
+    // Box-average down to 9×8, so every source pixel counts once and the hash
+    // doesn't depend on which pixels a nearest-neighbour pick happens to hit.
+    let mut grid = [[0u8; 9]; 8];
+    for (gy, row) in grid.iter_mut().enumerate() {
+        let (y0, y1) = span(gy, 8, height);
+        for (gx, cell) in row.iter_mut().enumerate() {
+            let (x0, x1) = span(gx, 9, width);
+            let mut sum = 0u64;
+            for y in y0..y1 {
+                sum += luma[y * width + x0..y * width + x1]
+                    .iter()
+                    .map(|&v| u64::from(v))
+                    .sum::<u64>();
+            }
+            *cell = (sum / ((y1 - y0) * (x1 - x0)) as u64) as u8;
+        }
+    }
+    let (min, max) = grid
+        .iter()
+        .flatten()
+        .fold((u8::MAX, 0), |(min, max), &v| (min.min(v), max.max(v)));
+    if max - min < FLAT {
+        return None;
+    }
+    let mut hash = 0u64;
+    for row in &grid {
+        for pair in row.windows(2) {
+            hash = hash << 1 | u64::from(pair[0] < pair[1]);
+        }
+    }
+    (hash != NO_HASH).then_some(hash)
+}
+
+/// The source pixels `from..to` that cell `i` of `cells` covers along an edge
+/// of `len` px; never empty, even for an edge shorter than the grid.
+fn span(i: usize, cells: usize, len: usize) -> (usize, usize) {
+    let from = (i * len / cells).min(len - 1);
+    let to = ((i + 1) * len / cells).max(from + 1);
+    (from, to)
+}
+
+/// Whether two hashes show the same picture.
+pub fn alike(a: u64, b: u64) -> bool {
+    (a ^ b).count_ones() <= SIMILAR_BITS
+}
+
+/// Group `hashes` into sets of alike ones, by index, each set in index order
+/// and the sets by their first index. A hash alike to none other is in no
+/// set. Alike is taken as catching: a burst of shots each a little different
+/// from the last is one set.
+pub fn sets(hashes: &[u64]) -> Vec<Vec<usize>> {
+    // Two hashes at most SIMILAR_BITS apart agree on at least one of their
+    // eight bytes, as long as SIMILAR_BITS < 8. So only hashes sharing some
+    // byte are compared, not every pair of the library.
+    const _: () = assert!(SIMILAR_BITS < 8);
+    let mut parent: Vec<usize> = (0..hashes.len()).collect();
+    fn root(parent: &mut [usize], mut i: usize) -> usize {
+        while parent[i] != i {
+            parent[i] = parent[parent[i]];
+            i = parent[i];
+        }
+        i
+    }
+    for byte in 0..8 {
+        let mut buckets: Vec<Vec<usize>> = vec![Vec::new(); 256];
+        for (i, &hash) in hashes.iter().enumerate() {
+            buckets[(hash >> (byte * 8)) as u8 as usize].push(i);
+        }
+        for bucket in &buckets {
+            for (n, &a) in bucket.iter().enumerate() {
+                for &b in &bucket[n + 1..] {
+                    if alike(hashes[a], hashes[b]) {
+                        let (ra, rb) = (root(&mut parent, a), root(&mut parent, b));
+                        if ra != rb {
+                            parent[ra.max(rb)] = ra.min(rb);
+                        }
+                    }
+                }
+            }
+        }
+    }
+    let mut by_root: std::collections::BTreeMap<usize, Vec<usize>> = Default::default();
+    for i in 0..hashes.len() {
+        let r = root(&mut parent, i);
+        by_root.entry(r).or_default().push(i);
+    }
+    by_root.into_values().filter(|set| set.len() > 1).collect()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// A 90×80 picture: a bright disc on a gradient.
+    fn picture(width: usize, height: usize) -> Vec<u8> {
+        let mut luma = Vec::with_capacity(width * height);
+        for y in 0..height {
+            for x in 0..width {
+                let (fx, fy) = (x as f64 / width as f64, y as f64 / height as f64);
+                let disc = ((fx - 0.3).powi(2) + (fy - 0.6).powi(2)).sqrt() < 0.2;
+                let v = if disc { 240.0 } else { 40.0 + 120.0 * fx * fy };
+                luma.push(v as u8);
+            }
+        }
+        luma
+    }
+
+    #[test]
+    fn a_resized_copy_hashes_alike() {
+        let big = difference_hash(900, 800, &picture(900, 800)).unwrap();
+        let small = difference_hash(91, 79, &picture(91, 79)).unwrap();
+        assert!(
+            alike(big, small),
+            "{} bits apart",
+            (big ^ small).count_ones()
+        );
+    }
+
+    #[test]
+    fn a_different_picture_hashes_apart() {
+        let one = difference_hash(90, 80, &picture(90, 80)).unwrap();
+        let mirrored: Vec<u8> = picture(90, 80)
+            .chunks(90)
+            .flat_map(|row| row.iter().rev().copied().collect::<Vec<_>>())
+            .collect();
+        let other = difference_hash(90, 80, &mirrored).unwrap();
+        assert!(!alike(one, other));
+    }
+
+    #[test]
+    fn a_flat_picture_has_no_hash() {
+        assert_eq!(difference_hash(64, 64, &[17; 64 * 64]), None);
+        assert_eq!(difference_hash(0, 0, &[]), None);
+        assert_eq!(difference_hash(4, 4, &[0; 3]), None);
+    }
+
+    #[test]
+    fn a_picture_smaller_than_the_grid_still_hashes() {
+        let luma: Vec<u8> = (0..4 * 3).map(|i| (i * 20) as u8).collect();
+        assert!(difference_hash(4, 3, &luma).is_some());
+    }
+
+    #[test]
+    fn alike_hashes_form_sets_and_catch_on() {
+        let a = 0x0123_4567_89ab_cdef_u64;
+        let hashes = [
+            a,
+            !a,
+            a ^ 0b111,
+            a ^ 0b111 ^ (0b1111 << 40),
+            0x5555_0000_ffff_aaaa,
+        ];
+        assert_eq!(sets(&hashes), vec![vec![0, 2, 3]]);
+    }
+
+    #[test]
+    fn hashes_apart_in_most_bytes_are_still_found() {
+        // One bit apart in each of six bytes: six bits in all, and only the
+        // top two bytes agree.
+        let a = 1u64 << 63;
+        let b = (0..6).fold(a, |h, byte| h | 1 << (byte * 8));
+        assert_eq!(sets(&[a, b]), vec![vec![0, 1]]);
+    }
+}

crates/pdfs-core/src/syncignore.rs

@@ -1,4 +1,4 @@
-//! Selective-sync ignore rules for mirror folders (features.md 1.1).
+//! Selective-sync ignore rules for mirror folders.
 //!
 //! A mirror folder otherwise uploads every path beneath its root, which for a
 //! project directory means `node_modules/`, `.git/`, and `target/` — quota
@@ -45,18 +45,30 @@ pub const DEFAULT_IGNORE_PATTERNS: &[&str] = &[
     "Thumbs.db",
 ];
 
-/// Suffixes of files that are, by convention, incomplete or throwaway while they
-/// wear the name: a download in flight, an editor's swap or backup, an atomic
-/// write's scratch copy. On an on-demand mount these should never seal a remote
-/// revision — the app renames the finished file to its real name, and only that
-/// belongs on Drive. See [`is_transient_name`] and docs/BUGS.md B70.
+/// Suffixes that mean a file is incomplete *by construction*: only the program
+/// writing it gives it this name, only while the write is in flight, and the
+/// finished article always arrives under a different one. On an on-demand mount
+/// these should never seal a remote revision — the app renames the finished file
+/// to its real name, and only that belongs on Drive. See [`is_transient_name`]
+/// and docs/BUGS.md B70.
+///
+/// `.tmp`/`.temp` are deliberately **not** here, though they are in
+/// [`DEFAULT_IGNORE_PATTERNS`]. The two lists answer different questions. Skipping
+/// a `report.tmp` during a mirror reconcile costs nothing — the file stays on
+/// disk, and a later pass picks it up if the rules change. Parking one on the
+/// on-demand mount is irreversible in practice: the only exit is a rename to a
+/// finished name, and a file that is simply *called* `index.tmp` — a static-site
+/// template, a tool's output — never gets one. It then never reaches Drive at
+/// all, silently, while its bytes pin `staging/`. That is what a real Takeout
+/// tree of `index.tmp`/`preview.tmp` gallery templates did: 43 creates parked
+/// forever. The generic-scratch case they were added for renames within
+/// milliseconds, so it never needed the park; the download suffixes below, which
+/// stay in flight for minutes, are what B70 was actually about.
 const TRANSIENT_SUFFIXES: &[&str] = &[
     ".crdownload", // Chromium/Brave/Edge
     ".part",       // Firefox, wget, aria2c
     ".partial",    // some downloaders / older IE
     ".download",   // Safari / macOS
-    ".tmp",        // generic scratch
-    ".temp",       // generic scratch
     ".swp",        // vim swap
     ".swx",        // vim swap (secondary)
 ];
@@ -84,6 +96,26 @@ pub fn is_transient_name(name: &str) -> bool {
     false
 }
 
+/// Whether `name` is scratch another program writes and throws away: the
+/// [`is_transient_name`] set, plus the save-by-rename temporaries of GTK, Qt,
+/// Chromium and Electron, generic `.tmp` files and Emacs lock files. None of
+/// them is a file the user made, so a front-end can hide them from a history
+/// of what happened to the user's files.
+pub fn is_scratch_name(name: &str) -> bool {
+    if is_transient_name(name) {
+        return true;
+    }
+    let lower = name.to_ascii_lowercase();
+    lower.ends_with(".tmp")
+        || lower.ends_with(".temp")
+        || name.starts_with(".goutputstream-")
+        || name.starts_with(".org.chromium.")
+        || name.starts_with(".com.google.chrome.")
+        || name.starts_with(".#")
+        || (name.starts_with('#') && name.ends_with('#') && name.len() > 1)
+        || (name.starts_with('.') && name.contains(".tmp."))
+}
+
 /// Filenames accepted as a folder's ignore file, in precedence order.
 const IGNORE_FILE_NAMES: &[&str] = &[".pdfsignore", ".protonignore"];
 
@@ -319,8 +351,6 @@ mod tests {
             "movie.mkv.part",
             "big.iso.partial",
             "photo.jpg.download",
-            "export.xml.tmp",
-            "notes.TEMP",
             ".notes.txt.swp",
             "doc.txt~",
             ".~lock.report.odt#",
@@ -331,6 +361,25 @@ mod tests {
         }
     }
 
+    #[test]
+    fn save_by_rename_temporaries_are_scratch() {
+        for name in [
+            ".goutputstream-4XK2B2",
+            ".org.chromium.Chromium.aBc123",
+            "~$report.docx",
+            ".~lock.budget.ods#",
+            ".#notes.org",
+            "#notes.org#",
+            "cache.tmp",
+            "page.html.crdownload",
+        ] {
+            assert!(is_scratch_name(name), "{name}");
+        }
+        for name in ["report.docx", "notes.org", "#1 hits.txt", ".bashrc"] {
+            assert!(!is_scratch_name(name), "{name}");
+        }
+    }
+
     #[test]
     fn finished_names_are_not_transient() {
         for name in [
@@ -340,6 +389,10 @@ mod tests {
             "archive.partly.zip", // ".partly" is not ".part"
             "template.dotx",
             "readme",
+            // A file that is *named* `.tmp` and never renamed: parking it would
+            // keep it off Drive for good.
+            "index.tmp",
+            "notes.TEMP",
         ] {
             assert!(!is_transient_name(name), "{name} should not be transient");
         }

crates/pdfs-core/src/syncroot.rs

@@ -0,0 +1,209 @@
+//! Which local folders may become a sync root.
+//!
+//! A mirror uploads everything under its root and deletes what Drive deleted,
+//! so a root that holds more than the user meant is dangerous. `$HOME` takes
+//! the decrypted content cache, `cache.db` and `~/.ssh` up to Drive, and walks
+//! into the mount, which downloads the whole Drive only to upload it again.
+//! Every place a root is added or restored goes through [`problem`].
+
+use std::path::{Component, Path, PathBuf};
+
+use crate::config::{AppConfig, AppDirs};
+
+/// The paths a sync root must stay clear of.
+#[derive(Debug, Clone, Default)]
+pub struct Reserved {
+    /// The user's home folder. A root may sit inside it, not be it or above it.
+    pub home: Option<PathBuf>,
+    /// The Proton Drive mount.
+    pub mountpoint: Option<PathBuf>,
+    /// This app's own configuration, state and cache folders.
+    pub app_dirs: Vec<PathBuf>,
+}
+
+impl Reserved {
+    /// The reserved paths of this user and `config`, resolved like a root.
+    pub fn current(config: &AppConfig) -> Self {
+        let Ok(dirs) = AppDirs::new() else {
+            return Self::default();
+        };
+        let resolved = |path: PathBuf| resolve(&path).unwrap_or(path);
+        let mut app_dirs = vec![dirs.state_dir(), dirs.cache_dir()];
+        if let Some(config_dir) = dirs.config_path().parent() {
+            app_dirs.push(config_dir.to_path_buf());
+        }
+        Self {
+            home: dirs.home_dir().map(resolved),
+            mountpoint: Some(resolved(dirs.resolved_mountpoint(config))),
+            app_dirs: app_dirs.into_iter().map(resolved).collect(),
+        }
+    }
+}
+
+/// `path` as a root is compared: absolute, with symlinks resolved as far as the
+/// path exists. A folder a restore is about to create does not exist yet, so
+/// the part past the deepest existing folder is kept as written.
+///
+/// A `..` is refused rather than resolved: before a symlink it means something
+/// else than after it, and a path recorded on another machine has no business
+/// holding one.
+pub fn resolve(path: &Path) -> Result<PathBuf, String> {
+    if !path.is_absolute() {
+        return Err(format!("{} is not an absolute path", path.display()));
+    }
+    if path.components().any(|c| c == Component::ParentDir) {
+        return Err(format!("{} goes up with `..`", path.display()));
+    }
+    let mut existing = path.to_path_buf();
+    let mut rest = Vec::new();
+    loop {
+        match existing.canonicalize() {
+            Ok(canonical) => {
+                return Ok(rest
+                    .into_iter()
+                    .rev()
+                    .fold(canonical, |path: PathBuf, part| path.join(part)));
+            }
+            Err(_) => {
+                let Some(name) = existing.file_name().map(|n| n.to_os_string()) else {
+                    return Ok(path.to_path_buf());
+                };
+                rest.push(name);
+                if !existing.pop() {
+                    return Ok(path.to_path_buf());
+                }
+            }
+        }
+    }
+}
+
+/// Why `root` cannot be synced, or `None` when it can. `root` is resolved
+/// ([`resolve`]); `others` are the roots already synced.
+pub fn problem(root: &Path, others: &[PathBuf], reserved: &Reserved) -> Option<String> {
+    let shown = root.display();
+    if root.parent().is_none() {
+        return Some(format!("{shown} is the root of the file system"));
+    }
+    if let Some(home) = &reserved.home {
+        if root == home {
+            return Some(format!(
+                "{shown} is your home folder; pick a folder inside it"
+            ));
+        }
+        if home.starts_with(root) {
+            return Some(format!("{shown} holds your home folder"));
+        }
+    }
+    if let Some(mountpoint) = &reserved.mountpoint
+        && overlaps(root, mountpoint)
+    {
+        return Some(format!(
+            "{shown} overlaps the Proton Drive folder {}",
+            mountpoint.display()
+        ));
+    }
+    if let Some(dir) = reserved.app_dirs.iter().find(|dir| overlaps(root, dir)) {
+        return Some(format!(
+            "{shown} overlaps this app's own data in {}",
+            dir.display()
+        ));
+    }
+    if others.iter().any(|other| other == root) {
+        return Some(format!("{shown} is already synced"));
+    }
+    if let Some(other) = others.iter().find(|other| overlaps(root, other)) {
+        return Some(format!(
+            "{shown} overlaps the synced folder {}",
+            other.display()
+        ));
+    }
+    None
+}
+
+/// Whether one of `a` and `b` is the other or inside it.
+fn overlaps(a: &Path, b: &Path) -> bool {
+    a.starts_with(b) || b.starts_with(a)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn reserved() -> Reserved {
+        Reserved {
+            home: Some("/home/u".into()),
+            mountpoint: Some("/home/u/ProtonDrive".into()),
+            app_dirs: vec![
+                "/home/u/.local/state/proton-drive-linux".into(),
+                "/home/u/.cache/proton-drive-linux".into(),
+                "/home/u/.config/proton-drive-linux".into(),
+            ],
+        }
+    }
+
+    fn refused(root: &str, others: &[&str]) -> bool {
+        let others: Vec<PathBuf> = others.iter().map(PathBuf::from).collect();
+        problem(Path::new(root), &others, &reserved()).is_some()
+    }
+
+    #[test]
+    fn a_folder_inside_home_can_be_synced() {
+        assert!(!refused("/home/u/Documents", &[]));
+        assert!(!refused("/home/u/Pictures", &["/home/u/Documents"]));
+        assert!(!refused("/mnt/data/photos", &[]));
+    }
+
+    #[test]
+    fn the_file_system_root_home_and_its_parents_are_refused() {
+        assert!(refused("/", &[]));
+        assert!(refused("/home/u", &[]));
+        assert!(refused("/home", &[]));
+    }
+
+    #[test]
+    fn the_mount_and_the_apps_own_data_are_refused_either_way_round() {
+        assert!(refused("/home/u/ProtonDrive", &[]));
+        assert!(refused("/home/u/ProtonDrive/Documents", &[]));
+        assert!(refused("/home/u/.cache", &[]));
+        assert!(refused(
+            "/home/u/.local/state/proton-drive-linux/staging",
+            &[]
+        ));
+        assert!(refused("/home/u/.config", &[]));
+    }
+
+    #[test]
+    fn a_root_inside_or_around_another_root_is_refused() {
+        assert!(refused("/home/u/Documents", &["/home/u/Documents"]));
+        assert!(refused("/home/u/Documents/Work", &["/home/u/Documents"]));
+        assert!(refused("/home/u/stuff", &["/home/u/stuff/Documents"]));
+        // A shared prefix of characters is not a shared folder.
+        assert!(!refused("/home/u/Documents2", &["/home/u/Documents"]));
+    }
+
+    #[test]
+    fn a_path_that_goes_up_or_is_relative_does_not_resolve() {
+        assert!(resolve(Path::new("/home/u/../../etc")).is_err());
+        assert!(resolve(Path::new("Documents")).is_err());
+    }
+
+    #[test]
+    fn a_folder_still_to_be_created_resolves_under_its_existing_parent() {
+        let base = std::env::temp_dir().join(format!("pdfs-syncroot-{}", std::process::id()));
+        std::fs::create_dir_all(&base).unwrap();
+        let canonical = base.canonicalize().unwrap();
+        assert_eq!(
+            resolve(&base.join("new/inner")).unwrap(),
+            canonical.join("new/inner")
+        );
+        let link = base.join("link");
+        let _ = std::fs::remove_file(&link);
+        std::os::unix::fs::symlink(&canonical, &link).unwrap();
+        assert_eq!(
+            resolve(&link.join("later")).unwrap(),
+            canonical.join("later"),
+            "a symlink is followed to where it points"
+        );
+        std::fs::remove_dir_all(&base).unwrap();
+    }
+}

crates/pdfs-core/src/takeout.rs

@@ -346,6 +346,11 @@ fn assemble(entries: Vec<RawEntry>, sidecars: &HashMap<String, Vec<u8>>) -> Take
             .insert(file.to_string(), body);
     }
 
+    let indexes: HashMap<&str, SidecarIndex<'_>> = per_folder_sidecars
+        .iter()
+        .map(|(folder, in_folder)| (folder.as_str(), SidecarIndex::new(in_folder)))
+        .collect();
+
     let mut scan = TakeoutScan::default();
     let mut seen_albums: HashMap<String, ()> = HashMap::new();
 
@@ -364,9 +369,13 @@ fn assemble(entries: Vec<RawEntry>, sidecars: &HashMap<String, Vec<u8>>) -> Take
             continue;
         };
 
-        let sidecar = per_folder_sidecars
-            .get(&folder)
-            .and_then(|in_folder| find_sidecar(&file_name, in_folder))
+        let found = indexes
+            .get(folder.as_str())
+            .and_then(|index| find_sidecar(&file_name, index));
+        // An edited copy keeps its own name: the title is its original's.
+        let title_is_ours = found.is_some_and(|((rule, _), _)| rule != EDITED_COPY);
+        let sidecar = found
+            .map(|(_, body)| body)
             .and_then(|body| serde_json::from_slice::<PhotoSidecar>(body).ok())
             .unwrap_or_default();
 
@@ -388,6 +397,7 @@ fn assemble(entries: Vec<RawEntry>, sidecars: &HashMap<String, Vec<u8>>) -> Take
         let name = sidecar
             .title
             .as_deref()
+            .filter(|_| title_is_ours)
             .map(str::trim)
             .filter(|title| !title.is_empty() && same_extension(title, &file_name))
             .unwrap_or(&file_name)
@@ -598,7 +608,7 @@ pub fn capture_time_from_name(name: &str) -> Option<i64> {
 
 /// Days between 1970-01-01 and a proleptic-Gregorian date (Howard Hinnant's
 /// `days_from_civil`). Avoids pulling a date crate in for one conversion.
-fn days_from_civil(year: i64, month: i64, day: i64) -> i64 {
+pub(crate) fn days_from_civil(year: i64, month: i64, day: i64) -> i64 {
     let year = year - i64::from(month <= 2);
     let era = if year >= 0 { year } else { year - 399 } / 400;
     let year_of_era = year - era * 400;
@@ -627,17 +637,106 @@ fn swap_duplicate_marker(file_name: &str) -> Option<String> {
     Some(format!("{}{}{}", &file_name[..marker_start], tail, marker))
 }
 
-/// Find the sidecar belonging to `file_name` among the JSON files of its folder.
+/// How long Google lets a sidecar's file name get, `.json` included. A longer
+/// one is cut, into its suffix first and then into the photo name.
+const SIDECAR_NAME_CAP: usize = 51;
+
+/// The rank [`sidecar_rank`] gives an edited copy that shares its original's
+/// sidecar. Its date is the original's, but its name is its own.
+const EDITED_COPY: u8 = 4;
+
+/// The sidecars of one folder by their `.json`-stripped stem, so a photo is
+/// ranked only against those that can match it.
+///
+/// Every rule in [`sidecar_rank`] needs the stem to start with a prefix of the
+/// photo name that the name itself determines, and a sorted map finds the
+/// stems with a given prefix as one range. Ranking every photo against every
+/// sidecar instead took 27 s for a folder of 10,000 photos.
+struct SidecarIndex<'a> {
+    by_stem: BTreeMap<&'a str, Vec<(&'a str, &'a Vec<u8>)>>,
+}
+
+impl<'a> SidecarIndex<'a> {
+    fn new(in_folder: &'a BTreeMap<String, &'a Vec<u8>>) -> Self {
+        let mut by_stem: BTreeMap<&'a str, Vec<(&'a str, &'a Vec<u8>)>> = BTreeMap::new();
+        for (name, body) in in_folder {
+            by_stem
+                .entry(strip_json_suffix(name))
+                .or_default()
+                .push((name.as_str(), *body));
+        }
+        Self { by_stem }
+    }
+
+    /// Add every sidecar whose stem starts with `prefix` to `out`.
+    fn starting_with(&self, prefix: &str, out: &mut BTreeMap<&'a str, &'a Vec<u8>>) {
+        let from = (
+            std::ops::Bound::Included(prefix),
+            std::ops::Bound::Unbounded,
+        );
+        for (stem, sidecars) in self.by_stem.range::<str, _>(from) {
+            if !stem.starts_with(prefix) {
+                break;
+            }
+            out.extend(sidecars.iter().copied());
+        }
+    }
+
+    /// Add every sidecar whose stem is `head`, or starts with `head` and a `.`:
+    /// those that have `head` as a boundary ([`suffix_boundaries`]).
+    fn bounded_by(&self, head: &str, out: &mut BTreeMap<&'a str, &'a Vec<u8>>) {
+        if let Some(sidecars) = self.by_stem.get(head) {
+            out.extend(sidecars.iter().copied());
+        }
+        self.starting_with(&format!("{head}."), out);
+    }
+
+    /// Every sidecar [`sidecar_rank`] could match `file_name` to, by name. A
+    /// superset: each one is still ranked.
+    fn candidates(&self, file_name: &str) -> BTreeMap<&'a str, &'a Vec<u8>> {
+        let mut out = BTreeMap::new();
+        // Rules 0 and 2: the stem starts with the name, or with its swapped form.
+        self.starting_with(file_name, &mut out);
+        if let Some(swapped) = swap_duplicate_marker(file_name) {
+            self.starting_with(&swapped, &mut out);
+        }
+        // Rule 1: the stem of `A(N).B` starts with `A.B(N)`, so with `A.`.
+        for (paren, _) in file_name.match_indices('(') {
+            self.starting_with(&format!("{}.", &file_name[..paren]), &mut out);
+        }
+        // Rule 3: a boundary of the stem is a prefix of the name.
+        for end in file_name
+            .char_indices()
+            .map(|(index, _)| index)
+            .skip(1)
+            .chain([file_name.len()])
+        {
+            self.bounded_by(&file_name[..end], &mut out);
+        }
+        // Rule 4: `<original>-<word>.<ext>` has `<original>.<ext>` as a boundary.
+        if let Some((_, extension)) = file_name.rsplit_once('.') {
+            for (dash, _) in file_name.match_indices('-') {
+                self.bounded_by(&format!("{}.{extension}", &file_name[..dash]), &mut out);
+            }
+        }
+        out
+    }
+}
+
+/// Find the sidecar belonging to `file_name` among the JSON files of its folder,
+/// with the rank it matched at.
 ///
 /// Tried in order of confidence: the exact name, the `(N)`-swapped name, an
-/// unknown trailing suffix, then a prefix match for the case where truncation
-/// cut into the photo name itself. The best-ranked candidate in the folder wins.
+/// unknown trailing suffix, a prefix match for the case where truncation cut
+/// into the photo name itself, and last an edited copy's original. The
+/// best-ranked candidate in the folder wins; of equal ranks, the first by
+/// sidecar name.
 fn find_sidecar<'a>(
     file_name: &str,
-    in_folder: &BTreeMap<String, &'a Vec<u8>>,
-) -> Option<&'a Vec<u8>> {
+    index: &SidecarIndex<'a>,
+) -> Option<((u8, usize), &'a Vec<u8>)> {
     let mut best: Option<((u8, usize), &'a Vec<u8>)> = None;
-    for (sidecar_name, body) in in_folder {
+    for (sidecar_name, body) in index.candidates(file_name) {
         let Some(rank) = sidecar_rank(strip_json_suffix(sidecar_name), file_name) else {
             continue;
         };
@@ -645,7 +744,7 @@ fn find_sidecar<'a>(
             best = Some((rank, body));
         }
     }
-    best.map(|(_, body)| body)
+    best
 }
 
 /// How well a sidecar's `.json`-stripped `stem` matches `file_name`, as a sort
@@ -683,19 +782,44 @@ fn sidecar_rank(stem: &str, file_name: &str) -> Option<(u8, usize)> {
     }
     // 3 — truncation ate into the photo name itself, leaving only a prefix of
     // it. Every suffix boundary in the stem is tried, so this holds for a
-    // localized suffix too; the longest surviving prefix wins, because a folder
-    // with `IMG_1.jpg` and `IMG_12.jpg` must not let the shorter stem claim the
-    // longer photo.
-    let mut longest: Option<usize> = None;
-    for head in suffix_boundaries(stem) {
-        if !head.is_empty()
-            && file_name.starts_with(head)
-            && longest.is_none_or(|len| head.len() > len)
-        {
-            longest = Some(head.len());
+    // localized suffix too; the longest surviving prefix wins.
+    //
+    // Only for a sidecar whose name is as long as Google lets one get: a name
+    // that short of the cap was not cut, so a prefix of it is the name of a
+    // different photo. Without this, `IMG_1.jpg`'s sidecar gave its name and
+    // date to an `IMG_12.jpg` that had none (`docs/BUGS.md` B225).
+    if stem.chars().count() + ".json".len() >= SIDECAR_NAME_CAP - 4 {
+        let mut longest: Option<usize> = None;
+        for head in suffix_boundaries(stem) {
+            if !head.is_empty()
+                && file_name.starts_with(head)
+                && longest.is_none_or(|len| head.len() > len)
+            {
+                longest = Some(head.len());
+            }
+        }
+        if let Some(len) = longest {
+            return Some((3, usize::MAX - len));
         }
     }
-    longest.map(|len| (3, usize::MAX - len))
+    // 4 — an edited copy, `IMG_1-edited.jpg` (the word is localized), which
+    // Google exports without a sidecar of its own next to `IMG_1.jpg`'s.
+    suffix_boundaries(stem)
+        .filter_map(|head| head.rsplit_once('.'))
+        .any(|(original, extension)| is_edited_copy_of(file_name, original, extension))
+        .then_some((EDITED_COPY, 0))
+}
+
+/// Whether `file_name` is `<original>-<word>.<extension>`, the name Google
+/// gives an edited copy of `<original>.<extension>`.
+fn is_edited_copy_of(file_name: &str, original: &str, extension: &str) -> bool {
+    !original.is_empty()
+        && file_name
+            .strip_prefix(original)
+            .and_then(|rest| rest.strip_prefix('-'))
+            .and_then(|rest| rest.strip_suffix(extension))
+            .and_then(|rest| rest.strip_suffix('.'))
+            .is_some_and(|word| !word.is_empty() && word.chars().all(char::is_alphabetic))
 }
 
 /// The stem plus each of its prefixes ending before a `.`, longest first — the
@@ -812,7 +936,7 @@ mod tests {
     }
 
     #[test]
-    fn sidecar_supplies_capture_time_and_favourite() {
+    fn sidecar_supplies_capture_time_and_favorite() {
         let mut sidecars = HashMap::new();
         sidecar(
             &mut sidecars,
@@ -994,21 +1118,25 @@ mod tests {
 
     #[test]
     fn longest_prefix_wins_when_the_photo_name_itself_was_truncated() {
+        // Both sidecar names are at Google's cap, so both were cut into the
+        // photo name; the longer surviving prefix is the better claim.
+        let short = "Takeout/Google Photos/Photos from 2019/PXL_20190101_120000000.PORTRAIT_ORIGINAL_E.json";
+        let long = "Takeout/Google Photos/Photos from 2019/PXL_20190101_120000000.PORTRAIT_ORIGINAL_EXPO.json";
         let mut sidecars = HashMap::new();
         sidecar(
             &mut sidecars,
-            "Takeout/Google Photos/Photos from 2019/IMG_1.json",
+            short,
             r#"{"photoTakenTime":{"timestamp":"1"}}"#,
         );
         sidecar(
             &mut sidecars,
-            "Takeout/Google Photos/Photos from 2019/IMG_12.json",
+            long,
             r#"{"photoTakenTime":{"timestamp":"2"}}"#,
         );
         let scan = assemble(
             vec![entry(
                 0,
-                "Takeout/Google Photos/Photos from 2019/IMG_12345.jpg",
+                "Takeout/Google Photos/Photos from 2019/PXL_20190101_120000000.PORTRAIT_ORIGINAL_EXPOSURE.jpg",
                 10,
             )],
             &sidecars,
@@ -1017,6 +1145,104 @@ mod tests {
         assert_eq!(scan.photos[0].capture_time, Some(2));
     }
 
+    /// A short sidecar name was not cut, so a prefix of it names another photo:
+    /// `IMG_1.jpg`'s sidecar gave its title and date to `IMG_12.jpg`
+    /// (`docs/BUGS.md` B225).
+    #[test]
+    fn a_photo_without_a_sidecar_does_not_take_a_shorter_named_siblings() {
+        let mut sidecars = HashMap::new();
+        sidecar(
+            &mut sidecars,
+            "Takeout/Google Photos/Photos from 2019/IMG_1.jpg.supplemental-metadata.json",
+            r#"{"title":"IMG_1.jpg","photoTakenTime":{"timestamp":"1"}}"#,
+        );
+        let scan = assemble(
+            vec![
+                entry(0, "Takeout/Google Photos/Photos from 2019/IMG_1.jpg", 10),
+                entry(0, "Takeout/Google Photos/Photos from 2019/IMG_12.jpg", 10),
+            ],
+            &sidecars,
+        );
+
+        assert_eq!(scan.photos[0].capture_time, Some(1));
+        assert_eq!(scan.photos[1].name, "IMG_12.jpg");
+        assert_eq!(scan.photos[1].capture_time, None);
+    }
+
+    /// An edited copy has no sidecar of its own. It takes its original's date,
+    /// but keeps its own name rather than the original's title.
+    #[test]
+    fn an_edited_copy_takes_its_originals_date_and_keeps_its_name() {
+        let mut sidecars = HashMap::new();
+        sidecar(
+            &mut sidecars,
+            "Takeout/Google Fotos/Urlaub/IMG_2.jpg.ergänzende-Metadaten.json",
+            r#"{"title":"Holiday.jpg","photoTakenTime":{"timestamp":"7"}}"#,
+        );
+        let scan = assemble(
+            vec![
+                entry(0, "Takeout/Google Fotos/Urlaub/IMG_2.jpg", 10),
+                entry(0, "Takeout/Google Fotos/Urlaub/IMG_2-bearbeitet.jpg", 10),
+            ],
+            &sidecars,
+        );
+
+        assert_eq!(scan.photos[0].name, "Holiday.jpg");
+        assert_eq!(scan.photos[1].name, "IMG_2-bearbeitet.jpg");
+        assert_eq!(scan.photos[1].capture_time, Some(7));
+    }
+
+    /// The index looks only at the sidecars that can match, and must find the
+    /// same one that ranking every sidecar in the folder finds.
+    #[test]
+    fn the_sidecar_index_finds_what_ranking_every_sidecar_finds() {
+        let long = "A_very_long_holiday_photo_name_from_the_beach_2019";
+        let names = [
+            "IMG_1.jpg".to_string(),
+            "IMG_12.jpg".to_string(),
+            "IMG_1(1).jpg".to_string(),
+            "IMG_1-edited.jpg".to_string(),
+            "IMG_1-bearbeitet.jpg".to_string(),
+            "my.photo(2).jpg".to_string(),
+            "x(1).jpg".to_string(),
+            "Ölbild.png".to_string(),
+            "PXL_20230101_101010.mp4".to_string(),
+            format!("{long}.jpg"),
+            format!("{long}_2.jpg"),
+            "nothing.heic".to_string(),
+        ];
+        let sidecar_names = [
+            "IMG_1.jpg.supplemental-metadata.json".to_string(),
+            "IMG_1.jpg(1).supplemental-metadata.json".to_string(),
+            "IMG_12.jpg.ergänzende-Metadaten.json".to_string(),
+            "my.photo.jpg(2).json".to_string(),
+            "x.jpg(1).json".to_string(),
+            "Ölbild.png.suppl.json".to_string(),
+            "PXL_20230101_101010.mp4.JSON".to_string(),
+            format!("{}.json", &long[..44]),
+            format!("{long}.jp.json"),
+            "IMG.json".to_string(),
+            "unrelated.json".to_string(),
+        ];
+        let bodies: Vec<Vec<u8>> = (0..sidecar_names.len())
+            .map(|i| i.to_string().into_bytes())
+            .collect();
+        let in_folder: BTreeMap<String, &Vec<u8>> =
+            sidecar_names.iter().cloned().zip(bodies.iter()).collect();
+        let index = SidecarIndex::new(&in_folder);
+        for name in &names {
+            let mut every: Option<((u8, usize), &Vec<u8>)> = None;
+            for (sidecar_name, body) in &in_folder {
+                if let Some(rank) = sidecar_rank(strip_json_suffix(sidecar_name), name)
+                    && every.as_ref().is_none_or(|(current, _)| rank < *current)
+                {
+                    every = Some((rank, *body));
+                }
+            }
+            assert_eq!(find_sidecar(name, &index), every, "{name}");
+        }
+    }
+
     #[test]
     fn sidecar_title_replaces_a_mangled_on_disk_name() {
         let mut sidecars = HashMap::new();

crates/pdfs-core/src/tray.rs

@@ -0,0 +1,20 @@
+//! Talking to a running `pdfs-tray`.
+//!
+//! The tray binds [`AppDirs::tray_socket`] to keep a single instance. The same
+//! socket takes one command, [`QUIT`], so the desktop app can take the icon
+//! away when the user hides it or stops Proton Drive.
+
+use std::io::Write;
+use std::os::unix::net::UnixStream;
+
+use crate::config::AppDirs;
+
+/// The line that asks the tray to exit.
+pub const QUIT: &str = "quit";
+
+/// Ask a running tray to exit. False when no tray answers.
+pub fn quit(dirs: &AppDirs) -> bool {
+    UnixStream::connect(dirs.tray_socket())
+        .and_then(|mut stream| stream.write_all(format!("{QUIT}\n").as_bytes()))
+        .is_ok()
+}

crates/pdfs-core/src/update.rs

@@ -0,0 +1,178 @@
+//! The opt-in check for a newer release ([`crate::config::AppConfig::update_check`]).
+//!
+//! Only the decisions live here: what GitHub's answer says, which version is
+//! newer, when to ask again, and whether a package manager already owns
+//! updates. The desktop app does the request itself.
+
+use std::path::Path;
+use std::time::Duration;
+
+/// GitHub's latest published release. Drafts and pre-releases never show up
+/// here, so a tagged release candidate does not reach anyone.
+pub const RELEASES_LATEST_API: &str =
+    "https://api.github.com/repos/narrrl/proton-drive-linux/releases/latest";
+
+/// How often the app asks, across restarts.
+pub const CHECK_INTERVAL: Duration = Duration::from_secs(24 * 60 * 60);
+
+/// Installed next to the binaries' `share/` by packages whose package manager
+/// brings new releases (COPR, the AUR, Nix). The release `.deb`, `.rpm` and
+/// `.tar.gz` do not ship it, since nothing updates them.
+pub const MANAGED_MARKER: &str = "share/proton-drive-linux/managed-updates";
+
+/// The version of the release in a `/releases/latest` answer, without the
+/// tag's `v`. `None` for an answer that does not parse or a tag that is not
+/// `X.Y.Z`, so a changed API can never claim an update.
+pub fn latest_version(body: &[u8]) -> Option<String> {
+    let json: serde_json::Value = serde_json::from_slice(body).ok()?;
+    let tag = json.get("tag_name")?.as_str()?;
+    let version = tag.strip_prefix('v').unwrap_or(tag);
+    parse(version).map(|_| version.to_owned())
+}
+
+/// Whether `version` is a later release than `current`. Anything that is not
+/// `X.Y.Z` on either side is not.
+pub fn is_newer(version: &str, current: &str) -> bool {
+    matches!((parse(version), parse(current)), (Some(v), Some(c)) if v > c)
+}
+
+/// The release page of `version` on GitHub.
+pub fn release_url(version: &str) -> String {
+    format!("https://github.com/narrrl/proton-drive-linux/releases/tag/v{version}")
+}
+
+/// `X.Y.Z` as numbers, so `3.10.0` sorts after `3.9.0`.
+fn parse(version: &str) -> Option<(u64, u64, u64)> {
+    let mut parts = version.split('.').map(|p| p.parse::<u64>().ok());
+    let triple = (parts.next()??, parts.next()??, parts.next()??);
+    parts.next().is_none().then_some(triple)
+}
+
+/// How long until the next check is due, given the last one in Unix seconds.
+/// Zero when it is due now. A clock set back past the last check counts as
+/// due, so a wrong clock cannot stop checks.
+pub fn next_check_in(last_checked: Option<i64>, now: i64) -> Duration {
+    let interval = CHECK_INTERVAL.as_secs() as i64;
+    match last_checked {
+        Some(last) if (0..interval).contains(&(now - last)) => {
+            Duration::from_secs((interval - (now - last)) as u64)
+        }
+        _ => Duration::ZERO,
+    }
+}
+
+/// Whether the binary at `exe` was installed by a package manager that updates
+/// it, which [`MANAGED_MARKER`] next to its `bin/` says.
+pub fn updates_managed(exe: &Path) -> bool {
+    exe.parent()
+        .and_then(Path::parent)
+        .is_some_and(|prefix| prefix.join(MANAGED_MARKER).exists())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn latest(tag: &str) -> Vec<u8> {
+        format!(
+            r#"{{"tag_name":"{tag}","html_url":"https://github.com/narrrl/proton-drive-linux/releases/tag/{tag}","draft":false}}"#
+        )
+        .into_bytes()
+    }
+
+    #[test]
+    fn the_latest_release_is_read_without_its_v() {
+        assert_eq!(latest_version(&latest("v3.6.0")).as_deref(), Some("3.6.0"));
+        assert_eq!(latest_version(&latest("3.6.0")).as_deref(), Some("3.6.0"));
+    }
+
+    #[test]
+    fn an_answer_that_does_not_parse_has_no_version() {
+        assert_eq!(latest_version(b"not json"), None);
+        assert_eq!(latest_version(b"{}"), None);
+        assert_eq!(latest_version(br#"{"tag_name":7}"#), None);
+        assert_eq!(latest_version(&latest("nightly")), None);
+        assert_eq!(latest_version(&latest("v4.0.0-rc1")), None);
+        assert_eq!(latest_version(&latest("v4.0")), None);
+        assert_eq!(latest_version(&latest("v4.0.0.1")), None);
+    }
+
+    #[test]
+    fn only_a_later_release_is_newer() {
+        assert!(is_newer("3.6.0", "3.5.0"));
+        assert!(!is_newer("3.5.0", "3.5.0"));
+        assert!(!is_newer("3.4.1", "3.5.0"));
+        assert!(!is_newer("2.99.99", "3.0.0"));
+    }
+
+    #[test]
+    fn versions_compare_as_numbers() {
+        assert!(is_newer("3.10.0", "3.9.0"));
+        assert!(is_newer("3.5.10", "3.5.9"));
+        assert!(is_newer("4.0.0", "3.99.99"));
+    }
+
+    #[test]
+    fn a_version_that_is_not_x_y_z_is_never_newer() {
+        assert!(!is_newer("4.0.0", "3.5.0-dev"));
+        assert!(!is_newer("garbage", "3.5.0"));
+    }
+
+    #[test]
+    fn the_release_page_uses_the_tag() {
+        assert_eq!(
+            release_url("3.6.0"),
+            "https://github.com/narrrl/proton-drive-linux/releases/tag/v3.6.0"
+        );
+    }
+
+    #[test]
+    fn a_check_is_due_once_a_day() {
+        let day = CHECK_INTERVAL.as_secs() as i64;
+        assert_eq!(next_check_in(None, 1_000), Duration::ZERO);
+        assert_eq!(next_check_in(Some(1_000), 1_000), CHECK_INTERVAL);
+        assert_eq!(
+            next_check_in(Some(1_000), 1_000 + day - 1),
+            Duration::from_secs(1)
+        );
+        assert_eq!(next_check_in(Some(1_000), 1_000 + day), Duration::ZERO);
+        assert_eq!(
+            next_check_in(Some(1_000), 999),
+            Duration::ZERO,
+            "a clock set back must not stop checks"
+        );
+    }
+
+    /// A unique temp directory removed on drop; avoids a dev-dependency, as in
+    /// [`crate::syncignore`]'s tests.
+    struct TempDir(std::path::PathBuf);
+    impl TempDir {
+        fn new() -> Self {
+            use std::sync::atomic::{AtomicU64, Ordering};
+            static N: AtomicU64 = AtomicU64::new(0);
+            let p = std::env::temp_dir().join(format!(
+                "pdfs-update-{}-{}",
+                std::process::id(),
+                N.fetch_add(1, Ordering::Relaxed)
+            ));
+            std::fs::create_dir_all(&p).unwrap();
+            TempDir(p)
+        }
+    }
+    impl Drop for TempDir {
+        fn drop(&mut self) {
+            let _ = std::fs::remove_dir_all(&self.0);
+        }
+    }
+
+    #[test]
+    fn the_marker_next_to_bin_means_a_package_manager_updates() {
+        let prefix = TempDir::new();
+        let exe = prefix.0.join("bin/pdfs-app");
+        assert!(!updates_managed(&exe));
+        let marker = prefix.0.join(MANAGED_MARKER);
+        std::fs::create_dir_all(marker.parent().unwrap()).unwrap();
+        std::fs::write(&marker, "").unwrap();
+        assert!(updates_managed(&exe));
+    }
+}

crates/pdfs-fuse/Cargo.toml

@@ -15,6 +15,7 @@ zip.workspace = true
 sha1.workspace = true
 proton-drive-rs.workspace = true
 futures.workspace = true
+async-trait.workspace = true
 tokio.workspace = true
 tracing.workspace = true
 fuser.workspace = true
@@ -24,3 +25,6 @@ base64.workspace = true
 image.workspace = true
 notify = "8"
 parking_lot.workspace = true
+
+[dev-dependencies]
+tracing-subscriber.workspace = true

crates/pdfs-fuse/src/albums.rs

@@ -15,7 +15,7 @@
 use std::collections::{HashMap, HashSet};
 use std::sync::atomic::Ordering;
 
-use pdfs_core::control::{AlbumInfo, PhotoItem};
+use pdfs_core::control::{AlbumInfo, PhotoItem, Topic};
 use pdfs_core::db::StoredAlbum;
 use pdfs_core::{CoreError, CoreResult};
 
@@ -137,10 +137,12 @@ impl Core {
         }
         let core = self.clone();
         self.rt.spawn(async move {
-            if let Err(e) = core.refresh_albums().await {
-                warn!(error = %e, "background album refresh failed");
-            }
+            let refreshed = core.refresh_albums().await;
             core.albums_refreshing.store(false, Ordering::SeqCst);
+            match refreshed {
+                Ok(_) => core.events.publish(&[Topic::Photos]),
+                Err(e) => warn!(error = %e, "background album refresh failed"),
+            }
         });
     }
 
@@ -175,6 +177,40 @@ impl Core {
         Ok(page.into_iter().map(|p| self.photo_item(p)).collect())
     }
 
+    /// Make sure every album of ours has its contents on disk, for a filter
+    /// that asks which photos no album holds. Contents are otherwise fetched
+    /// the first time an album is opened, so an album never opened would read
+    /// as empty and its photos as unfiled.
+    ///
+    /// Shared-with-me albums are skipped: their photos live on the sharer's
+    /// volume and are never in our timeline. The freshness stamps are no guide
+    /// here — any album change clears all of them — so only an album whose
+    /// server count disagrees with an empty local copy is waited for.
+    pub(crate) fn ensure_album_contents(&self) -> CoreResult<()> {
+        if self.db.albums_count().map_err(CoreError::from)? == 0
+            && !self.rt.block_on(self.refresh_albums())?
+        {
+            return Ok(());
+        }
+        let albums = self.db.albums_list().map_err(CoreError::from)?;
+        for album in albums.iter().filter(|a| !a.shared && a.photo_count > 0) {
+            let stored = self
+                .db
+                .album_photos_count(&album.uid)
+                .map_err(CoreError::from)?;
+            if stored > 0 {
+                continue;
+            }
+            let Some(uid) = parse_uid(&album.uid) else {
+                continue;
+            };
+            if let Err(error) = self.rt.block_on(self.refresh_album(&uid)) {
+                warn!(album = %album.uid, %error, "fetching an album's contents failed");
+            }
+        }
+        Ok(())
+    }
+
     /// Re-enumerate one album's photos and replace its persisted contents.
     pub(crate) async fn refresh_album(&self, album: &NodeUid) -> CoreResult<()> {
         let photos = self.photos();
@@ -221,6 +257,84 @@ impl Core {
         Ok(())
     }
 
+    /// Create an album and return its uid. The listing is refreshed before
+    /// answering, so the next [`Core::albums`] already shows it.
+    pub(crate) fn create_album(&self, name: &str) -> CoreResult<String> {
+        let uid = self
+            .rt
+            .block_on(self.photos().create_album(name))
+            .map_err(|e| CoreError::from_api(&e, "create album"))?;
+        self.after_album_change(None);
+        Ok(uid.to_string())
+    }
+
+    pub(crate) fn rename_album(&self, album: &NodeUid, name: &str) -> CoreResult<()> {
+        self.rt
+            .block_on(self.photos().rename_album(album, name))
+            .map_err(|e| CoreError::from_api(&e, "rename album"))?;
+        self.after_album_change(None);
+        Ok(())
+    }
+
+    /// Delete an album. Its photos stay in the timeline unless `delete_photos`,
+    /// which only matters for photos that exist in the album alone.
+    pub(crate) fn delete_album(&self, album: &NodeUid, delete_photos: bool) -> CoreResult<()> {
+        self.rt
+            .block_on(self.photos().delete_album(album, delete_photos))
+            .map_err(|e| CoreError::from_api(&e, "delete album"))?;
+        let key = album.to_string();
+        if let Err(error) = self.db.album_photos_replace(&key, &[]) {
+            warn!(%album, %error, "a deleted album's photos could not be dropped");
+        }
+        self.after_album_change(None);
+        Ok(())
+    }
+
+    /// Add photos to an album, per photo: `(added, failed)`.
+    pub(crate) fn add_to_album(
+        &self,
+        album: &NodeUid,
+        photos: &[NodeUid],
+    ) -> CoreResult<AlbumOutcome> {
+        let outcomes = self
+            .rt
+            .block_on(self.photos().add_photos_to_album(album, photos))
+            .map_err(|e| CoreError::from_api(&e, "add to album"))?;
+        self.after_album_change(Some(album));
+        Ok(album_outcome(outcomes))
+    }
+
+    /// Take photos out of an album, per photo: `(removed, failed)`. The photos
+    /// stay in the timeline.
+    pub(crate) fn remove_from_album(
+        &self,
+        album: &NodeUid,
+        photos: &[NodeUid],
+    ) -> CoreResult<AlbumOutcome> {
+        let outcomes = self
+            .rt
+            .block_on(self.photos().remove_photos_from_album(album, photos))
+            .map_err(|e| CoreError::from_api(&e, "remove from album"))?;
+        self.after_album_change(Some(album));
+        Ok(album_outcome(outcomes))
+    }
+
+    /// Bring the persisted listing, and `album`'s contents when given, up to
+    /// date after a change made here, so the reply is followed by a listing
+    /// that shows it. A failed refresh only leaves them stale: the change
+    /// itself went through.
+    fn after_album_change(&self, album: Option<&NodeUid>) {
+        self.invalidate_albums();
+        if let Err(error) = self.rt.block_on(self.refresh_albums()) {
+            warn!(%error, "album listing refresh after a change failed");
+        }
+        if let Some(album) = album
+            && let Err(error) = self.rt.block_on(self.refresh_album(album))
+        {
+            warn!(%album, %error, "album refresh after a change failed");
+        }
+    }
+
     /// Drop the album listing's and every album's freshness stamp, so the next
     /// request re-enumerates. Paired with [`Core::invalidate_photos`] — a change
     /// to the timeline can just as easily be a change to an album.
@@ -230,6 +344,20 @@ impl Core {
     }
 }
 
+/// Photos an album change applied to, and the ones it failed for with why.
+pub(crate) type AlbumOutcome = (Vec<String>, Vec<(String, String)>);
+
+fn album_outcome(outcomes: Vec<pdfs_core::batch::Outcome>) -> AlbumOutcome {
+    let (done, failed) = pdfs_core::batch::split(outcomes);
+    (
+        done.iter().map(|uid| uid.to_string()).collect(),
+        failed
+            .into_iter()
+            .map(|(uid, error)| (uid.to_string(), error.to_string()))
+            .collect(),
+    )
+}
+
 /// Project a persisted album into the wire item a front-end paints.
 fn album_info(album: StoredAlbum) -> AlbumInfo {
     AlbumInfo {

crates/pdfs-fuse/src/background.rs

@@ -2,6 +2,7 @@
 
 use super::*;
 use crate::shutdown::Shutdown;
+use pdfs_core::error::session_revoked;
 
 /// Drop `uid` from one mount's tree and tell that mount's kernel session the
 /// entry is gone. Shared by the trash and delete paths, which differ only in
@@ -9,7 +10,7 @@ use crate::shutdown::Shutdown;
 ///
 /// Returns whether this mount actually held the node, so the caller can evict
 /// its content blob once if *any* mount did.
-fn forget_and_notify(st: &mut State, notify: &mut NotifyBatch, uid: &NodeUid) -> bool {
+pub(crate) fn forget_and_notify(st: &mut State, notify: &mut NotifyBatch, uid: &NodeUid) -> bool {
     // Capture the inode before `forget` clears the uid mapping.
     let child = st.by_uid.get(uid).copied();
     let Some((parent, name)) = st.forget(uid) else {
@@ -97,15 +98,13 @@ fn event_serializes_shared_publication(event: &DriveEvent, own_volume: &VolumeId
 /// entry forces the next `lookup`/`readdir` to re-enumerate from the remote, so
 /// most events only need to invalidate listings rather than re-fetch eagerly.
 ///
-/// Applied to **every** mounted inode space, not just the primary one. This task
-/// is per-daemon while node state is per-mount, and a `DriveEvent` names a uid,
-/// not a mount — so a file trashed from another device has to be withdrawn from
-/// whichever of our sessions is showing it, which for anything under a sync
-/// folder is a [`Core::fork_state`] fork rather than `core.state`. Reaching for
-/// the primary state alone left forks serving a deleted file indefinitely, the
-/// read-side twin of `docs/BUGS.md` B74. Inode numbers are per-mount, so each
-/// mount must be notified through its **own** channel; that pairing is what
-/// [`Core::for_each_mount`] exists to preserve.
+/// Applied to the tree every mount shares, and told to **every** mounted
+/// session. A `DriveEvent` names a uid, not a mount, so a file trashed from
+/// another device has to be withdrawn from whichever of our sessions is showing
+/// it — which for anything under a sync folder is a [`Core::fork_state`] fork.
+/// Telling the primary session alone left forks serving a deleted file, the
+/// read-side twin of `docs/BUGS.md` B74. [`Core::for_each_mount`] sends what
+/// the kernel needs to hear to every session.
 fn apply_event(core: &Core, event: &DriveEvent, dirty: &mut DirtyParents) -> pdfs_core::Result<()> {
     let foreign_delete = is_foreign_node_delete(event, &core.primary_root_uid.volume_id);
     let serializes_shared_publication =
@@ -145,7 +144,7 @@ fn apply_event(core: &Core, event: &DriveEvent, dirty: &mut DirtyParents) -> pdf
             // create, and re-fetching would replace the size and mtime of the
             // write we just accepted with the stale revision's — making a file
             // that was copied in seconds ago read as empty until its upload
-            // lands (offline.md Phase 3).
+            // lands.
             //
             // A write that has already drained is the same story one step later:
             // the drain brought the tree level with the revision it sealed, so
@@ -233,6 +232,48 @@ fn acknowledge_applied_event(
     Ok(id)
 }
 
+/// How many applied node-update ids [`RecentEvents`] remembers. Far more than a
+/// replay can reach back: one comes from a batch whose last event was not its
+/// newest, so it repeats at most what that batch held.
+const RECENT_EVENTS: usize = 4096;
+
+/// The node updates applied lately, by event id, so one that comes round again
+/// is skipped rather than applied twice.
+///
+/// The feed can deliver the same event twice in a batch, and a batch whose
+/// events come out of order leaves the cursor short of its newest event, so the
+/// next poll replays what lay past it. An update applied twice is not harmless:
+/// the first copy of our own change claims the echo [`Core::note_self_change`]
+/// recorded, and the second, unclaimed, reads as someone else's edit and evicts
+/// the bytes a partial write is about to gap-fill from (docs/BUGS.md B123).
+///
+/// Only node updates: a delete is idempotent, and a continuity loss names the
+/// cursor it lost, which is an id already applied.
+#[derive(Default)]
+struct RecentEvents {
+    ids: HashSet<DriveEventId>,
+    order: VecDeque<DriveEventId>,
+}
+
+impl RecentEvents {
+    /// Record `event` as applied, reporting whether it was already.
+    fn repeat(&mut self, event: &DriveEvent) -> bool {
+        let DriveEvent::NodeUpdated { id, .. } = event else {
+            return false;
+        };
+        if !self.ids.insert(id.clone()) {
+            return true;
+        }
+        self.order.push_back(id.clone());
+        if self.order.len() > RECENT_EVENTS
+            && let Some(old) = self.order.pop_front()
+        {
+            self.ids.remove(&old);
+        }
+        false
+    }
+}
+
 /// The folders one batch of events changed, collected so the listings are
 /// dropped once at the end of the batch rather than once per event.
 ///
@@ -320,10 +361,10 @@ fn flush_dirty_parents(core: &Core, dirty: &DirtyParents) {
 /// only on fatal error.
 ///
 /// Takes the whole `Core` rather than the primary mount's pieces: an event has
-/// to be applied to every mounted inode space, and the registry that enumerates
-/// them lives on the `Core` (see [`apply_event`]).
+/// to reach every mounted session, and the registry that enumerates them lives
+/// on the `Core` (see [`apply_event`]).
 pub(super) async fn run_event_sync(
-    client: ProtonDriveClient,
+    client: Arc<dyn DriveApi>,
     scope: DriveEventScopeId,
     core: Core,
 ) {
@@ -335,7 +376,7 @@ pub(super) async fn run_event_sync(
         // server head; persist it so the next restart resumes instead of
         // reseeding (which would skip everything that changed offline).
         // Seeding needs the network, and this task also runs on mounts that
-        // started offline (offline.md Phase 1) — so retry rather than giving up,
+        // started offline — so retry rather than giving up,
         // which used to disable live sync for the life of the daemon.
         Ok(None) => {
             // Nothing says what changed before this cursor, so a persisted SDK
@@ -362,6 +403,12 @@ pub(super) async fn run_event_sync(
                         }
                         break Some(c.clone());
                     }
+                    // Refused for good: retrying would be refused the same way
+                    // for the life of the daemon. The next login restarts it.
+                    Err(e) if session_revoked(&e) => {
+                        core.mark_session_expired("seed event cursor", &e);
+                        return;
+                    }
                     Err(e) => {
                         warn!(error = %e, ?delay, "seed event cursor failed; retrying");
                         tokio::time::sleep(delay).await;
@@ -377,10 +424,12 @@ pub(super) async fn run_event_sync(
     };
     info!(?cursor, "event sync started");
 
+    let mut recent = RecentEvents::default();
     loop {
-        tokio::time::sleep(POLL_INTERVAL).await;
-        let events = match client.enumerate_events(&scope, cursor.as_ref()).await {
+        core.next_poll(POLL_INTERVAL).await;
+        let events = match link::bounded(client.enumerate_events(&scope, cursor.as_ref())).await {
             Ok(events) => events,
+            Err(e) if core.lost_link(&e, "poll events") => continue,
             Err(e) => {
                 warn!(error = %e, "event poll failed; retrying after interval");
                 continue;
@@ -396,6 +445,11 @@ pub(super) async fn run_event_sync(
         // did apply.
         let mut dirty = DirtyParents::default();
         for event in &events {
+            if recent.repeat(event) {
+                debug!(id = %event.id().as_str(), "skipping an event already applied");
+                cursor = Some(event.id().clone());
+                continue;
+            }
             // Converge the SDK's own caches (folder keys, entity cache) on the
             // server before applying the event to our tree. Without this, a node
             // re-keyed/moved by another client keeps a stale key in the SDK for
@@ -422,6 +476,10 @@ pub(super) async fn run_event_sync(
             cursor = Some(applied);
         }
         flush_dirty_parents(&core, &dirty);
+        // Another device changed the tree: whatever a front-end lists may have
+        // moved, including what is shared and what is in the trash.
+        core.events
+            .publish(&[Topic::Files, Topic::Trash, Topic::Shares, Topic::Conflicts]);
     }
 }
 
@@ -451,6 +509,10 @@ pub(super) async fn run_photos_event_sync(core: Core) {
                 debug!("no photos volume; photo event sync not started");
                 return;
             }
+            Err(error) if session_revoked(&error) => {
+                core.mark_session_expired("read the photos root", &error);
+                return;
+            }
             Err(error) => {
                 warn!(%error, "reading the photos root failed; retrying");
                 tokio::time::sleep(ONLINE_PROBE_MAX).await;
@@ -467,9 +529,10 @@ pub(super) async fn run_photos_event_sync(core: Core) {
     info!(?cursor, "photo event sync started");
 
     loop {
-        tokio::time::sleep(POLL_INTERVAL).await;
-        let events = match photos.enumerate_events(&scope, cursor.as_ref()).await {
+        core.next_poll(POLL_INTERVAL).await;
+        let events = match link::bounded(photos.enumerate_events(&scope, cursor.as_ref())).await {
             Ok(events) => events,
+            Err(error) if core.lost_link(&error, "poll photo events") => continue,
             Err(error) => {
                 warn!(%error, "photo event poll failed; retrying after interval");
                 continue;
@@ -479,6 +542,7 @@ pub(super) async fn run_photos_event_sync(core: Core) {
             continue;
         };
         apply_photo_events(&core, &events);
+        core.events.publish(&[Topic::Photos]);
         // One cursor write per batch, after the batch is applied: a crash in
         // between replays events whose effects are idempotent.
         if let Err(error) = core.db.set_state_str(PHOTOS_EVENT_CURSOR, head.as_str()) {
@@ -511,7 +575,7 @@ impl PhotoEventPlan {
 ///
 /// `is_echo` claims an event as this daemon's own write. The feed replays our
 /// writes back at us and a `NodeUpdated` carries no revision id, so without this
-/// every favourite the user toggles here would mark its own photo stale and buy
+/// every favorite the user toggles here would mark its own photo stale and buy
 /// a resolve of a node we just wrote.
 fn classify_photo_events(
     events: &[DriveEvent],
@@ -529,7 +593,7 @@ fn classify_photo_events(
                 if *is_trashed {
                     plan.gone.push(node_uid.to_string());
                 } else if !is_echo(node_uid) {
-                    // An untrash, a rename, a favourite toggled on the phone: the
+                    // An untrash, a rename, a favorite toggled on the phone: the
                     // event says which node changed but never what about it, so
                     // the uid is marked stale and the next refresh reads it.
                     plan.touched.push(node_uid.to_string());
@@ -607,7 +671,7 @@ pub(super) fn run_local_index(
             scan_local_once(&db, &indexing, &transfers, &mountpoint);
         }
         // Interruptible, so teardown joins this thread instead of leaving it to
-        // keep rewriting the index for a mount that is gone (bugs.md B44).
+        // keep rewriting the index for a mount that is gone (`docs/BUGS.md` B44).
         if !shutdown.sleep(LOCAL_INDEX_CHECK) {
             return;
         }
@@ -657,18 +721,26 @@ fn scan_local_once(
 
     // The walk has no idea how many files it will find, so the job counts what it
     // has seen and stays indeterminate.
-    let job = transfers.begin_job("Indexing this computer");
+    let job = transfers.begin_background_job("Indexing this computer");
     job.detail("Scanning your files");
+    let mut failed = false;
     let walked = localindex::scan(&[home], &excludes, |batch| {
         if let Err(e) = db.local_upsert_batch(generation, &batch) {
             warn!(error = %e, "local index: batch write failed");
+            failed = true;
         }
     });
 
-    // Prune what this scan did not see and rebuild the FTS index over the rest,
-    // even if some batches failed — a partial index still beats none.
+    // Prune what this scan did not see, unless it did not see everything:
+    // a path it never reached, or one whose batch failed to write, is not a
+    // path that is gone (`docs/BUGS.md` B217).
     job.detail("Building the search index");
-    match db.local_finish_scan(generation, now_secs()) {
+    let finished = if failed || walked >= localindex::MAX_ENTRIES {
+        db.local_finish_partial_scan(now_secs())
+    } else {
+        db.local_finish_scan(generation, now_secs())
+    };
+    match finished {
         Ok(indexed) => info!(
             walked,
             indexed,
@@ -710,6 +782,43 @@ mod tests {
         }
     }
 
+    #[test]
+    fn a_node_update_seen_again_is_skipped_but_other_events_are_not() {
+        let update = |id: &str| DriveEvent::NodeUpdated {
+            id: DriveEventId::from(id),
+            node_uid: uid("file"),
+            parent_node_uid: None,
+            is_trashed: false,
+            is_shared: false,
+        };
+        let mut recent = RecentEvents::default();
+        assert!(!recent.repeat(&update("e1")));
+        assert!(!recent.repeat(&update("e2")));
+        assert!(recent.repeat(&update("e1")), "a duplicate or replay");
+        let lost = DriveEvent::ContinuityLost {
+            id: DriveEventId::from("e2"),
+        };
+        assert!(!recent.repeat(&lost), "names the cursor, which was applied");
+    }
+
+    #[test]
+    fn recent_events_forget_the_oldest_past_their_bound() {
+        let update = |n: usize| DriveEvent::NodeUpdated {
+            id: DriveEventId::from(format!("e{n}")),
+            node_uid: uid("file"),
+            parent_node_uid: None,
+            is_trashed: false,
+            is_shared: false,
+        };
+        let mut recent = RecentEvents::default();
+        for n in 0..=RECENT_EVENTS {
+            assert!(!recent.repeat(&update(n)));
+        }
+        assert_eq!(recent.ids.len(), RECENT_EVENTS);
+        assert!(!recent.repeat(&update(0)), "the oldest was forgotten");
+        assert!(recent.repeat(&update(RECENT_EVENTS)));
+    }
+
     #[test]
     fn foreign_delete_invalidates_an_earlier_refresh_generation() {
         let own_volume = VolumeId::from("own");
@@ -974,7 +1083,7 @@ mod tests {
                 },
                 DriveEvent::NodeUpdated {
                     id: DriveEventId::from("e3"),
-                    node_uid: uid("favourited"),
+                    node_uid: uid("favorited"),
                     parent_node_uid: None,
                     is_trashed: false,
                     is_shared: false,
@@ -986,7 +1095,7 @@ mod tests {
             plan,
             PhotoEventPlan {
                 gone: vec![uid("deleted").to_string(), uid("trashed").to_string()],
-                touched: vec![uid("favourited").to_string()],
+                touched: vec![uid("favorited").to_string()],
                 full: false,
             }
         );

crates/pdfs-fuse/src/conflicts.rs

@@ -0,0 +1,179 @@
+//! Listing and resolving `(sync-conflict …)` copies for the user.
+//!
+//! The sweep ([`crate::sweep`]) removes copies it can prove identical and flags
+//! the rest once in the activity feed. That leaves divergent copies sitting in
+//! folders with nothing but a name to find them by. This module lists them with
+//! the file they are a copy of, and resolves one the way the user chooses:
+//!
+//! - **Keep the original**: trash the copy.
+//! - **Keep the copy**: trash the original, then give the copy its name.
+//! - **Keep both**: rename the copy to a name of the user's choosing.
+//!
+//! Every resolution goes through the Proton trash, never a delete, so a wrong
+//! choice is undone by restoring from Trash. Resolution refuses to touch a file
+//! that is open, being written, or still owed an upload, for the same reason the
+//! sweep does: the node is about to change under the decision.
+//!
+//! Copies under My Files are named by their mountpoint-relative path. Copies
+//! inside a synced folder are named by their absolute local path, so a resolve
+//! request routes to the on-demand mount that serves them. A copy inside a
+//! mirror folder is listed too, but it is a local file first: the sync engine
+//! owns it, so it is resolved by editing the folder, not through here.
+
+use super::*;
+use pdfs_core::control::{ConflictInfo, ConflictKeep};
+use sweep::conflict_base_name;
+
+impl Core {
+    /// Where the user finds a conflict copy: its mountpoint-relative path under
+    /// My Files, or its absolute local path inside a synced folder. `None` for a
+    /// node no location of this machine shows, such as one in another
+    /// computer's device tree.
+    ///
+    /// The sweep only flags what this can name, and the listing uses it too. A
+    /// flagged copy the listing cannot see reads as resolved in the activity
+    /// feed while it still sits in the folder (`docs/BUGS.md` B99).
+    pub(crate) fn conflict_path(&self, roots: &SearchRoots, uid: &NodeUid) -> Option<String> {
+        let uid = uid.to_string();
+        if let Ok(Some(path)) = self
+            .db
+            .path_relative_to(&self.primary_root_uid.to_string(), &uid)
+        {
+            return Some(path);
+        }
+        let drive_path = self.db.node_path(&uid).ok().flatten()?;
+        roots.resolve(&drive_path)
+    }
+
+    /// Every conflict copy this machine shows, with the file it is a copy of.
+    pub(crate) fn list_conflicts(&self) -> CoreResult<Vec<ConflictInfo>> {
+        let stored = self
+            .db
+            .load_all()
+            .map_err(|e| CoreError::internal(format!("listing conflicts: {e:?}")))?;
+        let nodes: Vec<Node> = stored
+            .into_iter()
+            .map(|stored| stored.node)
+            .filter(|node| !node.trashed && !node.is_folder() && self.is_own_or_virtual(&node.uid))
+            .collect();
+        let mut by_parent_name: HashMap<(Option<&NodeUid>, &str), &Node> = HashMap::new();
+        for node in &nodes {
+            by_parent_name.insert((node.parent_uid.as_ref(), node.name.as_str()), node);
+        }
+        let roots = self.search_roots();
+        let mut items = Vec::new();
+        for node in &nodes {
+            let Some(base) = conflict_base_name(&node.name) else {
+                continue;
+            };
+            let Some(path) = self.conflict_path(&roots, &node.uid) else {
+                continue;
+            };
+            let original = by_parent_name
+                .get(&(node.parent_uid.as_ref(), base.as_str()))
+                .copied();
+            items.push(ConflictInfo {
+                original_path: sibling_path(&path, &base),
+                original_exists: original.is_some(),
+                size: node_size(node),
+                modified: node.modification_time,
+                original_size: original.map(node_size),
+                original_modified: original.map(|o| o.modification_time),
+                identical: original.is_some_and(|o| {
+                    node_size(o) == node_size(node)
+                        && node_content_sha1(node)
+                            .is_some_and(|sha| node_content_sha1(o).as_deref() == Some(&sha))
+                }),
+                path,
+            });
+        }
+        items.sort_by(|a, b| a.path.cmp(&b.path));
+        Ok(items)
+    }
+
+    /// Resolve the conflict copy at mountpoint-relative `copy`. Returns a line
+    /// for the user saying what happened.
+    pub(crate) fn resolve_conflict(&self, copy: &Path, keep: &ConflictKeep) -> CoreResult<String> {
+        let copy_name = copy
+            .file_name()
+            .and_then(|n| n.to_str())
+            .ok_or_else(|| CoreError::invalid("the conflict copy has no file name"))?;
+        let base = conflict_base_name(copy_name)
+            .ok_or_else(|| CoreError::invalid(format!("{copy_name} is not a conflict copy")))?;
+        let original = copy.with_file_name(&base);
+        self.require_settled(copy)?;
+        let message = match keep {
+            ConflictKeep::Original => {
+                self.delete(copy)?;
+                self.log_activity(ActivityKind::Trash, copy_name, format!("kept {base}"), true);
+                format!("kept {base}; the copy is in Trash")
+            }
+            ConflictKeep::Copy => {
+                self.require_settled(&original)?;
+                self.delete(&original)?;
+                // The original is already in Trash. A failed rename leaves the
+                // copy under its conflict name, which loses nothing — say so
+                // rather than leave the user guessing where their file went.
+                self.rename(copy, &base).map_err(|e| {
+                    e.context(&format!(
+                        "{base} was moved to Trash, but the copy kept its conflict name"
+                    ))
+                })?;
+                self.log_activity(
+                    ActivityKind::Rename,
+                    &base,
+                    format!("kept the conflict copy {copy_name}"),
+                    true,
+                );
+                format!("kept the copy as {base}; the previous version is in Trash")
+            }
+            ConflictKeep::Both { name } => {
+                let name = name.trim();
+                if name == base || name == copy_name {
+                    return Err(CoreError::invalid("choose a name that no other file has"));
+                }
+                self.rename(copy, name)?;
+                self.log_activity(ActivityKind::Rename, name, format!("was {copy_name}"), true);
+                format!("kept both; the copy is now {name}")
+            }
+        };
+        Ok(message)
+    }
+
+    /// Refuse to resolve around a file that is open, being written, or still
+    /// owed an upload: the decision would be about content that is changing.
+    fn require_settled(&self, rel: &Path) -> CoreResult<()> {
+        let (_, uid) = self.resolve(rel)?;
+        let queued = self.db.has_any_op(&uid.to_string()).unwrap_or(true);
+        if queued || self.is_busy(&uid) {
+            return Err(CoreError::conflict(format!(
+                "{} is still being written or uploaded; try again once it has synced",
+                rel.display()
+            )));
+        }
+        self.conflict_notified.lock().remove(&uid);
+        Ok(())
+    }
+}
+
+/// The path of `name` next to the file at `path`.
+fn sibling_path(path: &str, name: &str) -> String {
+    match path.rsplit_once('/') {
+        Some((dir, _)) => format!("{dir}/{name}"),
+        None => name.to_string(),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn the_original_sits_next_to_its_copy() {
+        assert_eq!(
+            sibling_path("a/b/f (sync-conflict 1).txt", "f.txt"),
+            "a/b/f.txt"
+        );
+        assert_eq!(sibling_path("f (sync-conflict 1).txt", "f.txt"), "f.txt");
+    }
+}

crates/pdfs-fuse/src/devices.rs

@@ -1,5 +1,5 @@
 //! This machine as a Proton Drive Device, and the local folders it keeps in
-//! step with the remote (devices.md).
+//! step with the remote.
 //!
 //! A sync folder runs in one of two modes. `mirror` keeps a real local tree that
 //! the engine in [`super::sync`] reconciles both ways; `ondemand` evicts that
@@ -19,6 +19,7 @@ use pdfs_core::control::{
 };
 use pdfs_core::db::{StoredDevice, StoredSyncFolder};
 use pdfs_core::mounts::MountMode;
+use pdfs_core::profile::PROFILE_DIR_NAME;
 use pdfs_core::{CoreError, CoreResult};
 use proton_drive_rs::proton_sdk::ids::{DeviceUid, NodeUid};
 use proton_drive_rs::{DeviceType, Node};
@@ -27,7 +28,7 @@ use tracing::{info, warn};
 
 use super::sync::{self, base_name};
 use super::{
-    Core, SecondaryInsertRejection, SecondaryMount, State, SwitchBlocked, clear_stale_mount,
+    Core, SecondaryInsertRejection, SecondaryMount, SwitchBlocked, clear_stale_mount,
     device_type_str, dir_is_empty, evict_dir_contents, fuse_connection_id, is_stale_mount,
     now_secs, parse_uid, spawn_session, sync_folder_info, this_hostname,
 };
@@ -104,6 +105,15 @@ pub(crate) fn adopted_device_uid() -> Option<String> {
     (!uid.is_empty()).then_some(uid)
 }
 
+/// The paths no sync root may overlap, with the mountpoint as the config says
+/// now.
+pub(crate) fn sync_root_reserved() -> pdfs_core::syncroot::Reserved {
+    let config = pdfs_core::config::AppDirs::new()
+        .map(|dirs| dirs.load_config())
+        .unwrap_or_default();
+    pdfs_core::syncroot::Reserved::current(&config)
+}
+
 /// Pin (or, with `None`, unpin) the adopted device uid in `config.json`.
 pub(crate) fn set_adopted_device_uid(uid: Option<&str>) -> CoreResult<()> {
     let dirs = pdfs_core::config::AppDirs::new()
@@ -114,6 +124,22 @@ pub(crate) fn set_adopted_device_uid(uid: Option<&str>) -> CoreResult<()> {
         .map_err(|e| CoreError::internal(format!("write config: {e}")))
 }
 
+/// The name a folder moved in from `device` takes when `name` is already
+/// taken under this device's root, or `None` when it can keep its own.
+fn migrated_name(name: &str, device: &str, taken: &[String]) -> Option<String> {
+    let free = |candidate: &str| !taken.iter().any(|t| t == candidate);
+    if free(name) {
+        return None;
+    }
+    let base = format!("{name} ({device})");
+    if free(&base) {
+        return Some(base);
+    }
+    (2..)
+        .map(|n| format!("{base} {n}"))
+        .find(|candidate| free(candidate))
+}
+
 impl Core {
     // ---- devices ----------------------------------------------------------
 
@@ -128,6 +154,7 @@ impl Core {
             location.mounted = self.states.is_mounted_at(Path::new(&location.local_path));
             if let MountKind::Device { sync_folder_id } = &location.kind {
                 location.progress = progress.get(sync_folder_id).cloned();
+                location.paused = self.sync_folder_paused(*sync_folder_id);
             }
         }
         Ok(locations)
@@ -139,7 +166,7 @@ impl Core {
     pub(crate) fn list_devices(&self) -> CoreResult<Vec<DeviceInfo>> {
         let devices = self
             .rt
-            .block_on(self.client.enumerate_devices())
+            .block_on(self.drive.enumerate_devices())
             .map_err(|e| CoreError::from_api(&e, "list devices"))?;
         // No cached device row yet means this machine syncs nothing, so none of
         // the listed devices is ours.
@@ -183,8 +210,8 @@ impl Core {
     }
 
     /// Adopt an existing device as *this* machine's, pinning it in `config.json`
-    /// so hostname changes and reinstalls stop registering duplicates
-    /// (features.md 5.1). `None` clears the pin and returns to hostname matching.
+    /// so hostname changes and reinstalls stop registering duplicates.
+    /// `None` clears the pin and returns to hostname matching.
     ///
     /// Refuses a uid that is not in the account: the whole point of the pin is
     /// that it is authoritative, so accepting a bad one would defer the failure
@@ -196,7 +223,7 @@ impl Core {
         };
         let remote = self
             .rt
-            .block_on(self.client.enumerate_devices())
+            .block_on(self.drive.enumerate_devices())
             .map_err(|e| CoreError::from_api(&e, "enumerate devices"))?;
         let dev = remote
             .into_iter()
@@ -226,7 +253,141 @@ impl Core {
         Ok(format!("adopted device {name}"))
     }
 
-    // ---- device folder sync (devices.md, Phase 1) -------------------------
+    /// The device a restore reads from: this machine's own for `None`, else
+    /// the registered device with that uid.
+    ///
+    /// Another device is looked up without touching the cached device row —
+    /// restoring its folders here must not make this machine back up as it.
+    /// Its folders are then attached with that device's share id, so they sync
+    /// against the other computer's tree.
+    pub(crate) fn restore_source(&self, device: Option<&str>) -> CoreResult<StoredDevice> {
+        let Some(uid) = device else {
+            return self.ensure_device();
+        };
+        let remote = self
+            .rt
+            .block_on(self.drive.enumerate_devices())
+            .map_err(|e| CoreError::from_api(&e, "enumerate devices"))?;
+        let d = remote
+            .into_iter()
+            .find(|d| d.uid.to_string() == uid)
+            .ok_or_else(|| CoreError::not_found(format!("no device {uid} in this account")))?;
+        Ok(StoredDevice {
+            uid: d.uid.to_string(),
+            share_id: d.share_id.to_string(),
+            root_uid: d.root_folder_uid.to_string(),
+            name: d
+                .name
+                .as_deref()
+                .ok()
+                .unwrap_or("(unnamed device)")
+                .to_string(),
+            created: d.creation_time,
+        })
+    }
+
+    /// Move every folder another device backs up into this machine's own
+    /// device, so this machine carries on backing them up as itself. The other
+    /// device stays registered, without those folders.
+    ///
+    /// Device roots all live on the main volume, so each folder is one
+    /// server-side move: nothing is downloaded or uploaded, and no quota is
+    /// used twice. A folder whose name is taken here keeps its content under
+    /// "<name> (<device>)". A failed folder does not stop the others, and a
+    /// second run picks up whatever is still left on the other device.
+    ///
+    /// Nothing is synced to this disk here. The moved folders are now this
+    /// device's, so the ordinary restore offers them. Only folders that were
+    /// already synced here from the other device are touched: their uid
+    /// survives the move but their share does not.
+    pub(crate) fn migrate_device(&self, uid: &str) -> CoreResult<String> {
+        let source = self.restore_source(Some(uid))?;
+        let target = self.ensure_device()?;
+        if source.uid == target.uid {
+            return Err(CoreError::invalid(format!(
+                "{} is this computer's own device",
+                source.name
+            )));
+        }
+        let bad_root = |d: &StoredDevice| {
+            parse_uid(&d.root_uid)
+                .ok_or_else(|| CoreError::internal(format!("bad device root uid: {}", d.root_uid)))
+        };
+        let (source_root, target_root) = (bad_root(&source)?, bad_root(&target)?);
+        if source_root.volume_id != target_root.volume_id {
+            return Err(CoreError::invalid(
+                "the two devices are on different Proton Drive volumes; copy it instead",
+            ));
+        }
+
+        let mut taken: Vec<String> = self
+            .device_root_nodes(&target_root)?
+            .into_iter()
+            .filter(|n| !n.trashed)
+            .map(|n| n.name)
+            .collect();
+        let synced = self
+            .db
+            .sync_folder_list()
+            .map_err(|e| CoreError::internal(format!("db: {e:?}")))?;
+
+        let mut moved = 0usize;
+        let mut renamed: Vec<String> = Vec::new();
+        let mut left: Vec<String> = Vec::new();
+        let mut failed: Vec<String> = Vec::new();
+        for node in self.device_root_nodes(&source_root)? {
+            if node.trashed || node.name == PROFILE_DIR_NAME {
+                continue;
+            }
+            // A device root only ever holds folders (B68); anything else was
+            // put there by hand and is left for the user to deal with.
+            if !node.is_folder() {
+                left.push(node.name);
+                continue;
+            }
+            let new_name = migrated_name(&node.name, &source.name, &taken);
+            if let Err(e) = self.move_rename_remote(&node.uid, &target_root, new_name.as_deref()) {
+                warn!(name = %node.name, error = %e, "migrate: could not move folder");
+                failed.push(format!("{}: {e}", node.name));
+                continue;
+            }
+            let landed = new_name.clone().unwrap_or_else(|| node.name.clone());
+            if new_name.is_some() {
+                renamed.push(format!("{} -> {landed}", node.name));
+            }
+            taken.push(landed);
+            moved += 1;
+
+            let uid = node.uid.to_string();
+            for folder in synced.iter().filter(|f| f.remote_uid == uid) {
+                if let Err(e) = self.db.sync_folder_set_share(folder.id, &target.share_id) {
+                    warn!(id = folder.id, error = ?e, "migrate: could not re-point sync folder");
+                }
+                self.sync_now(Some(folder.id));
+            }
+        }
+        self.invalidate_children_of(&source_root);
+        self.invalidate_children_of(&target_root);
+
+        let mut message = format!("moved {moved} folder(s) from {}", source.name);
+        if !renamed.is_empty() {
+            message.push_str(&format!("; renamed: {}", renamed.join(", ")));
+        }
+        if !left.is_empty() {
+            message.push_str(&format!("; left behind: {}", left.join(", ")));
+        }
+        if !failed.is_empty() {
+            message.push_str(&format!("; failed: {}", failed.join(", ")));
+        }
+        info!(from = %source.name, to = %target.name, moved, failed = failed.len(),
+            "device migrated");
+        if moved == 0 && !failed.is_empty() {
+            return Err(CoreError::conflict(message));
+        }
+        Ok(message)
+    }
+
+    // ---- device folder sync -------------------------
 
     /// Auto-register (or recover) this machine as a Proton Drive Device, caching
     /// it so restarts reuse the same device. Recovery matches an existing remote
@@ -236,15 +397,14 @@ impl Core {
     /// Resolution order: the *adopted* uid pinned in `config.json`, then the
     /// cached DB row (validated remotely), then a hostname match, then create.
     /// The pin exists because the hostname heuristic silently registers a second
-    /// device after a rename or reinstall, orphaning the first one's folders
-    /// (features.md 5.1).
+    /// device after a rename or reinstall, orphaning the first one's folders.
     pub(crate) fn ensure_device(&self) -> CoreResult<StoredDevice> {
         let name = this_hostname();
         // Enumerate the remote devices once: used to validate the pin, validate
         // any cached record, and recover an existing device by name.
         let remote = self
             .rt
-            .block_on(self.client.enumerate_devices())
+            .block_on(self.drive.enumerate_devices())
             .map_err(|e| CoreError::from_api(&e, "enumerate devices"))?;
 
         // An adopted uid is an explicit instruction, so it outranks everything
@@ -303,7 +463,7 @@ impl Core {
             None => {
                 let d = self
                     .rt
-                    .block_on(self.client.create_device(&name, DeviceType::Linux))
+                    .block_on(self.drive.create_device(&name, DeviceType::Linux))
                     .map_err(|e| CoreError::from_api(&e, "create device"))?;
                 StoredDevice {
                     uid: d.uid.to_string(),
@@ -327,23 +487,27 @@ impl Core {
         root_uid: &NodeUid,
         name: &str,
     ) -> CoreResult<Option<NodeUid>> {
-        let uids = self
-            .rt
-            .block_on(self.client.enumerate_folder_children_node_uids(root_uid))
-            .map_err(|e| CoreError::from_api(&e, "list device root"))?;
-        if uids.is_empty() {
-            return Ok(None);
-        }
-        let nodes = self
-            .rt
-            .block_on(self.client.enumerate_nodes(&uids))
-            .map_err(|e| CoreError::from_api(&e, "resolve device root children"))?;
-        Ok(nodes
+        Ok(self
+            .device_root_nodes(root_uid)?
             .into_iter()
             .find(|n| n.is_folder() && !n.trashed && n.name == name)
             .map(|n| n.uid))
     }
 
+    /// Everything directly under a device root, trashed nodes included.
+    fn device_root_nodes(&self, root_uid: &NodeUid) -> CoreResult<Vec<Node>> {
+        let uids = self
+            .rt
+            .block_on(self.drive.enumerate_folder_children_node_uids(root_uid))
+            .map_err(|e| CoreError::from_api(&e, "list device root"))?;
+        if uids.is_empty() {
+            return Ok(Vec::new());
+        }
+        self.rt
+            .block_on(self.drive.enumerate_nodes(&uids))
+            .map_err(|e| CoreError::from_api(&e, "resolve device root children"))
+    }
+
     /// Add a local folder to this device's sync set: register the device if
     /// needed, create a matching folder under the device root, upload the local
     /// tree into it once, and record the mapping. Phase 1 is a one-shot upload —
@@ -362,18 +526,20 @@ impl Core {
             .map_err(|e| CoreError::internal(format!("canonicalize {}: {e}", local.display())))?;
         let local_str = local.to_string_lossy().to_string();
 
-        // Reject duplicates up front for a clear error (UNIQUE would also catch it).
-        if self
+        // `/`, the home folder, the mount, this app's own data, and a folder
+        // inside or around another root are refused (`syncroot::problem`). A
+        // duplicate is one of those, refused here for a clear error before
+        // the UNIQUE constraint would catch it.
+        let others = self
             .db
             .sync_folder_list()
             .map_err(|e| CoreError::internal(format!("db: {e:?}")))?
             .iter()
-            .any(|f| f.local_path == local_str)
-        {
-            return Err(CoreError::invalid(format!(
-                "{} is already synced",
-                local.display()
-            )));
+            .map(|f| PathBuf::from(&f.local_path))
+            .collect::<Vec<_>>();
+        let reserved = sync_root_reserved();
+        if let Some(problem) = pdfs_core::syncroot::problem(&local, &others, &reserved) {
+            return Err(CoreError::invalid(problem));
         }
 
         let name = local
@@ -412,10 +578,7 @@ impl Core {
             }
             None => self
                 .rt
-                .block_on(
-                    self.client
-                        .create_folder(&root_uid, &name, Some(now_secs())),
-                )
+                .block_on(self.drive.create_folder(&root_uid, &name, Some(now_secs())))
                 .map_err(|e| CoreError::from_api(&e, &format!("create device folder {name}")))?,
         };
 
@@ -448,7 +611,8 @@ impl Core {
             .into_iter()
             .map(|f| {
                 let live = progress.get(&f.id).cloned();
-                sync_folder_info(f, live)
+                let paused = self.sync_folder_paused(f.id);
+                sync_folder_info(f, live, paused)
             })
             .collect())
     }
@@ -504,12 +668,16 @@ impl Core {
                     } else {
                         p.total.max(p.done) as u64
                     },
+                    // Most passes find nothing; only one that queues work is
+                    // the drive syncing.
+                    background: true,
                 },
                 SyncPhase::Applying => JobItem {
                     title: format!("Syncing {folder}"),
                     detail: p.current.clone(),
                     done: p.done as u64,
                     total: p.total.max(p.done) as u64,
+                    background: false,
                 },
             });
         }
@@ -561,13 +729,14 @@ impl Core {
             return Err(CoreError::invalid(format!("no synced folder with id {id}")));
         }
         drop(_guard);
+        self.clear_sync_folder_paused(id);
         // Stop watching the folder we just dropped.
         let _ = self.sync_tx.send(sync::SyncMsg::Rewatch);
         if delete_remote
             && let Some(uid) = parse_uid(&folder.remote_uid)
             && let Err(e) = self
                 .rt
-                .block_on(self.client.trash_nodes(&[uid]))
+                .block_on(self.drive.trash_nodes(&[uid]))
                 .and_then(batch::into_unit)
         {
             warn!(id, error = %e, "delete remote device folder failed");
@@ -583,23 +752,16 @@ impl Core {
         };
     }
 
-    /// A sibling Core that shares this one's client/rt/cache/db (and transfer,
-    /// activity, mount registries) but gets a **fresh, empty `State`** — its own
-    /// inode space starting at [`ROOT_INO`]. Used to root a secondary FUSE session
-    /// at an `ondemand` sync folder without colliding with the main mount's inodes
-    /// (devices.md Phase 3).
+    /// A sibling Core for a secondary FUSE session rooted at an `ondemand` sync
+    /// folder. It shares everything with this one,
+    /// the tree included, except what belongs to a session: its root and its
+    /// kernel channel, both filled in when the session is built.
     pub(crate) fn fork_state(&self) -> Core {
         let mut fork = self.clone();
         fork.primary = false;
-        let share_access = self.state().share_access.clone();
-        fork.state = Arc::new(Mutex::new(State::new(self.db.clone(), share_access, 2)));
-        // A fresh inode space needs a fresh notification channel: this fork's
-        // session is the only one that knows these inodes, so it must be the one
-        // notified about them. Filled in by `spawn_ondemand_mount`.
+        fork.root = Arc::new(OnceLock::new());
         fork.notifier = Arc::new(OnceLock::new());
         fork.session_live = Arc::new(std::sync::atomic::AtomicBool::new(false));
-        // Size upgrades are keyed by inode, which is per-fork too.
-        fork.size_upgrades = Arc::new(Mutex::new(HashMap::new()));
         fork
     }
 
@@ -625,6 +787,13 @@ impl Core {
             .ok_or_else(|| CoreError::not_found(format!("no synced folder with id {id}")))?;
         let current_mode = ApplicableMountMode::parse(&folder.mode)
             .map_err(|error| CoreError::invalid(format!("unsupported current mode: {error}")))?;
+        // A switch waits on the folder's reconcile, which a pause skips; refuse
+        // it up front rather than queue it indefinitely.
+        if current_mode != target_mode && self.sync_folder_paused(id) {
+            return Err(CoreError::invalid(
+                "this folder is paused; resume it before changing its mode",
+            ));
+        }
         if current_mode == target_mode {
             let mode_str = target_mode.as_str();
             let lock = self.sync_lock(id);
@@ -810,12 +979,35 @@ impl Core {
                 if folder.state != "idle" {
                     return Err(SwitchBlocked::NotNow);
                 }
+                // `idle` only says the last pass applied everything it tried.
+                // It never tries an ignored path, a symlink or a folder it
+                // cannot read, it defers a file open for writing, and a file can
+                // change after it ends. The eviction below deletes all of those
+                // for good, so it waits until nothing here is missing from Drive
+                // (B198).
+                let baseline = self
+                    .db
+                    .sync_entries(id)
+                    .map_err(|e| SwitchBlocked::Failed(format!("db: {e:?}")))?;
+                let rules = self.ignore_rules(&local);
+                // Emptied before any pass saw it, the folder is still `idle`, and
+                // the check below would refuse the switch for good. The push pass
+                // of a queued switch forgets those paths instead (`docs/BUGS.md`
+                // B255).
+                if sync::mirror_tree_gone(&local, &baseline, &rules) {
+                    return Err(SwitchBlocked::NotNow);
+                }
+                if let Some(reason) = sync::mirror_tree_unsynced(&local, &baseline, &rules) {
+                    return Err(SwitchBlocked::Failed(format!(
+                        "going online-only would delete what is only on this computer: {reason}"
+                    )));
+                }
                 let root_uid = parse_uid(&folder.remote_uid).ok_or_else(|| {
                     SwitchBlocked::Failed(format!("bad remote uid: {}", folder.remote_uid))
                 })?;
                 let root = self
                     .rt
-                    .block_on(self.client.enumerate_nodes(std::slice::from_ref(&root_uid)))
+                    .block_on(self.drive.enumerate_nodes(std::slice::from_ref(&root_uid)))
                     .map_err(|e| SwitchBlocked::Failed(format!("fetch remote root: {e}")))?
                     .into_iter()
                     .next()
@@ -894,8 +1086,52 @@ impl Core {
         }
     }
 
-    /// Spawn a secondary FUSE session rooted at `root` over `local` on a forked
-    /// inode space. Clears any stale kernel mount first (a crashed run can leave
+    /// Forget what a mirror folder synced before, so its next pass compares the
+    /// two sides as on the first sync: what one side has alone is copied to the
+    /// other, and a file that differs is kept twice as a conflict. Nothing is
+    /// deleted on either side.
+    ///
+    /// This is the way out for a folder whose passes refuse to run because its
+    /// whole local copy, or its whole copy on Drive, is gone (#46). The baseline
+    /// is what makes that read as "delete everything", and it is what this
+    /// drops. A queued switch to on-demand is dropped with it: the push pass
+    /// that switch runs reads an unrecorded local file as new and would upload
+    /// a second copy of every file Drive already has.
+    pub(crate) fn reset_sync_folder(&self, id: i64) -> CoreResult<String> {
+        let lock = self.sync_lock(id);
+        let Some(_guard) = lock.try_lock() else {
+            return Err(CoreError::conflict(
+                "this folder is syncing; try again when the pass ends",
+            ));
+        };
+        let folder = self
+            .db
+            .sync_folder_get(id)
+            .map_err(|e| CoreError::internal(format!("db: {e:?}")))?
+            .ok_or_else(|| CoreError::not_found(format!("no synced folder with id {id}")))?;
+        if folder.mode != "mirror" {
+            return Err(CoreError::invalid(
+                "an online-only folder keeps nothing on this computer to sync from scratch",
+            ));
+        }
+        self.db
+            .sync_entries_clear(id)
+            .map_err(|e| CoreError::internal(format!("db: {e:?}")))?;
+        if folder.pending_mode.as_deref() == Some("ondemand") {
+            self.db
+                .sync_folder_set_pending_mode(id, None)
+                .map_err(|e| CoreError::internal(format!("db: {e:?}")))?;
+        }
+        let _ = self.sync_tx.send(sync::SyncMsg::Reconcile(id));
+        info!(id, path = %folder.local_path, "sync folder reset");
+        Ok(format!(
+            "{} will sync from scratch",
+            base_name(&folder.local_path)
+        ))
+    }
+
+    /// Spawn a secondary FUSE session rooted at `root` over `local` from a
+    /// [`Core::fork_state`] sibling. Clears any stale kernel mount first (a crashed run can leave
     /// one, which would fail the fresh mount with EBUSY).
     pub(crate) fn spawn_ondemand_mount(
         &self,
@@ -953,8 +1189,7 @@ impl Core {
     /// restart (their local dirs are empty on disk — the files live in the cloud).
     /// Best-effort per folder: a missing local path or a failed remote fetch marks
     /// the folder `error` and moves on rather than aborting the rest. Runs on its
-    /// own thread from `mount` so the network fetches never block startup
-    /// (devices.md Phase 4).
+    /// own thread from `mount` so the network fetches never block startup.
     pub(crate) fn restore_ondemand_mounts(&self) {
         let folders = match self.db.sync_folder_list() {
             Ok(f) => f,
@@ -974,7 +1209,7 @@ impl Core {
             };
             let root = match self
                 .rt
-                .block_on(self.client.enumerate_nodes(std::slice::from_ref(&root_uid)))
+                .block_on(self.drive.enumerate_nodes(std::slice::from_ref(&root_uid)))
             {
                 Ok(v) => match v.into_iter().next() {
                     Some(n) => n,
@@ -1076,10 +1311,38 @@ mod tests {
     use pdfs_core::db::StoredSyncFolder;
 
     use super::{
-        ApplicableMountMode, commit_after_teardown, pending_intent_matches,
+        ApplicableMountMode, commit_after_teardown, migrated_name, pending_intent_matches,
         restore_snapshot_matches, select_mode_transition,
     };
 
+    #[test]
+    fn migrated_folder_keeps_its_name_when_it_is_free() {
+        let taken = vec!["Pictures".to_string()];
+        assert_eq!(migrated_name("Documents", "DESKTOP-1", &taken), None);
+    }
+
+    #[test]
+    fn migrated_folder_is_named_after_its_device_on_a_clash() {
+        let taken = vec!["Documents".to_string()];
+        assert_eq!(
+            migrated_name("Documents", "DESKTOP-1", &taken).as_deref(),
+            Some("Documents (DESKTOP-1)")
+        );
+    }
+
+    #[test]
+    fn migrated_folder_is_numbered_when_the_device_name_is_taken_too() {
+        let taken = vec![
+            "Documents".to_string(),
+            "Documents (DESKTOP-1)".to_string(),
+            "Documents (DESKTOP-1) 2".to_string(),
+        ];
+        assert_eq!(
+            migrated_name("Documents", "DESKTOP-1", &taken).as_deref(),
+            Some("Documents (DESKTOP-1) 3")
+        );
+    }
+
     #[test]
     fn unknown_mode_cannot_select_a_destructive_transition() {
         let mut unmounts = 0;

crates/pdfs-fuse/src/diagnostics.rs

@@ -73,7 +73,29 @@ pub(crate) fn request_kind(line: &str) -> String {
         line.strip_prefix('"')
             .and_then(|rest| rest.split('"').next())
     };
-    name.unwrap_or("unknown").to_string()
+    let name = name.unwrap_or("unknown");
+    // A bare `Refresh` says nothing about which view wedged, and the three
+    // scopes cost wildly different amounts of work: a directory refresh is two
+    // cheap calls, a photos refresh can pull a whole library. The scope is one
+    // more peek into the same line.
+    if name == "Refresh"
+        && let Some(scope) = tagged_name_after(line, "\"scope\"")
+    {
+        return format!("Refresh({scope})");
+    }
+    name.to_string()
+}
+
+/// The variant name serde wrote for the value of `key`, whether that value is a
+/// bare string (a unit variant) or an object whose single key is the name.
+fn tagged_name_after(line: &str, key: &str) -> Option<String> {
+    let rest = line.split_once(key)?.1.trim_start();
+    let rest = rest.strip_prefix(':')?.trim_start();
+    let rest = rest.strip_prefix('{').map_or(rest, str::trim_start);
+    rest.strip_prefix('"')?
+        .split('"')
+        .next()
+        .map(str::to_string)
 }
 
 /// Resident set size in bytes, or `0` when `/proc` did not answer.
@@ -141,6 +163,7 @@ impl Core {
             control_handlers: super::control::active_handlers(),
             control_handler_limit: super::control::handler_limit(),
             pending_ops: self.pending_ops_quickly(),
+            sync: Some(self.sync_engine.report()),
         }
     }
 
@@ -181,6 +204,26 @@ mod tests {
         assert_eq!(request_kind("garbage"), "unknown");
     }
 
+    /// A stuck `Refresh` is only actionable if the report says which view it
+    /// was refreshing, so the scope rides along with the name.
+    #[test]
+    fn a_refresh_reports_its_scope() {
+        assert_eq!(
+            request_kind("{\"Refresh\":{\"scope\":{\"Photos\":{\"full\":true}}}}"),
+            "Refresh(Photos)"
+        );
+        assert_eq!(
+            request_kind("{\"Refresh\":{\"scope\":{\"Dir\":{\"path\":\"a\"}}}}"),
+            "Refresh(Dir)"
+        );
+        assert_eq!(
+            request_kind("{\"Refresh\":{\"scope\":\"Trash\"}}"),
+            "Refresh(Trash)"
+        );
+        // A shape this does not recognise still names the request.
+        assert_eq!(request_kind("{\"Refresh\":{}}"), "Refresh");
+    }
+
     /// A guard registers for its lifetime and nothing longer: a handler that
     /// returns must not leave a permanent "in flight" entry behind.
     #[test]

crates/pdfs-fuse/src/drive.rs

@@ -0,0 +1,524 @@
+//! The calls the sync core makes to Drive, behind a trait.
+//!
+//! `Core` used to hold a concrete [`ProtonDriveClient`], so no test could make
+//! Drive answer late, answer stale, or not answer at all, and the interleavings
+//! behind most recent bugs could only be found on a real account over a slow
+//! link. [`DriveApi`] is the part of the
+//! client the mount, the drain, the event feed and the mirror engine use; the
+//! real client implements it by forwarding, and the simulation tests implement
+//! it with an in-memory Drive that injects Drive's known faults.
+//!
+//! Sharing, Photos, invitations, public links and the rest of the device calls
+//! stay on the concrete client (`Core::client`) until a test needs them faked.
+//! Listing and registering devices are here, so a simulated daemon can sync a
+//! local folder.
+//!
+//! The method names match the client's own, so a call site reads the same
+//! whichever it goes through. The client's generic methods (a reader or writer
+//! of any type) cannot be part of an object-safe trait, so the trait takes them
+//! as trait objects and `impl dyn DriveApi` puts the generic signatures back.
+
+use std::io::{Read, Write};
+use std::sync::Arc;
+
+use async_trait::async_trait;
+use futures::StreamExt as _;
+use futures::stream::BoxStream;
+use proton_drive_rs::proton_sdk::account::Quota;
+use proton_drive_rs::proton_sdk::error::Result;
+use proton_drive_rs::proton_sdk::ids::{DriveEventId, NodeUid};
+use proton_drive_rs::{
+    AvailableNameTarget, Device, DeviceType, DriveEvent, DriveEventScopeId, Node, NodeMoveItem,
+    ProtonDriveClient, RevisionReader, Thumbnail,
+};
+
+/// One outcome per node of a batch request; see [`ProtonDriveClient::trash_nodes`].
+pub(crate) type NodeOutcomes = Vec<(NodeUid, Result<()>)>;
+
+/// The outcomes of a streaming batch request, one item per node as it lands.
+pub(crate) type OutcomeStream<'a> = BoxStream<'a, Result<(NodeUid, Result<()>)>>;
+
+/// Drive as the sync core sees it. See the module documentation.
+#[async_trait]
+pub(crate) trait DriveApi: Send + Sync {
+    async fn get_my_files_folder(&self) -> Result<Node>;
+
+    async fn get_node(&self, uid: &NodeUid) -> Result<Option<Node>>;
+
+    async fn enumerate_nodes(&self, uids: &[NodeUid]) -> Result<Vec<Node>>;
+
+    /// As [`DriveApi::enumerate_nodes`], but files carry no claimed size or
+    /// modification time; see [`ProtonDriveClient::enumerate_nodes_light`].
+    async fn enumerate_nodes_light(&self, uids: &[NodeUid]) -> Result<Vec<Node>>;
+
+    async fn enumerate_folder_children_node_uids(
+        &self,
+        folder_uid: &NodeUid,
+    ) -> Result<Vec<NodeUid>>;
+
+    async fn enumerate_events(
+        &self,
+        scope: &DriveEventScopeId,
+        cursor: Option<&DriveEventId>,
+    ) -> Result<Vec<DriveEvent>>;
+
+    async fn invalidate_caches_for_event(&self, event: &DriveEvent) -> Result<()>;
+
+    async fn create_folder(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        last_modification_time: Option<i64>,
+    ) -> Result<NodeUid>;
+
+    async fn upload_file(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        media_type: &str,
+        contents: &[u8],
+    ) -> Result<NodeUid>;
+
+    #[allow(clippy::too_many_arguments)]
+    async fn upload_file_from_dyn(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        media_type: &str,
+        reader: &mut (dyn Read + Send),
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+        aead: bool,
+    ) -> Result<NodeUid>;
+
+    #[allow(clippy::too_many_arguments)]
+    async fn upload_file_replacing_draft_from_dyn(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        media_type: &str,
+        reader: &mut (dyn Read + Send),
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+        aead: bool,
+    ) -> Result<NodeUid>;
+
+    async fn upload_new_revision_from_dyn(
+        &self,
+        file_uid: &NodeUid,
+        reader: &mut (dyn Read + Send),
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+    ) -> Result<()>;
+
+    async fn open_revision(&self, uid: &NodeUid) -> Result<Arc<dyn RevisionRead>>;
+
+    async fn download_file_to_dyn(
+        &self,
+        uid: &NodeUid,
+        output: &mut (dyn Write + Send),
+    ) -> Result<()>;
+
+    async fn download_revision_to_dyn(
+        &self,
+        file_uid: &NodeUid,
+        revision_id: &str,
+        writer: &mut (dyn Write + Send),
+    ) -> Result<()>;
+
+    async fn rename_node(
+        &self,
+        uid: &NodeUid,
+        new_name: &str,
+        new_media_type: Option<&str>,
+    ) -> Result<()>;
+
+    async fn move_node(&self, uid: &NodeUid, new_parent: &NodeUid) -> Result<()>;
+
+    /// See [`ProtonDriveClient::copy_node`].
+    async fn copy_node(
+        &self,
+        uid: &NodeUid,
+        new_parent: &NodeUid,
+        name: Option<&str>,
+    ) -> Result<NodeUid>;
+
+    /// See [`ProtonDriveClient::get_available_name`].
+    async fn get_available_name(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        target: AvailableNameTarget,
+    ) -> Result<String>;
+
+    fn move_nodes_streaming<'a>(
+        &'a self,
+        items: Vec<NodeMoveItem>,
+        new_parent: NodeUid,
+    ) -> OutcomeStream<'a>;
+
+    async fn trash_nodes(&self, uids: &[NodeUid]) -> Result<NodeOutcomes>;
+
+    async fn restore_nodes(&self, uids: &[NodeUid]) -> Result<NodeOutcomes>;
+
+    fn restore_nodes_streaming<'a>(&'a self, uids: &[NodeUid]) -> OutcomeStream<'a>;
+
+    fn delete_nodes_streaming<'a>(&'a self, uids: &[NodeUid]) -> OutcomeStream<'a>;
+
+    async fn enumerate_trash_node_uids(&self) -> Result<Vec<NodeUid>>;
+
+    /// The account's storage, across every Proton product.
+    async fn quota(&self) -> Result<Quota>;
+
+    /// Every computer that backs up to the account.
+    async fn enumerate_devices(&self) -> Result<Vec<Device>>;
+
+    /// Register a computer, with a root folder of its own named `name`.
+    async fn create_device(&self, name: &str, device_type: DeviceType) -> Result<Device>;
+
+    /// Called by the drain once it has let go of the blob of an upload to
+    /// `uid` that landed, or retired the create that made `uid`, before the
+    /// tree has the node, so a simulation can hold it there. Blocks the drain
+    /// worker it is called on.
+    #[cfg(test)]
+    fn landing(&self, _uid: &NodeUid) {}
+}
+
+/// The generic signatures of the client, over the object-safe ones above.
+impl dyn DriveApi {
+    #[allow(clippy::too_many_arguments)]
+    pub(crate) async fn upload_file_from<R: Read + Send>(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        media_type: &str,
+        mut reader: R,
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+        aead: bool,
+    ) -> Result<NodeUid> {
+        self.upload_file_from_dyn(
+            parent_uid,
+            name,
+            media_type,
+            &mut reader,
+            intended_size,
+            thumbnails,
+            last_modification_time,
+            aead,
+        )
+        .await
+    }
+
+    #[allow(clippy::too_many_arguments)]
+    pub(crate) async fn upload_file_replacing_draft_from<R: Read + Send>(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        media_type: &str,
+        mut reader: R,
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+        aead: bool,
+    ) -> Result<NodeUid> {
+        self.upload_file_replacing_draft_from_dyn(
+            parent_uid,
+            name,
+            media_type,
+            &mut reader,
+            intended_size,
+            thumbnails,
+            last_modification_time,
+            aead,
+        )
+        .await
+    }
+
+    pub(crate) async fn upload_new_revision_from<R: Read + Send>(
+        &self,
+        file_uid: &NodeUid,
+        mut reader: R,
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+    ) -> Result<()> {
+        self.upload_new_revision_from_dyn(
+            file_uid,
+            &mut reader,
+            intended_size,
+            thumbnails,
+            last_modification_time,
+        )
+        .await
+    }
+
+    pub(crate) async fn download_file_to<W: Write + Send>(
+        &self,
+        uid: &NodeUid,
+        output: &mut W,
+    ) -> Result<()> {
+        self.download_file_to_dyn(uid, output).await
+    }
+
+    pub(crate) async fn download_revision_to<W: Write + Send>(
+        &self,
+        file_uid: &NodeUid,
+        revision_id: &str,
+        writer: &mut W,
+    ) -> Result<()> {
+        self.download_revision_to_dyn(file_uid, revision_id, writer)
+            .await
+    }
+}
+
+/// An open revision that block reads are served from; see [`RevisionReader`].
+#[async_trait]
+pub(crate) trait RevisionRead: Send + Sync {
+    /// Plaintext size of the revision.
+    fn size(&self) -> u64;
+
+    /// Plaintext size of each block, in block order.
+    fn block_sizes(&self) -> &[u64];
+
+    async fn read_at(&self, offset: u64, length: u64) -> Result<Vec<u8>>;
+}
+
+#[async_trait]
+impl RevisionRead for RevisionReader {
+    fn size(&self) -> u64 {
+        RevisionReader::size(self)
+    }
+
+    fn block_sizes(&self) -> &[u64] {
+        RevisionReader::block_sizes(self)
+    }
+
+    async fn read_at(&self, offset: u64, length: u64) -> Result<Vec<u8>> {
+        RevisionReader::read_at(self, offset, length).await
+    }
+}
+
+#[async_trait]
+impl DriveApi for ProtonDriveClient {
+    async fn get_my_files_folder(&self) -> Result<Node> {
+        ProtonDriveClient::get_my_files_folder(self).await
+    }
+
+    async fn get_node(&self, uid: &NodeUid) -> Result<Option<Node>> {
+        ProtonDriveClient::get_node(self, uid).await
+    }
+
+    async fn enumerate_nodes(&self, uids: &[NodeUid]) -> Result<Vec<Node>> {
+        ProtonDriveClient::enumerate_nodes(self, uids).await
+    }
+
+    async fn enumerate_nodes_light(&self, uids: &[NodeUid]) -> Result<Vec<Node>> {
+        ProtonDriveClient::enumerate_nodes_light(self, uids).await
+    }
+
+    async fn enumerate_folder_children_node_uids(
+        &self,
+        folder_uid: &NodeUid,
+    ) -> Result<Vec<NodeUid>> {
+        ProtonDriveClient::enumerate_folder_children_node_uids(self, folder_uid).await
+    }
+
+    async fn enumerate_events(
+        &self,
+        scope: &DriveEventScopeId,
+        cursor: Option<&DriveEventId>,
+    ) -> Result<Vec<DriveEvent>> {
+        ProtonDriveClient::enumerate_events(self, scope, cursor).await
+    }
+
+    async fn invalidate_caches_for_event(&self, event: &DriveEvent) -> Result<()> {
+        ProtonDriveClient::invalidate_caches_for_event(self, event).await
+    }
+
+    async fn create_folder(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        last_modification_time: Option<i64>,
+    ) -> Result<NodeUid> {
+        ProtonDriveClient::create_folder(self, parent_uid, name, last_modification_time).await
+    }
+
+    async fn upload_file(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        media_type: &str,
+        contents: &[u8],
+    ) -> Result<NodeUid> {
+        ProtonDriveClient::upload_file(self, parent_uid, name, media_type, contents).await
+    }
+
+    async fn upload_file_from_dyn(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        media_type: &str,
+        reader: &mut (dyn Read + Send),
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+        aead: bool,
+    ) -> Result<NodeUid> {
+        ProtonDriveClient::upload_file_from(
+            self,
+            parent_uid,
+            name,
+            media_type,
+            reader,
+            intended_size,
+            thumbnails,
+            last_modification_time,
+            aead,
+        )
+        .await
+    }
+
+    async fn upload_file_replacing_draft_from_dyn(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        media_type: &str,
+        reader: &mut (dyn Read + Send),
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+        aead: bool,
+    ) -> Result<NodeUid> {
+        ProtonDriveClient::upload_file_replacing_draft_from(
+            self,
+            parent_uid,
+            name,
+            media_type,
+            reader,
+            intended_size,
+            thumbnails,
+            last_modification_time,
+            aead,
+        )
+        .await
+    }
+
+    async fn upload_new_revision_from_dyn(
+        &self,
+        file_uid: &NodeUid,
+        reader: &mut (dyn Read + Send),
+        intended_size: i64,
+        thumbnails: Vec<Thumbnail>,
+        last_modification_time: Option<i64>,
+    ) -> Result<()> {
+        ProtonDriveClient::upload_new_revision_from(
+            self,
+            file_uid,
+            reader,
+            intended_size,
+            thumbnails,
+            last_modification_time,
+        )
+        .await
+    }
+
+    async fn open_revision(&self, uid: &NodeUid) -> Result<Arc<dyn RevisionRead>> {
+        let reader = ProtonDriveClient::open_revision(self, uid).await?;
+        Ok(Arc::new(reader))
+    }
+
+    async fn download_file_to_dyn(
+        &self,
+        uid: &NodeUid,
+        output: &mut (dyn Write + Send),
+    ) -> Result<()> {
+        ProtonDriveClient::download_file_to(self, uid, &mut &mut *output).await
+    }
+
+    async fn download_revision_to_dyn(
+        &self,
+        file_uid: &NodeUid,
+        revision_id: &str,
+        writer: &mut (dyn Write + Send),
+    ) -> Result<()> {
+        ProtonDriveClient::download_revision_to(self, file_uid, revision_id, &mut &mut *writer)
+            .await
+    }
+
+    async fn rename_node(
+        &self,
+        uid: &NodeUid,
+        new_name: &str,
+        new_media_type: Option<&str>,
+    ) -> Result<()> {
+        ProtonDriveClient::rename_node(self, uid, new_name, new_media_type).await
+    }
+
+    async fn move_node(&self, uid: &NodeUid, new_parent: &NodeUid) -> Result<()> {
+        ProtonDriveClient::move_node(self, uid, new_parent).await
+    }
+
+    async fn copy_node(
+        &self,
+        uid: &NodeUid,
+        new_parent: &NodeUid,
+        name: Option<&str>,
+    ) -> Result<NodeUid> {
+        ProtonDriveClient::copy_node(self, uid, new_parent, name).await
+    }
+
+    async fn get_available_name(
+        &self,
+        parent_uid: &NodeUid,
+        name: &str,
+        target: AvailableNameTarget,
+    ) -> Result<String> {
+        ProtonDriveClient::get_available_name(self, parent_uid, name, target).await
+    }
+
+    fn move_nodes_streaming<'a>(
+        &'a self,
+        items: Vec<NodeMoveItem>,
+        new_parent: NodeUid,
+    ) -> OutcomeStream<'a> {
+        ProtonDriveClient::move_nodes_streaming(self, items, new_parent).boxed()
+    }
+
+    async fn trash_nodes(&self, uids: &[NodeUid]) -> Result<NodeOutcomes> {
+        ProtonDriveClient::trash_nodes(self, uids).await
+    }
+
+    async fn restore_nodes(&self, uids: &[NodeUid]) -> Result<NodeOutcomes> {
+        ProtonDriveClient::restore_nodes(self, uids).await
+    }
+
+    fn restore_nodes_streaming<'a>(&'a self, uids: &[NodeUid]) -> OutcomeStream<'a> {
+        ProtonDriveClient::restore_nodes_streaming(self, uids).boxed()
+    }
+
+    fn delete_nodes_streaming<'a>(&'a self, uids: &[NodeUid]) -> OutcomeStream<'a> {
+        ProtonDriveClient::delete_nodes_streaming(self, uids).boxed()
+    }
+
+    async fn enumerate_trash_node_uids(&self) -> Result<Vec<NodeUid>> {
+        ProtonDriveClient::enumerate_trash_node_uids(self).await
+    }
+
+    async fn quota(&self) -> Result<Quota> {
+        ProtonDriveClient::quota(self).await
+    }
+
+    async fn enumerate_devices(&self) -> Result<Vec<Device>> {
+        ProtonDriveClient::enumerate_devices(self).await
+    }
+
+    async fn create_device(&self, name: &str, device_type: DeviceType) -> Result<Device> {
+        ProtonDriveClient::create_device(self, name, device_type).await
+    }
+}

crates/pdfs-fuse/src/events.rs

@@ -0,0 +1,338 @@
+//! Push notifications for front-ends: the daemon's side of
+//! [`Request::Subscribe`](pdfs_core::control::Request::Subscribe).
+//!
+//! Front-ends used to poll. The app asked for the status every two seconds, the
+//! tray every three, and pages re-read themselves on a timer, whether or not
+//! anything had changed. Now each subscription owns a [`Mailbox`] in the
+//! [`EventHub`]. Code that changes something publishes a [`Topic`], and
+//! [`run_event_sampler`] watches the few things that change without passing
+//! through one place (the status, transfers, sync progress) and publishes them
+//! when they differ.
+
+use std::collections::{BTreeSet, HashMap};
+use std::sync::Arc;
+use std::time::Duration;
+
+use parking_lot::{Condvar, Mutex, MutexGuard};
+use pdfs_core::control::{Event, JobItem, Response as CtlResponse, Topic, TransferItem};
+use tracing::warn;
+
+use super::Core;
+
+/// Most subscriptions served at once. Each holds a control handler for as long
+/// as it lasts; the app and the tray need one apiece, and the cap keeps a
+/// misbehaving client from spending the handlers that requests need.
+const MAX_SUBSCRIBERS: usize = 8;
+
+/// How long a subscription waits after a change before it sends, so a burst
+/// (a folder of files trashed at once) goes out as one event.
+const COALESCE: Duration = Duration::from_millis(150);
+
+/// How often [`run_event_sampler`] looks at transfers. Status and sync
+/// progress are looked at every other time.
+const SAMPLE_INTERVAL: Duration = Duration::from_millis(500);
+
+/// Every open subscription's mailbox. Shared by every clone of [`Core`],
+/// on-demand forks included, so a change in any mount reaches every client.
+#[derive(Default)]
+pub(crate) struct EventHub {
+    mailboxes: Mutex<Vec<Arc<Mailbox>>>,
+}
+
+/// What one subscription has yet to send.
+#[derive(Default)]
+pub(crate) struct Mailbox {
+    inbox: Mutex<Inbox>,
+    ready: Condvar,
+}
+
+#[derive(Default)]
+struct Inbox {
+    topics: BTreeSet<Topic>,
+    /// The newest [`Event::Status`] line. A newer one replaces it unsent: a
+    /// client wants the current status, not every status in between.
+    status: Option<Arc<str>>,
+    /// The newest [`Event::Transfers`] line, replaced the same way.
+    transfers: Option<Arc<str>>,
+}
+
+impl Inbox {
+    fn is_empty(&self) -> bool {
+        self.topics.is_empty() && self.status.is_none() && self.transfers.is_none()
+    }
+}
+
+/// `event` as one line of the wire format, newline included.
+pub(crate) fn event_line(event: &Event) -> Option<Arc<str>> {
+    match serde_json::to_string(event) {
+        Ok(mut line) => {
+            line.push('\n');
+            Some(line.into())
+        }
+        Err(e) => {
+            warn!(error = %e, "events: serialize event failed");
+            None
+        }
+    }
+}
+
+impl EventHub {
+    /// Tell every subscriber that `topics` changed.
+    pub(crate) fn publish(&self, topics: &[Topic]) {
+        self.deliver(|inbox| inbox.topics.extend(topics.iter().copied()));
+    }
+
+    /// Hand every subscriber a fresh status.
+    fn publish_status(&self, line: Arc<str>) {
+        self.deliver(|inbox| inbox.status = Some(line.clone()));
+    }
+
+    /// Hand every subscriber a fresh transfers snapshot.
+    fn publish_transfers(&self, line: Arc<str>) {
+        self.deliver(|inbox| inbox.transfers = Some(line.clone()));
+    }
+
+    fn deliver(&self, put: impl Fn(&mut Inbox)) {
+        for mailbox in self.mailboxes.lock().iter() {
+            put(&mut mailbox.inbox.lock());
+            mailbox.ready.notify_one();
+        }
+    }
+
+    /// Open a mailbox, or `None` at [`MAX_SUBSCRIBERS`].
+    pub(crate) fn subscribe(&self) -> Option<Arc<Mailbox>> {
+        let mut mailboxes = self.mailboxes.lock();
+        if mailboxes.len() >= MAX_SUBSCRIBERS {
+            return None;
+        }
+        let mailbox = Arc::new(Mailbox::default());
+        mailboxes.push(mailbox.clone());
+        Some(mailbox)
+    }
+
+    pub(crate) fn unsubscribe(&self, mailbox: &Arc<Mailbox>) {
+        self.mailboxes
+            .lock()
+            .retain(|other| !Arc::ptr_eq(other, mailbox));
+    }
+
+    fn has_subscribers(&self) -> bool {
+        !self.mailboxes.lock().is_empty()
+    }
+}
+
+impl Mailbox {
+    /// Wait up to `timeout` for something to send, then take it as wire lines:
+    /// the status first, then transfers, then the changed topics. Empty when
+    /// nothing arrived in time.
+    pub(crate) fn take(&self, timeout: Duration) -> Vec<Arc<str>> {
+        let mut inbox = self.inbox.lock();
+        if inbox.is_empty() {
+            self.ready.wait_for(&mut inbox, timeout);
+            if inbox.is_empty() {
+                return Vec::new();
+            }
+            // Let the rest of a burst arrive before sending the first of it.
+            MutexGuard::unlocked(&mut inbox, || std::thread::sleep(COALESCE));
+        }
+        let mut lines = Vec::new();
+        lines.extend(inbox.status.take());
+        lines.extend(inbox.transfers.take());
+        if !inbox.topics.is_empty() {
+            let topics = std::mem::take(&mut inbox.topics).into_iter().collect();
+            lines.extend(event_line(&Event::Changed { topics }));
+        }
+        lines
+    }
+}
+
+/// The [`Event::Transfers`] line for a snapshot.
+pub(crate) fn transfers_line(items: Vec<TransferItem>, jobs: Vec<JobItem>) -> Option<Arc<str>> {
+    event_line(&Event::Transfers { items, jobs })
+}
+
+/// Watch what changes without a single place to publish it from, and publish it
+/// when it differs from the last look: the status (queue counts, cache use,
+/// online, pause), the transfers snapshot, the synced folders with their
+/// live progress, and which nodes have a sync issue. Looks only while someone
+/// is subscribed.
+pub(crate) fn run_event_sampler(core: Core, username: String, mountpoint: std::path::PathBuf) {
+    let mut transfers: Option<Arc<str>> = None;
+    let mut status: Option<String> = None;
+    let mut queue: Option<String> = None;
+    let mut locations: Option<String> = None;
+    let mut issues: Option<HashMap<i64, String>> = None;
+    let mut tick = 0u32;
+    while core.shutdown.sleep(SAMPLE_INTERVAL) {
+        if !core.events.has_subscribers() {
+            // A new subscription starts from a snapshot of its own, so there is
+            // nothing to compare against until then.
+            transfers = None;
+            status = None;
+            queue = None;
+            locations = None;
+            issues = None;
+            continue;
+        }
+        let line = transfers_line(core.transfers.snapshot(), core.jobs_snapshot());
+        if line.is_some() && line != transfers {
+            transfers = line.clone();
+            if let Some(line) = line {
+                core.events.publish_transfers(line);
+            }
+        }
+
+        tick = tick.wrapping_add(1);
+        if !tick.is_multiple_of(2) {
+            continue;
+        }
+        let now = super::control::status_response(&core, &username, &mountpoint);
+        let (key, queue_key) = status_keys(&now);
+        if status.as_ref().is_some_and(|last| *last != key)
+            && let Some(line) = event_line(&Event::Status(Box::new(now)))
+        {
+            core.events.publish_status(line);
+        }
+        status = Some(key);
+        if queue.as_ref().is_some_and(|last| *last != queue_key) {
+            core.events.publish(&[Topic::Queue]);
+        }
+        queue = Some(queue_key);
+
+        // An issue shows on its node in the file browser: raised, cleared,
+        // landed or discarded, the listing is out of date.
+        if let Ok(now) = core.db.node_issues() {
+            if issues.as_ref().is_some_and(|last| *last != now) {
+                core.events.publish(&[Topic::Files]);
+            }
+            issues = Some(now);
+        }
+
+        if let Ok(now) = core.list_locations() {
+            let key = serde_json::to_string(&now).unwrap_or_default();
+            if locations.as_ref().is_some_and(|last| *last != key) {
+                core.events.publish(&[Topic::Locations]);
+            }
+            locations = Some(key);
+        }
+    }
+}
+
+/// What makes two statuses different enough to send, and what makes their
+/// queues different. The age of the oldest staged write is left out of both:
+/// it grows every second while anything is staged, which is not news.
+fn status_keys(status: &CtlResponse) -> (String, String) {
+    let CtlResponse::Status {
+        parked_uploads,
+        failing_ops,
+        failing_error,
+        staged_bytes,
+        paused,
+        paused_until,
+        username,
+        mountpoint,
+        pinned,
+        used,
+        budget,
+        online,
+        session_expired,
+        tokens_unsaved,
+        pending_uploads,
+        pending_changes,
+        ..
+    } = status
+    else {
+        return (String::new(), String::new());
+    };
+    let queue = format!(
+        "{pending_uploads}/{pending_changes}/{parked_uploads}/{failing_ops}/{failing_error:?}"
+    );
+    let key = format!(
+        "{queue}/{staged_bytes}/{paused}/{paused_until:?}/{username}/{mountpoint}/{pinned}/\
+         {used}/{budget}/{online}/{session_expired}/{tokens_unsaved}"
+    );
+    (key, queue)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn changed(line: &str) -> Vec<Topic> {
+        match serde_json::from_str::<Event>(line.trim()) {
+            Ok(Event::Changed { topics }) => topics,
+            other => panic!("not a change event: {other:?}"),
+        }
+    }
+
+    #[test]
+    fn a_burst_of_changes_arrives_as_one_event() {
+        let hub = EventHub::default();
+        let mailbox = hub.subscribe().expect("room for a subscriber");
+        hub.publish(&[Topic::Trash, Topic::Files]);
+        hub.publish(&[Topic::Files]);
+        let lines = mailbox.take(Duration::from_secs(1));
+        assert_eq!(lines.len(), 1);
+        assert_eq!(changed(&lines[0]), vec![Topic::Files, Topic::Trash]);
+        assert!(mailbox.take(Duration::from_millis(10)).is_empty());
+    }
+
+    #[test]
+    fn a_newer_status_replaces_one_not_yet_sent() {
+        let hub = EventHub::default();
+        let mailbox = hub.subscribe().expect("room for a subscriber");
+        hub.publish_status("first\n".into());
+        hub.publish_status("second\n".into());
+        assert_eq!(
+            mailbox.take(Duration::from_secs(1)),
+            vec![Arc::<str>::from("second\n")]
+        );
+    }
+
+    #[test]
+    fn a_closed_subscription_gets_nothing_more() {
+        let hub = EventHub::default();
+        let mailbox = hub.subscribe().expect("room for a subscriber");
+        hub.unsubscribe(&mailbox);
+        hub.publish(&[Topic::Files]);
+        assert!(mailbox.take(Duration::from_millis(10)).is_empty());
+        assert!(!hub.has_subscribers());
+    }
+
+    #[test]
+    fn subscriptions_are_capped() {
+        let hub = EventHub::default();
+        let open: Vec<_> = (0..MAX_SUBSCRIBERS)
+            .map(|_| hub.subscribe().expect("room for a subscriber"))
+            .collect();
+        assert!(hub.subscribe().is_none());
+        hub.unsubscribe(&open[0]);
+        assert!(hub.subscribe().is_some());
+    }
+
+    #[test]
+    fn the_age_of_a_staged_write_is_not_a_change() {
+        let status = |staged_oldest_secs| CtlResponse::Status {
+            parked_uploads: 0,
+            failing_ops: 0,
+            failing_error: None,
+            staged_bytes: 10,
+            staged_oldest_secs,
+            paused: false,
+            paused_until: None,
+            username: "u".into(),
+            mountpoint: "/m".into(),
+            pinned: 0,
+            used: 0,
+            budget: 0,
+            pins: Vec::new(),
+            online: true,
+            session_expired: false,
+            tokens_unsaved: false,
+            version: "3.3.0".into(),
+            pending_uploads: 1,
+            pending_changes: 0,
+        };
+        assert_eq!(status_keys(&status(5)), status_keys(&status(6)));
+    }
+}

crates/pdfs-fuse/src/link.rs

@@ -0,0 +1,477 @@
+//! Noticing that the network went away mid-session, and that it came back.
+//!
+//! The mount used to learn it was offline only at startup: a daemon that had
+//! connected once kept `online` true for the rest of its life, so a Wi-Fi drop
+//! turned every create, mkdir, delete and rename into a two-minute wait for an
+//! `EIO`, while the queue that exists for exactly this sat unused. Now any
+//! remote call a handler waits on reports a failure to reach Proton through
+//! [`Core::lost_link`], which flips the mount offline; the handler then takes its
+//! queued path, and the probe thread here flips it back when Proton answers.
+
+use std::future::Future;
+use std::sync::atomic::{AtomicBool, Ordering};
+use std::time::{Duration, Instant};
+
+use parking_lot::{Condvar, Mutex};
+use pdfs_core::error::session_revoked;
+use proton_drive_rs::proton_sdk::api::ResponseCode;
+use proton_drive_rs::proton_sdk::error::{ProtonApiError, ProtonError};
+use tracing::{debug, info, warn};
+
+use crate::{Core, ONLINE_PROBE_MAX, ONLINE_PROBE_MIN};
+
+/// How long a remote call a FUSE handler is waiting on may take before the link
+/// is treated as down.
+///
+/// The SDK's own budget for one request is four attempts of up to 30 seconds
+/// each, so a dead link held the caller for about two minutes. A metadata call
+/// that has had no answer in 20 seconds is not going to be one the user wants to
+/// keep waiting for; queueing the change is the better answer.
+pub(crate) const INTERACTIVE_CALL_TIMEOUT: Duration = Duration::from_secs(20);
+
+/// How long a mount thread waits on a remote read that answers a batch: a
+/// chunk of file sizes, or the list of shares. Longer than an interactive call,
+/// since its answer grows with the batch, but well short of the SDK's own retry
+/// budget, which held a worker for minutes on a stalled link
+/// (`docs/BUGS.md` B220).
+pub(crate) const BATCH_CALL_TIMEOUT: Duration = Duration::from_secs(60);
+
+/// How long one probe may wait for an answer. A probe is a single small request,
+/// and a probe stuck in the SDK's retry budget would delay noticing the link is
+/// back by as long as it waits.
+const PROBE_TIMEOUT: Duration = Duration::from_secs(10);
+
+/// First delay between probes after a drop seen mid-session, and the ceiling it
+/// doubles to while the outage is young. A blip should end in seconds, not in
+/// the minutes [`ONLINE_PROBE_MAX`] allows.
+const PROBE_AFTER_DROP: Duration = Duration::from_secs(2);
+const PROBE_YOUNG_MAX: Duration = Duration::from_secs(30);
+/// How long an outage counts as young. Past it the probe backs off to
+/// [`ONLINE_PROBE_MAX`], because a laptop can sit offline for days and each
+/// probe is a real API round trip.
+const YOUNG_OUTAGE: Duration = Duration::from_secs(10 * 60);
+
+/// How often the idle probe thread checks for teardown while the link is up.
+const IDLE_CHECK: Duration = Duration::from_secs(1);
+
+/// The shared half of the online flag: when the link went down, and a condvar
+/// that fires on every change so waiters need not poll.
+#[derive(Default)]
+pub(crate) struct Link {
+    lost_at: Mutex<Option<Instant>>,
+    changed: Condvar,
+    /// Folders listed from the DB alone while offline, to relist once back.
+    pub(crate) stale_listings: Mutex<std::collections::HashSet<u64>>,
+    /// Proton refused the session itself ([`session_revoked`]). The mount is
+    /// offline then, but no probe brings it back: only a new login does, and
+    /// that restarts the daemon. Read by front-ends through `Response::Status`,
+    /// so they can ask the user to sign in again rather than say "offline".
+    pub(crate) session_expired: AtomicBool,
+    /// Proton refused the session, but only because another process rotated
+    /// it (`pdfs unlock`): the keyring holds the live tokens. The mount loop
+    /// ends the mount for the daemon to mount again from the keyring.
+    pub(crate) resume_stored: AtomicBool,
+}
+
+impl Link {
+    /// The link of a mount that comes up with Proton refusing its session, or
+    /// not.
+    pub(crate) fn starting(session_expired: bool) -> Self {
+        Self {
+            session_expired: AtomicBool::new(session_expired),
+            ..Default::default()
+        }
+    }
+}
+
+/// Whether a failed call failed because Proton could not be reached, rather
+/// than because it answered and refused.
+///
+/// A transport error is the network. So is a request timeout: the SDK reports
+/// Proton's own "try again" codes that way, and [`no_answer`] reports our
+/// deadline that way. Anything else Proton actually said, and says again on a
+/// retry, so it must reach the caller rather than be queued.
+pub(crate) fn is_network_error(e: &ProtonError) -> bool {
+    match e {
+        ProtonError::Transport(_) => true,
+        ProtonError::Api(api) => {
+            matches!(
+                api.code,
+                ResponseCode::RequestTimeout | ResponseCode::Timeout
+            )
+        }
+        _ => false,
+    }
+}
+
+/// The error a call reports when it runs past `after` without an answer.
+pub(crate) fn no_answer(after: Duration) -> ProtonError {
+    ProtonError::Api(ProtonApiError {
+        code: ResponseCode::RequestTimeout,
+        http_status: 408,
+        message: format!("no answer within {}s", after.as_secs()),
+        details: None,
+    })
+}
+
+/// The slowest upload rate an upload's deadline allows for, past the
+/// metadata budget it starts with. Far below any link that works, so only a
+/// transfer that has stopped moving runs into it.
+const SLOWEST_UPLOAD: u64 = 32 * 1024;
+
+/// How many uploads can share the upload cap at once: one per drain worker
+/// and one per mirror sync task.
+const UPLOAD_SLOTS: u64 = (crate::DRAIN_WORKERS + crate::sync::SYNC_CONCURRENCY) as u64;
+
+/// How long an upload of `len` bytes may take before the link is treated as
+/// down: the budget of a metadata call, plus the bytes at [`SLOWEST_UPLOAD`].
+///
+/// An upload cap of `cap` bytes per second (`0` for none) is shared by every
+/// upload running, so each may get as little as its share of it. Below
+/// [`SLOWEST_UPLOAD`] a share counts as the slowest rate instead: measured
+/// against the fixed rate, a large file under a low cap ran out of time on
+/// every attempt, took the mount offline each time, and started over from
+/// its first byte.
+pub(crate) fn upload_deadline(len: u64, cap: u64) -> Duration {
+    let slowest = match cap {
+        0 => SLOWEST_UPLOAD,
+        cap => SLOWEST_UPLOAD.min((cap / UPLOAD_SLOTS).max(1)),
+    };
+    INTERACTIVE_CALL_TIMEOUT + Duration::from_secs(len / slowest)
+}
+
+/// Run a remote call, giving up after [`INTERACTIVE_CALL_TIMEOUT`].
+pub(crate) async fn bounded<T>(
+    call: impl Future<Output = Result<T, ProtonError>>,
+) -> Result<T, ProtonError> {
+    bounded_by(INTERACTIVE_CALL_TIMEOUT, call).await
+}
+
+/// Run a remote call, giving up after `limit`.
+pub(crate) async fn bounded_by<T>(
+    limit: Duration,
+    call: impl Future<Output = Result<T, ProtonError>>,
+) -> Result<T, ProtonError> {
+    match tokio::time::timeout(limit, call).await {
+        Ok(result) => result,
+        Err(_) => Err(no_answer(limit)),
+    }
+}
+
+/// The next delay between probes, given the current one and how long the
+/// outage has lasted.
+fn next_probe_delay(delay: Duration, outage: Duration) -> Duration {
+    let cap = if outage < YOUNG_OUTAGE {
+        PROBE_YOUNG_MAX
+    } else {
+        ONLINE_PROBE_MAX
+    };
+    (delay * 2).min(cap)
+}
+
+impl Core {
+    /// Whether the mount currently believes Proton is reachable.
+    pub(crate) fn is_online(&self) -> bool {
+        self.online.load(Ordering::Relaxed)
+    }
+
+    /// Whether a namespace change through the mount goes to Drive inside the
+    /// syscall. Only with [`AppConfig::local_first`] off, and only online;
+    /// otherwise it is queued and the drain sends it.
+    ///
+    /// [`AppConfig::local_first`]: pdfs_core::config::AppConfig::local_first
+    pub(crate) fn sends_inline(&self) -> bool {
+        !self.local_first && self.is_online()
+    }
+
+    /// Run a remote call a caller is waiting on, giving up after
+    /// [`INTERACTIVE_CALL_TIMEOUT`]. The timer is built inside the runtime: a
+    /// FUSE worker has no reactor of its own.
+    pub(crate) fn block_on_bounded<T>(
+        &self,
+        call: impl Future<Output = Result<T, ProtonError>>,
+    ) -> Result<T, ProtonError> {
+        self.rt.block_on(bounded(call))
+    }
+
+    /// Run a remote call, giving up after `limit`: an upload, whose budget
+    /// grows with its size ([`upload_deadline`]).
+    pub(crate) fn block_on_within<T>(
+        &self,
+        limit: Duration,
+        call: impl Future<Output = Result<T, ProtonError>>,
+    ) -> Result<T, ProtonError> {
+        self.rt.block_on(bounded_by(limit, call))
+    }
+
+    /// Whether Proton refused the session itself, so nothing reaches it until
+    /// the user signs in again.
+    pub(crate) fn session_expired(&self) -> bool {
+        self.link.session_expired.load(Ordering::Relaxed)
+    }
+
+    /// Report a failed remote call. Returns whether it failed for want of a
+    /// network, or of a session, in which case the mount is now offline and the
+    /// caller should take its queued path instead of failing.
+    pub(crate) fn lost_link(&self, e: &ProtonError, what: &str) -> bool {
+        if session_revoked(e) {
+            self.mark_session_expired(what, e);
+            return true;
+        }
+        if !is_network_error(e) {
+            return false;
+        }
+        self.mark_offline(what, e);
+        true
+    }
+
+    /// Flip the mount offline for good: Proton refused the session, and only a
+    /// new login brings it back. Changes stay queued for the daemon that login
+    /// starts. Only the first caller logs it.
+    ///
+    /// Unless the keyring holds a newer session than the refused one: then
+    /// the mount ends, for the daemon to mount again on that one.
+    pub(crate) fn mark_session_expired(&self, what: &str, error: &dyn std::fmt::Display) {
+        if self.link.resume_stored.load(Ordering::Relaxed) {
+            return self.mark_offline(what, error);
+        }
+        if pdfs_core::auth::stored_session_is_newer() {
+            info!(during = what, %error,
+                  "Proton refused a session another process has since refreshed; resuming it");
+            self.link.resume_stored.store(true, Ordering::Relaxed);
+            return self.mark_offline(what, error);
+        }
+        if !self.link.session_expired.swap(true, Ordering::Relaxed) {
+            warn!(during = what, %error,
+                  "Proton refused the session; run `pdfs login` to sign in again");
+        }
+        self.mark_offline(what, error);
+    }
+
+    /// Flip the mount offline and wake the probe. Only the first caller of an
+    /// outage logs it; the rest are the same outage.
+    pub(crate) fn mark_offline(&self, what: &str, error: &dyn std::fmt::Display) {
+        let mut lost_at = self.link.lost_at.lock();
+        if self.online.swap(false, Ordering::Relaxed) {
+            *lost_at = Some(Instant::now());
+            warn!(during = what, %error,
+                  "lost the connection to Proton; queueing changes until it returns");
+            self.link.changed.notify_all();
+        }
+    }
+
+    /// Flip the mount back online, wake everything waiting for it, and let the
+    /// queue drain.
+    fn mark_online(&self) {
+        {
+            let mut lost_at = self.link.lost_at.lock();
+            self.online.store(true, Ordering::Relaxed);
+            match lost_at.take() {
+                Some(since) => {
+                    info!(
+                        offline_secs = since.elapsed().as_secs(),
+                        "connection to Proton is back"
+                    )
+                }
+                None => info!("back online"),
+            }
+            self.link.changed.notify_all();
+        }
+        self.relist_offline_listings();
+        // Anything written while offline is queued and waiting on exactly this,
+        // including ops that failed on the way down and are sitting out a
+        // backoff sized for a server that keeps refusing. Ops Drive refused
+        // keep theirs: the link was not why.
+        if let Err(e) = self.db.retry_unrefused_ops_now(crate::now_millis()) {
+            warn!(error = %e, "moving backed-off ops up after a reconnect failed");
+        }
+        self.wake_drain();
+    }
+
+    /// Wait up to `budget` for the link to come back. Returns whether it is up.
+    pub(crate) fn wait_online(&self, budget: Duration) -> bool {
+        let deadline = Instant::now() + budget;
+        let mut lost_at = self.link.lost_at.lock();
+        while !self.is_online() {
+            if self.shutdown.is_stopping()
+                || self
+                    .link
+                    .changed
+                    .wait_until(&mut lost_at, deadline)
+                    .timed_out()
+            {
+                return self.is_online();
+            }
+        }
+        true
+    }
+
+    /// Wait out `interval` before a background poll of Proton, and while the
+    /// link is down, until it is back. Returns as soon as it comes back, so the
+    /// poll catches up on what changed during the outage at once instead of
+    /// one interval later. A poll into a dead link would only wait out the
+    /// SDK's retries and fail.
+    pub(crate) async fn next_poll(&self, interval: Duration) {
+        let due = Instant::now() + interval;
+        let mut was_offline = false;
+        loop {
+            if !self.is_online() {
+                was_offline = true;
+            } else if was_offline || Instant::now() >= due {
+                return;
+            }
+            tokio::time::sleep(IDLE_CHECK).await;
+        }
+    }
+
+    /// Watch the link for the life of the mount: sleep while it is up, and
+    /// probe it back while it is down.
+    ///
+    /// A mount that started from the cache is down from the first moment; one
+    /// that started online goes down when [`Core::mark_offline`] says so.
+    pub(crate) fn run_online_probe(&self) {
+        // Only an offline start never had a chance to repair its share id.
+        let mut repair_share_id = !self.is_online();
+        loop {
+            {
+                let mut lost_at = self.link.lost_at.lock();
+                while self.is_online() {
+                    if self.shutdown.is_stopping() {
+                        return;
+                    }
+                    self.link.changed.wait_for(&mut lost_at, IDLE_CHECK);
+                }
+            }
+            if !self.probe_until_online() {
+                return;
+            }
+            if std::mem::take(&mut repair_share_id) {
+                self.repair_primary_share_id(&self.primary_root_uid);
+            }
+        }
+    }
+
+    /// Probe until Proton answers, then flip the mount online. Returns `false`
+    /// if the daemon is stopping.
+    ///
+    /// A refused session is not probed: every probe would be refused the same
+    /// way, forever (issue #35). The thread waits for the stop instead.
+    fn probe_until_online(&self) -> bool {
+        let started = self.link.lost_at.lock().unwrap_or_else(Instant::now);
+        let mut delay = if started.elapsed() < YOUNG_OUTAGE {
+            PROBE_AFTER_DROP
+        } else {
+            ONLINE_PROBE_MIN
+        };
+        loop {
+            if !self.shutdown.sleep(delay) {
+                return false;
+            }
+            if self.session_expired() {
+                delay = ONLINE_PROBE_MAX;
+                continue;
+            }
+            let probe = self.rt.block_on(async {
+                match tokio::time::timeout(PROBE_TIMEOUT, self.drive.get_my_files_folder()).await {
+                    Ok(result) => result,
+                    Err(_) => Err(no_answer(PROBE_TIMEOUT)),
+                }
+            });
+            match probe {
+                Ok(root) => {
+                    {
+                        let root_ino = self.root_ino();
+                        let mut st = self.state();
+                        if let Some(e) = st.entries.get_mut(&root_ino) {
+                            e.node = root.clone();
+                        }
+                    }
+                    if let Err(e) = self.db.upsert_node(&root) {
+                        warn!(error = %e, "refresh root after reconnect failed");
+                    }
+                    self.mark_online();
+                    return true;
+                }
+                Err(e) if session_revoked(&e) => self.mark_session_expired("online probe", &e),
+                Err(e) => {
+                    debug!(error = %e, ?delay, "online probe failed; still offline");
+                    delay = next_probe_delay(delay, started.elapsed());
+                }
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn api(code: ResponseCode) -> ProtonError {
+        ProtonError::Api(ProtonApiError {
+            code,
+            http_status: 422,
+            message: String::new(),
+            details: None,
+        })
+    }
+
+    #[test]
+    fn only_an_unanswered_call_counts_as_a_lost_link() {
+        assert!(is_network_error(&no_answer(INTERACTIVE_CALL_TIMEOUT)));
+        assert!(is_network_error(&api(ResponseCode::Timeout)));
+        assert!(!is_network_error(&api(ResponseCode::AlreadyExists)));
+        assert!(!is_network_error(&api(ResponseCode::InvalidRequirements)));
+        assert!(!is_network_error(&ProtonError::invalid_operation("bug")));
+    }
+
+    #[test]
+    fn an_upload_gets_longer_the_bigger_it_is() {
+        assert_eq!(upload_deadline(0, 0), INTERACTIVE_CALL_TIMEOUT);
+        assert_eq!(
+            upload_deadline(1 << 30, 0),
+            INTERACTIVE_CALL_TIMEOUT + Duration::from_secs(32_768)
+        );
+    }
+
+    #[test]
+    fn an_upload_under_a_low_cap_gets_the_time_its_share_of_the_cap_needs() {
+        // A generous cap leaves the deadline as it was.
+        assert_eq!(
+            upload_deadline(1 << 30, 100 << 20),
+            upload_deadline(1 << 30, 0)
+        );
+        // 512 KiB/s shared by every upload slot is below the slowest rate.
+        let share = (512 << 10) / UPLOAD_SLOTS;
+        assert!(share < SLOWEST_UPLOAD);
+        assert_eq!(
+            upload_deadline(1 << 30, 512 << 10),
+            INTERACTIVE_CALL_TIMEOUT + Duration::from_secs((1 << 30) / share)
+        );
+        // A cap below one byte per slot still gives a finite deadline.
+        assert!(upload_deadline(1 << 20, 1) > upload_deadline(1 << 20, 512 << 10));
+    }
+
+    #[test]
+    fn a_call_past_its_deadline_reads_as_a_lost_link() {
+        let rt = tokio::runtime::Builder::new_current_thread()
+            .enable_time()
+            .build()
+            .unwrap();
+        let limit = Duration::from_millis(20);
+        let result: Result<(), _> = rt.block_on(bounded_by(limit, std::future::pending()));
+        assert!(result.is_err_and(|e| is_network_error(&e)));
+    }
+
+    #[test]
+    fn a_young_outage_is_probed_often_and_an_old_one_rarely() {
+        let young = Duration::from_secs(60);
+        let old = YOUNG_OUTAGE + Duration::from_secs(1);
+        assert_eq!(
+            next_probe_delay(PROBE_AFTER_DROP, young),
+            2 * PROBE_AFTER_DROP
+        );
+        assert_eq!(next_probe_delay(PROBE_YOUNG_MAX, young), PROBE_YOUNG_MAX);
+        assert_eq!(next_probe_delay(PROBE_YOUNG_MAX, old), 2 * PROBE_YOUNG_MAX);
+        assert_eq!(next_probe_delay(ONLINE_PROBE_MAX, old), ONLINE_PROBE_MAX);
+    }
+}

crates/pdfs-fuse/src/mount.rs

@@ -12,6 +12,10 @@ pub enum MountOutcome {
     /// The kernel mount ended on its own (e.g. an external `fusermount -u`).
     /// The caller may want to remount.
     Unmounted,
+    /// Proton refused the session, and the keyring holds a newer one: another
+    /// process (`pdfs unlock`) refreshed it, or a new login replaced it. We
+    /// unmounted ourselves; the caller should mount again, which resumes it.
+    ResumeStored,
 }
 
 /// Whether `path` is a mountpoint whose FUSE connection is dead — the state a
@@ -193,7 +197,7 @@ pub(super) fn clear_stale_mount(mountpoint: &Path) {
 }
 
 /// `sync_state` key holding the uid of the My Files root, so a later run can
-/// recover it from `nodes` without the API (offline.md Phase 1).
+/// recover it from `nodes` without the API.
 const ROOT_UID_KEY: &str = "root_uid";
 
 /// The My Files root, and whether we got it from the API (`true`) or from the
@@ -207,33 +211,63 @@ const ROOT_UID_KEY: &str = "root_uid";
 /// means an empty tree, and mounting that would show the user an empty Drive
 /// rather than an honest error.
 fn fetch_or_recall_root(
-    client: &ProtonDriveClient,
+    drive: &dyn DriveApi,
     rt: &tokio::runtime::Handle,
     db: &Db,
-) -> std::io::Result<(Node, bool)> {
-    let err = match rt.block_on(client.get_my_files_folder()) {
+) -> std::io::Result<(Node, Start)> {
+    let err = match rt.block_on(drive.get_my_files_folder()) {
         Ok(root) => {
             if let Err(e) = db.set_state_str(ROOT_UID_KEY, &root.uid.to_string()) {
                 warn!(error = %e, "persist root uid failed");
             }
-            return Ok((root, true));
+            return Ok((root, Start::Online));
         }
         Err(e) => e,
     };
+    // Still mounted from the cache, so cached files stay readable and changes
+    // queue for the daemon the next login starts.
+    let start = if pdfs_core::error::session_revoked(&err) {
+        Start::SessionExpired
+    } else {
+        Start::Offline
+    };
+    if start == Start::SessionExpired && pdfs_core::auth::stored_session_is_newer() {
+        return Err(std::io::Error::other(format!(
+            "Proton refused a session another process has since refreshed ({err}); resuming it"
+        )));
+    }
     let cached = db
         .state_str(ROOT_UID_KEY)
         .ok()
         .flatten()
         .and_then(|uid| db.node_by_uid(&uid).ok().flatten());
     match cached {
+        Some(root) if start == Start::SessionExpired => {
+            warn!(error = %err,
+                  "Proton refused the session; mounting from cache until `pdfs login`");
+            Ok((root, start))
+        }
         Some(root) => {
             warn!(error = %err, "fetch My Files root failed; mounting from cache (offline)");
-            Ok((root, false))
+            Ok((root, start))
         }
+        None if start == Start::SessionExpired => Err(std::io::Error::other(format!(
+            "Proton refused the session ({err}); run `pdfs login`"
+        ))),
         None => Err(std::io::Error::other(format!("fetch My Files root: {err}"))),
     }
 }
 
+/// How a mount comes up, from what fetching its root said.
+#[derive(Clone, Copy, PartialEq, Eq)]
+enum Start {
+    Online,
+    /// Proton could not be reached: serve the cache and probe for the link.
+    Offline,
+    /// Proton refused the session: serve the cache until a new login.
+    SessionExpired,
+}
+
 /// Per-mount settings resolved from [`pdfs_core::config::AppConfig`] by the
 /// caller, so `mount` itself never reads config or the environment.
 pub struct MountOptions {
@@ -241,6 +275,12 @@ pub struct MountOptions {
     pub username: String,
     /// What the background conflict sweep may do. See `docs/BUGS.md` B71.
     pub sweep_mode: SweepMode,
+    /// Upload and download caps in bytes per second (`0` = no cap).
+    pub upload_limit: u64,
+    pub download_limit: u64,
+    /// Namespace changes through the mount are queued, not sent in the
+    /// syscall. See [`pdfs_core::config::AppConfig::local_first`].
+    pub local_first: bool,
 }
 
 /// Spawn one FUSE session rooted at an arbitrary remote node.
@@ -273,13 +313,14 @@ pub(super) fn spawn_session(
     info!(mountpoint = %mountpoint.display(), "mounting Proton Drive location");
     let fs = ProtonFs::new(core.clone(), root);
     if core.primary {
-        // Only My Files rehydrates the daemon-wide node cache. A secondary
-        // inode space is rooted at one device subtree and must not import every
-        // unrelated persisted node.
+        // Only My Files rehydrates the daemon-wide node cache, once. A
+        // secondary mount shares the tree it fills.
         core.hydrate();
     }
     let session = Session::new(fs, mountpoint, &config)?.spawn()?;
-    let _ = core.notifier.set(session.notifier());
+    let _ = core
+        .notifier
+        .set(KernelChannel::new(session.notifier(), core.root_ino()));
     core.session_live.store(true, Ordering::Release);
     Ok(session)
 }
@@ -336,15 +377,58 @@ pub fn mount(
     control_socket: &Path,
     db: Arc<Db>,
     options: MountOptions,
+) -> std::io::Result<MountOutcome> {
+    let drive = Arc::new(client.clone());
+    mount_with(
+        client,
+        drive,
+        rt,
+        mountpoint,
+        cache,
+        control_socket,
+        db,
+        options,
+        Host::Daemon,
+    )
+}
+
+/// What a mount runs in.
+pub(crate) enum Host {
+    /// The daemon: stopped by SIGTERM or SIGINT, and indexing the user's home
+    /// for the launcher prompt.
+    Daemon,
+    /// One of several simulated daemons in a test process (`sim`): stopped by
+    /// a message on this channel, and with no home to index.
+    #[cfg(test)]
+    Simulation(std::sync::mpsc::Receiver<()>),
+}
+
+/// [`mount`], reaching Drive through `drive` and hosted by `host`; `client`
+/// serves only what [`DriveApi`] does not cover.
+#[allow(clippy::too_many_arguments)]
+pub(crate) fn mount_with(
+    client: ProtonDriveClient,
+    drive: Arc<dyn DriveApi>,
+    rt: tokio::runtime::Handle,
+    mountpoint: &Path,
+    cache: ContentCache,
+    control_socket: &Path,
+    db: Arc<Db>,
+    options: MountOptions,
+    host: Host,
 ) -> std::io::Result<MountOutcome> {
     let MountOptions {
         username,
         sweep_mode,
+        upload_limit,
+        download_limit,
+        local_first,
     } = options;
     // Start the uptime clock here rather than at the first `pdfs diagnostics`,
     // so the age it reports is the daemon's own.
     super::diagnostics::start_clock();
-    let (root, online) = fetch_or_recall_root(&client, &rt, &db)?;
+    let (root, start) = fetch_or_recall_root(drive.as_ref(), &rt, &db)?;
+    let online = start == Start::Online;
     let scope = root.tree_event_scope_id();
     db.mount_upsert_my_files(&mountpoint.to_string_lossy(), &root.uid.to_string(), None)
         .map_err(|error| std::io::Error::other(format!("project My Files mount: {error}")))?;
@@ -352,18 +436,21 @@ pub fn mount(
         .all_share_access()
         .map_err(|error| std::io::Error::other(format!("load shared access: {error}")))?;
 
-    // The folder-sync engine (devices.md Phase 2) runs on its own thread and is
+    // The folder-sync engine runs on its own thread and is
     // nudged over this channel; the sender lives in Core so control-socket
     // handlers can trigger reconciles.
     let (sync_tx, sync_rx) = std::sync::mpsc::channel::<sync::SyncMsg>();
 
+    let paused_until = pause::load_paused_until(&db);
     let core = Core {
-        client: client.clone(),
+        drive,
+        client,
         rt: rt.clone(),
         maintenance: Arc::new(Mutex::new(Default::default())),
         primary_root_uid: root.uid.clone(),
         primary: true,
-        state: Arc::new(Mutex::new(State::new(db.clone(), share_access, 2))),
+        state: Arc::new(Mutex::new(State::new(db.clone(), share_access))),
+        root: Arc::new(OnceLock::new()),
         cache: Arc::new(cache),
         readers: Arc::new(Mutex::new(HashMap::new())),
         block_ring: Arc::new(Mutex::new(BlockRing::default())),
@@ -377,19 +464,28 @@ pub fn mount(
         shared_generation: Arc::new(AtomicU64::new(0)),
         shared_refresh_deadlines: Arc::new(Mutex::new(SharedRefreshDeadlines::default())),
         online: Arc::new(AtomicBool::new(online)),
+        link: Arc::new(link::Link::starting(start == Start::SessionExpired)),
+        sync_paused_until: Arc::new(AtomicI64::new(paused_until)),
         pending: Arc::new(Mutex::new(HashMap::new())),
         hidden: Arc::new(Mutex::new(HashSet::new())),
+        creating: Arc::new(Mutex::new(HashMap::new())),
+        sealed_unread: Arc::new(Mutex::new(HashMap::new())),
+        landed: Arc::new(Mutex::new(HashMap::new())),
+        unadopted: Arc::new(Mutex::new(HashMap::new())),
         drain_wake: Arc::new((Mutex::new(false), Condvar::new())),
         shutdown: Arc::new(crate::shutdown::Shutdown::default()),
         upload_times: Arc::new(Mutex::new(HashMap::new())),
         upload_cancel: Arc::new(Mutex::new(HashMap::new())),
+        landing_uploads: Arc::new(AtomicU64::new(0)),
         timeline_refreshing: Arc::new(AtomicBool::new(false)),
+        timeline_progress: Arc::new(Mutex::new(None)),
+        similar_hashing: Arc::new(AtomicBool::new(false)),
         albums_refreshing: Arc::new(AtomicBool::new(false)),
         trash_refreshing: Arc::new(AtomicBool::new(false)),
         trash_progress: Arc::new(tokio::sync::Notify::new()),
         conflict_notified: Arc::new(Mutex::new(HashSet::new())),
         sweep_mode,
-        own_sealed_revs: Arc::new(Mutex::new(HashMap::new())),
+        local_first,
         self_changes: Arc::new(Mutex::new(HashMap::new())),
         thumb_gen: Arc::new(Mutex::new(HashSet::new())),
         thumb_gen_budget: Arc::new(tokio::sync::Semaphore::new(
@@ -402,21 +498,26 @@ pub fn mount(
         thumbnail_build_cancel: Arc::new(tokio::sync::Notify::new()),
         thumbnail_misses: Arc::new(Mutex::new(Default::default())),
         quota: Arc::new(Mutex::new(None)),
+        quota_refreshing: Arc::new(AtomicBool::new(false)),
         size_upgrades: Arc::new(Mutex::new(HashMap::new())),
         size_waiters: Arc::new(Default::default()),
         notifier: Arc::new(OnceLock::new()),
         session_live: Arc::new(AtomicBool::new(false)),
         transfers: TransferRegistry::new(),
+        events: Arc::new(Default::default()),
         indexing: Arc::new(AtomicBool::new(false)),
         sync_progress: Arc::new(Mutex::new(HashMap::new())),
+        sync_engine: Arc::default(),
         sync_tx,
         mounts: Arc::new(Mutex::new(SecondaryMountRegistry::default())),
         sync_locks: Arc::new(Mutex::new(HashMap::new())),
+        sync_unscanned: Arc::new(Mutex::new(HashMap::new())),
         states: Arc::new(StateRegistry::default()),
     };
     // Before anything can queue work against it: the drain thread below reaches
-    // every mount's inode space through this registry, not through `core.state`.
+    // every mount's session through this registry.
     core.register_state(mountpoint);
+    core.transfers.set_limits(upload_limit, download_limit);
 
     // Writes queued by a previous run (or left behind by a crash) are still owed
     // an upload, and reads must be served from their staged blobs until they land.
@@ -437,20 +538,23 @@ pub fn mount(
     // Every long-lived thread started from here on is kept in `workers` and
     // joined at the end of this function. Leaving them running turned an
     // in-process remount into a second generation of drain workers, sync engines
-    // and sweeps all operating on a mount that no longer exists (bugs.md B44).
+    // and sweeps all operating on a mount that no longer exists (`docs/BUGS.md` B44).
     let mut workers: Vec<std::thread::JoinHandle<()>> = Vec::new();
     for worker in 0..DRAIN_WORKERS {
-        let core = core.clone();
-        workers.push(
-            std::thread::Builder::new()
-                .name(format!("pdfs-drain-{worker}"))
-                .spawn(move || core.run_pending_drain(worker == 0))?,
-        );
+        let drain_core = core.clone();
+        let spawned = std::thread::Builder::new()
+            .name(format!("pdfs-drain-{worker}"))
+            .spawn(move || drain_core.run_pending_drain(worker == 0));
+        keep_or_abandon(&core, control_socket, &mut workers, spawned)?;
     }
 
     // Start the folder-sync engine. It watches every mirror folder, polls the
     // remotes, and reconciles on its own thread — never in front of a FUSE call.
     workers.extend(sync::spawn(core.clone(), sync_rx));
+    // A timed pause persisted by an earlier run still has to end on time.
+    if pause::paused_at(paused_until, now_secs()) && paused_until != pause::PAUSED_INDEFINITELY {
+        core.arm_pause_timer(paused_until);
+    }
 
     // Reconcile leftover `(sync-conflict …)` copies: drop the ones that turned
     // out identical to the live file, surface the ones that genuinely diverge.
@@ -461,38 +565,36 @@ pub fn mount(
         info!("conflict sweep disabled by configuration");
     } else {
         info!(mode = ?sweep_mode, "conflict sweep enabled");
-        let core = core.clone();
-        workers.push(
-            std::thread::Builder::new()
-                .name("pdfs-conflict-sweep".into())
-                .spawn(move || core.run_conflict_sweep_loop())?,
-        );
+        let sweep_core = core.clone();
+        let spawned = std::thread::Builder::new()
+            .name("pdfs-conflict-sweep".into())
+            .spawn(move || sweep_core.run_conflict_sweep_loop());
+        keep_or_abandon(&core, control_socket, &mut workers, spawned)?;
     }
 
-    // Mounted from the cache: watch for the network coming back so the mount can
-    // stop being read-only-ish without the user restarting the daemon.
-    if !online {
-        let core = core.clone();
-        workers.push(
-            std::thread::Builder::new()
-                .name("pdfs-online-probe".into())
-                .spawn(move || core.run_online_probe())?,
-        );
+    // Watch the link for the life of the mount: a mount from the cache waits for
+    // the network to come back, and an online one can lose it at any moment and
+    // must notice it coming back just the same.
+    {
+        let probe_core = core.clone();
+        let spawned = std::thread::Builder::new()
+            .name("pdfs-online-probe".into())
+            .spawn(move || probe_core.run_online_probe());
+        keep_or_abandon(&core, control_socket, &mut workers, spawned)?;
     }
 
     // Keep the launcher prompt's "This computer" index fresh. Its own thread:
     // the walk is I/O-heavy and must never sit in front of a FUSE callback.
-    {
+    if matches!(host, Host::Daemon) {
         let db = core.db.clone();
         let indexing = core.indexing.clone();
         let transfers = core.transfers.clone();
         let mountpoint = mountpoint.to_path_buf();
         let shutdown = core.shutdown.clone();
-        workers.push(
-            std::thread::Builder::new()
-                .name("pdfs-localindex".into())
-                .spawn(move || run_local_index(db, indexing, transfers, mountpoint, shutdown))?,
-        );
+        let spawned = std::thread::Builder::new()
+            .name("pdfs-localindex".into())
+            .spawn(move || run_local_index(db, indexing, transfers, mountpoint, shutdown));
+        keep_or_abandon(&core, control_socket, &mut workers, spawned)?;
     }
 
     // Bind the control socket before the FUSE session takes over the thread. A
@@ -501,12 +603,18 @@ pub fn mount(
     let old_umask = unsafe { libc::umask(0o77) };
     let listener_res = UnixListener::bind(control_socket);
     unsafe { libc::umask(old_umask) };
-    let listener = listener_res?;
+    let listener = match listener_res {
+        Ok(listener) => listener,
+        Err(error) => {
+            abandon_workers(&core, control_socket, workers);
+            return Err(error);
+        }
+    };
     // Owner-only before anything can connect: a peer on this socket commands the
     // daemon's authenticated session without a credential of its own (B6).
     if let Err(e) = pdfs_core::config::restrict_socket(control_socket) {
         error!(error = %e, "cannot restrict control socket permissions; refusing to serve");
-        let _ = std::fs::remove_file(control_socket);
+        abandon_workers(&core, control_socket, workers);
         return Err(std::io::Error::other(format!(
             "control socket permissions: {e}"
         )));
@@ -517,7 +625,7 @@ pub fn mount(
     let bg = match spawn_session(&core, mountpoint, root) {
         Ok(session) => session,
         Err(error) => {
-            let _ = std::fs::remove_file(control_socket);
+            abandon_workers(&core, control_socket, workers);
             return Err(error);
         }
     };
@@ -537,7 +645,7 @@ pub fn mount(
             Err(error) => {
                 // A failure here abandons the mount, so the background threads
                 // started above have to be told to stop with it.
-                stop_workers(&core, control_socket, workers);
+                abandon_workers(&core, control_socket, workers);
                 let secondaries = core.mounts.lock().close_and_drain();
                 for (id, mount) in secondaries {
                     if let Err(unmount_error) = mount.teardown() {
@@ -554,6 +662,22 @@ pub fn mount(
             }
         }
     }
+    // Publishes what changes without one place to publish it from (status,
+    // transfers, sync progress) to the front-ends subscribed on that socket.
+    {
+        let sampler_core = core.clone();
+        let username = username.clone();
+        let mountpoint = mountpoint.to_path_buf();
+        match std::thread::Builder::new()
+            .name("pdfs-events".into())
+            .spawn(move || events::run_event_sampler(sampler_core, username, mountpoint))
+        {
+            Ok(handle) => workers.push(handle),
+            // Not fatal: front-ends still get every published change, only
+            // not the sampled ones.
+            Err(error) => warn!(error = %error, "could not start the event sampler"),
+        }
+    }
     // The supervisor watches the pool and answers the systemd watchdog. Started
     // after the listener, because its liveness probe is a round trip over that
     // socket.
@@ -571,8 +695,12 @@ pub fn mount(
         }
     }
     // The mount is live and the control socket answers: this is the moment the
-    // unit is actually usable, which is what `Type=notify` reports.
+    // unit is actually usable, which is what `Type=notify` reports. Whatever
+    // failure an earlier attempt recorded for the front ends is over.
     systemd::ready();
+    if let Some(state_dir) = control_socket.parent() {
+        pdfs_core::mountfail::clear(state_dir);
+    }
 
     // Re-establish on-demand mounts only after the primary session is live.
     // Keep the worker so shutdown can close publication, wait for any in-flight
@@ -590,11 +718,13 @@ pub fn mount(
             }
         }
     };
-    rt.spawn(run_event_sync(client, scope, core.clone()));
+    rt.spawn(run_event_sync(core.drive.clone(), scope, core.clone()));
     // The photos volume has its own event stream; without it the gallery only
     // learns about a phone-side deletion when the timeline goes stale.
     rt.spawn(run_photos_event_sync(core.clone()));
-    if online {
+    // Not in a simulation: `client` there reaches no API, and the thread, still
+    // waiting on it after the stop, held the database a test opens next.
+    if online && matches!(host, Host::Daemon) {
         let repair_core = core.clone();
         let root_uid = core.primary_root_uid.clone();
         if let Err(error) = std::thread::Builder::new()
@@ -609,6 +739,82 @@ pub fn mount(
     // are delivered onto the async runtime; bridge them onto a sync channel so
     // the loop below can react without blocking a worker thread. A bounded
     // channel of 1 is enough — we only need to know that *a* stop arrived.
+    let sig_rx = match host {
+        Host::Daemon => signal_stop(&rt),
+        #[cfg(test)]
+        Host::Simulation(rx) => rx,
+    };
+
+    // Wait for whichever happens first: a stop signal (→ we unmount ourselves
+    // via the lazy MNT_DETACH path, clean even mid-download), or the kernel
+    // mount ending on its own (→ the session thread finishes). Poll instead of
+    // blocking on `join` so we can also notice the signal.
+    let outcome = loop {
+        match sig_rx.recv_timeout(Duration::from_millis(500)) {
+            Ok(()) => {
+                info!("stop requested; unmounting");
+                core.mounts.lock().close();
+                if let Err(e) = teardown_session(&core.session_live, || {
+                    umount_session_unblocked(bg, main_conn)
+                }) {
+                    warn!(error = %e, "umount_and_join failed");
+                }
+                break MountOutcome::Shutdown;
+            }
+            Err(std::sync::mpsc::RecvTimeoutError::Timeout)
+                if core.link.resume_stored.load(Ordering::Relaxed) =>
+            {
+                info!("unmounting to resume the stored session");
+                core.mounts.lock().close();
+                if let Err(e) = teardown_session(&core.session_live, || {
+                    umount_session_unblocked(bg, main_conn)
+                }) {
+                    warn!(error = %e, "umount_and_join failed");
+                }
+                break MountOutcome::ResumeStored;
+            }
+            Err(std::sync::mpsc::RecvTimeoutError::Timeout) => {
+                if bg.guard.is_finished() {
+                    info!("mount ended externally");
+                    core.mounts.lock().close();
+                    if let Err(e) = teardown_session(&core.session_live, || bg.join()) {
+                        warn!(error = %e, "session join failed");
+                    }
+                    break MountOutcome::Unmounted;
+                }
+            }
+            Err(std::sync::mpsc::RecvTimeoutError::Disconnected) => {
+                // Signal task gone (failed to install); fall back to join.
+                core.mounts.lock().close();
+                let _ = teardown_session(&core.session_live, || bg.join());
+                break MountOutcome::Unmounted;
+            }
+        }
+    };
+
+    // Unmount every on-demand sync folder too, or the kernel mounts linger as
+    // stale and the next start fails with EBUSY.
+    if let Some(worker) = restore_worker
+        && worker.join().is_err()
+    {
+        warn!("on-demand restore worker panicked");
+    }
+    let secondaries = core.mounts.lock().close_and_drain();
+    for (id, mount) in secondaries {
+        if let Err(e) = mount.teardown() {
+            warn!(id, error = %e, "unmount on-demand folder failed");
+        }
+    }
+    core.queue_unreleased_writes();
+
+    // Last, because the workers may legitimately be finishing an upload, and
+    // nothing they can still do is harmful once the mounts are down.
+    stop_workers(&core, control_socket, workers);
+    Ok(outcome)
+}
+
+/// A channel that receives once SIGTERM or SIGINT arrives.
+fn signal_stop(rt: &tokio::runtime::Handle) -> std::sync::mpsc::Receiver<()> {
     let (sig_tx, sig_rx) = std::sync::mpsc::sync_channel::<()>(1);
     rt.spawn(async move {
         let mut sigterm =
@@ -633,65 +839,13 @@ pub fn mount(
         }
         let _ = sig_tx.try_send(());
     });
-
-    // Wait for whichever happens first: a stop signal (→ we unmount ourselves
-    // via the lazy MNT_DETACH path, clean even mid-download), or the kernel
-    // mount ending on its own (→ the session thread finishes). Poll instead of
-    // blocking on `join` so we can also notice the signal.
-    let outcome = loop {
-        match sig_rx.recv_timeout(Duration::from_millis(500)) {
-            Ok(()) => {
-                info!("stop requested; unmounting");
-                core.mounts.lock().close();
-                if let Err(e) = teardown_session(&core.session_live, || {
-                    umount_session_unblocked(bg, main_conn)
-                }) {
-                    warn!(error = %e, "umount_and_join failed");
-                }
-                break MountOutcome::Shutdown;
-            }
-            Err(std::sync::mpsc::RecvTimeoutError::Timeout) => {
-                if bg.guard.is_finished() {
-                    info!("mount ended externally");
-                    core.mounts.lock().close();
-                    if let Err(e) = teardown_session(&core.session_live, || bg.join()) {
-                        warn!(error = %e, "session join failed");
-                    }
-                    break MountOutcome::Unmounted;
-                }
-            }
-            Err(std::sync::mpsc::RecvTimeoutError::Disconnected) => {
-                // Signal task gone (failed to install); fall back to join.
-                core.mounts.lock().close();
-                let _ = teardown_session(&core.session_live, || bg.join());
-                break MountOutcome::Unmounted;
-            }
-        }
-    };
-
-    // Unmount every on-demand sync folder too, or the kernel mounts linger as
-    // stale and the next start fails with EBUSY (devices.md Phase 3).
-    if let Some(worker) = restore_worker
-        && worker.join().is_err()
-    {
-        warn!("on-demand restore worker panicked");
-    }
-    let secondaries = core.mounts.lock().close_and_drain();
-    for (id, mount) in secondaries {
-        if let Err(e) = mount.teardown() {
-            warn!(id, error = %e, "unmount on-demand folder failed");
-        }
-    }
-
-    // Last, because the workers may legitimately be finishing an upload, and
-    // nothing they can still do is harmful once the mounts are down.
-    stop_workers(&core, control_socket, workers);
-    Ok(outcome)
+    sig_rx
 }
 
-/// How long teardown waits for the FUSE worker pool before proceeding without
-/// it. Long enough for an in-flight block to finish, short enough that
-/// `systemctl --user stop` still returns inside `TimeoutStopSec`.
+/// How long teardown waits for the background workers and then the FUSE worker
+/// pool before proceeding without them. Long enough for an in-flight block to
+/// finish, short enough that `systemctl --user stop` still returns inside
+/// `TimeoutStopSec`.
 const WORKER_JOIN_DEADLINE: Duration = Duration::from_secs(10);
 
 /// Signal every background loop, wake the ones that are blocked, and join them.
@@ -699,18 +853,56 @@ const WORKER_JOIN_DEADLINE: Duration = Duration::from_secs(10);
 /// The mount used to return with its drain workers, sync engine, sweep, online
 /// probe, indexer and control listener all still running on `Core` clones. The
 /// process usually exited straight afterwards and hid it; an in-process remount
-/// did not, and started a second full set on top of the first (bugs.md B44).
+/// did not, and started a second full set on top of the first (`docs/BUGS.md` B44).
 ///
 /// Two of the waits need waking rather than merely flagging: the drain workers
 /// block on their own condvar (`wake_drain`), and the control listener blocks
 /// inside `accept`, which only returns when a connection arrives — so it gets
 /// one, from us, after the flag is already set. A join that takes a while is
 /// expected and correct: a worker part-way through an upload finishes it rather
-/// than abandoning the user's bytes mid-flight.
+/// than abandoning the user's bytes mid-flight. Up to [`WORKER_JOIN_DEADLINE`]:
+/// past it, systemd kills the daemon anyway, and a queued op is sent again.
 fn stop_workers(core: &Core, control_socket: &Path, workers: Vec<std::thread::JoinHandle<()>>) {
     // Say so before the joins: a teardown that takes a while must not be read as
     // a missed watchdog ping and turned into a kill.
     systemd::stopping();
+    halt_workers(core, control_socket, workers);
+}
+
+/// Stop the threads a mount attempt started before it failed.
+///
+/// The early returns used to leave them running: every retry started another
+/// set beside the last one's, and their `Core` kept the database open, so the
+/// next attempt could fail on its lock instead of the real error
+/// (`docs/BUGS.md` B264). Unlike [`stop_workers`] this does not tell systemd
+/// the daemon is stopping, because it is about to try again, and a manager
+/// told so would wait for the process to exit and then kill it.
+fn abandon_workers(core: &Core, control_socket: &Path, workers: Vec<std::thread::JoinHandle<()>>) {
+    halt_workers(core, control_socket, workers);
+}
+
+/// Keep a worker the mount just spawned, or, if the spawn failed, stop the
+/// ones already running and hand the error back.
+fn keep_or_abandon(
+    core: &Core,
+    control_socket: &Path,
+    workers: &mut Vec<std::thread::JoinHandle<()>>,
+    spawned: std::io::Result<std::thread::JoinHandle<()>>,
+) -> std::io::Result<()> {
+    match spawned {
+        Ok(handle) => {
+            workers.push(handle);
+            Ok(())
+        }
+        Err(error) => {
+            abandon_workers(core, control_socket, std::mem::take(workers));
+            Err(error)
+        }
+    }
+}
+
+/// Signal every worker to stop, then join them up to [`WORKER_JOIN_DEADLINE`].
+fn halt_workers(core: &Core, control_socket: &Path, workers: Vec<std::thread::JoinHandle<()>>) {
     core.shutdown.stop();
     core.wake_drain();
     let _ = core.sync_tx.send(sync::SyncMsg::Stop);
@@ -720,25 +912,71 @@ fn stop_workers(core: &Core, control_socket: &Path, workers: Vec<std::thread::Jo
     // unlink cannot put it back to sleep.
     let _ = std::os::unix::net::UnixStream::connect(control_socket);
     let _ = std::fs::remove_file(control_socket);
-    for worker in workers {
-        if worker.join().is_err() {
-            warn!("a background worker panicked before shutdown");
-        }
+    let deadline = Instant::now() + WORKER_JOIN_DEADLINE;
+    let running = join_until(workers, deadline);
+    if !running.is_empty() {
+        warn!(
+            running = running.join(","),
+            "background workers did not finish in time; shutting down around them"
+        );
     }
     // The FUSE worker pool last, and *before* the caller drops the tokio
     // runtime: a worker still inside a job when the runtime goes away panics on
     // its next timer with "A Tokio 1.x context was found, but it is being
     // shutdown" (seen 2026-09-18 09:58:35). Bounded, so a worker wedged on the
     // network cannot hold the stop the user asked for.
-    core.workers.stop_and_join(WORKER_JOIN_DEADLINE);
+    core.workers
+        .stop_and_join(deadline.saturating_duration_since(Instant::now()));
     debug!("background workers stopped");
 }
 
+/// Join `workers` until `deadline`, and name the ones still running then.
+///
+/// A stop used to wait for each one however long it took. The sync engine's
+/// mirror pass waits on Drive with no deadline, and a stop that waited for it
+/// was killed by systemd (`docs/BUGS.md` B165).
+fn join_until(workers: Vec<std::thread::JoinHandle<()>>, deadline: Instant) -> Vec<String> {
+    let mut running = Vec::new();
+    for worker in workers {
+        while !worker.is_finished() && Instant::now() < deadline {
+            std::thread::sleep(Duration::from_millis(10));
+        }
+        if !worker.is_finished() {
+            running.push(worker.thread().name().unwrap_or("unnamed").to_string());
+        } else if worker.join().is_err() {
+            warn!("a background worker panicked before shutdown");
+        }
+    }
+    running
+}
+
 #[cfg(test)]
 mod tests {
     use std::sync::atomic::{AtomicBool, Ordering};
+    use std::time::{Duration, Instant};
 
-    use super::{SecondaryInsertRejection, SecondaryMountRegistry, teardown_session};
+    use super::{SecondaryInsertRejection, SecondaryMountRegistry, join_until, teardown_session};
+
+    #[test]
+    fn a_stop_names_the_workers_still_running_at_its_deadline() {
+        let (release, wait) = std::sync::mpsc::channel::<()>();
+        let done = std::thread::Builder::new()
+            .name("done".into())
+            .spawn(|| {})
+            .unwrap();
+        let stuck = std::thread::Builder::new()
+            .name("stuck".into())
+            .spawn(move || {
+                let _ = wait.recv();
+            })
+            .unwrap();
+        let running = join_until(
+            vec![done, stuck],
+            Instant::now() + Duration::from_millis(100),
+        );
+        assert_eq!(running, ["stuck"]);
+        drop(release);
+    }
 
     #[test]
     fn session_is_unroutable_before_teardown_starts() {

crates/pdfs-fuse/src/pause.rs

@@ -0,0 +1,235 @@
+//! Pausing sync: the user's "stop uploading for a while".
+//!
+//! A pause holds back the two things that push local state to the remote — the
+//! drain worker's queued uploads and changes, and the mirror-folder reconcile.
+//! Nothing else stops: reads through the mount keep hydrating on demand (a read
+//! is not a sync), writes keep being accepted and staged, and the watcher and
+//! poll keep noticing changes. Resuming lets all of it catch up at once.
+//!
+//! The pause is persisted in the database, so a restart does not quietly undo
+//! it. A timed pause ends by itself; [`Core::set_sync_paused`] arms a timer for
+//! that, and [`Core::sync_paused`] compares against the clock anyway, so a
+//! deadline that passed while the daemon was down is already over on mount.
+//!
+//! A single mirror folder can be paused on its own as well: its reconcile is
+//! skipped, and with it any mode switch queued behind that reconcile, until the
+//! folder is resumed. On-demand folders cannot be paused alone — their writes
+//! go through the shared queue, which only the global pause holds back.
+
+use super::*;
+
+/// The `state` key holding the resume time, in Unix seconds.
+const PAUSED_UNTIL_KEY: &str = "sync_paused_until";
+
+/// The `state` key prefix marking one synced folder paused; its id follows.
+const FOLDER_PAUSED_PREFIX: &str = "sync_folder_paused:";
+
+fn folder_paused_key(id: i64) -> String {
+    format!("{FOLDER_PAUSED_PREFIX}{id}")
+}
+
+/// Whether synced folder `id` is paused on its own. A row that cannot be read
+/// counts as running, like the global pause.
+fn load_folder_paused(db: &Db, id: i64) -> bool {
+    match db.state_i64(&folder_paused_key(id)) {
+        Ok(value) => value.is_some_and(|v| v != 0),
+        Err(error) => {
+            warn!(id, %error, "reading a folder pause failed; treating it as running");
+            false
+        }
+    }
+}
+
+/// Persist synced folder `id`'s own pause, or clear it.
+fn store_folder_paused(db: &Db, id: i64, paused: bool) -> pdfs_core::Result<()> {
+    match paused {
+        true => db.set_state_i64(&folder_paused_key(id), 1),
+        false => db.clear_state(&folder_paused_key(id)),
+    }
+}
+
+/// A pause with no end time: it lasts until the user resumes.
+pub(crate) const PAUSED_INDEFINITELY: i64 = i64::MAX;
+
+/// The persisted pause, or `0` when there is none (or it cannot be read — a
+/// broken state row must not keep sync paused forever).
+pub(crate) fn load_paused_until(db: &Db) -> i64 {
+    match db.state_i64(PAUSED_UNTIL_KEY) {
+        Ok(Some(until)) => until,
+        Ok(None) => 0,
+        Err(error) => {
+            warn!(%error, "reading the sync pause failed; treating sync as running");
+            0
+        }
+    }
+}
+
+/// Whether a pause lasting until `until` is still in force at `now` (both Unix
+/// seconds). `0` is "not paused".
+pub(crate) fn paused_at(until: i64, now: i64) -> bool {
+    until != 0 && now < until
+}
+
+impl Core {
+    /// True while the user has sync paused.
+    pub(crate) fn sync_paused(&self) -> bool {
+        paused_at(self.sync_paused_until.load(Ordering::Relaxed), now_secs())
+    }
+
+    /// The pause's end, for status reporting: `None` when not paused,
+    /// `Some(None)` when paused until resumed, `Some(Some(t))` when it ends at
+    /// Unix second `t`.
+    pub(crate) fn sync_pause(&self) -> Option<Option<i64>> {
+        let until = self.sync_paused_until.load(Ordering::Relaxed);
+        match paused_at(until, now_secs()) {
+            false => None,
+            true if until == PAUSED_INDEFINITELY => Some(None),
+            true => Some(Some(until)),
+        }
+    }
+
+    /// Pause sync until Unix second `until` (`None`: until resumed), or resume
+    /// it when `paused` is false.
+    pub(crate) fn set_sync_paused(
+        &self,
+        paused: bool,
+        until: Option<i64>,
+    ) -> Result<(), CoreError> {
+        let value = match (paused, until) {
+            (false, _) => 0,
+            (true, None) => PAUSED_INDEFINITELY,
+            (true, Some(until)) if until <= now_secs() => {
+                return Err(CoreError::invalid("the pause would already be over"));
+            }
+            (true, Some(until)) => until,
+        };
+        let persisted = match value {
+            0 => self.db.clear_state(PAUSED_UNTIL_KEY),
+            value => self.db.set_state_i64(PAUSED_UNTIL_KEY, value),
+        };
+        if let Err(error) = persisted {
+            return Err(CoreError::internal(format!(
+                "saving the sync pause: {error}"
+            )));
+        }
+        self.sync_paused_until.store(value, Ordering::Relaxed);
+        match value {
+            0 => {
+                info!("sync resumed");
+                self.resume_sync_work();
+            }
+            PAUSED_INDEFINITELY => info!("sync paused until resumed"),
+            until => {
+                info!(until, "sync paused");
+                self.arm_pause_timer(until);
+            }
+        }
+        Ok(())
+    }
+
+    /// True while the user has this one synced folder paused. A row that cannot
+    /// be read counts as running, like the global pause.
+    pub(crate) fn sync_folder_paused(&self, id: i64) -> bool {
+        load_folder_paused(&self.db, id)
+    }
+
+    /// Pause or resume one mirror folder's reconcile. Resuming reconciles it
+    /// straight away to catch up on what the watcher saw meanwhile.
+    pub(crate) fn set_sync_folder_paused(&self, id: i64, paused: bool) -> Result<(), CoreError> {
+        let folder = self
+            .db
+            .sync_folder_get(id)
+            .map_err(|e| CoreError::internal(format!("db: {e:?}")))?
+            .ok_or_else(|| CoreError::not_found(format!("no synced folder with id {id}")))?;
+        if paused && folder.mode != "mirror" {
+            return Err(CoreError::invalid(
+                "only a mirrored folder can be paused on its own; an on-demand folder uploads \
+                 through the shared queue, so pause all sync instead",
+            ));
+        }
+        if let Err(error) = store_folder_paused(&self.db, id, paused) {
+            return Err(CoreError::internal(format!(
+                "saving the folder pause: {error}"
+            )));
+        }
+        if paused {
+            info!(id, path = %folder.local_path, "folder sync paused");
+        } else {
+            info!(id, path = %folder.local_path, "folder sync resumed");
+            self.sync_now(Some(id));
+        }
+        Ok(())
+    }
+
+    /// Forget a removed folder's pause, so a later folder reusing its id does
+    /// not start out paused.
+    pub(crate) fn clear_sync_folder_paused(&self, id: i64) {
+        if let Err(error) = store_folder_paused(&self.db, id, false) {
+            warn!(id, %error, "clearing a removed folder's pause failed");
+        }
+    }
+
+    /// Wake everything a pause held back: the drain worker, and a full
+    /// reconcile to pick up whatever the watcher saw meanwhile.
+    fn resume_sync_work(&self) {
+        self.wake_drain();
+        self.sync_now(None);
+    }
+
+    /// End a timed pause when its time comes. A later pause or resume replaces
+    /// the stored deadline, which the timer checks before acting, so a stale
+    /// timer does nothing.
+    pub(crate) fn arm_pause_timer(&self, until: i64) {
+        let core = self.clone();
+        let wait = Duration::from_secs(until.saturating_sub(now_secs()).max(0) as u64);
+        let spawned = std::thread::Builder::new()
+            .name("pdfs-pause".into())
+            .spawn(move || {
+                if !core.shutdown.sleep(wait) {
+                    return;
+                }
+                if core
+                    .sync_paused_until
+                    .compare_exchange(until, 0, Ordering::Relaxed, Ordering::Relaxed)
+                    .is_ok()
+                {
+                    if let Err(error) = core.db.clear_state(PAUSED_UNTIL_KEY) {
+                        warn!(%error, "clearing an expired sync pause failed");
+                    }
+                    info!("sync pause ended");
+                    core.resume_sync_work();
+                }
+            });
+        if let Err(error) = spawned {
+            // The drain and the reconcile still read the clock, so the pause
+            // still ends — just at the next idle poll instead of on time.
+            warn!(%error, "starting the sync pause timer failed");
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn a_pause_is_in_force_until_its_deadline() {
+        assert!(!paused_at(0, 100));
+        assert!(paused_at(200, 100));
+        assert!(!paused_at(200, 200));
+        assert!(!paused_at(200, 300));
+        assert!(paused_at(PAUSED_INDEFINITELY, i64::MAX - 1));
+    }
+
+    #[test]
+    fn a_folder_pause_is_kept_per_folder_until_cleared() {
+        let db = Db::open(std::path::Path::new(":memory:")).unwrap();
+        assert!(!load_folder_paused(&db, 3));
+        store_folder_paused(&db, 3, true).unwrap();
+        assert!(load_folder_paused(&db, 3));
+        assert!(!load_folder_paused(&db, 4));
+        assert!(!load_folder_paused(&db, 33));
+        store_folder_paused(&db, 3, false).unwrap();
+        assert!(!load_folder_paused(&db, 3));
+    }
+}

crates/pdfs-fuse/src/profile.rs

@@ -1,5 +1,5 @@
 //! Reading and writing this machine's `profile.json` in its device root, and
-//! the restore that rebuilds a machine from one (features.md 5.2 and 5.3).
+//! the restore that rebuilds a machine from one.
 //!
 //! The document itself is defined in [`pdfs_core::profile`]; this is the half
 //! that talks to Drive. Saving is best-effort and never blocks a user action:
@@ -15,6 +15,7 @@ use pdfs_core::control::RestorableFolder;
 use pdfs_core::mounts::MountMode;
 use pdfs_core::profile::{
     PROFILE_DIR_NAME, PROFILE_FILE_NAME, PROFILE_VERSION, Profile, ProfileFolder, ProfilePin,
+    is_profile_file_name, profile_file_name, saved_folder,
 };
 use pdfs_core::{CoreError, CoreResult};
 use proton_drive_rs::proton_sdk::ids::NodeUid;
@@ -66,9 +67,14 @@ impl Core {
         });
     }
 
-    /// Serialize this machine's arrangement and upload it to `profile.json`
-    /// inside the device's `.proton-drive-linux` folder, replacing any existing
-    /// one.
+    /// Serialize this machine's arrangement and upload it to its own
+    /// `profile-<machine id>.json` inside the device's `.proton-drive-linux`
+    /// folder, replacing any existing one.
+    ///
+    /// Each machine writes its own file because several computers may sync
+    /// into one device. They used to share `profile.json`, and each save
+    /// replaced the other computers' folders with its own (B252). That file is
+    /// left as it is, so a client before 3.5.0 still finds it.
     ///
     /// Does nothing when no device is registered: a machine that has never
     /// synced a folder has nothing to describe, and registering a device just to
@@ -85,7 +91,11 @@ impl Core {
             CoreError::internal(format!("bad device root uid: {}", device.root_uid))
         })?;
 
-        let profile = self.build_profile(&device.uid)?;
+        let machine_id = pdfs_core::config::AppDirs::new()
+            .and_then(|dirs| dirs.machine_id())
+            .map_err(|e| CoreError::internal(format!("machine id: {e}")))?;
+        let file_name = profile_file_name(&machine_id);
+        let profile = self.build_profile(&device.uid, &machine_id)?;
         let bytes = profile.to_bytes().map_err(CoreError::internal)?;
         let len = bytes.len() as i64;
 
@@ -97,11 +107,11 @@ impl Core {
         // same name: the profile is a single mutable record, and a folder with
         // `profile.json` and `profile (1).json` in it teaches the restore
         // nothing about which to trust.
-        let existing = self.find_device_child_file(&dir_uid, PROFILE_FILE_NAME)?;
+        let existing = self.find_device_child_file(&dir_uid, &file_name)?;
         match existing {
             Some(uid) => self
                 .rt
-                .block_on(self.client.upload_new_revision_from(
+                .block_on(self.drive.upload_new_revision_from(
                     &uid,
                     Cursor::new(bytes),
                     len,
@@ -112,9 +122,9 @@ impl Core {
                 .map_err(|e| CoreError::from_api(&e, "upload profile revision"))?,
             None => {
                 self.rt
-                    .block_on(self.client.upload_file_replacing_draft_from(
+                    .block_on(self.drive.upload_file_replacing_draft_from(
                         &dir_uid,
-                        PROFILE_FILE_NAME,
+                        &file_name,
                         "application/json",
                         Cursor::new(bytes),
                         len,
@@ -130,7 +140,7 @@ impl Core {
     }
 
     /// Snapshot the local arrangement into a [`Profile`].
-    fn build_profile(&self, device_uid: &str) -> CoreResult<Profile> {
+    fn build_profile(&self, device_uid: &str, machine_id: &str) -> CoreResult<Profile> {
         let folders = self
             .db
             .sync_folder_list()
@@ -162,6 +172,7 @@ impl Core {
             version: PROFILE_VERSION,
             device_uid: device_uid.to_string(),
             hostname: this_hostname(),
+            machine_id: Some(machine_id.to_string()),
             saved_at: now_secs(),
             folders,
             pins,
@@ -182,14 +193,14 @@ impl Core {
     ) -> CoreResult<Option<NodeUid>> {
         let uids = self
             .rt
-            .block_on(self.client.enumerate_folder_children_node_uids(parent_uid))
+            .block_on(self.drive.enumerate_folder_children_node_uids(parent_uid))
             .map_err(|e| CoreError::from_api(&e, "list device root"))?;
         if uids.is_empty() {
             return Ok(None);
         }
         let nodes = self
             .rt
-            .block_on(self.client.enumerate_nodes(&uids))
+            .block_on(self.drive.enumerate_nodes(&uids))
             .map_err(|e| CoreError::from_api(&e, "resolve device root children"))?;
         Ok(nodes
             .into_iter()
@@ -197,6 +208,31 @@ impl Core {
             .map(|n| n.uid))
     }
 
+    /// The untrashed files directly under `parent_uid` whose name passes
+    /// `keep`.
+    fn device_child_files(
+        &self,
+        parent_uid: &NodeUid,
+        keep: impl Fn(&str) -> bool,
+    ) -> CoreResult<Vec<NodeUid>> {
+        let uids = self
+            .rt
+            .block_on(self.drive.enumerate_folder_children_node_uids(parent_uid))
+            .map_err(|e| CoreError::from_api(&e, "list device root"))?;
+        if uids.is_empty() {
+            return Ok(Vec::new());
+        }
+        let nodes = self
+            .rt
+            .block_on(self.drive.enumerate_nodes(&uids))
+            .map_err(|e| CoreError::from_api(&e, "resolve device root children"))?;
+        Ok(nodes
+            .into_iter()
+            .filter(|n| !n.is_folder() && !n.trashed && keep(&n.name))
+            .map(|n| n.uid)
+            .collect())
+    }
+
     /// The device's `.proton-drive-linux` folder, creating it if this is the
     /// first save on this device.
     ///
@@ -210,7 +246,7 @@ impl Core {
         let uid = self
             .rt
             .block_on(
-                self.client
+                self.drive
                     .create_folder(root_uid, PROFILE_DIR_NAME, Some(now_secs())),
             )
             .map_err(|e| CoreError::from_api(&e, "create profile folder"))?;
@@ -218,68 +254,84 @@ impl Core {
         Ok(uid)
     }
 
-    /// Download and parse this device's profile, if there is one.
+    /// Download and parse the profiles of every computer that syncs this
+    /// device.
     ///
-    /// Looks in `.proton-drive-linux/` first and falls back to the device root,
-    /// where clients before the B78 fix wrote it. The fallback is read-only: the
-    /// next save writes to the folder, and the legacy document is left alone
-    /// rather than trashed, so an older client on another machine keeps working.
-    fn load_profile(&self, root_uid: &NodeUid) -> CoreResult<Option<Profile>> {
-        let current = match self.find_device_child_folder(root_uid, PROFILE_DIR_NAME)? {
-            Some(dir_uid) => self.find_device_child_file(&dir_uid, PROFILE_FILE_NAME)?,
-            None => None,
+    /// Reads every profile in `.proton-drive-linux/`: one per machine, and the
+    /// shared `profile.json` of clients before 3.5.0. Falls back to the device
+    /// root, where clients before the B78 fix wrote it. The fallback is
+    /// read-only: the next save writes to the folder, and the legacy document
+    /// is left alone rather than trashed, so an older client on another
+    /// machine keeps working. A profile that cannot be read is left out and
+    /// logged; the others still count.
+    fn load_profiles(&self, root_uid: &NodeUid) -> CoreResult<Vec<Profile>> {
+        let mut uids = match self.find_device_child_folder(root_uid, PROFILE_DIR_NAME)? {
+            Some(dir_uid) => self.device_child_files(&dir_uid, is_profile_file_name)?,
+            None => Vec::new(),
         };
-        let uid = match current {
-            Some(uid) => uid,
-            None => match self.find_device_child_file(root_uid, PROFILE_FILE_NAME)? {
-                Some(uid) => {
-                    info!("reading legacy profile from the device root");
-                    uid
+        if uids.is_empty() {
+            uids = self.device_child_files(root_uid, |name| name == PROFILE_FILE_NAME)?;
+            if !uids.is_empty() {
+                info!("reading legacy profile from the device root");
+            }
+        }
+        let mut profiles = Vec::new();
+        for uid in uids {
+            let mut buf: Vec<u8> = Vec::new();
+            let parsed = self
+                .rt
+                .block_on(self.drive.download_file_to(&uid, &mut buf))
+                .map_err(|e| CoreError::from_api(&e, "download profile"))
+                .and_then(|()| Profile::parse(&buf).map_err(CoreError::invalid));
+            match parsed {
+                Ok(profile) => profiles.push(profile),
+                Err(e) => {
+                    warn!(%uid, error = %e.message, "could not read a profile; leaving it out")
                 }
-                None => return Ok(None),
-            },
-        };
-        let mut buf: Vec<u8> = Vec::new();
-        self.rt
-            .block_on(self.client.download_file_to(&uid, &mut buf))
-            .map_err(|e| CoreError::from_api(&e, "download profile"))?;
-        Profile::parse(&buf).map(Some).map_err(CoreError::invalid)
+            }
+        }
+        Ok(profiles)
     }
 
-    // ---- restore (features.md 5.2) ----------------------------------------
+    // ---- restore ----------------------------------------
 
-    /// The folders under this machine's device that could be synced here, each
-    /// carrying a proposed local path.
+    /// The folders under a device that could be synced here, each carrying a
+    /// proposed local path. `device` names another registered device by uid;
+    /// `None` is this machine's own.
     ///
     /// The proposal comes from the profile when one exists and its path is
     /// usable on this machine, and from `~/<name>` otherwise — a restored
     /// `/home/other-user/Documents` would be wrong on a machine where that user
     /// does not exist, and silently creating it would be worse.
-    pub(crate) fn list_restorable_folders(&self) -> CoreResult<Vec<RestorableFolder>> {
-        let device = self.ensure_device()?;
+    pub(crate) fn list_restorable_folders(
+        &self,
+        device: Option<&str>,
+    ) -> CoreResult<Vec<RestorableFolder>> {
+        let device = self.restore_source(device)?;
         let root_uid = parse_uid(&device.root_uid).ok_or_else(|| {
             CoreError::internal(format!("bad device root uid: {}", device.root_uid))
         })?;
 
-        let profile = match self.load_profile(&root_uid) {
+        let profiles = match self.load_profiles(&root_uid) {
             Ok(p) => p,
             // A missing or unreadable profile is not fatal: the remote tree
             // alone is enough to offer a restore, just with plainer defaults.
             Err(e) => {
-                warn!(error = %e.message, "could not read profile; proposing defaults");
-                None
+                warn!(error = %e.message, "could not read profiles; proposing defaults");
+                Vec::new()
             }
         };
+        let hostname = this_hostname();
 
         let uids = self
             .rt
-            .block_on(self.client.enumerate_folder_children_node_uids(&root_uid))
+            .block_on(self.drive.enumerate_folder_children_node_uids(&root_uid))
             .map_err(|e| CoreError::from_api(&e, "list device root"))?;
         let nodes = if uids.is_empty() {
             Vec::new()
         } else {
             self.rt
-                .block_on(self.client.enumerate_nodes(&uids))
+                .block_on(self.drive.enumerate_nodes(&uids))
                 .map_err(|e| CoreError::from_api(&e, "resolve device root children"))?
         };
 
@@ -300,11 +352,12 @@ impl Core {
             // offering it as a restorable folder would sync the bookkeeping
             // back onto the machine it describes.
             .filter(|n| n.is_folder() && !n.trashed && n.name != PROFILE_DIR_NAME)
+            // The name becomes a folder under the home folder: one with a `/`
+            // or a `..` in it would put the restore somewhere else.
+            .filter(|n| crate::remote_name_is_safe(&n.name))
             .map(|n| {
                 let uid = n.uid.to_string();
-                let saved = profile
-                    .as_ref()
-                    .and_then(|p| p.folders.iter().find(|f| f.remote_uid == uid));
+                let saved = saved_folder(&profiles, &uid, &hostname);
                 let proposed = saved
                     .map(|f| PathBuf::from(&f.local_path))
                     .filter(|p| path_is_usable_here(p))
@@ -334,33 +387,40 @@ impl Core {
     pub(crate) fn restore_sync_folders(
         &self,
         items: &[pdfs_core::control::RestoreItem],
-    ) -> CoreResult<String> {
+        device: Option<&str>,
+    ) -> CoreResult<RestoreOutcome> {
         if items.is_empty() {
             return Err(CoreError::invalid("nothing to restore"));
         }
-        let device = self.ensure_device()?;
+        let device = self.restore_source(device)?;
         let existing = self
             .db
             .sync_folder_list()
             .map_err(|e| CoreError::internal(format!("db: {e:?}")))?;
 
+        let reserved = crate::devices::sync_root_reserved();
+        let mut roots: Vec<PathBuf> = existing
+            .iter()
+            .map(|f| PathBuf::from(&f.local_path))
+            .collect();
         let mut restored = 0usize;
         let mut skipped: Vec<String> = Vec::new();
         for item in items {
-            let local = PathBuf::from(&item.local_path);
-            if !local.is_absolute() {
-                skipped.push(format!("{}: local path must be absolute", item.local_path));
-                continue;
-            }
+            // The path is proposed from the profile another machine wrote, and
+            // may hold a folder name from Drive: checked like any new root.
+            let local = match pdfs_core::syncroot::resolve(Path::new(&item.local_path)) {
+                Ok(local) => local,
+                Err(e) => {
+                    skipped.push(format!("{}: {e}", item.local_path));
+                    continue;
+                }
+            };
             if existing.iter().any(|f| f.remote_uid == item.remote_uid) {
                 skipped.push(format!("{}: already synced", item.local_path));
                 continue;
             }
-            if existing
-                .iter()
-                .any(|f| Path::new(&f.local_path) == local.as_path())
-            {
-                skipped.push(format!("{}: path already in use", item.local_path));
+            if let Some(problem) = pdfs_core::syncroot::problem(&local, &roots, &reserved) {
+                skipped.push(format!("{}: {problem}", item.local_path));
                 continue;
             }
             if let Err(e) = std::fs::create_dir_all(&local) {
@@ -374,10 +434,19 @@ impl Core {
                     continue;
                 }
             };
-            let id = self
-                .db
-                .sync_folder_add(&local.to_string_lossy(), &item.remote_uid, &device.share_id)
-                .map_err(|e| CoreError::internal(format!("db: {e:?}")))?;
+            let id = match self.db.sync_folder_add(
+                &local.to_string_lossy(),
+                &item.remote_uid,
+                &device.share_id,
+            ) {
+                Ok(id) => id,
+                Err(e) => {
+                    // The folders restored before this one still need their
+                    // watches.
+                    let _ = self.sync_tx.send(sync::SyncMsg::Rewatch);
+                    return Err(CoreError::internal(format!("db: {e:?}")));
+                }
+            };
             // `ondemand` is applied through the normal request path, which
             // mounts the FUSE session and handles the folder being busy — the
             // restore does not get its own copy of that logic.
@@ -385,16 +454,44 @@ impl Core {
                 let _ = self.request_sync_folder_mode(id, "ondemand");
             }
             let _ = self.sync_tx.send(sync::SyncMsg::Reconcile(id));
+            roots.push(local);
             restored += 1;
         }
         let _ = self.sync_tx.send(sync::SyncMsg::Rewatch);
 
-        let mut message = format!("restored {restored} folder(s)");
-        if !skipped.is_empty() {
-            message.push_str(&format!("; skipped: {}", skipped.join(", ")));
+        // The reply is gone once the app has shown it; the log keeps why.
+        for reason in &skipped {
+            info!(reason, "restore skipped a folder");
         }
         info!(restored, skipped = skipped.len(), "restore complete");
-        Ok(message)
+        Ok(RestoreOutcome { restored, skipped })
+    }
+}
+
+/// What a restore did: how many folders it attached, and why it left out the
+/// others.
+pub(crate) struct RestoreOutcome {
+    pub(crate) restored: usize,
+    pub(crate) skipped: Vec<String>,
+}
+
+impl RestoreOutcome {
+    /// The activity entry's headline.
+    pub(crate) fn summary(&self) -> String {
+        match (self.restored, self.skipped.len()) {
+            (n, 0) => format!("restored {n} folder(s)"),
+            (0, k) => format!("restored no folders; skipped {k}"),
+            (n, k) => format!("restored {n} folder(s); skipped {k}"),
+        }
+    }
+
+    /// Whether the restore did all it was asked. A skipped folder is a
+    /// failure the user has to see: the reply went out as "restoring folders"
+    /// before any work, and a restore that skipped every folder used to be
+    /// logged as a success, so nothing anywhere said that nothing would
+    /// download (`docs/BUGS.md` B269).
+    pub(crate) fn complete(&self) -> bool {
+        self.skipped.is_empty()
     }
 }
 
@@ -477,7 +574,7 @@ mod tests {
     #[test]
     fn loading_falls_back_to_a_legacy_device_root_profile() {
         let source = include_str!("profile.rs");
-        let load = function_source(source, "fn load_profile(");
+        let load = function_source(source, "fn load_profiles(");
         assert!(load.contains("PROFILE_DIR_NAME"), "prefers the folder");
         assert!(
             load.find("PROFILE_DIR_NAME").unwrap()
@@ -486,6 +583,16 @@ mod tests {
         );
     }
 
+    /// Several computers may share a device, so each saves to a file named
+    /// after its own machine id and never to the shared `profile.json`
+    /// (B252).
+    #[test]
+    fn each_machine_saves_its_own_profile() {
+        let save = function_source(include_str!("profile.rs"), "fn save_profile(");
+        assert!(save.contains("profile_file_name(&machine_id)"));
+        assert!(!save.contains("PROFILE_FILE_NAME"));
+    }
+
     /// The profile folder is the client's own bookkeeping: neither the restore
     /// picker nor `add_sync_folder` may treat it as a user folder.
     #[test]
@@ -499,6 +606,19 @@ mod tests {
         assert_before(add, "PROFILE_DIR_NAME", "ensure_device()");
     }
 
+    /// Restoring another computer's folders reads that device, but must not
+    /// rewrite the cached device row: this machine would then back up as the
+    /// other computer.
+    #[test]
+    fn restoring_from_another_device_keeps_this_machines_identity() {
+        let source = function_source(include_str!("devices.rs"), "fn restore_source(");
+        assert!(source.contains("ensure_device()"), "None is this machine");
+        assert!(
+            !source.contains("device_set"),
+            "the cached row is untouched"
+        );
+    }
+
     fn function_source<'a>(source: &'a str, signature: &str) -> &'a str {
         let start = source.find(signature).expect("function signature exists");
         let body = &source[start..];

crates/pdfs-fuse/src/queue.rs

@@ -0,0 +1,302 @@
+//! The user's view of the pending-op queue: what has not reached the remote
+//! yet, why, and a way to stop waiting out a backoff.
+//!
+//! Read-only apart from "retry now", which only moves a row's next-attempt time,
+//! "export", which copies a staged blob out, and "discard". Only a discard the
+//! user asks for drops a queued op: its staged blob is the only copy of the
+//! user's write (see `docs/RECOVERY.md`), so it is the user's call, made after
+//! an export or for a change with no content.
+
+use super::*;
+use std::io::Write;
+
+use pdfs_core::control::{PendingOpInfo, SyncIssue};
+use pdfs_core::db::{FAILING_ATTEMPTS, PARK_UNTIL};
+
+impl Core {
+    /// Every queued op, oldest first, with a path a person can read.
+    pub(crate) fn pending_op_infos(&self) -> CoreResult<Vec<PendingOpInfo>> {
+        let ops = self
+            .db
+            .pending_ops()
+            .map_err(|e| CoreError::internal(format!("listing the queue: {e}")))?;
+        let issues = self
+            .db
+            .op_issues()
+            .map_err(|e| CoreError::internal(format!("listing sync issues: {e}")))?;
+        Ok(ops
+            .into_iter()
+            .map(|op| {
+                let parked = op.next_attempt_at >= PARK_UNTIL;
+                let exportable = exportable(&op).is_some();
+                let path = self
+                    .db
+                    .node_path(&op.uid)
+                    .ok()
+                    .flatten()
+                    .filter(|path| !path.is_empty())
+                    .or_else(|| op.name.clone())
+                    .unwrap_or_else(|| op.uid.clone());
+                PendingOpInfo {
+                    id: op.id,
+                    kind: op.kind,
+                    path,
+                    attempts: op.attempts,
+                    last_error: op.last_error,
+                    queued_at: op.created_at / 1000,
+                    next_attempt_at: (!parked).then_some(op.next_attempt_at / 1000),
+                    parked,
+                    failing: op.attempts >= FAILING_ATTEMPTS || issues.contains_key(&op.id),
+                    issue: issues.get(&op.id).map(|issue| SyncIssue::parse(issue)),
+                    exportable,
+                }
+            })
+            .collect())
+    }
+
+    /// Retry one backed-off op now, or every failed one when `id` is `None`.
+    /// Returns how many ops were moved up.
+    pub(crate) fn retry_pending_ops(&self, id: Option<i64>) -> CoreResult<usize> {
+        let now = now_millis();
+        let moved = match id {
+            Some(id) => self.db.retry_op_now(id, now).map(usize::from),
+            None => self.db.retry_failed_ops_now(now),
+        }
+        .map_err(|e| CoreError::internal(format!("retrying the queue: {e}")))?;
+        if moved > 0 {
+            self.wake_drain();
+        }
+        Ok(moved)
+    }
+
+    /// Copy the content of queued op `id` to `dest`, which must not exist yet.
+    /// The op stays queued. Returns the bytes written.
+    pub(crate) fn export_pending_op(&self, id: i64, dest: &Path) -> CoreResult<u64> {
+        if !dest.is_absolute() {
+            return Err(CoreError::invalid(format!(
+                "destination must be an absolute path: {}",
+                dest.display()
+            )));
+        }
+        let ops = self
+            .db
+            .pending_ops()
+            .map_err(|e| CoreError::internal(format!("listing the queue: {e}")))?;
+        let op = ops
+            .iter()
+            .find(|op| op.id == id)
+            .ok_or_else(|| CoreError::not_found(format!("no queued change {id}")))?;
+        let blob = exportable(op).ok_or_else(|| {
+            CoreError::invalid(format!("queued change {id} carries no file content"))
+        })?;
+        let written = copy_to_new_file(blob, dest)?;
+        info!(id, dest = %dest.display(), written, "exported a queued change's content");
+        Ok(written)
+    }
+
+    /// Drop queued op `id` and undo the local change it stood for. Returns the
+    /// node's path.
+    ///
+    /// A node only made here goes, as if it had been deleted before it was
+    /// sent. A node Drive has goes back to what Drive has: every change queued
+    /// on it goes, since each was made on top of the one before. What is queued
+    /// below it stays.
+    pub(crate) fn discard_pending_op(&self, id: i64) -> CoreResult<String> {
+        let ops = self
+            .db
+            .pending_ops()
+            .map_err(|e| CoreError::internal(format!("listing the queue: {e}")))?;
+        let op = ops
+            .into_iter()
+            .find(|op| op.id == id)
+            .ok_or_else(|| CoreError::not_found(format!("no queued change {id}")))?;
+        let uid = parse_node_uid(&op.uid)
+            .ok_or_else(|| CoreError::internal(format!("queued change {id} names {}", op.uid)))?;
+        let path = self
+            .db
+            .node_path(&op.uid)
+            .ok()
+            .flatten()
+            .filter(|path| !path.is_empty())
+            .or(op.name)
+            .unwrap_or_else(|| op.uid.clone());
+        if self.is_open_anywhere(&uid) {
+            return Err(CoreError::conflict(format!(
+                "{path} is open; close it first"
+            )));
+        }
+        if is_local_uid(&uid) {
+            self.discard_local_node(&uid, &path)?;
+        } else {
+            self.discard_remote_change(&uid, &path)?;
+        }
+        info!(id, %uid, path, "discarded a queued change");
+        Ok(path)
+    }
+
+    /// Forget a node made here, and the ops that would have sent it.
+    fn discard_local_node(&self, uid: &NodeUid, path: &str) -> CoreResult<()> {
+        // Its children's ops go with its own, and their content with them.
+        if self
+            .db
+            .has_children(uid)
+            .map_err(|e| CoreError::internal(format!("listing {path}: {e}")))?
+        {
+            return Err(CoreError::conflict(format!(
+                "{path} is not empty; discard what is in it first"
+            )));
+        }
+        self.discard_queued_ops(uid)
+            .map_err(|_| CoreError::internal(format!("dropping the changes to {path}")))?;
+        self.for_each_mount(|st, notify| {
+            background::forget_and_notify(st, notify, uid);
+        });
+        self.cache.evict(uid);
+        self.evict_reader(uid);
+        Ok(())
+    }
+
+    /// Put a node Drive has back as Drive has it.
+    fn discard_remote_change(&self, uid: &NodeUid, path: &str) -> CoreResult<()> {
+        // Asked first: the change is only dropped once what replaces it is known.
+        let node = self
+            .fetch_node_remote(uid)
+            .map_err(|e| CoreError::from_api(&e, "reading the node from Drive"))?
+            .filter(|node| !node.trashed);
+        self.cancel_upload(uid);
+        let blobs = self
+            .db
+            .discard_ops_on_node(&uid.to_string())
+            .map_err(|e| CoreError::internal(format!("dropping the changes to {path}: {e}")))?
+            .ok_or_else(|| {
+                CoreError::conflict(format!("{path} is being sent right now; try again"))
+            })?;
+        self.release_dropped_ops(uid, blobs);
+        self.hidden.lock().remove(uid);
+        match node {
+            Some(node) => self.for_each_mount(|st, notify| {
+                put_back(st, notify, node);
+            }),
+            None => self.for_each_mount(|st, notify| {
+                background::forget_and_notify(st, notify, uid);
+            }),
+        }
+        self.cache.evict(uid);
+        self.evict_reader(uid);
+        Ok(())
+    }
+}
+
+/// Show `node` as Drive has it, where Drive has it, under the inode it has.
+fn put_back(st: &mut State, notify: &mut NotifyBatch, node: Node) {
+    let parent = node.parent_uid.as_ref().and_then(|parent| {
+        st.by_uid
+            .get(parent)
+            .copied()
+            .map(|ino| (ino, parent.clone()))
+    });
+    let here = st.by_uid.get(&node.uid).and_then(|ino| {
+        st.entries
+            .get(ino)
+            .map(|entry| (*ino, entry.parent, entry.node.name.clone()))
+    });
+    match (here, parent) {
+        (Some((ino, from, name)), Some((to, to_uid))) => {
+            st.relocate(ino, from, to, &to_uid, &node.name);
+            st.write_through(node.clone());
+            st.intern_mem(to, node);
+            notify.inval_entry(from, name);
+            notify.inval_inode(ino);
+        }
+        (here, parent) => {
+            if let Some((_, from, name)) = here {
+                st.forget_mem(&node.uid);
+                st.invalidate_listing(from);
+                notify.inval_entry(from, name);
+            }
+            if let Some((to, _)) = parent {
+                st.invalidate_listing(to);
+                notify.inval_entry(to, node.name.clone());
+            }
+            st.write_through(node);
+        }
+    }
+}
+
+/// Copy `blob` to `dest`, which must not exist: never replace something the
+/// user already has.
+fn copy_to_new_file(blob: &Path, dest: &Path) -> CoreResult<u64> {
+    let mut source = File::open(blob)
+        .map_err(|e| CoreError::internal(format!("open {}: {e}", blob.display())))?;
+    let mut out = std::fs::OpenOptions::new()
+        .write(true)
+        .create_new(true)
+        .open(dest)
+        .map_err(|e| match e.kind() {
+            std::io::ErrorKind::AlreadyExists => {
+                CoreError::invalid(format!("{} already exists", dest.display()))
+            }
+            _ => CoreError::internal(format!("create {}: {e}", dest.display())),
+        })?;
+    std::io::copy(&mut source, &mut out)
+        .and_then(|written| out.flush().and(out.sync_all()).map(|()| written))
+        .map_err(|e| CoreError::internal(format!("write {}: {e}", dest.display())))
+}
+
+/// The staged blob of an op that carries file content.
+fn exportable(op: &PendingOp) -> Option<&Path> {
+    matches!(op.kind.as_str(), OP_CREATE | OP_REVISION)
+        .then_some(op.blob_path.as_deref())
+        .flatten()
+        .map(Path::new)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn op(kind: &str, blob: Option<&str>) -> PendingOp {
+        PendingOp {
+            id: 1,
+            kind: kind.to_string(),
+            uid: "vol~a".to_string(),
+            parent_uid: None,
+            name: None,
+            blob_path: blob.map(str::to_string),
+            meta_json: None,
+            created_at: 0,
+            attempts: 0,
+            last_error: None,
+            next_attempt_at: 0,
+        }
+    }
+
+    #[test]
+    fn only_a_change_that_carries_file_content_can_be_exported() {
+        assert!(exportable(&op(OP_CREATE, Some("/staging/a"))).is_some());
+        assert!(exportable(&op(OP_REVISION, Some("/staging/a"))).is_some());
+        assert!(exportable(&op(OP_CREATE, None)).is_none());
+        assert!(exportable(&op(OP_RENAME, None)).is_none());
+        assert!(exportable(&op(OP_TRASH, None)).is_none());
+    }
+
+    #[test]
+    fn an_export_copies_the_bytes_and_never_replaces_a_file() {
+        let dir = std::env::temp_dir().join(format!("pdfs-export-{}", std::process::id()));
+        std::fs::create_dir_all(&dir).unwrap();
+        let blob = dir.join("blob");
+        std::fs::write(&blob, b"queued bytes").unwrap();
+
+        let dest = dir.join("copy.txt");
+        assert_eq!(copy_to_new_file(&blob, &dest).unwrap(), 12);
+        assert_eq!(std::fs::read(&dest).unwrap(), b"queued bytes");
+
+        std::fs::write(&dest, b"the user's own").unwrap();
+        assert!(copy_to_new_file(&blob, &dest).is_err());
+        assert_eq!(std::fs::read(&dest).unwrap(), b"the user's own");
+        // The staged blob is the only copy of the change: an export leaves it.
+        assert_eq!(std::fs::read(&blob).unwrap(), b"queued bytes");
+
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+}

crates/pdfs-fuse/src/reads.rs

@@ -36,6 +36,21 @@ pub(super) const REPAIR_MAX: u64 = 512 * 1024 * 1024;
 /// demand read in the SDK's flat block semaphore.
 pub(super) const PREFETCH_BUDGET: usize = 8;
 
+/// How long one read may wait on the network before it fails with `EIO`.
+///
+/// The SDK bounds each request, not a read: a block that times out is retried,
+/// and a read can need several blocks, a block table and a repair. On a link
+/// that stalled without dropping, one `read` held a FUSE worker for 34 minutes
+/// and the application behind it with it (`docs/BUGS.md` B107). An error the
+/// caller can retry beats a process stuck in `D` state.
+pub(super) const READ_FETCH_TIMEOUT: Duration = Duration::from_secs(120);
+
+/// How long a read of blocks nothing has cached waits for a lost link to come
+/// back before failing. Long enough to ride out a Wi-Fi blip, short enough
+/// that a program reading while the laptop is really offline gets its `EIO`
+/// before the user gives up on it.
+const READ_OFFLINE_WAIT: Duration = Duration::from_secs(30);
+
 /// Cap on tracked sequential streams. Only reads in progress matter, so the map
 /// is cleared wholesale rather than aged — losing the ramp costs one read.
 const MAX_PREFETCH_STREAMS: usize = 256;
@@ -55,7 +70,7 @@ const MAX_OPEN_READERS: usize = 64;
 /// `(mtime, size)` — the same validity pair the content cache uses (a new
 /// revision bumps mtime). On a mismatch the reader is dropped and reopened.
 pub(super) struct CachedReader {
-    reader: Arc<RevisionReader>,
+    reader: Arc<dyn RevisionRead>,
     mtime: i64,
     size: u64,
     /// For LRU eviction.
@@ -86,7 +101,7 @@ pub(super) struct PendingOpen {
     /// must not join this open — it would get a reader for the wrong bytes.
     mtime: i64,
     size: u64,
-    rx: tokio::sync::watch::Receiver<Option<Result<Arc<RevisionReader>, Errno>>>,
+    rx: tokio::sync::watch::Receiver<Option<Result<Arc<dyn RevisionRead>, Errno>>>,
 }
 
 /// Distinguishes `PendingOpen`s. Only uniqueness matters, so a plain counter
@@ -298,16 +313,16 @@ impl Core {
         uid: &NodeUid,
         mtime: i64,
         fsize: u64,
-    ) -> Result<Arc<RevisionReader>, Errno> {
+    ) -> Result<Arc<dyn RevisionRead>, Errno> {
         loop {
             // Decide under the lock, act outside it — network I/O must never run
             // under a std `Mutex`, and an await must never hold one.
             enum Act {
-                Hit(Arc<RevisionReader>),
-                Join(tokio::sync::watch::Receiver<Option<Result<Arc<RevisionReader>, Errno>>>),
+                Hit(Arc<dyn RevisionRead>),
+                Join(tokio::sync::watch::Receiver<Option<Result<Arc<dyn RevisionRead>, Errno>>>),
                 Lead(
                     u64,
-                    tokio::sync::watch::Sender<Option<Result<Arc<RevisionReader>, Errno>>>,
+                    tokio::sync::watch::Sender<Option<Result<Arc<dyn RevisionRead>, Errno>>>,
                 ),
             }
             let act = {
@@ -361,9 +376,10 @@ impl Core {
                 }
                 Act::Lead(id, tx) => {
                     debug!(%uid, mtime, fsize, "opening revision");
-                    let result = match self.client.open_revision(uid).await {
-                        Ok(reader) => Ok(Arc::new(reader)),
+                    let result = match self.drive.open_revision(uid).await {
+                        Ok(reader) => Ok(reader),
                         Err(e) => {
+                            self.lost_link(&e, "open a revision");
                             warn!(%uid, error = %e, "open_revision failed");
                             Err(Errno::EIO)
                         }
@@ -448,7 +464,7 @@ impl Core {
     /// it, and the [`BLOCK_SIZE`] assumption when it has not — which is every
     /// first read, because learning it means opening a `RevisionReader`, and
     /// opening one per read is precisely the per-file key derivation the block
-    /// cache exists to avoid (bugs.md B12, B85).
+    /// cache exists to avoid (`docs/BUGS.md` B12, B85).
     ///
     /// Being wrong here is not a correctness problem: `read_at` plans over the
     /// real sizes and clamps to what was asked for either way. It costs a
@@ -477,15 +493,20 @@ impl Core {
     ) -> Result<Vec<u8>, Errno> {
         // A queued write has not reached the remote yet, so the remote's current
         // revision is stale and the staged blob is the truth. Serve from it until
-        // the drain worker lands the upload (offline.md Phase 3).
+        // the drain worker lands the upload.
         if let Some(pending) = self.pending.lock().get(uid).cloned() {
             return self.read_pending(&pending, offset, len);
         }
         // A node created offline and never written has no blob and no remote: it
         // is an empty file, and asking the API about a `local~` uid would only
-        // earn a 404 (offline.md Phase 3b).
+        // earn a 404. Unless its create landed since the
+        // read looked: the landing moved the queued write to the uid Drive
+        // gave it, and the row says which (`docs/BUGS.md` B179).
         if is_local_uid(uid) {
-            return Ok(Vec::new());
+            return match self.landed_stand_in_uid(uid) {
+                Some(real) => self.read_range(&real, mtime, fsize, offset, len, cache_blocks),
+                None => Ok(Vec::new()),
+            };
         }
         self.read_range_remote(uid, mtime, fsize, offset, len, cache_blocks)
     }
@@ -552,7 +573,7 @@ impl Core {
     /// leader's claim is gone by then, so this cannot recurse.
     async fn fetch_block(
         &self,
-        reader: &Arc<RevisionReader>,
+        reader: &Arc<dyn RevisionRead>,
         uid: &NodeUid,
         mtime: i64,
         fsize: u64,
@@ -615,10 +636,10 @@ impl Core {
     ///
     /// A block that comes back shorter than the file's size says it must be is
     /// never served, cached or promoted: it is repaired from the whole-file
-    /// download, or the read fails. See [`Core::repair_block`] (bugs.md B84).
+    /// download, or the read fails. See [`Core::repair_block`] (`docs/BUGS.md` B84).
     async fn read_block(
         &self,
-        reader: &Arc<RevisionReader>,
+        reader: &Arc<dyn RevisionRead>,
         uid: &NodeUid,
         mtime: i64,
         fsize: u64,
@@ -630,7 +651,7 @@ impl Core {
         // file's boundaries are wrong, and the failure is not confined to short
         // reads: sizes that *overstate* a block shift every block after it and
         // return full-length reads of the wrong bytes. So a disagreement of any
-        // kind takes the file off the range path entirely (bugs.md B84).
+        // kind takes the file off the range path entirely (`docs/BUGS.md` B84).
         let bytes = if reader.size() != fsize {
             warn!(
                 %uid, bidx = span.idx, fsize,
@@ -643,12 +664,13 @@ impl Core {
             // Learned here rather than on a path of its own: this is where a
             // reader is open anyway, and recording the real geometry is what
             // lets the *next* read plan its blocks on the revision's own
-            // boundaries instead of the 4 MiB fallback (bugs.md B85).
+            // boundaries instead of the 4 MiB fallback (`docs/BUGS.md` B85).
             if cache_blocks {
                 self.cache
                     .store_block_geometry(uid, mtime, fsize, reader.block_sizes());
             }
             let bytes = reader.read_at(span.start, span.len).await.map_err(|e| {
+                self.lost_link(&e, "read a block");
                 warn!(%uid, start = span.start, len = span.len, error = %e, "block read failed");
                 Errno::EIO
             })?;
@@ -684,7 +706,7 @@ impl Core {
     /// The range path derives every block boundary from the revision's recorded
     /// block sizes. When those understate the file, reads come back short with
     /// no error anywhere, and the caller hands userspace a truncated file that
-    /// looks like a clean EOF (bugs.md B84). `download_file_to` does not do that
+    /// looks like a clean EOF (`docs/BUGS.md` B84). `download_file_to` does not do that
     /// arithmetic — it streams the block list — so it returns the whole file
     /// where the range path cannot, which is why pinning was the workaround.
     /// This does the same thing on the user's behalf, once, and the resulting
@@ -761,7 +783,7 @@ impl Core {
             error!(%uid, %error, "creating a scratch file for a repair download failed");
             Errno::EIO
         })?;
-        let outcome = self.client.download_file_to(uid, &mut file).await;
+        let outcome = self.drive.download_file_to(uid, &mut file).await;
         let stored = (|| {
             outcome.map_err(|error| {
                 warn!(%uid, %error, "repair download failed");
@@ -896,7 +918,7 @@ impl Core {
         let end = offset.saturating_add(len).min(fsize);
         let mut out = Vec::with_capacity((end - offset) as usize);
         // On the revision's own boundaries where they are known, and on the
-        // 4 MiB fallback where they are not (bugs.md B85).
+        // 4 MiB fallback where they are not (`docs/BUGS.md` B85).
         let wanted = self.block_geometry(uid, mtime, fsize).spans(offset, end);
         let (Some(&head), Some(&tail)) = (wanted.first(), wanted.last()) else {
             return Ok(out);
@@ -939,28 +961,67 @@ impl Core {
         }
 
         if !misses.is_empty() {
-            let fetched = self.rt.block_on(async {
-                // Resolve the file's keys and block table once, then read every
-                // missing block through the shared reader. Previously each block
-                // called `download_range`, which redid that resolution per block.
-                let reader = self.revision_reader(uid, mtime, fsize).await?;
+            // One deadline for the whole read, including any wait for a lost
+            // link. A fetch that failed because the link went down is tried
+            // once more when it is back, so a blip is a slow read, not an `EIO`.
+            let deadline = Instant::now() + READ_FETCH_TIMEOUT;
+            let mut retried = false;
+            let fetched = loop {
+                if !self.is_online() {
+                    let wait =
+                        READ_OFFLINE_WAIT.min(deadline.saturating_duration_since(Instant::now()));
+                    if !self.wait_online(wait) {
+                        debug!(%uid, offset, len, "read of uncached blocks while offline; failing it");
+                        return Err(Errno::EIO);
+                    }
+                }
+                let fetch = async {
+                    // Resolve the file's keys and block table once, then read every
+                    // missing block through the shared reader. Previously each block
+                    // called `download_range`, which redid that resolution per block.
+                    let reader = self.revision_reader(uid, mtime, fsize).await?;
 
-                let mut set = tokio::task::JoinSet::new();
-                for &span in &misses {
-                    let (core, reader, uid) = (self.clone(), reader.clone(), uid.clone());
-                    set.spawn(async move {
-                        core.fetch_block(&reader, &uid, mtime, fsize, span, cache_blocks)
-                            .await
-                            .map(|bytes| (span.idx, bytes))
-                    });
+                    let mut set = tokio::task::JoinSet::new();
+                    for &span in &misses {
+                        let (core, reader, uid) = (self.clone(), reader.clone(), uid.clone());
+                        set.spawn(async move {
+                            core.fetch_block(&reader, &uid, mtime, fsize, span, cache_blocks)
+                                .await
+                                .map(|bytes| (span.idx, bytes))
+                        });
+                    }
+                    let mut out = Vec::with_capacity(misses.len());
+                    while let Some(joined) = set.join_next().await {
+                        // A join error means the task panicked; surface it as EIO.
+                        out.push(joined.map_err(|_| Errno::EIO)??);
+                    }
+                    Ok::<_, Errno>(out)
+                };
+                // Dropping the fetch on the deadline aborts the block tasks with
+                // it; blocks that already landed stay cached for the retry. The
+                // timer is built inside the runtime: a FUSE worker has no reactor
+                // of its own, and building it out here panicked every remote read.
+                let budget = deadline.saturating_duration_since(Instant::now());
+                match self
+                    .rt
+                    .block_on(async { tokio::time::timeout(budget, fetch).await })
+                {
+                    Ok(Ok(fetched)) => break fetched,
+                    // The fetch saw the link go down; the top of the loop waits
+                    // for it within what is left of the deadline.
+                    Ok(Err(_)) if !retried && !self.is_online() => retried = true,
+                    Ok(Err(e)) => return Err(e),
+                    Err(_) => {
+                        warn!(
+                            %uid, offset, len, secs = READ_FETCH_TIMEOUT.as_secs(),
+                            "read got no answer from the network in time; failing it"
+                        );
+                        self.evict_reader(uid);
+                        self.mark_offline("read", &crate::link::no_answer(READ_FETCH_TIMEOUT));
+                        return Err(Errno::EIO);
+                    }
                 }
-                let mut out = Vec::with_capacity(misses.len());
-                while let Some(joined) = set.join_next().await {
-                    // A join error means the task panicked; surface it as EIO.
-                    out.push(joined.map_err(|_| Errno::EIO)??);
-                }
-                Ok::<_, Errno>(out)
-            })?;
+            };
             for (bidx, bytes) in fetched {
                 blocks[(bidx - first) as usize] = Some(bytes);
             }
@@ -977,7 +1038,7 @@ impl Core {
             // — the cache validates it on read, the network path in `read_block`
             // — so this refuses rather than quietly contributing fewer bytes
             // than the range covers, which is what turned a bad block into a
-            // truncated file (bugs.md B84).
+            // truncated file (`docs/BUGS.md` B84).
             if block.len() as u64 != span.len {
                 error!(
                     %uid, bidx = span.idx, len = block.len(), expected = span.len, fsize,
@@ -1009,37 +1070,121 @@ impl Core {
         offset: u64,
         size: u64,
     ) -> Result<Vec<u8>, Errno> {
-        if offset >= len || size == 0 {
-            return Ok(Vec::new());
+        let read = assemble_open_read(file, len, uid, base_size, written, offset, size, |s, n| {
+            self.read_range(uid, base_mtime, base_size, s, n, true)
+                .map(Some)
+        })?;
+        Ok(read.unwrap_or_default())
+    }
+
+    /// [`Core::serve_open_read`] from bytes already on this machine only:
+    /// `Ok(None)` the moment a range of the base would have to be downloaded.
+    ///
+    /// The dispatch loop answers a read on a write handle with this, and hands
+    /// it to a worker only when it says `None`. Fetching the base there held
+    /// every request of the mount for as long as the download took, up to the
+    /// read deadline (`docs/BUGS.md` B227).
+    #[allow(clippy::too_many_arguments)]
+    pub(super) fn serve_open_read_local(
+        &self,
+        file: &Arc<File>,
+        len: u64,
+        uid: &NodeUid,
+        base_mtime: i64,
+        base_size: u64,
+        written: &Intervals,
+        offset: u64,
+        size: u64,
+    ) -> Result<Option<Vec<u8>>, Errno> {
+        assemble_open_read(file, len, uid, base_size, written, offset, size, |s, n| {
+            Ok(self.cached_range(uid, base_mtime, base_size, s, n))
+        })
+    }
+
+    /// Bytes `[offset, offset + size)` of file `ino` as `read` would serve
+    /// them, but only from this machine: an open write handle, a queued write,
+    /// or the caches. `Ok(None)` the moment any of it would have to be
+    /// downloaded, and for a size still provisional (B100), which only the
+    /// network settles. Short or empty past the end of the file.
+    ///
+    /// `copy_file_range` answers with this and leaves anything else to the
+    /// kernel's read/write copy.
+    pub(super) fn read_local(
+        &self,
+        ino: u64,
+        offset: u64,
+        size: u64,
+    ) -> Result<Option<Vec<u8>>, Errno> {
+        let handle = self.state().active_writes.get(&ino).map(|h| {
+            (
+                h.file.clone(),
+                h.len,
+                h.uid.clone(),
+                h.base_mtime,
+                h.base_size,
+                h.written.clone(),
+            )
+        });
+        if let Some((file, len, uid, base_mtime, base_size, written)) = handle {
+            return self.serve_open_read_local(
+                &file, len, &uid, base_mtime, base_size, &written, offset, size,
+            );
         }
-        let end = offset.saturating_add(size).min(len);
-        let mut out = Vec::with_capacity((end - offset) as usize);
-        for (s, e, authored) in written.segments(offset, end) {
-            if authored {
-                let mut buf = vec![0u8; (e - s) as usize];
-                file.read_exact_at(&mut buf, s).map_err(|err| {
-                    warn!(%uid, error = %err, "scratch read failed");
-                    Errno::EIO
-                })?;
-                out.extend_from_slice(&buf);
-            } else {
-                let bend = e.min(base_size);
-                if s < bend {
-                    out.extend_from_slice(&self.read_range(
-                        uid,
-                        base_mtime,
-                        base_size,
-                        s,
-                        bend - s,
-                        true,
-                    )?);
-                }
-                // Anything past the base is a hole: zero-fill.
-                let zeros = e.saturating_sub(s.max(base_size));
-                out.resize(out.len() + zeros as usize, 0);
+        let (uid, mtime, fsize, provisional) = {
+            let st = self.state();
+            match st.entries.get(&ino) {
+                Some(e) if e.node.is_file() => (
+                    e.uid.clone(),
+                    e.node.modification_time,
+                    node_size(&e.node),
+                    matches!(
+                        &e.node.kind,
+                        NodeKind::File {
+                            claimed_size: None,
+                            ..
+                        }
+                    ),
+                ),
+                Some(_) => return Err(Errno::EISDIR),
+                None => return Err(Errno::ENOENT),
             }
+        };
+        let uid = match is_local_uid(&uid) {
+            true => self.landed_stand_in_uid(&uid).unwrap_or(uid),
+            false => uid,
+        };
+        if let Some(pending) = self.pending.lock().get(&uid).cloned() {
+            let m = &pending.meta;
+            let file = File::open(&pending.path).map_err(|e| {
+                error!(%uid, path = %pending.path.display(), error = %e, "open staged blob failed");
+                Errno::EIO
+            })?;
+            let mut written = Intervals::default();
+            for &(s, e) in &m.authored {
+                written.add(s, e);
+            }
+            return self.serve_open_read_local(
+                &Arc::new(file),
+                m.len,
+                &uid,
+                m.base_mtime,
+                m.base_size,
+                &written,
+                offset,
+                size,
+            );
         }
-        Ok(out)
+        // Created here and never written: an empty file.
+        if is_local_uid(&uid) {
+            return Ok(Some(Vec::new()));
+        }
+        if provisional {
+            return Ok(None);
+        }
+        if offset >= fsize || size == 0 {
+            return Ok(Some(Vec::new()));
+        }
+        Ok(self.cached_range(&uid, mtime, fsize, offset, size.min(fsize - offset)))
     }
 
     /// The largest amount [`Core::fill_gaps_cached`] will copy out of the local
@@ -1181,6 +1326,49 @@ impl Core {
     }
 }
 
+/// The bytes `[offset, offset + size)` of a file open for writing: authored
+/// ranges from its scratch `file`, the rest from its base through `base`,
+/// and zeros past the base. `Ok(None)` when `base` says it cannot answer.
+#[allow(clippy::too_many_arguments)]
+fn assemble_open_read(
+    file: &File,
+    len: u64,
+    uid: &NodeUid,
+    base_size: u64,
+    written: &Intervals,
+    offset: u64,
+    size: u64,
+    mut base: impl FnMut(u64, u64) -> Result<Option<Vec<u8>>, Errno>,
+) -> Result<Option<Vec<u8>>, Errno> {
+    if offset >= len || size == 0 {
+        return Ok(Some(Vec::new()));
+    }
+    let end = offset.saturating_add(size).min(len);
+    let mut out = Vec::with_capacity((end - offset) as usize);
+    for (s, e, authored) in written.segments(offset, end) {
+        if authored {
+            let mut buf = vec![0u8; (e - s) as usize];
+            file.read_exact_at(&mut buf, s).map_err(|err| {
+                warn!(%uid, error = %err, "scratch read failed");
+                Errno::EIO
+            })?;
+            out.extend_from_slice(&buf);
+        } else {
+            let bend = e.min(base_size);
+            if s < bend {
+                match base(s, bend - s)? {
+                    Some(bytes) => out.extend_from_slice(&bytes),
+                    None => return Ok(None),
+                }
+            }
+            // Anything past the base is a hole: zero-fill.
+            let zeros = e.saturating_sub(s.max(base_size));
+            out.resize(out.len() + zeros as usize, 0);
+        }
+    }
+    Ok(Some(out))
+}
+
 #[cfg(test)]
 mod block_ring_tests {
     use super::*;
@@ -1333,3 +1521,54 @@ mod prefetch_tests {
         assert!(prefetch.streams.lock().len() <= MAX_PREFETCH_STREAMS + 1);
     }
 }
+
+#[cfg(test)]
+mod open_read_tests {
+    use super::*;
+    use proton_drive_rs::proton_sdk::ids::{LinkId, VolumeId};
+
+    fn scratch(bytes: &[u8]) -> (File, PathBuf) {
+        let path = std::env::temp_dir().join(format!(
+            "pdfs-open-read-{}-{}",
+            std::process::id(),
+            std::time::SystemTime::now()
+                .duration_since(std::time::UNIX_EPOCH)
+                .unwrap()
+                .as_nanos()
+        ));
+        let mut file = File::options()
+            .create_new(true)
+            .read(true)
+            .write(true)
+            .open(&path)
+            .unwrap();
+        file.write_all(bytes).unwrap();
+        (file, path)
+    }
+
+    /// The dispatch loop answers from what is on this machine and says so
+    /// when the base would have to be downloaded, for a worker to do it
+    /// (`docs/BUGS.md` B227).
+    #[test]
+    fn a_read_needing_the_base_says_so_instead_of_fetching() {
+        let uid = NodeUid::new(VolumeId::from("v"), LinkId::from("f"));
+        let (file, path) = scratch(b"ABCDEFGH");
+        let mut written = Intervals::default();
+        written.add(0, 4);
+
+        let authored = assemble_open_read(&file, 8, &uid, 8, &written, 0, 4, |_, _| {
+            panic!("an authored range never asks the base")
+        });
+        assert_eq!(authored.unwrap().as_deref(), Some(&b"ABCD"[..]));
+
+        let missing = assemble_open_read(&file, 8, &uid, 8, &written, 2, 6, |_, _| Ok(None));
+        assert_eq!(missing.unwrap(), None);
+
+        let base = assemble_open_read(&file, 10, &uid, 8, &written, 2, 8, |s, n| {
+            assert_eq!((s, n), (4, 4));
+            Ok(Some(b"efgh".to_vec()))
+        });
+        assert_eq!(base.unwrap().as_deref(), Some(&b"CDefgh\0\0"[..]));
+        let _ = std::fs::remove_file(path);
+    }
+}

crates/pdfs-fuse/src/redate.rs

@@ -8,7 +8,7 @@
 //!
 //! Proton has no API for editing a sealed capture time, so the repair is a
 //! rewrite: download the photo, upload it again with the capture time its file
-//! name implies, put the new copy back into the albums and favourites the old
+//! name implies, put the new copy back into the albums and favorites the old
 //! one was in, then **trash** (not delete) the original. Trash rather than
 //! delete so a bad run is recoverable from Proton's own trash for 30 days.
 //!
@@ -190,6 +190,7 @@ impl Core {
             .map_err(|e| CoreError::internal(format!("stat {}: {e}", path.display())))?
             .len();
         let media_type = media_type_of(&candidate.name).unwrap_or("application/octet-stream");
+        let thumbnails = self.upload_thumbnails(&path, &candidate.name);
 
         let guard =
             self.transfers
@@ -206,14 +207,14 @@ impl Core {
             media_type,
             reader,
             size as i64,
-            Vec::new(),
+            thumbnails,
             metadata,
             false,
         ));
         drop(guard);
         let new_uid = uploaded.map_err(|e| CoreError::from_api(&e, "re-upload photo"))?;
 
-        // Favourite and albums before the trash: a failure here leaves both
+        // Favorite and albums before the trash: a failure here leaves both
         // copies on the account, which is recoverable by hand. Losing the
         // original first would not be.
         if candidate.favorite {
@@ -227,7 +228,7 @@ impl Core {
                 // just set does not come back as a remote change.
                 Ok(_) => self.note_self_change(&new_uid),
                 Err(e) => {
-                    warn!(name = candidate.name, error = %e, "cannot re-favourite re-dated photo")
+                    warn!(name = candidate.name, error = %e, "cannot re-favorite re-dated photo")
                 }
             }
         }
@@ -248,10 +249,7 @@ impl Core {
 
         let outcomes = self
             .rt
-            .block_on(
-                self.client
-                    .trash_nodes(std::slice::from_ref(&candidate.uid)),
-            )
+            .block_on(self.drive.trash_nodes(std::slice::from_ref(&candidate.uid)))
             .map_err(|e| CoreError::from_api(&e, "trash the mis-dated original"))?;
         if let Some((_, Err(e))) = outcomes.into_iter().next() {
             return Err(CoreError::from_api(&e, "trash the mis-dated original"));
@@ -294,7 +292,14 @@ impl Core {
             let mut state = REDATE.lock();
             state.running = false;
             state.cancel = None;
-            state.last = outcome.as_ref().ok().cloned();
+            state.last = Some(match &outcome {
+                Ok(summary) => summary.clone(),
+                Err(e) => RedateSummary {
+                    error: Some(e.to_string()),
+                    dry_run,
+                    ..Default::default()
+                },
+            });
             match &outcome {
                 Ok(summary) => info!(
                     candidates = summary.candidates,

crates/pdfs-fuse/src/relocate.rs

@@ -0,0 +1,638 @@
+//! Moving a file or folder between two locations without moving its content.
+//!
+//! `~/ProtonDrive` and every on-demand synced folder are separate FUSE
+//! sessions, and a mirror folder is a plain local directory. The kernel refuses
+//! `rename(2)` between two mounts with `EXDEV` before the daemon hears of it, so
+//! `mv` falls back to copy and delete: every file is downloaded, uploaded again,
+//! and the original is trashed.
+//!
+//! On Proton Drive the same move is one metadata change, because My files and
+//! this computer's device folders all live on the main volume. So a move
+//! request that names two absolute paths is resolved here, against every
+//! location this daemon knows, and done as a server-side move. The local side
+//! is then brought level: a mount forgets the node and re-lists, and a mirror
+//! folder either renames its copy along or drops it once Drive has it.
+
+use std::collections::HashMap;
+use std::path::{Component, Path, PathBuf};
+use std::sync::Arc;
+use std::time::Duration;
+
+use parking_lot::Mutex;
+use pdfs_core::db::{StoredSyncEntry, StoredSyncFolder};
+use pdfs_core::{CoreError, CoreResult};
+use proton_drive_rs::NodeKind;
+use proton_drive_rs::proton_sdk::ids::NodeUid;
+use tracing::{info, warn};
+
+use super::state::State;
+use super::sync::mirror_subtree_unsynced;
+use super::{Core, is_local_uid, parse_uid};
+
+/// How long a move waits for a mirror folder's sync pass to finish before it
+/// reports the folder busy. A pass can run for hours on a first sync, so the
+/// move does not wait it out.
+const MIRROR_LOCK_WAIT: Duration = Duration::from_secs(5);
+
+/// Where one side of a move lives.
+enum Place {
+    /// My files or an on-demand folder: a live FUSE mount. `core` is rooted at
+    /// that mount and `rel` is relative to its mountpoint.
+    Mounted { core: Box<Core>, rel: PathBuf },
+    /// A mirror folder: a local directory kept in step by the sync engine.
+    /// `rel` is `/`-separated, relative to the folder, as the baseline keys it.
+    Mirror {
+        folder: StoredSyncFolder,
+        rel: String,
+    },
+}
+
+impl Place {
+    fn mirror_id(&self) -> Option<i64> {
+        match self {
+            Place::Mirror { folder, .. } => Some(folder.id),
+            Place::Mounted { .. } => None,
+        }
+    }
+
+    fn is_root(&self) -> bool {
+        match self {
+            Place::Mounted { rel, .. } => rel.as_os_str().is_empty(),
+            Place::Mirror { rel, .. } => rel.is_empty(),
+        }
+    }
+}
+
+/// The mirror folder that holds `abs`, and `abs` relative to it. The deepest
+/// folder wins, although folders are not expected to nest.
+fn mirror_covering(
+    folders: Vec<StoredSyncFolder>,
+    abs: &Path,
+) -> Option<(StoredSyncFolder, String)> {
+    let folder = folders
+        .into_iter()
+        .filter(|f| f.mode == "mirror" && abs.starts_with(&f.local_path))
+        .max_by_key(|f| Path::new(&f.local_path).components().count())?;
+    let rel = abs.strip_prefix(&folder.local_path).ok()?;
+    let parts = rel
+        .components()
+        .map(|c| match c {
+            Component::Normal(part) => part.to_str(),
+            _ => None,
+        })
+        .collect::<Option<Vec<_>>>()?;
+    Some((folder, parts.join("/")))
+}
+
+/// `name` inside the folder at `parent`, in baseline form.
+fn join_rel(parent: &str, name: &str) -> String {
+    if parent.is_empty() {
+        name.to_string()
+    } else {
+        format!("{parent}/{name}")
+    }
+}
+
+/// Rename `from` to `to`, failing with `EEXIST` rather than replacing
+/// whatever appeared at `to` since it was checked. A filesystem that cannot
+/// promise that gets a plain rename after one more look.
+fn rename_noreplace(from: &Path, to: &Path) -> std::io::Result<()> {
+    use std::ffi::CString;
+    use std::os::unix::ffi::OsStrExt;
+
+    let cstr = |p: &Path| {
+        CString::new(p.as_os_str().as_bytes())
+            .map_err(|_| std::io::Error::from_raw_os_error(libc::EINVAL))
+    };
+    let (c_from, c_to) = (cstr(from)?, cstr(to)?);
+    // SAFETY: both are valid NUL-terminated paths that outlive the call.
+    let rc = unsafe {
+        libc::syscall(
+            libc::SYS_renameat2,
+            libc::AT_FDCWD,
+            c_from.as_ptr(),
+            libc::AT_FDCWD,
+            c_to.as_ptr(),
+            libc::RENAME_NOREPLACE,
+        )
+    };
+    if rc == 0 {
+        return Ok(());
+    }
+    let err = std::io::Error::last_os_error();
+    match err.raw_os_error() {
+        Some(libc::EINVAL | libc::ENOSYS) => {
+            if std::fs::symlink_metadata(to).is_ok() {
+                return Err(std::io::Error::from_raw_os_error(libc::EEXIST));
+            }
+            std::fs::rename(from, to)
+        }
+        _ => Err(err),
+    }
+}
+
+/// Whether inode `ino` is `root` or lies under it in the tree. Bounded, so a
+/// parent cycle in corrupt data cannot hang the caller.
+pub(crate) fn ino_is_under(st: &State, mut ino: u64, root: u64) -> bool {
+    for _ in 0..MAX_TREE_DEPTH {
+        if ino == root {
+            return true;
+        }
+        match st.entries.get(&ino) {
+            Some(entry) if entry.parent != ino => ino = entry.parent,
+            _ => return false,
+        }
+    }
+    false
+}
+
+/// How deep [`ino_is_under`] walks before it gives up.
+const MAX_TREE_DEPTH: usize = 1024;
+
+/// Whether `uid` is `root` or lies under it, as far as the nodes table knows.
+fn under(core: &Core, root: &NodeUid, uid: &str) -> bool {
+    let root = root.to_string();
+    uid == root || matches!(core.db.path_relative_to(&root, uid), Ok(Some(_)))
+}
+
+impl Core {
+    /// Move the file or folder at `src` into the folder at `dest_parent`. Both
+    /// are absolute local paths and may be in different locations: My files,
+    /// an on-demand folder, or a mirror folder. Returns the moved name.
+    ///
+    /// Within one mount this is [`Core::move_to`]. Across locations it
+    /// is a server-side move followed by local bookkeeping; see the module
+    /// documentation. Anything the move could lose is refused up front.
+    pub(crate) fn move_between(&self, src: &Path, dest_parent: &Path) -> CoreResult<String> {
+        let source = self.locate(src)?;
+        let dest = self.locate(dest_parent)?;
+        if let (
+            Place::Mounted { core, rel },
+            Place::Mounted {
+                core: dest_core,
+                rel: dest_rel,
+            },
+        ) = (&source, &dest)
+            && Arc::ptr_eq(&core.root, &dest_core.root)
+        {
+            return core.move_to(rel, dest_rel);
+        }
+        if source.is_root() {
+            return Err(CoreError::invalid(format!(
+                "{} is the root of a synced location and cannot be moved",
+                src.display()
+            )));
+        }
+        if dest_parent.starts_with(src) {
+            return Err(CoreError::invalid("cannot move a folder into itself"));
+        }
+        let name = src
+            .file_name()
+            .and_then(|n| n.to_str())
+            .ok_or_else(|| CoreError::invalid(format!("{} has no usable name", src.display())))?
+            .to_string();
+
+        // Hold every mirror folder involved, lowest id first, so no sync pass
+        // sees the move half done and two moves cannot deadlock.
+        let mut ids: Vec<i64> = [source.mirror_id(), dest.mirror_id()]
+            .into_iter()
+            .flatten()
+            .collect();
+        ids.sort_unstable();
+        ids.dedup();
+        let locks: Vec<Arc<Mutex<()>>> = ids.iter().map(|&id| self.sync_lock(id)).collect();
+        let mut guards = Vec::with_capacity(locks.len());
+        for lock in &locks {
+            guards.push(lock.try_lock_for(MIRROR_LOCK_WAIT).ok_or_else(|| {
+                CoreError::conflict("a synced folder is busy syncing; try again when it is done")
+            })?);
+        }
+
+        let mut baselines: HashMap<i64, HashMap<String, StoredSyncEntry>> = HashMap::new();
+        for &id in &ids {
+            let baseline = self
+                .db
+                .sync_entries(id)
+                .map_err(|e| CoreError::internal(format!("db: {e}")))?;
+            baselines.insert(id, baseline);
+        }
+
+        let src_uid = self.source_uid(&source, &baselines)?;
+        let dest_uid = self.dest_uid(&dest, &baselines, &name)?;
+        if src_uid.volume_id != dest_uid.volume_id {
+            return Err(CoreError::invalid(
+                "the two locations are on different Proton Drive volumes; copy it instead",
+            ));
+        }
+        if under(self, &src_uid, &dest_uid.to_string()) {
+            return Err(CoreError::invalid("cannot move a folder into itself"));
+        }
+
+        // Mirror to mirror renames the local copy along, so it needs nothing
+        // proven about it. Another filesystem makes that impossible, and the
+        // pair then drops the source copy like any other move out of a mirror.
+        let mut renamed = None;
+        if let (
+            Place::Mirror { folder, rel },
+            Place::Mirror {
+                folder: dest_folder,
+                rel: dest_rel,
+            },
+        ) = (&source, &dest)
+        {
+            let from = src.to_path_buf();
+            let to = dest_parent.join(&name);
+            match rename_noreplace(&from, &to) {
+                Ok(()) => renamed = Some((from, to)),
+                Err(e) if e.raw_os_error() == Some(libc::EXDEV) => {
+                    info!(from = %folder.local_path, to = %dest_folder.local_path,
+                        "move: mirror folders are on different filesystems");
+                }
+                Err(e) if e.raw_os_error() == Some(libc::EEXIST) => {
+                    return Err(CoreError::conflict(format!(
+                        "{name} already exists in {}",
+                        dest_parent.display()
+                    )));
+                }
+                Err(e) => {
+                    return Err(CoreError::internal(format!(
+                        "could not move {} to {}: {e}",
+                        rel,
+                        join_rel(dest_rel, &name)
+                    )));
+                }
+            }
+        }
+        if renamed.is_none()
+            && let Place::Mirror { folder, rel } = &source
+            && let Some(reason) =
+                mirror_subtree_unsynced(Path::new(&folder.local_path), rel, &baselines[&folder.id])
+        {
+            return Err(CoreError::conflict(format!(
+                "{name} is not fully synced yet ({reason}); wait for the sync to finish"
+            )));
+        }
+
+        if let Err(e) = self.move_remote(&src_uid, &dest_uid) {
+            if let Some((from, to)) = &renamed
+                && let Err(back) = std::fs::rename(to, from)
+            {
+                warn!(from = %to.display(), to = %from.display(), error = %back,
+                    "move: could not put the local copy back after the move failed");
+            }
+            return Err(CoreError::from_api(&e, "move"));
+        }
+
+        self.settle_source(
+            &source,
+            &src_uid,
+            renamed.is_some(),
+            &dest,
+            &dest_uid,
+            &name,
+            src,
+        );
+        self.settle_dest(&dest, &dest_uid, &name);
+        drop(guards);
+        if renamed.is_none()
+            && let Place::Mirror { folder, .. } = &dest
+        {
+            // Nothing local exists yet; the folder's next pass downloads it.
+            self.sync_now(Some(folder.id));
+        }
+        Ok(name)
+    }
+
+    /// Which location `abs` is in. A mount is looked for first: a mirror folder
+    /// cannot live inside a FUSE mount, and a mount is what the kernel sees.
+    fn locate(&self, abs: &Path) -> CoreResult<Place> {
+        if let Some((core, rel)) = self.rooted_at(abs) {
+            if !rel
+                .components()
+                .all(|c| matches!(c, Component::Normal(_) | Component::CurDir))
+            {
+                return Err(CoreError::invalid(format!(
+                    "{} escapes its location",
+                    abs.display()
+                )));
+            }
+            return Ok(Place::Mounted {
+                core: Box::new(core),
+                rel,
+            });
+        }
+        let folders = self
+            .db
+            .sync_folder_list()
+            .map_err(|e| CoreError::internal(format!("db: {e}")))?;
+        mirror_covering(folders, abs)
+            .map(|(folder, rel)| Place::Mirror { folder, rel })
+            .ok_or_else(|| {
+                CoreError::invalid(format!(
+                    "{} is not in a Proton Drive location",
+                    abs.display()
+                ))
+            })
+    }
+
+    /// The node being moved, once everything that must hold before it leaves
+    /// its location has been checked.
+    fn source_uid(
+        &self,
+        source: &Place,
+        baselines: &HashMap<i64, HashMap<String, StoredSyncEntry>>,
+    ) -> CoreResult<NodeUid> {
+        match source {
+            Place::Mounted { core, rel } => {
+                let (ino, uid) = core.resolve(rel)?;
+                if is_local_uid(&uid) {
+                    return Err(CoreError::conflict(
+                        "it is not uploaded yet; wait for the upload to finish",
+                    ));
+                }
+                let old_parent = core.source_parent_uid(ino, rel)?;
+                for authority in [&uid, &old_parent] {
+                    core.require_uid_writable(authority)
+                        .map_err(|e| core.errno_error(e, "move access"))?;
+                }
+                if self.has_pending_work_under(ino, &uid)? {
+                    return Err(CoreError::conflict(
+                        "it has changes that are not uploaded yet; wait for the upload to finish",
+                    ));
+                }
+                Ok(uid)
+            }
+            Place::Mirror { folder, rel } => baselines[&folder.id]
+                .get(rel)
+                .and_then(|entry| entry.remote_uid.as_deref())
+                .and_then(parse_uid)
+                .ok_or_else(|| {
+                    CoreError::conflict(
+                        "it is not on Proton Drive yet; wait for the sync to finish",
+                    )
+                }),
+        }
+    }
+
+    /// The folder the node moves into. Refuses one that already holds `name`
+    /// on this disk; Drive refuses a clash on its side by itself.
+    fn dest_uid(
+        &self,
+        dest: &Place,
+        baselines: &HashMap<i64, HashMap<String, StoredSyncEntry>>,
+        name: &str,
+    ) -> CoreResult<NodeUid> {
+        let uid = match dest {
+            Place::Mounted { core, rel } => {
+                let (ino, uid) = core.resolve(rel)?;
+                let is_folder = core
+                    .state()
+                    .entries
+                    .get(&ino)
+                    .is_some_and(|entry| matches!(entry.node.kind, NodeKind::Folder));
+                if !is_folder {
+                    return Err(CoreError::invalid("the destination is not a folder"));
+                }
+                core.require_uid_writable(&uid)
+                    .map_err(|e| core.errno_error(e, "move access"))?;
+                uid
+            }
+            Place::Mirror { folder, rel } => {
+                let local = Path::new(&folder.local_path).join(rel);
+                if !local.is_dir() {
+                    return Err(CoreError::invalid("the destination is not a folder"));
+                }
+                if std::fs::symlink_metadata(local.join(name)).is_ok() {
+                    return Err(CoreError::conflict(format!(
+                        "{name} already exists in the destination"
+                    )));
+                }
+                let uid = if rel.is_empty() {
+                    Some(folder.remote_uid.as_str())
+                } else {
+                    baselines[&folder.id]
+                        .get(rel)
+                        .and_then(|entry| entry.remote_uid.as_deref())
+                };
+                uid.and_then(parse_uid).ok_or_else(|| {
+                    CoreError::conflict(
+                        "the destination is not on Proton Drive yet; wait for the sync to finish",
+                    )
+                })?
+            }
+        };
+        if is_local_uid(&uid) {
+            return Err(CoreError::conflict(
+                "the destination is not uploaded yet; wait for the upload to finish",
+            ));
+        }
+        Ok(uid)
+    }
+
+    /// Whether the queue owes Drive anything for `uid` (inode `ino`) or a node
+    /// under it, or a file under it is open for writing. Moving such a node away
+    /// would strand the pending change against a path that no longer leads to
+    /// it.
+    fn has_pending_work_under(&self, ino: u64, uid: &NodeUid) -> CoreResult<bool> {
+        let ops = self
+            .db
+            .pending_ops()
+            .map_err(|e| CoreError::internal(format!("db: {e}")))?;
+        if ops.iter().any(|op| {
+            under(self, uid, &op.uid)
+                || op
+                    .parent_uid
+                    .as_deref()
+                    .is_some_and(|p| under(self, uid, p))
+        }) {
+            return Ok(true);
+        }
+        // Open handles are placed in the tree by inode, not by uid in the nodes
+        // table. A file opened just after its create landed holds the
+        // placeholder uid until the tree catches up, and the nodes table no
+        // longer has that uid, so the move went ahead under an open write.
+        let mut writing = false;
+        self.for_each_state(|st| {
+            writing = st
+                .active_writes
+                .keys()
+                .chain(&st.releasing)
+                .any(|&open| ino_is_under(st, open, ino));
+        });
+        Ok(writing)
+    }
+
+    /// Bring the source location level with a move Drive has made.
+    #[allow(clippy::too_many_arguments)]
+    fn settle_source(
+        &self,
+        source: &Place,
+        uid: &NodeUid,
+        renamed: bool,
+        dest: &Place,
+        dest_uid: &NodeUid,
+        name: &str,
+        src: &Path,
+    ) {
+        match source {
+            Place::Mounted { core, rel } => {
+                // Moved everywhere, so no mount keeps serving it under the old
+                // path and its row stays (`docs/BUGS.md` B117). The content
+                // cache is keyed by uid and stays valid.
+                self.relocate_everywhere(uid, dest_uid, name);
+                core.invalidate_parent_listing(rel);
+            }
+            Place::Mirror { folder, rel } if renamed => {
+                let Place::Mirror {
+                    folder: dest_folder,
+                    rel: dest_rel,
+                } = dest
+                else {
+                    unreachable!("a local rename needs a mirror destination");
+                };
+                if let Err(e) = self.db.sync_entries_move(
+                    folder.id,
+                    rel,
+                    dest_folder.id,
+                    &join_rel(dest_rel, name),
+                ) {
+                    // The next passes see a copy with no baseline on one side
+                    // and a baseline with no copy on the other. That uploads
+                    // the files again and trashes the moved node, but loses
+                    // nothing.
+                    warn!(error = ?e, "move: could not carry the sync baseline along");
+                }
+            }
+            Place::Mirror { folder, rel } => {
+                // Drive had all of it when the move was checked, so the local
+                // copy is only a stale duplicate now. Something written into it
+                // since would go with it, so look again first; if anything is
+                // new, the copy stays and the next pass uploads it as new.
+                let changed = match self.db.sync_entries(folder.id) {
+                    Ok(baseline) => {
+                        mirror_subtree_unsynced(Path::new(&folder.local_path), rel, &baseline)
+                    }
+                    Err(e) => Some(format!("db: {e}")),
+                };
+                if let Err(e) = self.db.sync_entries_remove_subtree(folder.id, rel) {
+                    warn!(error = ?e, "move: could not drop the sync baseline");
+                }
+                if let Some(reason) = changed {
+                    warn!(path = %src.display(), reason,
+                        "move: the local copy changed during the move; keeping it");
+                    return;
+                }
+                let removed = if src.is_dir() {
+                    std::fs::remove_dir_all(src)
+                } else {
+                    std::fs::remove_file(src)
+                };
+                if let Err(e) = removed {
+                    warn!(path = %src.display(), error = %e,
+                        "move: could not remove the local copy after the move");
+                }
+            }
+        }
+    }
+
+    /// Bring the destination location level with a move Drive has made.
+    fn settle_dest(&self, dest: &Place, uid: &NodeUid, name: &str) {
+        if let Place::Mounted { .. } = dest {
+            self.invalidate_children_of(uid);
+            // A lookup that missed before the move may be cached as absent.
+            self.for_each_mount(|st, notify| {
+                if let Some(&ino) = st.by_uid.get(uid) {
+                    notify.inval_entry(ino, name.to_string());
+                }
+            });
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn folder(id: i64, local_path: &str, mode: &str) -> StoredSyncFolder {
+        StoredSyncFolder {
+            id,
+            local_path: local_path.to_string(),
+            remote_uid: format!("vol~root{id}"),
+            remote_share_id: String::new(),
+            mode: mode.to_string(),
+            pending_mode: None,
+            state: "idle".to_string(),
+            last_sync: 0,
+        }
+    }
+
+    #[test]
+    fn mirror_covering_finds_the_mirror_folder_and_the_path_inside_it() {
+        let folders = vec![folder(1, "/home/u/Docs", "mirror")];
+        let (found, rel) = mirror_covering(folders, Path::new("/home/u/Docs/a/b.txt")).unwrap();
+        assert_eq!(found.id, 1);
+        assert_eq!(rel, "a/b.txt");
+    }
+
+    #[test]
+    fn mirror_covering_gives_an_empty_path_for_the_folder_itself() {
+        let folders = vec![folder(1, "/home/u/Docs", "mirror")];
+        let (_, rel) = mirror_covering(folders, Path::new("/home/u/Docs")).unwrap();
+        assert_eq!(rel, "");
+    }
+
+    #[test]
+    fn mirror_covering_prefers_the_deepest_folder() {
+        let folders = vec![
+            folder(1, "/home/u/Docs", "mirror"),
+            folder(2, "/home/u/Docs/Work", "mirror"),
+        ];
+        let (found, rel) = mirror_covering(folders, Path::new("/home/u/Docs/Work/x")).unwrap();
+        assert_eq!(found.id, 2);
+        assert_eq!(rel, "x");
+    }
+
+    #[test]
+    fn mirror_covering_skips_on_demand_folders_and_name_prefixed_siblings() {
+        let folders = || {
+            vec![
+                folder(1, "/home/u/Docs", "mirror"),
+                folder(2, "/home/u/Media", "ondemand"),
+            ]
+        };
+        assert!(mirror_covering(folders(), Path::new("/home/u/Media/x")).is_none());
+        assert!(mirror_covering(folders(), Path::new("/home/u/Docs2/x")).is_none());
+    }
+
+    #[test]
+    fn mirror_covering_refuses_a_path_that_climbs_out() {
+        let folders = vec![folder(1, "/home/u/Docs", "mirror")];
+        assert!(mirror_covering(folders, Path::new("/home/u/Docs/../etc")).is_none());
+    }
+
+    #[test]
+    fn join_rel_names_a_child_of_the_folder_root_without_a_slash() {
+        assert_eq!(join_rel("", "a"), "a");
+        assert_eq!(join_rel("x/y", "a"), "x/y/a");
+    }
+
+    #[test]
+    fn a_local_rename_never_replaces_what_is_already_there() {
+        let dir = std::env::temp_dir().join(format!("pdfs-relocate-{}", std::process::id()));
+        let _ = std::fs::remove_dir_all(&dir);
+        std::fs::create_dir_all(&dir).unwrap();
+        let (a, b, c) = (dir.join("a"), dir.join("b"), dir.join("c"));
+        std::fs::write(&a, b"moved").unwrap();
+        std::fs::write(&b, b"kept").unwrap();
+
+        let err = rename_noreplace(&a, &b).unwrap_err();
+        assert_eq!(err.raw_os_error(), Some(libc::EEXIST));
+        assert_eq!(std::fs::read(&b).unwrap(), b"kept");
+        assert!(a.exists());
+
+        rename_noreplace(&a, &c).unwrap();
+        assert!(!a.exists());
+        assert_eq!(std::fs::read(&c).unwrap(), b"moved");
+        std::fs::remove_dir_all(&dir).unwrap();
+    }
+}

crates/pdfs-fuse/src/revisions.rs

@@ -232,7 +232,7 @@ impl Core {
         let mut out = std::io::BufWriter::new(file);
         let written = match self
             .rt
-            .block_on(self.client.download_revision_to(uid, revision_id, &mut out))
+            .block_on(self.drive.download_revision_to(uid, revision_id, &mut out))
             .map_err(|e| CoreError::from_api(&e, "download revision"))
             .and_then(|()| {
                 use std::io::Write as _;

crates/pdfs-fuse/src/sharing.rs

@@ -9,7 +9,8 @@
 use std::path::Path;
 
 use pdfs_core::control::{
-    BookmarkInfo, DirEntry, InvitationInfo, PublicLinkInfo, ShareEntry, ShareEntryKind, SharedItem,
+    BookmarkInfo, DirEntry, InvitationInfo, LinkChange, PublicLinkInfo, ShareEntry, ShareEntryKind,
+    SharedItem,
 };
 use pdfs_core::{CoreError, CoreResult};
 use tracing::debug;
@@ -21,7 +22,7 @@ use proton_drive_rs::{MemberRole, NodeKind};
 
 use proton_drive_rs::Node;
 
-use super::{Core, ROOT_INO, node_size, parse_uid, public_link_info, role_from_str, role_to_str};
+use super::{Core, node_size, parse_uid, public_link_info, role_from_str, role_to_str};
 
 /// A node from a *shared* listing as a [`DirEntry`].
 ///
@@ -66,6 +67,7 @@ fn shared_entry_of(core: &Core, n: Node) -> DirEntry {
         size,
         modified: n.modification_time,
         pinned: false,
+        pinned_by_folder: false,
         cached: false,
         path: core.rel_path_for_uid(&n.uid).unwrap_or_default(),
         uid: n.uid.to_string(),
@@ -73,6 +75,9 @@ fn shared_entry_of(core: &Core, n: Node) -> DirEntry {
         shared_by,
         shared_at,
         shared_by_unverified,
+        trashed_at: 0,
+        trashed_from: None,
+        issue: None,
     }
 }
 
@@ -363,6 +368,89 @@ impl Core {
         Ok(public_link_info(link))
     }
 
+    /// Change the public link `id` on the node at `rel`, keeping its URL.
+    pub(crate) fn update_public_link(
+        &self,
+        rel: &Path,
+        id: &str,
+        role: Option<&str>,
+        password: LinkChange<String>,
+        expires: LinkChange<i64>,
+    ) -> CoreResult<PublicLinkInfo> {
+        let (_ino, uid) = self.resolve(rel)?;
+        self.update_public_link_for_uid(&uid, id, role, password, expires)
+    }
+
+    /// [`Core::update_public_link`] for a node addressed through any daemon
+    /// location.
+    pub(crate) fn update_public_link_by_uid(
+        &self,
+        uid: &str,
+        id: &str,
+        role: Option<&str>,
+        password: LinkChange<String>,
+        expires: LinkChange<i64>,
+    ) -> CoreResult<PublicLinkInfo> {
+        let uid = self.resolve_anywhere(uid)?;
+        self.update_public_link_for_uid(&uid, id, role, password, expires)
+    }
+
+    /// The SDK sets all three options on every update, so each `Keep` is
+    /// filled in from the link as it stands.
+    fn update_public_link_for_uid(
+        &self,
+        uid: &NodeUid,
+        id: &str,
+        role: Option<&str>,
+        password: LinkChange<String>,
+        expires: LinkChange<i64>,
+    ) -> CoreResult<PublicLinkInfo> {
+        let link = self
+            .rt
+            .block_on(self.client.get_public_link(uid))
+            .map_err(|e| CoreError::from_api(&e, "get public link"))?
+            .filter(|l| l.public_link_id == id)
+            .ok_or_else(|| CoreError::not_found("public link not found"))?;
+        let role = match role {
+            Some(role) => role_from_str(role)?,
+            None => link.role,
+        };
+        let unchanged = role == link.role
+            && matches!(password, LinkChange::Keep)
+            && matches!(expires, LinkChange::Keep);
+        if unchanged {
+            return Ok(public_link_info(link));
+        }
+        let password = match password {
+            LinkChange::Keep if link.has_custom_password => {
+                // Sending `None` here would quietly remove the password.
+                Some(link.custom_password.clone().ok_or_else(|| {
+                    CoreError::conflict(
+                        "the link's password could not be read; set a new one or remove it",
+                    )
+                })?)
+            }
+            LinkChange::Keep | LinkChange::Remove => None,
+            LinkChange::Set(password) if password.is_empty() => {
+                return Err(CoreError::invalid("the new password is empty"));
+            }
+            LinkChange::Set(password) => Some(password),
+        };
+        let expires = match expires {
+            LinkChange::Keep => link.expiration_time,
+            LinkChange::Remove => None,
+            LinkChange::Set(at) => Some(at),
+        };
+        let link = self
+            .rt
+            .block_on(
+                self.client
+                    .update_public_link(uid, role, password.as_deref(), expires),
+            )
+            .map_err(|e| CoreError::from_api(&e, "update public link"))?;
+        Ok(public_link_info(link))
+    }
+
     /// Remove the public link `id` from the node at `rel`.
     pub(crate) fn remove_public_link(&self, rel: &Path, id: &str) -> CoreResult<()> {
         let (_ino, uid) = self.resolve(rel)?;
@@ -388,6 +476,66 @@ impl Core {
             .map_err(|e| CoreError::from_api(&e, "remove public link"))
     }
 
+    /// Stop sharing the node at `rel`: remove its public link, then every
+    /// external invitation, Proton invitation and member. The SDK has no call
+    /// that deletes a share outright, so this composes the per-entry ones.
+    pub(crate) fn stop_sharing(&self, rel: &Path) -> CoreResult<StopSharingOutcome> {
+        let (_ino, uid) = self.resolve(rel)?;
+        self.stop_sharing_for_uid(&uid)
+    }
+
+    /// [`Core::stop_sharing`] for a node addressed through any daemon location.
+    pub(crate) fn stop_sharing_by_uid(&self, uid: &str) -> CoreResult<StopSharingOutcome> {
+        let uid = self.resolve_anywhere(uid)?;
+        self.stop_sharing_for_uid(&uid)
+    }
+
+    /// Each step runs even when an earlier one failed, so a single stuck
+    /// member does not leave the public link up. The link goes first: it is the
+    /// widest exposure. Only a failure to list what is there fails the whole
+    /// request; a failed removal is counted in the outcome.
+    fn stop_sharing_for_uid(&self, uid: &NodeUid) -> CoreResult<StopSharingOutcome> {
+        let mut outcome = StopSharingOutcome::default();
+        let link = self
+            .rt
+            .block_on(self.client.get_public_link(uid))
+            .map_err(|e| CoreError::from_api(&e, "get public link"))?;
+        if let Some(link) = link {
+            let result = self.rt.block_on(self.client.remove_public_link(&link));
+            outcome.record("the public link", result);
+        }
+        let external = self
+            .rt
+            .block_on(self.client.list_external_invitations(uid))
+            .map_err(|e| CoreError::from_api(&e, "list external invitations"))?;
+        for ext in external {
+            let result = self
+                .rt
+                .block_on(self.client.delete_external_invitation(&ext));
+            outcome.record(&ext.invitee_email, result);
+        }
+        let invitations = self
+            .rt
+            .block_on(self.client.list_share_invitations(uid))
+            .map_err(|e| CoreError::from_api(&e, "list invitations"))?;
+        for inv in invitations {
+            let result = self.rt.block_on(self.client.delete_invitation(&inv));
+            outcome.record(&inv.invitee_email, result);
+        }
+        let members = self
+            .rt
+            .block_on(self.client.list_share_members(uid))
+            .map_err(|e| CoreError::from_api(&e, "list members"))?;
+        for member in members {
+            let result = self.rt.block_on(self.client.remove_member(&member));
+            outcome.record(&member.email, result);
+        }
+        if !outcome.failed.is_empty() {
+            debug!(%uid, failed = ?outcome.failed, "stop sharing: some entries remain");
+        }
+        Ok(outcome)
+    }
+
     // ---- shared with me ---------------------------------------------------
 
     /// List nodes shared with me that I have accepted.
@@ -401,7 +549,7 @@ impl Core {
         }
         let nodes = self
             .rt
-            .block_on(self.client.enumerate_nodes(&uids))
+            .block_on(self.drive.enumerate_nodes(&uids))
             .map_err(|e| CoreError::from_api(&e, "enumerate nodes"))?;
         Ok(nodes
             .into_iter()
@@ -419,14 +567,14 @@ impl Core {
             parse_uid(uid).ok_or_else(|| CoreError::invalid(format!("invalid uid: {uid}")))?;
         let child_uids = self
             .rt
-            .block_on(self.client.enumerate_folder_children_node_uids(&uid))
+            .block_on(self.drive.enumerate_folder_children_node_uids(&uid))
             .map_err(|e| CoreError::from_api(&e, "enumerate shared children"))?;
         if child_uids.is_empty() {
             return Ok(Vec::new());
         }
         let nodes = self
             .rt
-            .block_on(self.client.enumerate_nodes(&child_uids))
+            .block_on(self.drive.enumerate_nodes(&child_uids))
             .map_err(|e| CoreError::from_api(&e, "enumerate nodes"))?;
         Ok(nodes
             .into_iter()
@@ -447,7 +595,7 @@ impl Core {
             parse_uid(uid).ok_or_else(|| CoreError::invalid(format!("invalid uid: {uid}")))?;
         let node = self
             .rt
-            .block_on(self.client.get_node(&uid))
+            .block_on(self.drive.get_node(&uid))
             .map_err(|e| CoreError::from_api(&e, "get shared node"))?
             .ok_or_else(|| CoreError::not_found("shared node not found"))?;
         if !node.is_file() {
@@ -548,10 +696,10 @@ impl Core {
     /// Proton products (not Drive-only). A remote round-trip; nothing here is
     /// cached.
     pub(crate) fn account_quota(&self) -> CoreResult<(i64, i64)> {
-        let q = self
-            .rt
-            .block_on(self.client.quota())
-            .map_err(|e| CoreError::from_api(&e, "account quota"))?;
+        let q = self.block_on_bounded(self.drive.quota()).map_err(|e| {
+            self.lost_link(&e, "account quota");
+            CoreError::from_api(&e, "account quota")
+        })?;
         *self.quota.lock() = Some((std::time::Instant::now(), q.max_space, q.used_space));
         Ok((q.max_space, q.used_space))
     }
@@ -570,11 +718,22 @@ impl Core {
         {
             return Some((max_space, used_space));
         }
-        match self.account_quota() {
+        let last = || self.quota.lock().map(|(_, max, used)| (max, used));
+        if !self.is_online()
+            || self
+                .quota_refreshing
+                .swap(true, std::sync::atomic::Ordering::AcqRel)
+        {
+            return last();
+        }
+        let refreshed = self.account_quota();
+        self.quota_refreshing
+            .store(false, std::sync::atomic::Ordering::Release);
+        match refreshed {
             Ok(q) => Some(q),
             Err(error) => {
                 debug!(%error, "account quota refresh failed; serving the last known figures");
-                self.quota.lock().map(|(_, max, used)| (max, used))
+                last()
             }
         }
     }
@@ -586,6 +745,12 @@ impl Core {
     /// the shared uids; the per-node detail is then gathered best-effort — a single
     /// node racing with an unshare drops from the list rather than failing the whole
     /// request.
+    ///
+    /// A share with no member, no invitation and no link is left out: it is
+    /// shared with no one. Stop Sharing leaves exactly that behind, since the SDK
+    /// cannot delete a share, and Drive keeps listing it until its backend gets
+    /// round to removing it, minutes later. A share whose details could not all
+    /// be read stays listed, so a failed lookup never hides a live share.
     pub(crate) fn list_shared_by_me(&self) -> CoreResult<Vec<SharedItem>> {
         let uids = self
             .rt
@@ -596,7 +761,7 @@ impl Core {
         }
         let nodes = self
             .rt
-            .block_on(self.client.enumerate_nodes(&uids))
+            .block_on(self.drive.enumerate_nodes(&uids))
             .map_err(|e| CoreError::from_api(&e, "enumerate nodes"))?;
         let mut items = Vec::with_capacity(nodes.len());
         for n in nodes {
@@ -604,24 +769,26 @@ impl Core {
             let members = self
                 .rt
                 .block_on(self.client.list_share_members(&uid))
-                .map(|m| m.len())
-                .unwrap_or(0);
+                .map(|m| m.len());
             let proton_invites = self
                 .rt
                 .block_on(self.client.list_share_invitations(&uid))
-                .map(|i| i.len())
-                .unwrap_or(0);
+                .map(|i| i.len());
             let external_invites = self
                 .rt
                 .block_on(self.client.list_external_invitations(&uid))
-                .map(|i| i.len())
-                .unwrap_or(0);
-            let link = self
-                .rt
-                .block_on(self.client.get_public_link(&uid))
-                .ok()
-                .flatten()
-                .map(public_link_info);
+                .map(|i| i.len());
+            let link = self.rt.block_on(self.client.get_public_link(&uid));
+            if let (Ok(0), Ok(0), Ok(0), Ok(None)) =
+                (&members, &proton_invites, &external_invites, &link)
+            {
+                debug!(%uid, "shared by me: skipping an emptied share");
+                continue;
+            }
+            let members = members.unwrap_or(0);
+            let proton_invites = proton_invites.unwrap_or(0);
+            let external_invites = external_invites.unwrap_or(0);
+            let link = link.ok().flatten().map(public_link_info);
             items.push(SharedItem {
                 uid: uid.to_string(),
                 is_dir: n.is_folder(),
@@ -641,33 +808,103 @@ impl Core {
     /// been seen through the mount (e.g. shared but not browsed to this session) —
     /// the caller then leaves the path empty.
     pub(crate) fn rel_path_for_uid(&self, uid: &NodeUid) -> Option<String> {
-        // Searched across every mount: a node browsed through a sync folder is
-        // interned in that fork's inode space, so the primary state alone had no
-        // entry for it and the caller silently showed an empty path
-        // (`docs/BUGS.md` B74). The path is relative to whichever mount holds
-        // it, which is the one the user reached it through.
-        let mut path = None;
-        self.for_each_state(|st| {
-            if path.is_some() {
-                return;
-            }
-            path = walk_to_root(st, uid);
-        });
-        path
+        // Relative to whichever root the node sits under: a node browsed
+        // through a sync folder is under that fork's root, not My Files
+        // (`docs/BUGS.md` B74).
+        walk_to_root(&self.state(), uid)
     }
 }
 
-/// Walk one inode space from `uid` up to its root, building the path. `None` if
-/// that mount has never interned the node, or if the chain breaks partway —
+/// What [`Core::stop_sharing`] managed to remove.
+#[derive(Debug, Default, PartialEq, Eq)]
+pub(crate) struct StopSharingOutcome {
+    /// Entries removed (the link, invitations and members alike).
+    pub(crate) removed: usize,
+    /// What could not be removed, by email (or "the public link"), with why.
+    pub(crate) failed: Vec<(String, String)>,
+}
+
+impl StopSharingOutcome {
+    fn record<E: std::fmt::Display>(&mut self, what: &str, result: Result<(), E>) {
+        match result {
+            Ok(()) => self.removed += 1,
+            Err(e) => self.failed.push((what.to_string(), e.to_string())),
+        }
+    }
+
+    /// The reply for the control socket: a sentence for a full success, an
+    /// error naming what is left otherwise.
+    pub(crate) fn into_result(self) -> Result<String, String> {
+        if self.failed.is_empty() {
+            return Ok(match self.removed {
+                0 => "it was not shared".to_string(),
+                1 => "stopped sharing: removed 1 entry".to_string(),
+                n => format!("stopped sharing: removed {n} entries"),
+            });
+        }
+        let left = self
+            .failed
+            .iter()
+            .map(|(what, why)| format!("{what} ({why})"))
+            .collect::<Vec<_>>()
+            .join(", ");
+        Err(format!(
+            "removed {} of {}; still shared: {left}",
+            self.removed,
+            self.removed + self.failed.len()
+        ))
+    }
+}
+
+/// Walk the tree from `uid` up to the root it sits under, building the path.
+/// `None` if the node has never been interned, or if the chain breaks partway —
 /// a broken chain is a missing path, never a partial one.
 fn walk_to_root(st: &super::State, uid: &NodeUid) -> Option<String> {
     let mut ino = *st.by_uid.get(uid)?;
     let mut parts = Vec::new();
-    while ino != ROOT_INO {
+    while !st.roots.contains(&ino) {
         let entry = st.entries.get(&ino)?;
+        // An orphan is its own parent without being a root.
+        if entry.parent == ino {
+            return None;
+        }
         parts.push(entry.node.name.clone());
         ino = entry.parent;
     }
     parts.reverse();
     Some(parts.join("/"))
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn stopping_a_share_that_is_gone_says_so() {
+        let outcome = StopSharingOutcome::default();
+        assert_eq!(outcome.into_result().unwrap(), "it was not shared");
+    }
+
+    #[test]
+    fn a_failed_removal_names_what_is_still_shared() {
+        let mut outcome = StopSharingOutcome::default();
+        outcome.record::<String>("the public link", Ok(()));
+        outcome.record("bob@example.com", Err("forbidden"));
+        outcome.record::<String>("carol@example.com", Ok(()));
+        assert_eq!(
+            outcome.into_result().unwrap_err(),
+            "removed 2 of 3; still shared: bob@example.com (forbidden)"
+        );
+    }
+
+    #[test]
+    fn a_full_removal_counts_the_entries() {
+        let mut outcome = StopSharingOutcome::default();
+        outcome.record::<String>("the public link", Ok(()));
+        outcome.record::<String>("bob@example.com", Ok(()));
+        assert_eq!(
+            outcome.into_result().unwrap(),
+            "stopped sharing: removed 2 entries"
+        );
+    }
+}

crates/pdfs-fuse/src/shutdown.rs

@@ -6,7 +6,7 @@
 //! invisible; in an in-process remount — which is how the daemon recovers from a
 //! mount going away underneath it — it leaks a whole generation of threads that
 //! keep writing to the database and calling the API on behalf of a mount that no
-//! longer exists (bugs.md B44).
+//! longer exists (`docs/BUGS.md` B44).
 //!
 //! The primitive is deliberately small: a flag plus a condvar. Every loop that
 //! would otherwise call `thread::sleep` calls [`Shutdown::sleep`] instead, which

crates/pdfs-fuse/src/sim/mod.rs

@@ -0,0 +1,9 @@
+//! The simulation tests of `docs/TESTING.md`: a fake Drive with
+//! the faults the real one has, and the tools to drive daemons against it.
+
+pub(crate) mod daemon;
+pub(crate) mod fake_drive;
+pub(crate) mod model;
+pub(crate) mod rng;
+pub(crate) mod run;
+pub(crate) mod watchdog;

crates/pdfs-fuse/src/sim/model.rs

@@ -0,0 +1,445 @@
+//! The reference file system of the simulation (`docs/TESTING.md`,
+//! "Simulation runs", property 1): what each syscall a simulated client makes should
+//! answer, and the tree it should leave, on a POSIX file system.
+//!
+//! A [`Model`] covers one client's own folder; paths are relative to it. The
+//! same type is used for the trees read back from a mount and from Drive.
+
+use std::collections::BTreeMap;
+
+use super::rng::Rng;
+
+/// A tree by path: `None` is a folder, `Some` a file's content.
+pub(crate) type Tree = BTreeMap<String, Option<Vec<u8>>>;
+
+/// One file system call, or the few a program makes in a row.
+#[derive(Clone, Debug)]
+pub(crate) enum FsOp {
+    /// `open(O_CREAT | O_TRUNC)`, write `data`, close.
+    Create {
+        path: String,
+        data: Vec<u8>,
+    },
+    /// Open an existing file, write `data` at `offset` (past the end leaves
+    /// a hole), close.
+    Write {
+        path: String,
+        offset: u64,
+        data: Vec<u8>,
+    },
+    /// Open an existing file and `ftruncate` it to `len`.
+    Truncate {
+        path: String,
+        len: u64,
+    },
+    Mkdir {
+        path: String,
+    },
+    Rename {
+        from: String,
+        to: String,
+    },
+    Unlink {
+        path: String,
+    },
+    /// Open a file, unlink it, append `data`, read it all back, close.
+    UnlinkOpen {
+        path: String,
+        data: Vec<u8>,
+    },
+    Rmdir {
+        path: String,
+    },
+}
+
+impl FsOp {
+    /// The paths the op names, for the log.
+    pub(crate) fn describe(&self) -> String {
+        match self {
+            FsOp::Create { path, data } => format!("create {path} ({} bytes)", data.len()),
+            FsOp::Write { path, offset, data } => {
+                format!("write {path} at {offset} ({} bytes)", data.len())
+            }
+            FsOp::Truncate { path, len } => format!("truncate {path} to {len}"),
+            FsOp::Mkdir { path } => format!("mkdir {path}"),
+            FsOp::Rename { from, to } => format!("rename {from} -> {to}"),
+            FsOp::Unlink { path } => format!("unlink {path}"),
+            FsOp::UnlinkOpen { path, data } => {
+                format!("unlink {path} while open, then append {} bytes", data.len())
+            }
+            FsOp::Rmdir { path } => format!("rmdir {path}"),
+        }
+    }
+}
+
+/// The tree one client should have, and what its syscalls should answer.
+#[derive(Clone, Debug, Default)]
+pub(crate) struct Model {
+    tree: Tree,
+}
+
+/// Names are drawn from small pools, so ops collide: a create over an existing
+/// file, a rename over another, a mkdir of a name that is taken.
+const FILE_NAMES: [&str; 6] = ["a.txt", "b.txt", "c.bin", "d", "e.md", "f.txt"];
+const DIR_NAMES: [&str; 3] = ["x", "y", "z"];
+
+impl Model {
+    pub(crate) fn tree(&self) -> &Tree {
+        &self.tree
+    }
+
+    pub(crate) fn content(&self, path: &str) -> Option<&Vec<u8>> {
+        self.tree.get(path)?.as_ref()
+    }
+
+    /// What `op` should answer on a POSIX file system, applying it on success.
+    pub(crate) fn apply(&mut self, op: &FsOp) -> Result<(), i32> {
+        match op {
+            FsOp::Create { path, data } => {
+                self.check_parent(path)?;
+                match self.tree.get(path) {
+                    Some(None) => Err(libc::EISDIR),
+                    _ => {
+                        self.tree.insert(path.clone(), Some(data.clone()));
+                        Ok(())
+                    }
+                }
+            }
+            FsOp::Write { path, offset, data } => {
+                let content = self.file_mut(path)?;
+                // Writing nothing changes nothing, even past the end.
+                if data.is_empty() {
+                    return Ok(());
+                }
+                let end = *offset as usize + data.len();
+                if content.len() < end {
+                    content.resize(end, 0);
+                }
+                content[*offset as usize..end].copy_from_slice(data);
+                Ok(())
+            }
+            FsOp::Truncate { path, len } => {
+                self.file_mut(path)?.resize(*len as usize, 0);
+                Ok(())
+            }
+            FsOp::Mkdir { path } => {
+                self.check_parent(path)?;
+                if self.tree.contains_key(path) {
+                    return Err(libc::EEXIST);
+                }
+                self.tree.insert(path.clone(), None);
+                Ok(())
+            }
+            FsOp::Rename { from, to } => self.rename(from, to),
+            FsOp::Unlink { path } | FsOp::UnlinkOpen { path, .. } => {
+                self.check_parent(path)?;
+                match self.tree.get(path) {
+                    None => Err(libc::ENOENT),
+                    Some(None) => Err(libc::EISDIR),
+                    Some(Some(_)) => {
+                        self.tree.remove(path);
+                        Ok(())
+                    }
+                }
+            }
+            FsOp::Rmdir { path } => {
+                self.check_parent(path)?;
+                match self.tree.get(path) {
+                    None => Err(libc::ENOENT),
+                    Some(Some(_)) => Err(libc::ENOTDIR),
+                    Some(None) if self.has_children(path) => Err(libc::ENOTEMPTY),
+                    Some(None) => {
+                        self.tree.remove(path);
+                        Ok(())
+                    }
+                }
+            }
+        }
+    }
+
+    fn rename(&mut self, from: &str, to: &str) -> Result<(), i32> {
+        self.check_parent(from)?;
+        let Some(moving) = self.tree.get(from).cloned() else {
+            return Err(libc::ENOENT);
+        };
+        self.check_parent(to)?;
+        if from == to {
+            return Ok(());
+        }
+        if is_below(to, from) {
+            return Err(libc::EINVAL);
+        }
+        match (&moving, self.tree.get(to)) {
+            (None, Some(Some(_))) => return Err(libc::ENOTDIR),
+            (Some(_), Some(None)) => return Err(libc::EISDIR),
+            (None, Some(None)) if self.has_children(to) => return Err(libc::ENOTEMPTY),
+            _ => {}
+        }
+        let below: Vec<(String, Option<Vec<u8>>)> = self
+            .tree
+            .iter()
+            .filter(|(path, _)| is_below(path, from))
+            .map(|(path, entry)| (path.clone(), entry.clone()))
+            .collect();
+        self.tree.remove(from);
+        self.tree.insert(to.to_owned(), moving);
+        for (path, entry) in below {
+            self.tree.remove(&path);
+            self.tree
+                .insert(format!("{to}{}", &path[from.len()..]), entry);
+        }
+        Ok(())
+    }
+
+    fn file_mut(&mut self, path: &str) -> Result<&mut Vec<u8>, i32> {
+        self.check_parent(path)?;
+        match self.tree.get_mut(path) {
+            None => Err(libc::ENOENT),
+            Some(None) => Err(libc::EISDIR),
+            Some(Some(content)) => Ok(content),
+        }
+    }
+
+    fn check_parent(&self, path: &str) -> Result<(), i32> {
+        let mut at = String::new();
+        let mut names = path.split('/').peekable();
+        while let Some(name) = names.next() {
+            if names.peek().is_none() {
+                break;
+            }
+            if !at.is_empty() {
+                at.push('/');
+            }
+            at.push_str(name);
+            match self.tree.get(&at) {
+                None => return Err(libc::ENOENT),
+                Some(Some(_)) => return Err(libc::ENOTDIR),
+                Some(None) => {}
+            }
+        }
+        Ok(())
+    }
+
+    fn has_children(&self, path: &str) -> bool {
+        self.tree.keys().any(|other| is_below(other, path))
+    }
+
+    pub(crate) fn files(&self) -> Vec<&String> {
+        self.tree
+            .iter()
+            .filter(|(_, entry)| entry.is_some())
+            .map(|(path, _)| path)
+            .collect()
+    }
+
+    /// Folders, the client's own root (`""`) first.
+    fn dirs(&self) -> Vec<String> {
+        std::iter::once(String::new())
+            .chain(
+                self.tree
+                    .iter()
+                    .filter(|(_, entry)| entry.is_none())
+                    .map(|(path, _)| path.clone()),
+            )
+            .collect()
+    }
+
+    /// A random op, mostly one that should succeed.
+    pub(crate) fn next_op(&self, rng: &mut Rng) -> FsOp {
+        let dirs = self.dirs();
+        let files = self.files();
+        let in_dir = |rng: &mut Rng, names: &[&str]| {
+            let parent = rng.pick(&dirs).expect("the root is always there");
+            join(parent, rng.pick(names).expect("names"))
+        };
+        let a_file = |rng: &mut Rng| rng.pick(&files).map(|path| (*path).clone());
+        let roll = rng.below(100);
+        let op = match roll {
+            0..30 => Some(FsOp::Create {
+                path: in_dir(rng, &FILE_NAMES),
+                data: data(rng),
+            }),
+            30..45 => a_file(rng).map(|path| {
+                let len = self.content(&path).map_or(0, Vec::len) as u64;
+                FsOp::Write {
+                    offset: rng.below(len + 8),
+                    path,
+                    data: data(rng),
+                }
+            }),
+            45..50 => a_file(rng).map(|path| FsOp::Truncate {
+                len: rng.below(32),
+                path,
+            }),
+            50..60 => Some(FsOp::Mkdir {
+                path: in_dir(rng, &DIR_NAMES),
+            }),
+            // A rename in place, over another file as often as not.
+            60..72 => a_file(rng).map(|from| FsOp::Rename {
+                to: join(parent_of(&from), rng.pick(&FILE_NAMES).expect("names")),
+                from,
+            }),
+            // A move to another folder, renamed or not.
+            72..82 => a_file(rng).map(|from| {
+                let name = if rng.chance(0.5) {
+                    name_of(&from)
+                } else {
+                    rng.pick(&FILE_NAMES).expect("names")
+                };
+                let parent = rng.pick(&dirs).expect("the root is always there");
+                FsOp::Rename {
+                    to: join(parent, name),
+                    from,
+                }
+            }),
+            // A folder renamed or moved, with what is in it.
+            82..86 => {
+                let from = rng.pick(&dirs[1..]).cloned();
+                from.map(|from| FsOp::Rename {
+                    to: in_dir(rng, &DIR_NAMES),
+                    from,
+                })
+            }
+            86..94 => a_file(rng).map(|path| FsOp::Unlink { path }),
+            94..97 => a_file(rng).map(|path| FsOp::UnlinkOpen {
+                path,
+                data: data(rng),
+            }),
+            _ => rng
+                .pick(&dirs[1..])
+                .cloned()
+                .map(|path| FsOp::Rmdir { path }),
+        };
+        op.unwrap_or_else(|| FsOp::Create {
+            path: in_dir(rng, &FILE_NAMES),
+            data: data(rng),
+        })
+    }
+}
+
+/// Content for a write: short and recognisable, now and then empty.
+pub(crate) fn data(rng: &mut Rng) -> Vec<u8> {
+    let len = if rng.chance(0.1) {
+        0
+    } else {
+        rng.between(1, 48)
+    };
+    let tag = rng.next_u64();
+    (0..len).map(|i| b'a' + ((tag + i) % 26) as u8).collect()
+}
+
+pub(crate) fn join(parent: &str, name: &str) -> String {
+    if parent.is_empty() {
+        name.to_owned()
+    } else {
+        format!("{parent}/{name}")
+    }
+}
+
+pub(crate) fn parent_of(path: &str) -> &str {
+    path.rsplit_once('/').map_or("", |(parent, _)| parent)
+}
+
+pub(crate) fn name_of(path: &str) -> &str {
+    path.rsplit_once('/').map_or(path, |(_, name)| name)
+}
+
+fn is_below(path: &str, ancestor: &str) -> bool {
+    path.len() > ancestor.len() + 1
+        && path.starts_with(ancestor)
+        && path.as_bytes()[ancestor.len()] == b'/'
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn create(path: &str, data: &[u8]) -> FsOp {
+        FsOp::Create {
+            path: path.into(),
+            data: data.to_vec(),
+        }
+    }
+
+    #[test]
+    fn a_rename_over_a_file_replaces_it_and_a_folder_moves_with_its_contents() {
+        let mut model = Model::default();
+        model.apply(&FsOp::Mkdir { path: "x".into() }).unwrap();
+        model.apply(&create("x/a", b"1")).unwrap();
+        model.apply(&create("b", b"2")).unwrap();
+        model
+            .apply(&FsOp::Rename {
+                from: "b".into(),
+                to: "x/a".into(),
+            })
+            .unwrap();
+        assert_eq!(model.content("x/a"), Some(&b"2".to_vec()));
+        model
+            .apply(&FsOp::Rename {
+                from: "x".into(),
+                to: "y".into(),
+            })
+            .unwrap();
+        assert_eq!(model.tree().keys().collect::<Vec<_>>(), vec!["y", "y/a"]);
+    }
+
+    #[test]
+    fn calls_fail_the_way_posix_says() {
+        let mut model = Model::default();
+        model.apply(&FsOp::Mkdir { path: "x".into() }).unwrap();
+        model.apply(&create("x/a", b"1")).unwrap();
+        assert_eq!(
+            model.apply(&FsOp::Mkdir { path: "x".into() }),
+            Err(libc::EEXIST)
+        );
+        assert_eq!(model.apply(&create("x", b"")), Err(libc::EISDIR));
+        assert_eq!(model.apply(&create("x/a/b", b"")), Err(libc::ENOTDIR));
+        assert_eq!(model.apply(&create("q/a", b"")), Err(libc::ENOENT));
+        assert_eq!(
+            model.apply(&FsOp::Rmdir { path: "x".into() }),
+            Err(libc::ENOTEMPTY)
+        );
+        assert_eq!(
+            model.apply(&FsOp::Unlink { path: "x".into() }),
+            Err(libc::EISDIR)
+        );
+        let into_itself = FsOp::Rename {
+            from: "x".into(),
+            to: "x/y".into(),
+        };
+        assert_eq!(model.apply(&into_itself), Err(libc::EINVAL));
+    }
+
+    #[test]
+    fn a_write_past_the_end_leaves_a_hole() {
+        let mut model = Model::default();
+        model.apply(&create("a", b"ab")).unwrap();
+        let write = FsOp::Write {
+            path: "a".into(),
+            offset: 4,
+            data: b"z".to_vec(),
+        };
+        model.apply(&write).unwrap();
+        assert_eq!(model.content("a"), Some(&b"ab\0\0z".to_vec()));
+        let nothing = FsOp::Write {
+            path: "a".into(),
+            offset: 9,
+            data: Vec::new(),
+        };
+        model.apply(&nothing).unwrap();
+        assert_eq!(model.content("a"), Some(&b"ab\0\0z".to_vec()));
+    }
+
+    #[test]
+    fn generated_ops_mostly_succeed() {
+        let mut rng = Rng::new(9);
+        let mut model = Model::default();
+        let ok = (0..500)
+            .filter(|_| {
+                let op = model.next_op(&mut rng);
+                model.apply(&op).is_ok()
+            })
+            .count();
+        assert!(ok > 350, "{ok} of 500 succeeded");
+    }
+}

crates/pdfs-fuse/src/sim/rng.rs

@@ -0,0 +1,79 @@
+//! A small seeded random number generator, so a failing simulation is a seed
+//! that can be replayed. SplitMix64: fast, good enough for choosing operations
+//! and faults, and no dependency.
+
+#[derive(Clone, Debug)]
+pub(crate) struct Rng(u64);
+
+impl Rng {
+    pub(crate) fn new(seed: u64) -> Self {
+        Self(seed)
+    }
+
+    pub(crate) fn next_u64(&mut self) -> u64 {
+        self.0 = self.0.wrapping_add(0x9E37_79B9_7F4A_7C15);
+        let mut z = self.0;
+        z = (z ^ (z >> 30)).wrapping_mul(0xBF58_476D_1CE4_E5B9);
+        z = (z ^ (z >> 27)).wrapping_mul(0x94D0_49BB_1331_11EB);
+        z ^ (z >> 31)
+    }
+
+    /// A value in `0..n`; `0` when `n` is `0`.
+    pub(crate) fn below(&mut self, n: u64) -> u64 {
+        if n == 0 { 0 } else { self.next_u64() % n }
+    }
+
+    /// A value in `lo..=hi`.
+    pub(crate) fn between(&mut self, lo: u64, hi: u64) -> u64 {
+        lo + self.below(hi.saturating_sub(lo) + 1)
+    }
+
+    /// True with probability `p`.
+    pub(crate) fn chance(&mut self, p: f64) -> bool {
+        p > 0.0 && (self.next_u64() >> 11) as f64 / (1u64 << 53) as f64 <= p
+    }
+
+    pub(crate) fn pick<'a, T>(&mut self, items: &'a [T]) -> Option<&'a T> {
+        items.get(self.below(items.len() as u64) as usize)
+    }
+
+    /// A child generator, so one consumer's draws do not shift another's.
+    pub(crate) fn fork(&mut self) -> Self {
+        Self::new(self.next_u64())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::Rng;
+
+    #[test]
+    fn a_seed_replays_the_same_sequence() {
+        let (mut a, mut b) = (Rng::new(7), Rng::new(7));
+        for _ in 0..100 {
+            assert_eq!(a.next_u64(), b.next_u64());
+        }
+        assert_ne!(Rng::new(7).next_u64(), Rng::new(8).next_u64());
+    }
+
+    #[test]
+    fn bounded_draws_stay_in_range() {
+        let mut rng = Rng::new(1);
+        for _ in 0..1000 {
+            assert!(rng.below(5) < 5);
+            assert!((3..=9).contains(&rng.between(3, 9)));
+        }
+        assert_eq!(rng.below(0), 0);
+        assert!(!rng.chance(0.0));
+        assert!(rng.chance(1.0));
+        assert!([1, 2, 3].contains(rng.pick(&[1, 2, 3]).unwrap()));
+        assert_eq!(rng.pick::<u8>(&[]), None);
+    }
+
+    #[test]
+    fn a_fork_draws_apart_from_its_parent() {
+        let mut parent = Rng::new(3);
+        let mut child = parent.fork();
+        assert_ne!(parent.next_u64(), child.next_u64());
+    }
+}

crates/pdfs-fuse/src/sim/run.rs

@@ -0,0 +1,766 @@
+//! The simulation runs described in `docs/TESTING.md`.
+//!
+//! A run starts one or more daemons on a shared [`FakeDrive`], each owning a
+//! folder `c<n>` that only it writes, so there is one writer per file. A seed
+//! draws a sequence of steps: syscalls through a client's mount, the link of a
+//! client going down or up, a client restarting, a pause, a settle (every
+//! queue must drain within a budget), or a file deleted while the drain holds
+//! a write to it. Every syscall is checked against the
+//! client's [`Model`] as it happens. After the sequence every link comes up
+//! and the run checks:
+//!
+//! 1. **No loss** and **POSIX**: Drive's copy of each client's folder is what
+//!    its model says.
+//! 2. **No false conflicts**: no conflict copy anywhere, and the drain never
+//!    takes a client's own delete for another device's.
+//! 3. **Convergence**: each mount shows Drive's tree.
+//! 4. **Liveness**: no syscall took longer than its budget, settles drained in
+//!    theirs, and every queue drained once the links were up. A call that
+//!    never returns is caught by the [`Watchdog`], which keeps the stacks.
+//!
+//! A restart here is a stop and a start: the next run finds what the last
+//! one wrote down, but in-flight work is let finish rather than cut. A kill
+//! at an arbitrary point needs the daemon in a process of its own.
+//!
+//! Any of these fails the seed. No open bug is tolerated: the matchers that
+//! counted the ones still open instead (B125 to B134) went once a run on
+//! 812eb3d passed every profile without them.
+//!
+//! The runs are ignored by default because they mount FUSE; run them with
+//! `cargo test -p pdfs-fuse --lib sim::run -- --ignored --test-threads=1`.
+//! `PDFS_SIM_SEEDS=n` sets how many seeds each runs, `PDFS_SIM_SEED=s` replays
+//! one, and `RUST_LOG=pdfs_fuse=debug` with `--nocapture` shows the daemons'
+//! logs.
+
+use std::fs::OpenOptions;
+use std::io::{self, Read, Seek, SeekFrom, Write};
+use std::path::{Path, PathBuf};
+use std::time::{Duration, Instant};
+
+use super::daemon::{Daemon, scratch, take_logged, wait_until};
+use super::fake_drive::{Entry, FakeDrive, Faults};
+use super::model::{FsOp, Model, Tree, data, join};
+use super::rng::Rng;
+use super::watchdog::Watchdog;
+
+/// What a run looks like.
+pub(crate) struct Profile {
+    pub(crate) name: &'static str,
+    pub(crate) clients: usize,
+    pub(crate) steps: usize,
+    pub(crate) faults: fn() -> Faults,
+    /// Whether links go down and come back during the sequence.
+    pub(crate) link_flaps: bool,
+    pub(crate) restarts: bool,
+    /// Whether files are deleted while the drain holds a write to them.
+    pub(crate) deletes_mid_upload: bool,
+    /// How long a settle may take; `None` leaves settles out.
+    pub(crate) settle_budget: Option<Duration>,
+    /// How long one syscall may take.
+    pub(crate) syscall_budget: Duration,
+    /// How long the queues may take to drain at the end.
+    pub(crate) drain_budget: Duration,
+}
+
+/// How long the mounts may take to show Drive's tree at the end: one event
+/// poll (`POLL_INTERVAL`) and change.
+const CONVERGE_BUDGET: Duration = Duration::from_secs(30);
+
+/// How long the drain may take to pick up a write: its debounce, and then
+/// some.
+const PICKUP_BUDGET: Duration = Duration::from_secs(8);
+
+/// Steps shown before a failure. All of them are written to `steps.log` in
+/// the state the failure keeps.
+const LOG_TAIL: usize = 40;
+
+/// How long past its own budget a step, or the checks after the last one, may
+/// run before the watchdog takes it for hung. Longer than a restart may wait
+/// for the last run and the mount.
+const STALL: Duration = Duration::from_secs(120);
+
+struct Client {
+    daemon: Option<Daemon>,
+    model: Model,
+    online: bool,
+}
+
+impl Client {
+    fn daemon(&self) -> &Daemon {
+        self.daemon
+            .as_ref()
+            .expect("a client always has a daemon between steps")
+    }
+
+    fn root(&self, index: usize) -> PathBuf {
+        self.daemon().mountpoint.join(owned_folder(index))
+    }
+}
+
+struct Run {
+    seed: u64,
+    profile: &'static Profile,
+    drive: std::sync::Arc<FakeDrive>,
+    clients: Vec<Client>,
+    log: Vec<String>,
+    started: Instant,
+    /// What the daemons logged at `INFO` and above so far.
+    logged: Vec<String>,
+    watchdog: Watchdog,
+}
+
+/// Run `profile` on `seed`. It answers a report of what went wrong, if
+/// anything did.
+pub(crate) fn run(seed: u64, profile: &'static Profile) -> Result<(), String> {
+    let dir = scratch(&format!("{}-{seed}", profile.name));
+    let mut rng = Rng::new(seed);
+    let drive = FakeDrive::new();
+    let mut run = Run {
+        seed,
+        profile,
+        drive: drive.clone(),
+        clients: Vec::new(),
+        log: Vec::new(),
+        started: Instant::now(),
+        logged: Vec::new(),
+        watchdog: Watchdog::new(&dir),
+    };
+    let outcome = run.execute(&dir, &mut rng);
+    run.watchdog.stop();
+    // A hang is what failed, whatever the aborted call answered.
+    let outcome = match run.watchdog.fired() {
+        Some(hung) => Err(format!("liveness: {}", hung.what)),
+        None => outcome,
+    };
+    for client in &mut run.clients {
+        if let Some(daemon) = client.daemon.take() {
+            let _ = daemon.stop();
+        }
+    }
+    match outcome {
+        Ok(()) => {
+            let _ = std::fs::remove_dir_all(&dir);
+            Ok(())
+        }
+        Err(failure) => Err(run.report(&failure, &dir)),
+    }
+}
+
+impl Run {
+    fn execute(&mut self, dir: &Path, rng: &mut Rng) -> Result<(), String> {
+        for index in 0..self.profile.clients {
+            let client = self.drive.client(rng.next_u64(), (self.profile.faults)());
+            let daemon = Daemon::start(&dir.join(format!("client{index}")), client)
+                .map_err(|error| format!("client {index} did not mount: {error}"))?;
+            let folder = daemon.mountpoint.join(owned_folder(index));
+            std::fs::create_dir(&folder)
+                .map_err(|error| format!("client {index}: mkdir {}: {error}", folder.display()))?;
+            self.clients.push(Client {
+                daemon: Some(daemon),
+                model: Model::default(),
+                online: true,
+            });
+        }
+        for step in 0..self.profile.steps {
+            let _watching = self.watchdog.arm(self.step_limit(), format!("step {step}"));
+            self.step(rng)?;
+        }
+        let _watching = self.watchdog.arm(
+            self.profile.drain_budget + STALL,
+            "the checks after the last step",
+        );
+        self.finish()
+    }
+
+    /// How long a step may run before the watchdog takes it for hung.
+    fn step_limit(&self) -> Duration {
+        let budget = self.profile.settle_budget.unwrap_or_default();
+        budget.max(self.profile.syscall_budget) + STALL
+    }
+
+    fn step(&mut self, rng: &mut Rng) -> Result<(), String> {
+        let index = rng.below(self.clients.len() as u64) as usize;
+        let roll = rng.below(100);
+        match roll {
+            0..4 if self.profile.link_flaps => {
+                let client = &mut self.clients[index];
+                client.online = !client.online;
+                client.daemon().client.set_online(client.online);
+                let state = if client.online { "up" } else { "down" };
+                self.note(format!("client {index}: link {state}"));
+            }
+            4..6 if self.profile.restarts => {
+                self.note(format!("client {index}: restart"));
+                let daemon = self.clients[index].daemon.take().expect("daemon");
+                let daemon = daemon
+                    .restart()
+                    .map_err(|error| format!("client {index} did not come back: {error}"))?;
+                self.clients[index].daemon = Some(daemon);
+            }
+            6..12 => {
+                let ms = rng.below(400);
+                self.note(format!("pause {ms} ms"));
+                std::thread::sleep(Duration::from_millis(ms));
+            }
+            12..15 => {
+                if let Some(budget) = self.profile.settle_budget
+                    && self.clients.iter().all(|client| client.online)
+                {
+                    self.settle(budget)?;
+                }
+            }
+            15..25 if self.profile.deletes_mid_upload => self.delete_mid_upload(index, rng)?,
+            _ => {
+                let op = self.clients[index].model.next_op(rng);
+                self.syscall(index, op)?;
+            }
+        }
+        Ok(())
+    }
+
+    /// Wait until every queue is empty and Drive holds what the models say.
+    fn settle(&mut self, budget: Duration) -> Result<(), String> {
+        self.note("settle".into());
+        let start = Instant::now();
+        // What the drain logs meanwhile is what explains a known bug's damage.
+        if !wait_until(budget, || {
+            self.logged.extend(take_logged());
+            self.landed()
+        }) {
+            return Err(format!(
+                "liveness: the queues did not drain within {budget:?}\n{}",
+                self.difference()
+            ));
+        }
+        self.note(format!("settled in {:?}", start.elapsed()));
+        Ok(())
+    }
+
+    /// Write a file Drive holds and delete it once the drain has picked the
+    /// write up. The drain's read of the node is held until the delete has
+    /// landed and a read shows it, so the drain finds the file trashed while
+    /// it holds a write that is no longer queued (B105). Half the time the
+    /// file is made again meanwhile, which queues a write under the op id the
+    /// delete freed (B133).
+    fn delete_mid_upload(&mut self, index: usize, rng: &mut Rng) -> Result<(), String> {
+        let file = rng
+            .pick(&self.clients[index].model.files())
+            .map(|path| (*path).clone());
+        let (Some(path), Some(budget), true) = (
+            file,
+            self.profile.settle_budget,
+            self.clients.iter().all(|client| client.online),
+        ) else {
+            let op = self.clients[index].model.next_op(rng);
+            return self.syscall(index, op);
+        };
+        self.settle(budget)?;
+        let Some(uid) = self.drive.lookup(&join(&owned_folder(index), &path)) else {
+            return Err(format!(
+                "no loss: client {index}'s {path} is not on Drive after a settle"
+            ));
+        };
+        let held = self.clients[index].daemon().client.hold_next_read(&uid);
+        self.note(format!("client {index}: hold the next read of {path}"));
+        let mut bytes = data(rng);
+        bytes.push(b'.');
+        self.syscall(
+            index,
+            FsOp::Write {
+                path: path.clone(),
+                offset: 0,
+                data: bytes,
+            },
+        )?;
+        if !wait_until(PICKUP_BUDGET, || held.reached()) {
+            self.note(format!("client {index}: nothing read {path} back"));
+            return Ok(());
+        }
+        self.syscall(index, FsOp::Unlink { path: path.clone() })?;
+        if rng.chance(0.5) {
+            self.syscall(
+                index,
+                FsOp::Create {
+                    path: path.clone(),
+                    data: data(rng),
+                },
+            )?;
+        }
+        let faults = (self.profile.faults)();
+        let latency = Duration::from_millis(faults.latency_ms.1);
+        std::thread::sleep(faults.listing_lag + latency);
+        drop(held);
+        self.note(format!("client {index}: the read of {path} goes on"));
+        // Nothing else is queued before the drain has answered.
+        std::thread::sleep(2 * latency + Duration::from_millis(50));
+        Ok(())
+    }
+
+    fn syscall(&mut self, index: usize, op: FsOp) -> Result<(), String> {
+        let client = &mut self.clients[index];
+        let root = client.root(index);
+        let before = client.model.content(target(&op)).cloned();
+        let expected = client.model.apply(&op);
+        self.logged.extend(take_logged());
+        let start = Instant::now();
+        let actual = perform(&root, &op, before.as_deref());
+        let took = start.elapsed();
+        self.note(format!(
+            "client {index}: {} = {} in {took:?}",
+            op.describe(),
+            outcome(&actual)
+        ));
+        let actual = actual?;
+        self.logged.extend(take_logged());
+        if actual != expected {
+            return Err(format!(
+                "posix: client {index}: {} answered {}, expected {}",
+                op.describe(),
+                errno(actual),
+                errno(expected)
+            ));
+        }
+        if took > self.profile.syscall_budget {
+            return Err(format!(
+                "liveness: client {index}: {} took {took:?}, over {:?}",
+                op.describe(),
+                self.profile.syscall_budget
+            ));
+        }
+        Ok(())
+    }
+
+    fn finish(&mut self) -> Result<(), String> {
+        for client in &mut self.clients {
+            client.online = true;
+            client.daemon().client.set_online(true);
+        }
+        self.note("every link up; waiting for the queues".into());
+        if !wait_until(self.profile.drain_budget, || {
+            self.logged.extend(take_logged());
+            self.landed()
+        }) {
+            return Err(format!(
+                "no loss: Drive does not hold what the clients wrote, {:?} after the links came up\n{}",
+                self.profile.drain_budget,
+                self.difference()
+            ));
+        }
+        // Each client's folder is its own, so a file of it gone from Drive was
+        // deleted by the client.
+        if let Some(line) = self.logged.iter().find(|line| is_own_delete_kept(line)) {
+            return Err(format!(
+                "no false conflicts: the drain took a client's own delete for another device's: {line}"
+            ));
+        }
+        let drive = self.drive_tree();
+        let conflicts: Vec<String> = drive
+            .into_keys()
+            .filter(|path| is_conflict_copy(path))
+            .collect();
+        if !conflicts.is_empty() {
+            return Err(format!("no false conflicts: {conflicts:?}"));
+        }
+        for index in 0..self.clients.len() {
+            let mut last = Tree::new();
+            let converged = wait_until(CONVERGE_BUDGET, || {
+                last = read_tree(&self.clients[index].daemon().mountpoint);
+                last == self.drive_tree()
+            });
+            if !converged {
+                return Err(format!(
+                    "convergence: client {index}'s mount differs from Drive\n{}",
+                    diff(&last, &self.drive_tree(), "mount", "Drive")
+                ));
+            }
+        }
+        Ok(())
+    }
+
+    /// Whether every client's queue is empty and Drive holds what its model
+    /// says.
+    fn landed(&self) -> bool {
+        let drive = self.drive_tree();
+        self.clients.iter().enumerate().all(|(index, client)| {
+            client
+                .daemon()
+                .pending()
+                .is_ok_and(|items| items.is_empty())
+                && subtree(&drive, &owned_folder(index)) == *client.model.tree()
+        })
+    }
+
+    fn difference(&self) -> String {
+        let drive = self.drive_tree();
+        let mut out = String::new();
+        for (index, client) in self.clients.iter().enumerate() {
+            let pending = client.daemon().pending().unwrap_or_default();
+            for item in pending {
+                out.push_str(&format!(
+                    "  client {index} still owes: {} {} (attempts {}, last error {:?})\n",
+                    item.kind, item.path, item.attempts, item.last_error
+                ));
+            }
+            out.push_str(&diff(
+                client.model.tree(),
+                &subtree(&drive, &owned_folder(index)),
+                &format!("client {index}'s model"),
+                "Drive",
+            ));
+        }
+        out
+    }
+
+    fn drive_tree(&self) -> Tree {
+        self.drive
+            .tree()
+            .into_iter()
+            .map(|(path, entry)| {
+                let content = match entry {
+                    Entry::Folder => None,
+                    Entry::File(content) => Some(content.as_ref().clone()),
+                };
+                (path, content)
+            })
+            .collect()
+    }
+
+    fn note(&mut self, line: String) {
+        let line = format!("{:>8.3}s {line}", self.started.elapsed().as_secs_f64());
+        self.log.push(line);
+    }
+
+    fn report(&self, failure: &str, dir: &Path) -> String {
+        let _ = std::fs::write(dir.join("steps.log"), self.log.join("\n") + "\n");
+        let tail = &self.log[self.log.len().saturating_sub(LOG_TAIL)..];
+        format!(
+            "{} seed {} failed: {failure}\nreplay with PDFS_SIM_SEED={}; state kept in {}\nlast steps:\n{}",
+            self.profile.name,
+            self.seed,
+            self.seed,
+            dir.display(),
+            tail.join("\n")
+        )
+    }
+}
+
+/// Make `op` under `root`. The outer error is a check that failed on the
+/// way (the content an open file reads back), the inner one the errno.
+fn perform(root: &Path, op: &FsOp, before: Option<&[u8]>) -> Result<Result<(), i32>, String> {
+    let at = |path: &str| root.join(path);
+    let result: io::Result<()> = match op {
+        FsOp::Create { path, data } => std::fs::write(at(path), data),
+        FsOp::Write { path, offset, data } => OpenOptions::new()
+            .write(true)
+            .open(at(path))
+            .and_then(|mut file| {
+                file.seek(SeekFrom::Start(*offset))?;
+                file.write_all(data)
+            }),
+        FsOp::Truncate { path, len } => OpenOptions::new()
+            .write(true)
+            .open(at(path))
+            .and_then(|file| file.set_len(*len)),
+        FsOp::Mkdir { path } => std::fs::create_dir(at(path)),
+        FsOp::Rename { from, to } => std::fs::rename(at(from), at(to)),
+        FsOp::Unlink { path } => std::fs::remove_file(at(path)),
+        FsOp::Rmdir { path } => std::fs::remove_dir(at(path)),
+        FsOp::UnlinkOpen { path, data } => {
+            let mut read = Vec::new();
+            let result = OpenOptions::new()
+                .read(true)
+                .write(true)
+                .open(at(path))
+                .and_then(|mut file| {
+                    std::fs::remove_file(at(path))?;
+                    file.seek(SeekFrom::End(0))?;
+                    file.write_all(data)?;
+                    file.seek(SeekFrom::Start(0))?;
+                    file.read_to_end(&mut read).map(|_| ())
+                });
+            if result.is_ok() {
+                let expected = [before.unwrap_or_default(), data].concat();
+                if read != expected {
+                    return Err(format!(
+                        "posix: {} read back {:?}, expected {:?}",
+                        op.describe(),
+                        String::from_utf8_lossy(&read),
+                        String::from_utf8_lossy(&expected)
+                    ));
+                }
+            }
+            result
+        }
+    };
+    Ok(result.map_err(|error| error.raw_os_error().unwrap_or(libc::EIO)))
+}
+
+fn target(op: &FsOp) -> &str {
+    match op {
+        FsOp::Create { path, .. }
+        | FsOp::Write { path, .. }
+        | FsOp::Truncate { path, .. }
+        | FsOp::Mkdir { path }
+        | FsOp::Unlink { path }
+        | FsOp::UnlinkOpen { path, .. }
+        | FsOp::Rmdir { path } => path,
+        FsOp::Rename { from, .. } => from,
+    }
+}
+
+/// The tree a mount shows, read through it, without the folders it makes up
+/// that are not on Drive.
+fn read_tree(mountpoint: &Path) -> Tree {
+    fn walk(dir: &Path, prefix: &str, tree: &mut Tree) {
+        let Ok(entries) = std::fs::read_dir(dir) else {
+            return;
+        };
+        for entry in entries.flatten() {
+            let name = entry.file_name().to_string_lossy().into_owned();
+            if prefix.is_empty() && name == crate::r#virtual::SHARED_WITH_ME_BASE {
+                continue;
+            }
+            let path = super::model::join(prefix, &name);
+            if entry.file_type().is_ok_and(|kind| kind.is_dir()) {
+                tree.insert(path.clone(), None);
+                walk(&entry.path(), &path, tree);
+            } else {
+                tree.insert(path, std::fs::read(entry.path()).ok());
+            }
+        }
+    }
+    let mut tree = Tree::new();
+    walk(mountpoint, "", &mut tree);
+    tree
+}
+
+/// The part of `tree` below `folder`, relative to it.
+fn subtree(tree: &Tree, folder: &str) -> Tree {
+    let prefix = format!("{folder}/");
+    tree.iter()
+        .filter_map(|(path, entry)| Some((path.strip_prefix(&prefix)?.to_owned(), entry.clone())))
+        .collect()
+}
+
+fn owned_folder(index: usize) -> String {
+    format!("c{index}")
+}
+
+/// Whether `line` is the drain keeping a queued write as a conflict copy
+/// because Drive no longer holds the file, or holds it in the trash.
+fn is_own_delete_kept(line: &str) -> bool {
+    line.starts_with("queued write conflicts; keeping a conflict copy")
+        && (line.contains(" reason=\"the file was trashed remotely\"")
+            || line.contains(" reason=\"the file no longer exists remotely\""))
+}
+
+fn is_conflict_copy(path: &str) -> bool {
+    path.split('/')
+        .any(|name| name.contains("(sync-conflict") || name.starts_with("recovered-"))
+}
+
+fn diff(left: &Tree, right: &Tree, left_name: &str, right_name: &str) -> String {
+    let show = |entry: &Option<Vec<u8>>| match entry {
+        None => "folder".to_owned(),
+        Some(content) => format!("{:?}", String::from_utf8_lossy(content)),
+    };
+    let mut out = String::new();
+    for (path, entry) in left {
+        match right.get(path) {
+            None => out.push_str(&format!(
+                "  {path}: only in {left_name} ({})\n",
+                show(entry)
+            )),
+            Some(other) if other != entry => out.push_str(&format!(
+                "  {path}: {left_name} has {}, {right_name} has {}\n",
+                show(entry),
+                show(other)
+            )),
+            Some(_) => {}
+        }
+    }
+    for (path, entry) in right {
+        if !left.contains_key(path) {
+            out.push_str(&format!(
+                "  {path}: only in {right_name} ({})\n",
+                show(entry)
+            ));
+        }
+    }
+    out
+}
+
+fn errno(outcome: Result<(), i32>) -> String {
+    match outcome {
+        Ok(()) => "ok".into(),
+        Err(code) => io::Error::from_raw_os_error(code).to_string(),
+    }
+}
+
+fn outcome(actual: &Result<Result<(), i32>, String>) -> String {
+    match actual {
+        Ok(result) => errno(*result),
+        Err(check) => check.clone(),
+    }
+}
+
+/// Run `profile` over the seeds the environment asks for: `PDFS_SIM_SEED`
+/// alone, or `PDFS_SIM_SEEDS` of them (default `default_seeds`) from 1.
+pub(crate) fn run_seeds(profile: &'static Profile, default_seeds: u64) {
+    let seeds: Vec<u64> = match std::env::var("PDFS_SIM_SEED") {
+        Ok(seed) => vec![seed.parse().expect("PDFS_SIM_SEED is a number")],
+        Err(_) => {
+            let count = std::env::var("PDFS_SIM_SEEDS")
+                .ok()
+                .and_then(|count| count.parse().ok())
+                .unwrap_or(default_seeds);
+            (1..=count).collect()
+        }
+    };
+    let failures: Vec<String> = seeds
+        .into_iter()
+        .filter_map(|seed| run(seed, profile).err())
+        .collect();
+    assert!(failures.is_empty(), "\n{}", failures.join("\n\n"));
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    static ONE_CLIENT_LAN: Profile = Profile {
+        name: "one-client-lan",
+        clients: 1,
+        steps: 80,
+        faults: Faults::lan,
+        link_flaps: false,
+        restarts: false,
+        deletes_mid_upload: false,
+        settle_budget: Some(Duration::from_secs(4)),
+        syscall_budget: Duration::from_secs(2),
+        drain_budget: Duration::from_secs(20),
+    };
+
+    static ONE_CLIENT_DELETES: Profile = Profile {
+        name: "one-client-deletes",
+        clients: 1,
+        steps: 30,
+        faults: Faults::lan,
+        link_flaps: false,
+        restarts: false,
+        deletes_mid_upload: true,
+        settle_budget: Some(Duration::from_secs(4)),
+        syscall_budget: Duration::from_secs(2),
+        drain_budget: Duration::from_secs(20),
+    };
+
+    static ONE_CLIENT_ECHOES: Profile = Profile {
+        name: "one-client-echoes",
+        clients: 1,
+        steps: 60,
+        faults: Faults::echoes,
+        link_flaps: false,
+        restarts: false,
+        deletes_mid_upload: false,
+        settle_budget: Some(Duration::from_secs(6)),
+        syscall_budget: Duration::from_secs(2),
+        drain_budget: Duration::from_secs(20),
+    };
+
+    static ONE_CLIENT_FLAKY: Profile = Profile {
+        name: "one-client-flaky",
+        clients: 1,
+        steps: 80,
+        faults: Faults::wifi,
+        link_flaps: true,
+        restarts: true,
+        deletes_mid_upload: false,
+        settle_budget: None,
+        syscall_budget: Duration::from_secs(10),
+        drain_budget: Duration::from_secs(120),
+    };
+
+    /// Wi-Fi's latency, lag and lost replies without the outages: the queue
+    /// has to drain within a settle, not merely by the end.
+    static ONE_CLIENT_WIFI: Profile = Profile {
+        name: "one-client-wifi",
+        clients: 1,
+        steps: 80,
+        faults: Faults::wifi,
+        link_flaps: false,
+        restarts: false,
+        deletes_mid_upload: false,
+        settle_budget: Some(Duration::from_secs(30)),
+        syscall_budget: Duration::from_secs(10),
+        drain_budget: Duration::from_secs(120),
+    };
+
+    static THREE_CLIENTS: Profile = Profile {
+        name: "three-clients",
+        clients: 3,
+        steps: 90,
+        faults: Faults::lan,
+        link_flaps: true,
+        restarts: true,
+        deletes_mid_upload: false,
+        settle_budget: Some(Duration::from_secs(4)),
+        syscall_budget: Duration::from_secs(2),
+        drain_budget: Duration::from_secs(60),
+    };
+
+    #[test]
+    #[ignore = "mounts FUSE: run with `cargo test -p pdfs-fuse --lib sim::run -- --ignored`"]
+    fn one_client_on_a_good_link() {
+        run_seeds(&ONE_CLIENT_LAN, 12);
+    }
+
+    #[test]
+    #[ignore = "mounts FUSE: run with `cargo test -p pdfs-fuse --lib sim::run -- --ignored`"]
+    fn one_client_deleting_files_as_they_upload() {
+        run_seeds(&ONE_CLIENT_DELETES, 3);
+    }
+
+    #[test]
+    #[ignore = "mounts FUSE: run with `cargo test -p pdfs-fuse --lib sim::run -- --ignored`"]
+    fn one_client_with_stale_hashes_and_reordered_echoes() {
+        run_seeds(&ONE_CLIENT_ECHOES, 6);
+    }
+
+    #[test]
+    #[ignore = "mounts FUSE: run with `cargo test -p pdfs-fuse --lib sim::run -- --ignored`"]
+    fn one_client_on_a_flaky_link() {
+        run_seeds(&ONE_CLIENT_FLAKY, 4);
+    }
+
+    #[test]
+    #[ignore = "mounts FUSE: run with `cargo test -p pdfs-fuse --lib sim::run -- --ignored`"]
+    fn one_client_on_a_slow_link() {
+        run_seeds(&ONE_CLIENT_WIFI, 4);
+    }
+
+    #[test]
+    #[ignore = "mounts FUSE: run with `cargo test -p pdfs-fuse --lib sim::run -- --ignored`"]
+    fn three_clients_one_writer_each() {
+        run_seeds(&THREE_CLIENTS, 3);
+    }
+
+    #[test]
+    fn conflict_copies_are_recognised_by_name() {
+        assert!(is_conflict_copy("c0/x/a (sync-conflict 1790000000).txt"));
+        assert!(is_conflict_copy("recovered-XrPUSW5W"));
+        assert!(!is_conflict_copy("c0/x/a.txt"));
+    }
+
+    #[test]
+    fn a_write_kept_because_its_file_is_gone_is_recognised() {
+        let kept = "queued write conflicts; keeping a conflict copy uid=simvol~L10 \
+                    name=\"f.txt\" reason=\"the file was trashed remotely\"";
+        assert!(is_own_delete_kept(kept));
+        assert!(is_own_delete_kept(
+            &kept.replace("was trashed", "no longer exists")
+        ));
+        assert!(!is_own_delete_kept(
+            &kept.replace("the file was trashed remotely", "revision changed")
+        ));
+    }
+}

crates/pdfs-fuse/src/sim/watchdog.rs

@@ -0,0 +1,322 @@
+//! Turns a call through a mount that never returns into a failed seed.
+//!
+//! A deadlock in a daemon leaves the run's own thread blocked in the kernel,
+//! where no budget check is reached, and a CI job then waits for its timeout.
+//! The watchdog runs beside the run. When what it was armed for outlives its
+//! limit, it aborts the FUSE connections of the run's mounts, so the blocked
+//! call fails and the run reports the hang, and writes where every thread was
+//! to `stacks.txt`.
+//!
+//! The stacks are taken in the process: no tracer can stop a thread that
+//! waits on a FUSE request, so `gdb` and `eu-stack` wait as long as the hang.
+//! Each thread is sent [`stack_signal`] and captures its own backtrace in the
+//! handler. A thread in the kernel takes the signal only once it leaves, so
+//! the mounts are aborted after the others have answered, and its backtrace
+//! is the call it was blocked in. Its kernel wait channel is recorded first.
+
+use std::collections::BTreeMap;
+use std::path::{Path, PathBuf};
+use std::sync::Arc;
+use std::sync::atomic::{AtomicBool, Ordering};
+use std::thread::JoinHandle;
+use std::time::{Duration, Instant};
+
+use parking_lot::Mutex;
+
+use super::daemon::wait_until;
+
+/// How long the threads that are not blocked in the kernel have to answer,
+/// and the blocked ones once the mounts are aborted.
+const ANSWER: Duration = Duration::from_secs(5);
+
+pub(crate) struct Watchdog {
+    shared: Arc<Shared>,
+    thread: Option<JoinHandle<()>>,
+}
+
+struct Shared {
+    /// The run's directory: its mounts are below it, and the stacks go in it.
+    dir: PathBuf,
+    armed: Mutex<Option<Armed>>,
+    /// What hung, once something has.
+    fired: Mutex<Option<Hung>>,
+    stop: AtomicBool,
+}
+
+struct Armed {
+    what: String,
+    limit: Duration,
+    since: Instant,
+}
+
+/// What hung, and where every thread was then.
+#[derive(Clone, Debug, PartialEq)]
+pub(crate) struct Hung {
+    pub(crate) what: String,
+    pub(crate) stacks: String,
+}
+
+/// Disarms the watchdog when dropped.
+pub(crate) struct Watching(Arc<Shared>);
+
+impl Watchdog {
+    pub(crate) fn new(dir: &Path) -> Self {
+        let shared = Arc::new(Shared {
+            dir: dir.to_path_buf(),
+            armed: Mutex::new(None),
+            fired: Mutex::new(None),
+            stop: AtomicBool::new(false),
+        });
+        let thread = {
+            let shared = shared.clone();
+            std::thread::Builder::new()
+                .name("sim-watchdog".into())
+                .spawn(move || watch(&shared))
+                .expect("spawn the simulation watchdog")
+        };
+        Self {
+            shared,
+            thread: Some(thread),
+        }
+    }
+
+    /// Watch `what` until the guard is dropped, and abort the mounts if it is
+    /// still going after `limit`.
+    pub(crate) fn arm(&self, limit: Duration, what: impl Into<String>) -> Watching {
+        *self.shared.armed.lock() = Some(Armed {
+            what: what.into(),
+            limit,
+            since: Instant::now(),
+        });
+        Watching(self.shared.clone())
+    }
+
+    /// What hung, if something did. Call [`Watchdog::stop`] first, or the
+    /// stacks may still be on their way.
+    pub(crate) fn fired(&self) -> Option<Hung> {
+        self.shared.fired.lock().clone()
+    }
+
+    /// Stop watching, once the stacks of a hang that was caught are taken.
+    pub(crate) fn stop(&mut self) {
+        self.shared.stop.store(true, Ordering::Relaxed);
+        if let Some(thread) = self.thread.take() {
+            let _ = thread.join();
+        }
+    }
+}
+
+impl Drop for Watching {
+    fn drop(&mut self) {
+        *self.0.armed.lock() = None;
+    }
+}
+
+impl Drop for Watchdog {
+    fn drop(&mut self) {
+        self.stop();
+    }
+}
+
+fn watch(shared: &Shared) {
+    while !shared.stop.load(Ordering::Relaxed) {
+        std::thread::sleep(Duration::from_millis(200));
+        let hung = shared
+            .armed
+            .lock()
+            .as_ref()
+            .filter(|armed| armed.since.elapsed() > armed.limit)
+            .map(|armed| format!("{} has not returned after {:?}", armed.what, armed.limit));
+        let Some(hung) = hung else { continue };
+        *shared.armed.lock() = None;
+        let path = shared.dir.join("stacks.txt");
+        // Before anything is let go: the run reports the hang, not whatever
+        // the aborted call answers.
+        *shared.fired.lock() = Some(Hung {
+            what: format!("{hung}; the stacks are in {}", path.display()),
+            stacks: String::new(),
+        });
+        let stacks = take_stacks(|| abort_mounts_below(&shared.dir));
+        let what = match std::fs::write(&path, &stacks) {
+            Ok(()) => format!("{hung}; the stacks are in {}", path.display()),
+            Err(error) => format!("{hung}; the stacks could not be kept: {error}"),
+        };
+        *shared.fired.lock() = Some(Hung { what, stacks });
+    }
+}
+
+/// One thread of this process as the kernel sees it.
+struct Task {
+    tid: i32,
+    name: String,
+    /// `D` while it waits in the kernel uninterruptibly.
+    state: String,
+    wchan: String,
+}
+
+/// Where every other thread of this process is, with `release` called once
+/// those not blocked in the kernel have answered.
+fn take_stacks(release: impl FnOnce()) -> String {
+    let own = gettid();
+    let tasks = tasks();
+    STACKS.lock().clear();
+    install_stack_handler();
+    for task in &tasks {
+        if task.tid != own {
+            // SAFETY: tgkill only sends a signal, whose handler is installed.
+            unsafe {
+                libc::syscall(
+                    libc::SYS_tgkill,
+                    std::process::id() as i32,
+                    task.tid,
+                    stack_signal(),
+                )
+            };
+        }
+    }
+    let answered = |blocked: bool| {
+        let stacks = STACKS.lock();
+        tasks
+            .iter()
+            .filter(|task| task.tid != own && (blocked || task.state != "D"))
+            .all(|task| stacks.contains_key(&task.tid))
+    };
+    wait_until(ANSWER, || answered(false));
+    release();
+    wait_until(ANSWER, || answered(true));
+    let stacks = std::mem::take(&mut *STACKS.lock());
+    let mut out = String::new();
+    for task in tasks.iter().filter(|task| task.tid != own) {
+        out.push_str(&format!(
+            "thread {} {:?}, state {}, waiting in {}\n",
+            task.tid, task.name, task.state, task.wchan
+        ));
+        match stacks.get(&task.tid) {
+            Some(stack) => out.push_str(&format!("{stack}\n")),
+            None => out.push_str("  (no answer)\n\n"),
+        }
+    }
+    out
+}
+
+fn tasks() -> Vec<Task> {
+    let Ok(entries) = std::fs::read_dir("/proc/self/task") else {
+        return Vec::new();
+    };
+    let read = |dir: &Path, file: &str| {
+        std::fs::read_to_string(dir.join(file))
+            .unwrap_or_default()
+            .trim()
+            .to_owned()
+    };
+    entries
+        .flatten()
+        .filter_map(|entry| {
+            let tid = entry.file_name().to_str()?.parse().ok()?;
+            let dir = entry.path();
+            // The state follows the name in parentheses, which may hold spaces.
+            let stat = read(&dir, "stat");
+            let state = stat.rsplit_once(") ")?.1.split(' ').next()?.to_owned();
+            Some(Task {
+                tid,
+                name: read(&dir, "comm"),
+                state,
+                wchan: read(&dir, "wchan"),
+            })
+        })
+        .collect()
+}
+
+/// The backtraces the threads captured, by thread id.
+static STACKS: Mutex<BTreeMap<i32, std::backtrace::Backtrace>> = Mutex::new(BTreeMap::new());
+
+/// The signal that asks a thread for its stack: one of the real-time signals
+/// that nothing in the daemon uses.
+fn stack_signal() -> i32 {
+    libc::SIGRTMIN() + 7
+}
+
+fn install_stack_handler() {
+    static INSTALLED: std::sync::Once = std::sync::Once::new();
+    extern "C" fn handler(_: i32) {
+        // Not async-signal-safe: the capture allocates. The process is being
+        // taken down for a hang anyway, and a thread interrupted inside the
+        // allocator only fails to answer.
+        let stack = std::backtrace::Backtrace::force_capture();
+        STACKS.lock().insert(gettid(), stack);
+    }
+    INSTALLED.call_once(|| {
+        // SAFETY: a zeroed sigaction with a handler and SA_RESTART is valid,
+        // and the handler only touches `STACKS`.
+        unsafe {
+            let mut action: libc::sigaction = std::mem::zeroed();
+            action.sa_sigaction = handler as extern "C" fn(i32) as usize;
+            action.sa_flags = libc::SA_RESTART;
+            libc::sigemptyset(&mut action.sa_mask);
+            libc::sigaction(stack_signal(), &action, std::ptr::null_mut());
+        }
+    });
+}
+
+fn gettid() -> i32 {
+    // SAFETY: gettid has no arguments and cannot fail.
+    unsafe { libc::gettid() }
+}
+
+/// Abort the FUSE connection of every mount below `dir`. Its connection is
+/// named by the mount's device minor, and the mount's owner may abort it.
+fn abort_mounts_below(dir: &Path) {
+    let Ok(mountinfo) = std::fs::read_to_string("/proc/self/mountinfo") else {
+        return;
+    };
+    for line in mountinfo.lines() {
+        let fields: Vec<&str> = line.split(' ').collect();
+        let (Some(device), Some(mountpoint)) = (fields.get(2), fields.get(4)) else {
+            continue;
+        };
+        if !Path::new(mountpoint).starts_with(dir) {
+            continue;
+        }
+        if let Some((_, minor)) = device.split_once(':') {
+            let _ = std::fs::write(format!("/sys/fs/fuse/connections/{minor}/abort"), "1");
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::sim::daemon::scratch;
+
+    #[test]
+    #[ignore = "signals every thread of the test process: run with `cargo test -p pdfs-fuse sim:: -- --ignored`"]
+    fn a_call_past_its_limit_is_reported_with_the_stacks() {
+        let dir = scratch("watchdog");
+        let mut watchdog = Watchdog::new(&dir);
+        {
+            let _watching = watchdog.arm(Duration::from_millis(100), "a nap");
+            std::thread::sleep(Duration::from_secs(1));
+        }
+        watchdog.stop();
+        let hung = watchdog.fired().expect("the watchdog fired");
+        assert!(
+            hung.what.starts_with("a nap has not returned after"),
+            "{hung:?}"
+        );
+        assert!(hung.stacks.contains("a_call_past_its_limit"), "{hung:?}");
+        let kept = std::fs::read_to_string(dir.join("stacks.txt")).unwrap_or_default();
+        assert_eq!(kept, hung.stacks);
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+
+    #[test]
+    fn a_call_within_its_limit_is_not() {
+        let dir = scratch("watchdog-quiet");
+        let watchdog = Watchdog::new(&dir);
+        drop(watchdog.arm(Duration::from_millis(100), "a quick call"));
+        std::thread::sleep(Duration::from_millis(500));
+        assert_eq!(watchdog.fired(), None);
+        drop(watchdog);
+        let _ = std::fs::remove_dir_all(&dir);
+    }
+}

crates/pdfs-fuse/src/state.rs

@@ -128,7 +128,62 @@ pub(crate) struct WriteHandle {
     pub(crate) open_count: usize,
 }
 
-/// A released write whose upload has not happened yet (offline.md Phase 3).
+impl WriteHandle {
+    /// A handle on a file `create` just made: an empty base, so everything
+    /// written through it is authored.
+    ///
+    /// The base is `node` as it actually exists — the empty file the server just
+    /// minted, or a queued create's placeholder — never the local clock or a
+    /// blank. `queue_revision` turns it into `StagedWrite::based_on` and the
+    /// drain compares the remote against that. A `now_secs()` mtime differs from
+    /// the server's stamp by the create round-trip. A missing revision id drops
+    /// the check to `(mtime, size)`, and moving or renaming the file advances
+    /// its mtime on the server. Either way the first write to a brand-new file
+    /// lands as a conflict copy of its own empty create (`docs/BUGS.md` B113).
+    pub(crate) fn created(ino: u64, node: &Node, file: File, path: PathBuf) -> Self {
+        Self {
+            ino,
+            uid: node.uid.clone(),
+            file: Arc::new(file),
+            path,
+            written: Intervals::default(),
+            len: 0,
+            base_size: 0,
+            base_mtime: node.modification_time,
+            base_revision_id: crate::node_revision_id(node),
+            dirty: false,
+            open_count: 0,
+        }
+    }
+
+    /// Write over `node` from now on: a revision this daemon has just put on the
+    /// server under the handle, by sealing an upload or by landing the queued
+    /// create the handle was opened on. The base the handle opened over is gone,
+    /// and keeping it would make the drain read our own revision as another
+    /// device's edit.
+    pub(crate) fn rebase_onto(&mut self, node: &Node) {
+        self.base_mtime = node.modification_time;
+        self.base_size = crate::node_size(node);
+        self.base_revision_id = crate::node_revision_id(node);
+    }
+
+    /// Account for `[offset, end)` having been written through this handle.
+    ///
+    /// A write past the end leaves a hole that reads as zeros, exactly like a
+    /// grow through `setattr`, so the hole is authored too. Left unauthored, a
+    /// hole below the base size was refilled from the base at commit: a file
+    /// shrunk and then written past its new end got its old bytes back.
+    pub(crate) fn record_write(&mut self, offset: u64, end: u64) {
+        if offset > self.len {
+            self.written.add(self.len, offset);
+        }
+        self.written.add(offset, end);
+        self.len = self.len.max(end);
+        self.dirty = true;
+    }
+}
+
+/// A released write whose upload has not happened yet.
 ///
 /// The bytes live in the content cache's staging dir and the intent lives in the
 /// `pending_op` table; this pairs them in memory so a read can be served without
@@ -146,8 +201,54 @@ pub(crate) struct PendingRevision {
 /// time so this module stays free of `fuser`.
 pub(crate) type DirListing = Vec<(u64, bool, String)>;
 
+/// Where [`State::next_ino`] starts: far above any local id a database hands
+/// out, so a fallback inode never names a node that has a row.
+pub(crate) const FALLBACK_INO: u64 = 1 << 48;
+
+/// The inode of the node with local id `lid` ([`Db::lids_for`]). One above the
+/// lid, so no node takes [`crate::ROOT_INO`], which each session uses for its
+/// own root. The same in every mount and after a restart.
+pub(crate) fn lid_ino(lid: i64) -> u64 {
+    lid.unsigned_abs() + crate::ROOT_INO
+}
+
+/// The inodes of the stored nodes among `nodes`, from their local ids, for
+/// [`State::intern_batch_with_inos`]. Called before the state lock is taken;
+/// a node that already has a row keeps its id. Empty when the database fails,
+/// which leaves the allocation to the lock holder, as before.
+pub(crate) fn inos_for(db: &Db, nodes: &[Node]) -> HashMap<NodeUid, u64> {
+    let stored: Vec<&Node> = nodes
+        .iter()
+        .filter(|node| !crate::r#virtual::is_virtual_uid(&node.uid))
+        .collect();
+    if stored.is_empty() {
+        return HashMap::new();
+    }
+    match db.lids_for(&stored) {
+        Ok(lids) => stored
+            .iter()
+            .zip(lids)
+            .map(|(node, lid)| (node.uid.clone(), lid_ino(lid)))
+            .collect(),
+        Err(error) => {
+            warn!(%error, "no local ids before the listing; allocating them under the lock");
+            HashMap::new()
+        }
+    }
+}
+
+/// The local id of the node at inode `ino`, the inverse of [`lid_ino`]:
+/// `None` for the root and a fallback inode.
+pub(crate) fn ino_lid(ino: u64) -> Option<i64> {
+    (ino > crate::ROOT_INO && ino < FALLBACK_INO).then(|| (ino - crate::ROOT_INO) as i64)
+}
+
 /// Mutable inode bookkeeping, guarded by a mutex because fuser drives the
 /// `Filesystem` trait through `&self`.
+///
+/// One tree for every mount of the daemon. Each FUSE session sees it from its
+/// own root ([`State::roots`]) and calls that root inode 1; `ProtonFs`
+/// translates at the kernel boundary.
 pub(crate) struct State {
     /// inode -> node metadata.
     pub(crate) entries: HashMap<u64, Entry>,
@@ -156,9 +257,20 @@ pub(crate) struct State {
     /// Cached directory listings: parent inode -> child inodes. Presence of a
     /// key means the directory has been enumerated.
     pub(crate) children: HashMap<u64, Vec<u64>>,
+    /// The inodes some session has been rooted at ([`State::install_root`]).
+    /// A root is its own parent. Kept after its session unmounts: the node is
+    /// still one of ours.
+    pub(crate) roots: HashSet<u64>,
+    /// The next inode for a node the database could not give a local id, from
+    /// a range lids never reach ([`FALLBACK_INO`]).
     pub(crate) next_ino: u64,
     /// Shared write state keyed by inode. Concurrent writers share the scratch file.
     pub(crate) active_writes: HashMap<u64, WriteHandle>,
+    /// Inodes whose last write handle `release` has taken out of
+    /// [`State::active_writes`] and is still queueing. `close(2)` returns
+    /// before that, and in between the edit is in neither the handles nor the
+    /// queue, so a check for changes not on Drive yet looks here too.
+    pub(crate) releasing: HashSet<u64>,
     /// Maps file handle id (fh) to inode (ino). Read-only opens use fh 0 and
     /// have no entry here.
     pub(crate) handles: HashMap<u64, u64>,
@@ -171,13 +283,13 @@ pub(crate) struct State {
     /// two pages shifted every later entry, and the caller silently skipped or
     /// repeated whatever crossed the page boundary. Each entry is
     /// `(ino, is_dir, name)` — the kernel type is mapped at reply time so this
-    /// module stays free of `fuser` (bugs.md B43).
+    /// module stays free of `fuser` (`docs/BUGS.md` B43).
     pub(crate) dir_snapshots: HashMap<u64, Arc<DirListing>>,
     /// Resident inode attributes whose effective access changed during an
     /// intern/root refresh. Core drains this set and notifies the matching
     /// kernel only after releasing the State lock.
     pub(crate) access_changes: HashSet<u64>,
-    /// Persisted shared-root authority, loaded once per inode space and updated
+    /// Persisted shared-root authority, loaded once and updated
     /// on every explicit role/tombstone change. Descendant inheritance reads
     /// this map instead of issuing one SQLite query per interned inode.
     pub(crate) share_access: HashMap<NodeUid, Access>,
@@ -245,15 +357,6 @@ pub(crate) struct StateGuard<'a> {
     db: &'a Arc<Db>,
 }
 
-/// Take a mount's inode lock through a [`StateGuard`], for the loops that walk
-/// every live mount rather than this `Core`'s own.
-pub(crate) fn lock_state<'a>(
-    state: &'a parking_lot::Mutex<State>,
-    db: &'a Arc<Db>,
-) -> StateGuard<'a> {
-    StateGuard::new(state.lock(), db)
-}
-
 impl<'a> StateGuard<'a> {
     pub(crate) fn new(guard: parking_lot::MutexGuard<'a, State>, db: &'a Arc<Db>) -> Self {
         Self {
@@ -294,9 +397,6 @@ impl Drop for StateGuard<'_> {
 impl State {
     /// A fresh, empty inode space over `db`.
     ///
-    /// `next_ino` differs by caller: a real mount hands out [`crate::ROOT_INO`]
-    /// to its root and starts allocating above it, while a bare state under test
-    /// has no root and starts at 1.
     /// Apply the write-throughs this state owes SQLite, here and now.
     ///
     /// Production code never calls this: [`StateGuard`] releases the inode lock
@@ -323,13 +423,17 @@ impl State {
         self.db.clone()
     }
 
-    pub(crate) fn new(db: Arc<Db>, share_access: HashMap<NodeUid, Access>, next_ino: u64) -> Self {
+    /// A fresh, empty tree over `db`. A node's inode comes from its local id
+    /// ([`lid_ino`]).
+    pub(crate) fn new(db: Arc<Db>, share_access: HashMap<NodeUid, Access>) -> Self {
         Self {
             entries: HashMap::new(),
             by_uid: HashMap::new(),
             children: HashMap::new(),
-            next_ino,
+            roots: HashSet::new(),
+            next_ino: FALLBACK_INO,
             active_writes: HashMap::new(),
+            releasing: HashSet::new(),
             handles: HashMap::new(),
             next_fh: 1,
             dir_snapshots: HashMap::new(),
@@ -339,18 +443,55 @@ impl State {
             outbox: Vec::new(),
         }
     }
-    /// Whether `uid` names an owned node currently reachable from this mount's
-    /// root. Resident entries can outlive their dentries for open-handle and
-    /// access-revocation semantics, so `by_uid` membership alone is not proof
-    /// that a control-socket operation may address the node.
-    pub(crate) fn owns_visible_uid(&self, uid: &NodeUid) -> bool {
-        let Some(root) = self.entries.get(&crate::ROOT_INO) else {
-            return false;
-        };
-        if uid.volume_id != root.uid.volume_id {
-            return false;
+    /// Make `root` the root of a session and return its inode, which comes
+    /// from its local id like any other node's.
+    ///
+    /// The root is always owned. It must never be classified as a shared root,
+    /// even when the API returns membership data (which happens when the owner
+    /// shares the folder with someone else), so a stale share row a downgrade
+    /// may have left for it is dropped. A node already in the tree keeps its
+    /// inode and its lookups, and stops naming a parent: a session's `..` at
+    /// its root is the root.
+    pub(crate) fn install_root(&mut self, root: Node) -> u64 {
+        if let Err(e) = self.db.upsert_node(&root) {
+            warn!(uid = %root.uid, error = %e, "db upsert root failed");
         }
+        self.share_access.remove(&root.uid);
+        let _ = self.db.delete_share_access(&root.uid);
+        let ino = match self.by_uid.get(&root.uid) {
+            Some(&ino) => ino,
+            None => self.new_inos(&[&root])[&root.uid],
+        };
+        if let Some(old) = self.entries.get(&ino).map(|e| e.parent)
+            && old != ino
+            && let Some(kids) = self.children.get_mut(&old)
+        {
+            kids.retain(|&k| k != ino);
+        }
+        self.by_uid.insert(root.uid.clone(), ino);
+        let lookup_count = self.entries.get(&ino).map_or(1, |e| e.lookup_count.max(1));
+        self.entries.insert(
+            ino,
+            Entry {
+                uid: root.uid.clone(),
+                parent: ino,
+                node: root,
+                access: Access::Owner,
+                lookup_count,
+                open_count: 0,
+                unlinked: false,
+            },
+        );
+        self.roots.insert(ino);
+        ino
+    }
 
+    /// Whether `uid` names an owned node currently reachable from one of
+    /// `roots`, the roots of the sessions that are mounted. Resident entries
+    /// can outlive their dentries for open-handle and access-revocation
+    /// semantics, so `by_uid` membership alone is not proof that a
+    /// control-socket operation may address the node.
+    pub(crate) fn owns_visible_uid(&self, uid: &NodeUid, roots: &[u64]) -> bool {
         let Some(&target) = self.by_uid.get(uid) else {
             return false;
         };
@@ -363,7 +504,9 @@ impl State {
             let Some(entry) = self.entries.get(&current) else {
                 return false;
             };
-            if entry.uid.volume_id != root.uid.volume_id
+            // The node's root scopes it to its own volume, excluding foreign
+            // shared-with-me residents from uid-addressed sharing.
+            if entry.uid.volume_id != uid.volume_id
                 || entry.unlinked
                 || entry.node.trashed
                 || entry.uid != entry.node.uid
@@ -371,7 +514,7 @@ impl State {
             {
                 return false;
             }
-            if current == crate::ROOT_INO {
+            if roots.contains(&current) {
                 return true;
             }
 
@@ -413,15 +556,17 @@ impl State {
         self.access_changes.drain().collect()
     }
 
-    /// Whether `uid` lives on the volume this mount is rooted in.
+    /// Whether `uid` lives on a volume some mount is rooted in.
     ///
     /// Used to decide how a node with no resolvable parent is classified:
     /// our own volume fails open (it is owned content), a foreign volume
     /// fails closed (it can only be there because of a share).
     pub(crate) fn is_own_volume(&self, uid: &NodeUid) -> bool {
-        self.entries
-            .get(&crate::ROOT_INO)
-            .is_some_and(|root| root.uid.volume_id == uid.volume_id)
+        self.roots.iter().any(|ino| {
+            self.entries
+                .get(ino)
+                .is_some_and(|root| root.uid.volume_id == uid.volume_id)
+        })
     }
 
     fn stored_share_access(&self, uid: &NodeUid) -> Option<Access> {
@@ -770,11 +915,59 @@ impl State {
         }
     }
 
-    pub(crate) fn intern(&mut self, parent: u64, node: Node) -> u64 {
+    /// Write `node` through as it is, in order with the tree's other changes.
+    pub(crate) fn write_through(&mut self, node: Node) {
+        self.outbox.push(DbWrite::Upsert(vec![node]));
+    }
+
+    pub(crate) fn intern(&mut self, parent: u64, mut node: Node) -> u64 {
+        self.keep_revision_mtime(&mut node);
         self.outbox.push(DbWrite::Upsert(vec![node.clone()]));
         self.intern_mem(parent, node)
     }
 
+    /// Keep what is already known of a file whose content has not changed.
+    ///
+    /// Drive stamps a link's `ModifyTime` on a rename or a move as well, while
+    /// POSIX leaves a file's mtime alone there. The content cache is keyed by
+    /// it, too: taking the new stamp made the cached bytes unreachable, so the
+    /// next partial write could not fill its gaps (docs/BUGS.md B123). The
+    /// active revision id says whether the content is still the same, and so
+    /// whether the size and digest a listing left out are still the ones known.
+    fn keep_revision_mtime(&self, node: &mut Node) {
+        let Some(entry) = self
+            .by_uid
+            .get(&node.uid)
+            .and_then(|ino| self.entries.get(ino))
+        else {
+            return;
+        };
+        if let (
+            NodeKind::File {
+                active_revision_id: Some(now),
+                claimed_size,
+                content_sha1,
+                ..
+            },
+            NodeKind::File {
+                active_revision_id: Some(was),
+                claimed_size: size_was,
+                content_sha1: sha1_was,
+                ..
+            },
+        ) = (&mut node.kind, &entry.node.kind)
+            && now == was
+        {
+            node.modification_time = entry.node.modification_time;
+            if claimed_size.is_none() {
+                *claimed_size = *size_was;
+            }
+            if content_sha1.is_none() {
+                content_sha1.clone_from(sha1_was);
+            }
+        }
+    }
+
     /// Allocate (or reuse) a stable inode for a node, updating the hot-cache maps
     /// only. Every caller owes the DB a write-through — see the callers below;
     /// the split exists so a batch can pay for one transaction instead of `n`.
@@ -784,11 +977,24 @@ impl State {
     }
 
     fn intern_mem_with_access(&mut self, parent: u64, node: Node, access: Access) -> u64 {
+        self.intern_mem_at(parent, node, access, None)
+    }
+
+    /// [`State::intern_mem_with_access`], with the inode a new entry takes when
+    /// the caller already has it ([`State::new_inos`]).
+    fn intern_mem_at(
+        &mut self,
+        parent: u64,
+        mut node: Node,
+        access: Access,
+        ino: Option<u64>,
+    ) -> u64 {
         if let Some(&ino) = self.by_uid.get(&node.uid) {
             let changed = self
                 .entries
                 .get(&ino)
                 .is_some_and(|entry| entry.access != access);
+            self.keep_open_write_size(ino, &mut node);
             if let Some(e) = self.entries.get_mut(&ino) {
                 e.node = node;
                 e.parent = parent;
@@ -801,8 +1007,13 @@ impl State {
             }
             return ino;
         }
-        let ino = self.next_ino;
-        self.next_ino += 1;
+        let ino = match ino {
+            Some(ino) => ino,
+            None => self
+                .new_inos(&[&node])
+                .remove(&node.uid)
+                .unwrap_or_default(),
+        };
         self.by_uid.insert(node.uid.clone(), ino);
         self.entries.insert(
             ino,
@@ -819,6 +1030,41 @@ impl State {
         ino
     }
 
+    /// The inodes for nodes that have no entry yet, from their local ids, in
+    /// one transaction. Should the database fail, they come from
+    /// [`State::next_ino`] instead, so the mount keeps working, with inodes
+    /// that last only this run. So do the inodes of virtual folders, which are
+    /// never stored.
+    fn new_inos(&mut self, nodes: &[&Node]) -> HashMap<NodeUid, u64> {
+        let (virtual_nodes, stored): (Vec<&Node>, Vec<&Node>) = nodes
+            .iter()
+            .partition(|node| crate::r#virtual::is_virtual_uid(&node.uid));
+        let mut inos = match self.db.lids_for(&stored) {
+            Ok(lids) => stored
+                .iter()
+                .zip(lids)
+                .map(|(node, lid)| (node.uid.clone(), lid_ino(lid)))
+                .collect(),
+            Err(error) => {
+                warn!(%error, "no local ids for new inodes; numbering them for this run");
+                self.run_inos(&stored)
+            }
+        };
+        inos.extend(self.run_inos(&virtual_nodes));
+        inos
+    }
+
+    fn run_inos(&mut self, nodes: &[&Node]) -> HashMap<NodeUid, u64> {
+        nodes
+            .iter()
+            .map(|node| {
+                let ino = self.next_ino;
+                self.next_ino += 1;
+                (node.uid.clone(), ino)
+            })
+            .collect()
+    }
+
     /// Decrement lookup count for an inode and prune if lookup_count == 0 && open_count == 0 && unlinked.
     pub(crate) fn forget_lookup(&mut self, ino: u64, nlookup: u64) -> Option<(u64, String)> {
         if let Some(entry) = self.entries.get_mut(&ino) {
@@ -833,6 +1079,23 @@ impl State {
 
     /// Allocate (or reuse) a stable inode for a node that came *from* the
     /// database, which is why nothing is written back.
+    /// Carry the size and mtime of a file open for writing over onto `node`,
+    /// which is about to replace its entry.
+    ///
+    /// The open handle is ahead of anything Drive can report until release
+    /// queues it. Taking the remote's size shrank a SQLite database to its empty
+    /// create while SQLite held it open, and the kernel then capped every read
+    /// at that size (docs/BUGS.md B109).
+    pub(crate) fn keep_open_write_size(&self, ino: u64, node: &mut Node) {
+        let (Some(aw), Some(entry)) = (self.active_writes.get(&ino), self.entries.get(&ino)) else {
+            return;
+        };
+        if let NodeKind::File { claimed_size, .. } = &mut node.kind {
+            *claimed_size = Some(aw.len as i64);
+            node.modification_time = entry.node.modification_time;
+        }
+    }
+
     pub(crate) fn intern_from_db(&mut self, parent: u64, node: Node) -> u64 {
         self.intern_mem(parent, node)
     }
@@ -842,14 +1105,51 @@ impl State {
     ///
     /// This is what keeps `ls` on a large folder quick: one commit for the
     /// listing, rather than one autocommit — and one fsync — per child.
+    #[cfg(test)]
     pub(crate) fn intern_batch(&mut self, parent: u64, nodes: Vec<Node>) -> Vec<u64> {
+        self.intern_batch_with_inos(parent, nodes, HashMap::new())
+    }
+
+    /// [`State::intern_batch`], with inodes for new entries that the caller
+    /// allocated before it took the lock, from [`inos_for`]. Allocating them
+    /// is a write transaction, and under the lock every inline handler of
+    /// every mount waited for it (`docs/BUGS.md` B221). A node missing from
+    /// `inos` still gets one here.
+    pub(crate) fn intern_batch_with_inos(
+        &mut self,
+        parent: u64,
+        mut nodes: Vec<Node>,
+        mut inos: HashMap<NodeUid, u64>,
+    ) -> Vec<u64> {
+        for node in &mut nodes {
+            self.keep_revision_mtime(node);
+        }
         self.outbox.push(DbWrite::Upsert(nodes.clone()));
+        let fresh: Vec<&Node> = nodes
+            .iter()
+            .filter(|node| !self.by_uid.contains_key(&node.uid) && !inos.contains_key(&node.uid))
+            .collect();
+        if !fresh.is_empty() {
+            let allocated = self.new_inos(&fresh);
+            inos.extend(allocated);
+        }
         nodes
             .into_iter()
-            .map(|node| self.intern_mem(parent, node))
+            .map(|node| {
+                let access = self.access_for_node(parent, &node);
+                let ino = inos.remove(&node.uid);
+                self.intern_mem_at(parent, node, access, ino)
+            })
             .collect()
     }
 
+    /// Record `uid`'s listing as complete once the lock is released, after the
+    /// rows of its children: written the other way round, a crash between the
+    /// two left a folder marked complete with no children to show.
+    pub(crate) fn record_listed(&mut self, uid: NodeUid) {
+        self.outbox.push(DbWrite::SetListed(uid, true));
+    }
+
     /// Check if a directory inode has any child nodes in memory or in the database.
     pub(crate) fn has_children(&self, parent: u64) -> bool {
         // A resident listing is authoritative even when empty. Falling through
@@ -950,7 +1250,7 @@ impl State {
     /// re-enumerate, which is the cheaper way to stay honest about what the
     /// server did. A queued rename cannot: re-enumerating needs the network, and
     /// the server has not been told yet in any case — so this *is* the tree's
-    /// new truth until the op drains (offline.md Phase 3b).
+    /// new truth until the op drains.
     pub(crate) fn rename_in_place(
         &mut self,
         ino: u64,
@@ -1069,7 +1369,7 @@ impl State {
     /// created offline, nothing at all), so this row is the only record that the
     /// file is as long as the caller was told it is. Without it a restart serves
     /// the stale size and the file reads as truncated — or empty — while its
-    /// bytes sit safely in staging (offline.md Phase 3).
+    /// bytes sit safely in staging.
     /// Returns the node to persist rather than queueing it in the outbox, and is
     /// the one mutation here whose write-through the caller must do itself:
     /// failure has to be reported, not logged. Dropping the row silently means
@@ -1088,6 +1388,13 @@ mod tests {
     use proton_drive_rs::ShareMembership;
     use proton_drive_rs::proton_sdk::ids::{LinkId, ShareId, ShareMembershipId, VolumeId};
 
+    #[test]
+    fn an_inode_gives_back_the_lid_it_was_made_from() {
+        assert_eq!(ino_lid(lid_ino(7)), Some(7));
+        assert_eq!(ino_lid(crate::ROOT_INO), None);
+        assert_eq!(ino_lid(FALLBACK_INO + 3), None);
+    }
+
     fn uid(link: &str) -> NodeUid {
         NodeUid::new(VolumeId::from("vol"), LinkId::from(link))
     }
@@ -1164,7 +1471,7 @@ mod tests {
         let dir = TempDir::new();
         let db = Db::open(&dir.0.join("cache.db")).unwrap();
         let share_access = db.all_share_access().unwrap();
-        let st = State::new(Arc::new(db), share_access, 1);
+        let st = State::new(Arc::new(db), share_access);
         (st, dir)
     }
 
@@ -1182,6 +1489,74 @@ mod tests {
         (st, dir, src, dst)
     }
 
+    /// A node's inode comes from its row's local id, so it is the same in
+    /// every mount, after the hot cache drops it, and after a restart.
+    #[test]
+    fn a_node_keeps_its_inode_across_mounts_evictions_and_restarts() {
+        let (mut st, _dir) = state();
+        let inos = st.intern_batch(
+            0,
+            vec![
+                node("a", "root", "a.txt", false),
+                node("b", "root", "b.txt", false),
+            ],
+        );
+        st.flush_outbox();
+        let db = st.db.clone();
+        assert_ne!(inos[0], inos[1]);
+        assert!(!inos.contains(&crate::ROOT_INO));
+
+        let mut other = State::new(db.clone(), HashMap::new());
+        assert_eq!(other.intern(0, node("b", "root", "b.txt", false)), inos[1]);
+
+        st.forget_mem(&uid("a"));
+        assert_eq!(st.intern(0, node("a", "root", "a.txt", false)), inos[0]);
+
+        let mut restarted = State::new(db, HashMap::new());
+        assert_eq!(
+            restarted.intern_batch(0, vec![node("a", "root", "a.txt", false)]),
+            [inos[0]]
+        );
+    }
+
+    /// A listing's inodes are allocated before the lock is taken, and are the
+    /// ones a node would get under it (`docs/BUGS.md` B221).
+    #[test]
+    fn inodes_allocated_before_the_lock_are_the_ones_the_node_keeps() {
+        let (mut st, _dir) = state();
+        let db = st.db.clone();
+        let listing = vec![
+            node("a", "root", "a.txt", false),
+            node("b", "root", "b.txt", false),
+        ];
+        let early = inos_for(&db, &listing);
+        assert_eq!(early.len(), 2);
+        let inos = st.intern_batch_with_inos(0, listing, early.clone());
+        assert_eq!(inos, [early[&uid("a")], early[&uid("b")]]);
+        st.flush_outbox();
+
+        let mut restarted = State::new(db, HashMap::new());
+        assert_eq!(
+            restarted.intern(0, node("b", "root", "b.txt", false)),
+            early[&uid("b")]
+        );
+    }
+
+    /// The listing is recorded complete with its children's rows, when the
+    /// lock goes, never ahead of them.
+    #[test]
+    fn a_listing_is_recorded_complete_after_its_children() {
+        let (mut st, _dir) = state();
+        let root = st.intern(0, node("root", "none", "root", true));
+        st.flush_outbox();
+        st.intern_batch(root, vec![node("a", "root", "a.txt", false)]);
+        st.record_listed(uid("root"));
+        assert!(st.db.children_if_listed(&uid("root")).unwrap().is_none());
+        st.flush_outbox();
+        let kids = st.db.children_if_listed(&uid("root")).unwrap().unwrap();
+        assert_eq!(kids.len(), 1);
+    }
+
     /// Audit A5. Moving a file out of a folder leaves that folder's DB row
     /// claiming a complete listing, so once the hot cache drops it, remote
     /// changes under it stop being seen. `relocate` has to clear both sides.
@@ -1247,7 +1622,7 @@ mod tests {
         let root = uid("root").to_string();
         let state = parking_lot::Mutex::new(st);
         {
-            let mut guard = lock_state(&state, &db);
+            let mut guard = StateGuard::new(state.lock(), &db);
             guard.intern(0, node("root", "none", "My Files", true));
             assert!(
                 db.node_by_uid(&root).unwrap().is_none(),
@@ -1490,6 +1865,7 @@ mod tests {
             })
             .collect();
         st.next_ino = crate::ROOT_INO + nodes.len() as u64;
+        st.roots.insert(crate::ROOT_INO);
         // Reverse order so entries are inserted before their parents, matching
         // the hydration path's non-topological materialization.
         for (i, node) in nodes.iter().enumerate().rev() {
@@ -1863,4 +2239,137 @@ mod tests {
         let shared = st.intern(owned_root, shared);
         assert_eq!(st.entries[&shared].access, Access::Viewer);
     }
+
+    fn at_revision(mut node: Node, revision: &str, mtime: i64) -> Node {
+        node.modification_time = mtime;
+        if let NodeKind::File {
+            active_revision_id, ..
+        } = &mut node.kind
+        {
+            *active_revision_id = Some(revision.into());
+        }
+        node
+    }
+
+    #[test]
+    fn a_renamed_file_keeps_its_mtime_until_its_content_changes() {
+        let (mut st, _dir) = state();
+        let root = st.intern(0, node("root", "none", "My Files", true));
+        let mut file = node("file", "root", "a.txt", false);
+        if let NodeKind::File {
+            claimed_size,
+            total_size_on_storage,
+            ..
+        } = &mut file.kind
+        {
+            *claimed_size = Some(7);
+            *total_size_on_storage = 519;
+        }
+        let ino = st.intern(root, at_revision(file.clone(), "r1", 100));
+        // Drive stamps the link on a rename; the content is still r1.
+        let mut renamed = at_revision(file.clone(), "r1", 160);
+        renamed.name = "b.txt".into();
+        st.intern_batch(root, vec![renamed]);
+        assert_eq!(st.entries[&ino].node.modification_time, 100);
+        assert_eq!(st.entries[&ino].node.name, "b.txt");
+        // A listing that leaves the size out still describes revision r1.
+        let mut bare = at_revision(file.clone(), "r1", 170);
+        assert_eq!(crate::node_size(&st.entries[&ino].node), 7);
+        if let NodeKind::File { claimed_size, .. } = &mut bare.kind {
+            *claimed_size = None;
+        }
+        st.intern(root, bare);
+        assert_eq!(st.entries[&ino].node.modification_time, 100);
+        assert_eq!(crate::node_size(&st.entries[&ino].node), 7);
+        st.intern(root, at_revision(file, "r2", 200));
+        assert_eq!(st.entries[&ino].node.modification_time, 200);
+    }
+
+    fn handle(len: u64, base_size: u64) -> WriteHandle {
+        let path = std::env::temp_dir().join(format!("pdfs-handle-test-{}", std::process::id()));
+        let file = File::create(&path).unwrap();
+        let _ = std::fs::remove_file(&path);
+        WriteHandle {
+            ino: 2,
+            uid: uid("file"),
+            file: Arc::new(file),
+            path,
+            written: Intervals::default(),
+            len,
+            base_size,
+            base_mtime: 0,
+            base_revision_id: None,
+            dirty: false,
+            open_count: 1,
+        }
+    }
+
+    /// A file created offline is written while its queued create lands. The
+    /// placeholder had no revision, so the handle takes the empty file the
+    /// create made as its base (docs/BUGS.md B113).
+    #[test]
+    fn a_handle_on_a_placeholder_is_based_on_the_create_that_landed() {
+        let mut h = handle(0, 0);
+        assert_eq!(h.base_revision_id, None);
+        let mut landed = node("real", "root", "f.mkv", false);
+        landed.modification_time = 160;
+        if let NodeKind::File {
+            active_revision_id, ..
+        } = &mut landed.kind
+        {
+            *active_revision_id = Some("empty".into());
+        }
+        h.rebase_onto(&landed);
+        assert_eq!(h.base_mtime, 160);
+        assert_eq!(h.base_size, 0);
+        assert_eq!(h.base_revision_id.as_deref(), Some("empty"));
+    }
+
+    #[test]
+    fn a_listing_does_not_shrink_a_file_open_for_writing() {
+        let (mut st, _dir) = state();
+        let ino = st.intern(0, node("file", "root", "workload.db", false));
+        let mut h = handle(16384, 0);
+        h.ino = ino;
+        st.active_writes.insert(ino, h);
+        st.set_size(ino, 16384);
+        st.touch_mtime(ino, 500);
+
+        // The remote still holds the empty file the create sealed.
+        st.intern(0, node("file", "root", "workload.db", false));
+        assert_eq!(crate::node_size(&st.entries[&ino].node), 16384);
+        assert_eq!(st.entries[&ino].node.modification_time, 500);
+
+        // Once the handle is gone the remote is authoritative again.
+        st.active_writes.remove(&ino);
+        st.intern(0, node("file", "root", "workload.db", false));
+        assert_eq!(crate::node_size(&st.entries[&ino].node), 0);
+    }
+
+    #[test]
+    fn a_write_past_the_end_authors_the_hole_it_leaves() {
+        // Opened over a 100-byte base, shrunk to 10, then written at 50: bytes
+        // 10..50 are zeros now, not the base's bytes 10..50.
+        let mut h = handle(100, 100);
+        h.written.clip(10);
+        h.len = 10;
+        h.record_write(50, 60);
+        assert_eq!(h.len, 60);
+        assert_eq!(
+            h.written.segments(0, 60),
+            vec![(0, 10, false), (10, 60, true)],
+            "only the head still comes from the base"
+        );
+    }
+
+    #[test]
+    fn a_write_inside_the_file_leaves_the_rest_to_the_base() {
+        let mut h = handle(100, 100);
+        h.record_write(40, 50);
+        assert_eq!(h.len, 100);
+        assert_eq!(
+            h.written.segments(0, 100),
+            vec![(0, 40, false), (40, 50, true), (50, 100, false)]
+        );
+    }
 }

crates/pdfs-fuse/src/supervisor.rs

@@ -36,6 +36,11 @@ const TICK: Duration = Duration::from_secs(15);
 /// would train everyone to ignore the real one.
 const STALL_AFTER: Duration = Duration::from_secs(120);
 
+/// How long a sync pass may go without reporting progress before it is called
+/// out. Much longer than [`STALL_AFTER`]: one large file in a pass reports
+/// nothing until its whole transfer is done.
+const SYNC_STALL_AFTER: Duration = Duration::from_secs(900);
+
 /// How often resident size goes into the log.
 const RSS_EVERY: Duration = Duration::from_secs(300);
 
@@ -137,6 +142,26 @@ fn report_stalls(report: &pdfs_core::control::Diagnostics, last_completed: (u64,
             "the transfer lane has not completed a job since the last check"
         );
     }
+    // The engine is one thread for every synced folder, so a dead or stuck one
+    // stops all of them while the worker pool looks idle (B259).
+    if let Some(sync) = &report.sync {
+        if !sync.running {
+            stalled = true;
+            warn!(
+                "the sync engine is not running; synced folders do not sync until the daemon restarts"
+            );
+        } else if let Some(folder) = sync.pass_folder
+            && sync.pass_idle_secs >= SYNC_STALL_AFTER.as_secs()
+        {
+            stalled = true;
+            warn!(
+                folder,
+                age_secs = sync.pass_age_secs,
+                idle_secs = sync.pass_idle_secs,
+                "a sync pass has not reported progress for a long time"
+            );
+        }
+    }
     stalled
 }
 
@@ -169,7 +194,7 @@ fn probe(control_socket: &Path) -> bool {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use pdfs_core::control::{Diagnostics, InflightRequest, WorkerState};
+    use pdfs_core::control::{Diagnostics, InflightRequest, SyncEngineState, WorkerState};
 
     fn busy_worker(age_secs: u64) -> WorkerState {
         WorkerState {
@@ -226,4 +251,45 @@ mod tests {
         assert!(!report_stalls(&report, (0, 30)));
         assert!(report_stalls(&report, (0, 40)));
     }
+
+    fn with_sync(sync: SyncEngineState) -> Diagnostics {
+        Diagnostics {
+            sync: Some(sync),
+            ..Default::default()
+        }
+    }
+
+    /// A sync engine that has ended stops every synced folder while the
+    /// workers look idle, which is what issue #53 looked like (B259).
+    #[test]
+    fn a_stopped_sync_engine_is_a_stall() {
+        assert!(report_stalls(
+            &with_sync(SyncEngineState::default()),
+            (0, 0)
+        ));
+    }
+
+    /// A pass that keeps reporting progress is working, however long it runs;
+    /// one that has gone quiet past the threshold is not.
+    #[test]
+    fn only_a_quiet_sync_pass_is_a_stall() {
+        let busy = SyncEngineState {
+            running: true,
+            pass_folder: Some(3),
+            pass_age_secs: SYNC_STALL_AFTER.as_secs() * 4,
+            pass_idle_secs: 5,
+            panics: 0,
+        };
+        assert!(!report_stalls(&with_sync(busy.clone()), (0, 0)));
+        let quiet = SyncEngineState {
+            pass_idle_secs: SYNC_STALL_AFTER.as_secs(),
+            ..busy
+        };
+        assert!(report_stalls(&with_sync(quiet), (0, 0)));
+        let idle = SyncEngineState {
+            running: true,
+            ..Default::default()
+        };
+        assert!(!report_stalls(&with_sync(idle), (0, 0)));
+    }
 }

crates/pdfs-fuse/src/sweep.rs

@@ -37,10 +37,12 @@ const CONFLICT_SWEEP_INTERVAL: Duration = Duration::from_secs(300);
 /// fresh mount does at startup (hydrate, recover, initial enumeration).
 const CONFLICT_SWEEP_WARMUP: Duration = Duration::from_secs(30);
 
-/// Strip a ` (sync-conflict <stamp>)` (optionally `-<n>`) marker from `name`,
-/// returning the original file name it was a copy of. `None` when `name` is not a
-/// conflict copy. Inverse of the `conflict_name` / `conflict_path_with_suffix`
-/// formats, so the two halves of the product agree on what a conflict is.
+/// Strip a ` (sync-conflict <label> <stamp>)` (optionally `-<n>`) marker from
+/// `name`, returning the original file name it was a copy of. The label names
+/// the computer that made the copy (B251); copies made before it have the bare
+/// stamp. `None` when `name` is not a conflict copy. Inverse of the
+/// `conflict_name` / `conflict_path_with_suffix` formats, so the two halves of
+/// the product agree on what a conflict is.
 pub(crate) fn conflict_base_name(name: &str) -> Option<String> {
     const MARKER: &str = " (sync-conflict ";
     let start = name.find(MARKER)?;
@@ -48,6 +50,20 @@ pub(crate) fn conflict_base_name(name: &str) -> Option<String> {
     let close_rel = name[after..].find(')')?;
     let close = after + close_rel;
     let inner = &name[after..close];
+    // A label is one token of the characters `conflict_label_of` keeps.
+    let inner = match inner.rsplit_once(' ') {
+        Some((label, token)) => {
+            let label_ok = !label.is_empty()
+                && label
+                    .chars()
+                    .all(|c| c.is_ascii_alphanumeric() || matches!(c, '-' | '_' | '.'));
+            if !label_ok {
+                return None;
+            }
+            token
+        }
+        None => inner,
+    };
     // The token is `<digits>` or `<digits>-<digits>`; reject anything else so a
     // file that merely happens to contain the phrase is not mistaken for a copy.
     let mut parts = inner.splitn(2, '-');
@@ -68,7 +84,7 @@ impl Core {
     /// docs for the identical/divergent policy.
     pub(crate) fn run_conflict_sweep_loop(&self) {
         // Both waits are interruptible, so teardown does not have to sit out a
-        // five-minute interval to join this thread (bugs.md B44).
+        // five-minute interval to join this thread (`docs/BUGS.md` B44).
         if !self.shutdown.sleep(CONFLICT_SWEEP_WARMUP) {
             return;
         }
@@ -109,6 +125,7 @@ impl Core {
             by_parent_name.insert((parent, node.name.as_str()), node);
         }
 
+        let roots = self.search_roots();
         let online = self.online.load(std::sync::atomic::Ordering::Relaxed);
         for node in &nodes {
             if node.trashed || node.is_folder() {
@@ -132,7 +149,16 @@ impl Core {
                         self.remove_conflict_copy(node, &base);
                     }
                 }
-                SweepAction::Flag(detail) => self.flag_conflict(&node.uid, &node.name, detail),
+                // Only flag what the conflict listing can show. Anything else
+                // would sit in the feed as "resolved" with no way to act on
+                // it (`docs/BUGS.md` B99).
+                SweepAction::Flag(detail) => {
+                    if self.conflict_path(&roots, &node.uid).is_some() {
+                        self.flag_conflict(&node.uid, &node.name, detail);
+                    } else {
+                        debug!(uid = %node.uid, name = %node.name, "conflict sweep: copy is outside every location; not flagged");
+                    }
+                }
             }
         }
     }
@@ -155,7 +181,7 @@ impl Core {
         }
         match self
             .rt
-            .block_on(self.client.trash_nodes(std::slice::from_ref(uid)))
+            .block_on(self.drive.trash_nodes(std::slice::from_ref(uid)))
             .and_then(batch::into_unit)
         {
             Ok(()) => {}
@@ -165,10 +191,9 @@ impl Core {
                 return;
             }
         }
-        // Every mount, not just the primary: a conflict copy inside a sync
-        // folder lives in that fork's inode space, and leaving it hooked up
-        // there keeps a node we have just trashed remotely visible and
-        // readable there (`docs/BUGS.md` B74).
+        // Leaving a conflict copy hooked up in the tree keeps a node we have
+        // just trashed remotely visible and readable in whichever mount shows
+        // it (`docs/BUGS.md` B74).
         self.for_each_state(|st| {
             st.forget_or_unlink(uid);
         });
@@ -261,7 +286,7 @@ impl Core {
     /// is tracked in that fork's `State` — so consulting `self.state` alone
     /// answered "idle" for every such file and let the sweep delete one with a
     /// writer attached (`docs/BUGS.md` B74). Busy anywhere means busy.
-    fn is_busy(&self, uid: &NodeUid) -> bool {
+    pub(crate) fn is_busy(&self, uid: &NodeUid) -> bool {
         let mut busy = false;
         self.for_each_state(|st| {
             let Some(&ino) = st.by_uid.get(uid) else {
@@ -273,12 +298,21 @@ impl Core {
         busy
     }
 
-    /// Surface a conflict copy that needs the user's attention, at most once per
-    /// daemon run so the activity feed is not spammed each pass.
+    /// Surface a conflict copy that needs the user's attention, once. The
+    /// in-memory set spares the database each pass; the activity log itself is
+    /// what keeps a restart from logging the same copy again (`docs/BUGS.md`
+    /// B99). A copy's name carries its own timestamp, so the name identifies it.
     fn flag_conflict(&self, uid: &NodeUid, name: &str, detail: String) {
         if !self.conflict_notified.lock().insert(uid.clone()) {
             return;
         }
+        match self.db.activity_has(ActivityKind::Conflict, name) {
+            Ok(false) => {}
+            Ok(true) => return,
+            // Cannot tell: a repeat entry is harmless, a missing one hides a
+            // conflict.
+            Err(e) => warn!(%uid, name, error = ?e, "conflict sweep: activity check failed"),
+        }
         debug!(%uid, name, detail, "conflict sweep: flagged divergent conflict copy");
         self.log_activity(ActivityKind::Conflict, name, detail, false);
     }
@@ -468,6 +502,24 @@ mod tests {
             conflict_base_name("notes (sync-conflict 1700-3).txt").as_deref(),
             Some("notes.txt")
         );
+        // Named after the computer that made it (B251).
+        assert_eq!(
+            conflict_base_name("notes (sync-conflict laptop 1700).txt").as_deref(),
+            Some("notes.txt")
+        );
+        assert_eq!(
+            conflict_base_name("notes (sync-conflict my-pc.local 1700-2).txt").as_deref(),
+            Some("notes.txt")
+        );
+        assert_eq!(
+            conflict_base_name(&crate::conflict_name("notes.txt", 1700)).as_deref(),
+            Some("notes.txt")
+        );
+        assert_eq!(
+            conflict_base_name("x (sync-conflict two words 1).txt"),
+            None
+        );
+        assert_eq!(conflict_base_name("x (sync-conflict  1).txt"), None);
         // Not a conflict copy.
         assert_eq!(conflict_base_name("report.xml"), None);
         // Looks close but the token is not a stamp.

crates/pdfs-fuse/src/sync/engine.rs

@@ -1,5 +1,11 @@
 //! Filesystem watcher, polling, and debounce control plane.
 
+use std::panic::AssertUnwindSafe;
+use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
+
+use pdfs_core::control::SyncEngineState;
+use tracing::error;
+
 use super::*;
 
 /// How long the watcher must go quiet before a burst counts as finished.
@@ -22,7 +28,10 @@ const POLL_INTERVAL: Duration = Duration::from_secs(120);
 /// per-item work is a network round-trip, so a folder with thousands of files is
 /// unusable done one-at-a-time; this bounds concurrency like the bulk-upload
 /// engine does.
-pub(super) const SYNC_CONCURRENCY: usize = 8;
+pub(crate) const SYNC_CONCURRENCY: usize = 8;
+/// How long after a failed watch the engine tries the watches again. Each try
+/// walks the folder's whole tree, so not on every burst.
+const REWATCH_RETRY: Duration = Duration::from_secs(600);
 
 /// A message to the sync engine's control thread.
 pub(crate) enum SyncMsg {
@@ -33,7 +42,7 @@ pub(crate) enum SyncMsg {
     /// Re-read the folder set and adjust the filesystem watches (after add/remove).
     Rewatch,
     /// The daemon is tearing down: finish the current pass and end the engine
-    /// thread, so the mount can join it (bugs.md B44).
+    /// thread, so the mount can join it (`docs/BUGS.md` B44).
     Stop,
 }
 
@@ -53,30 +62,159 @@ pub(crate) fn spawn(core: Core, rx: Receiver<SyncMsg>) -> Option<std::thread::Jo
     }
 }
 
+/// Whether the engine is alive and which pass it is on, for
+/// [`Core::diagnostics`] and the supervisor.
+///
+/// A panic in a pass used to end the engine thread. Every later add, sync-now,
+/// file event and poll then went into a channel nobody read, the folder stayed
+/// `syncing`, and nothing said so until a restart (`docs/BUGS.md` B259).
+#[derive(Default)]
+pub(crate) struct EngineHealth {
+    running: AtomicBool,
+    pass: Mutex<Option<PassClock>>,
+    panics: AtomicU64,
+}
+
+#[derive(Clone, Copy)]
+struct PassClock {
+    folder: i64,
+    started: Instant,
+    moved: Instant,
+}
+
+impl EngineHealth {
+    /// Note that the running pass reported progress.
+    pub(crate) fn moved(&self) {
+        if let Some(pass) = self.pass.lock().as_mut() {
+            pass.moved = Instant::now();
+        }
+    }
+
+    /// The engine's state now. `try_lock`, like the rest of the diagnostics:
+    /// a report must not wait on the thing it reports about.
+    pub(crate) fn report(&self) -> SyncEngineState {
+        let pass = self.pass.try_lock().and_then(|pass| *pass);
+        SyncEngineState {
+            running: self.running.load(Ordering::Relaxed),
+            pass_folder: pass.map(|p| p.folder),
+            pass_age_secs: pass.map_or(0, |p| p.started.elapsed().as_secs()),
+            pass_idle_secs: pass.map_or(0, |p| p.moved.elapsed().as_secs()),
+            panics: self.panics.load(Ordering::Relaxed),
+        }
+    }
+
+    /// Run `pass` over `folder` as the current pass. Returns `false` when it
+    /// panicked: the panic ends that pass, not the engine.
+    pub(super) fn contain(&self, folder: i64, pass: impl FnOnce()) -> bool {
+        let now = Instant::now();
+        *self.pass.lock() = Some(PassClock {
+            folder,
+            started: now,
+            moved: now,
+        });
+        let finished = std::panic::catch_unwind(AssertUnwindSafe(pass)).is_ok();
+        *self.pass.lock() = None;
+        if !finished {
+            self.panics.fetch_add(1, Ordering::Relaxed);
+        }
+        finished
+    }
+}
+
+/// Marks the engine running for as long as it lives.
+pub(super) struct Running<'a>(&'a EngineHealth);
+
+impl<'a> Running<'a> {
+    pub(super) fn start(health: &'a EngineHealth) -> Self {
+        health.running.store(true, Ordering::Relaxed);
+        Self(health)
+    }
+}
+
+impl Drop for Running<'_> {
+    fn drop(&mut self) {
+        self.0.running.store(false, Ordering::Relaxed);
+        if std::thread::panicking() {
+            error!("sync engine stopped by a panic; nothing syncs until the daemon restarts");
+        }
+    }
+}
+
+/// One folder's pass, contained: a panic in it sets the folder to `error` and
+/// leaves the engine running for the next one.
+fn run_pass(core: &Core, folder: &StoredSyncFolder) {
+    if core
+        .sync_engine
+        .contain(folder.id, || core.reconcile_folder(folder))
+    {
+        return;
+    }
+    core.progress_end(folder.id);
+    let _ = core
+        .db
+        .sync_folder_set_state(folder.id, "error", folder.last_sync);
+    error!(
+        id = folder.id,
+        path = %folder.local_path,
+        "sync: pass panicked; the folder is set to error and tried again on the next poll"
+    );
+    core.log_activity(
+        ActivityKind::Sync,
+        format!("sync failed for {}", base_name(&folder.local_path)),
+        "an internal error stopped the pass; it is tried again on the next poll",
+        false,
+    );
+}
+
+/// The pass a `Reconcile(id)` asked for. A folder removed since, or one the
+/// database cannot read, is said so rather than dropped in silence.
+fn run_pass_by_id(core: &Core, id: i64) {
+    match core.db.sync_folder_get(id) {
+        Ok(Some(folder)) => run_pass(core, &folder),
+        Ok(None) => debug!(id, "sync: folder removed before its pass; skipping"),
+        Err(e) => warn!(id, error = ?e, "sync: cannot read folder; skipping its pass"),
+    }
+}
+
 /// The engine control loop. Single-threaded so reconcile passes never overlap.
 fn engine_loop(core: Core, rx: Receiver<SyncMsg>) {
+    let health = core.sync_engine.clone();
+    let _running = Running::start(&health);
     // Paths the watcher currently covers, mapped to their folder id, so an event
     // path can be resolved back to the folder it belongs to.
-    let watched: Mutex<Vec<(PathBuf, i64)>> = Mutex::new(Vec::new());
+    let watched: Arc<Mutex<Vec<(PathBuf, i64)>>> = Arc::new(Mutex::new(Vec::new()));
 
     let tx_events = core.sync_tx.clone();
-    let mut watcher =
-        match notify::recommended_watcher(move |res: notify::Result<notify::Event>| {
-            let Ok(event) = res else { return };
-            // Ignore pure metadata/access noise; content changes are what matter.
-            if matches!(event.kind, notify::EventKind::Access(_)) {
-                return;
-            }
-            let _ = tx_events.send(SyncMsg::ReconcileAll);
-        }) {
-            Ok(w) => w,
+    let watched_events = watched.clone();
+    let handler = move |res: notify::Result<notify::Event>| {
+        let event = match res {
+            Ok(event) => event,
+            // Running out of inotify watches for a new subfolder, say: that
+            // subfolder's changes are then found by the poll only (B262).
             Err(e) => {
-                warn!(error = %e, "filesystem watcher unavailable; sync is poll-only");
-                // Still run the loop so polling and explicit SyncNow work.
-                poll_only_loop(&core, rx);
+                warn!(error = %e, "sync: filesystem watcher error");
                 return;
             }
         };
+        // Ignore pure metadata/access noise; content changes are what matter.
+        if matches!(event.kind, notify::EventKind::Access(_)) {
+            return;
+        }
+        let msg = event_target(&watched_events.lock(), &event.paths);
+        let _ = tx_events.send(msg);
+    };
+    // The local walk skips symlinks, so following them here would only watch
+    // trees outside the folder, and run passes over it for changes made there.
+    let config = notify::Config::default().with_follow_symlinks(false);
+    let mut watcher = match notify::RecommendedWatcher::new(handler, config) {
+        Ok(w) => w,
+        Err(e) => {
+            warn!(error = %e, "filesystem watcher unavailable; sync is poll-only");
+            // Still run the loop so polling and explicit SyncNow work.
+            poll_only_loop(&core, rx);
+            return;
+        }
+    };
 
     // Periodic remote poll.
     {
@@ -94,8 +232,9 @@ fn engine_loop(core: Core, rx: Receiver<SyncMsg>) {
             .ok();
     }
 
-    rewatch(&core, &mut watcher, &watched);
-    reconcile_all(&core);
+    let mut unwatched = rewatch(&core, &mut watcher, &watched);
+    let mut last_rewatch = Instant::now();
+    run_burst(&core, &HashSet::new(), true);
 
     while let Ok(msg) = rx.recv() {
         let mut ids: HashSet<i64> = HashSet::new();
@@ -108,18 +247,16 @@ fn engine_loop(core: Core, rx: Receiver<SyncMsg>) {
             return;
         }
 
+        // A folder whose watch failed is tried again now and then, not only
+        // when the folder set changes.
+        if unwatched && last_rewatch.elapsed() >= REWATCH_RETRY {
+            do_rewatch = true;
+        }
         if do_rewatch {
-            rewatch(&core, &mut watcher, &watched);
-        }
-        if all {
-            reconcile_all(&core);
-        } else {
-            for id in ids {
-                if let Ok(Some(folder)) = core.db.sync_folder_get(id) {
-                    core.reconcile_folder(&folder);
-                }
-            }
+            unwatched = rewatch(&core, &mut watcher, &watched);
+            last_rewatch = Instant::now();
         }
+        run_burst(&core, &ids, all);
     }
 }
 
@@ -136,7 +273,7 @@ fn poll_only_loop(core: &Core, rx: Receiver<SyncMsg>) {
             }
         });
     }
-    reconcile_all(core);
+    run_burst(core, &HashSet::new(), true);
     while let Ok(msg) = rx.recv() {
         let mut ids: HashSet<i64> = HashSet::new();
         let mut all = false;
@@ -147,18 +284,59 @@ fn poll_only_loop(core: &Core, rx: Receiver<SyncMsg>) {
             info!("sync engine stopping");
             return;
         }
-        if all {
-            reconcile_all(core);
-        } else {
-            for id in ids {
-                if let Ok(Some(folder)) = core.db.sync_folder_get(id) {
-                    core.reconcile_folder(&folder);
-                }
+        run_burst(core, &ids, all);
+    }
+}
+
+/// Run the passes one burst asked for: the folders named by id, then, when the
+/// burst also holds a `ReconcileAll`, every other folder.
+///
+/// The named folders go first. A poll's `ReconcileAll` in the same burst used
+/// to take their place, and a folder just added, which has the highest id,
+/// waited behind a full pass over every other folder (`docs/BUGS.md` B260).
+///
+/// Skipped while sync is paused, and cut short when a pause lands between
+/// passes: the resume sends a fresh [`SyncMsg::ReconcileAll`], which walks
+/// everything the watcher reported meanwhile.
+fn run_burst(core: &Core, ids: &HashSet<i64>, all: bool) {
+    if core.sync_paused() {
+        debug!("sync: paused, skipping reconcile");
+        return;
+    }
+    let folders = if all {
+        match core.db.sync_folder_list() {
+            Ok(folders) => folders,
+            Err(e) => {
+                warn!(error = ?e, "sync: cannot list folders");
+                Vec::new()
             }
         }
+    } else {
+        Vec::new()
+    };
+    let listed: Vec<i64> = folders.iter().map(|f| f.id).collect();
+    for id in burst_order(ids, &listed) {
+        if core.sync_paused() {
+            break;
+        }
+        // A folder named by id is read again: it may have been added or
+        // changed after the list was read.
+        match folders.iter().find(|f| f.id == id) {
+            Some(folder) if !ids.contains(&id) => run_pass(core, folder),
+            _ => run_pass_by_id(core, id),
+        }
     }
 }
 
+/// The order of a burst's passes: the folders named by id, lowest first, then
+/// the rest of `listed` in its own order.
+pub(super) fn burst_order(ids: &HashSet<i64>, listed: &[i64]) -> Vec<i64> {
+    let mut order: Vec<i64> = ids.iter().copied().collect();
+    order.sort_unstable();
+    order.extend(listed.iter().copied().filter(|id| !ids.contains(id)));
+    order
+}
+
 /// Absorb the rest of an event burst, returning once the watcher has been quiet
 /// for `quiet` or `cap` has elapsed since the first event — whichever comes
 /// first. The caller has already classified the event that opened the burst.
@@ -230,27 +408,37 @@ pub(super) fn classify(
     true
 }
 
-/// Reconcile every mirror folder in turn.
-fn reconcile_all(core: &Core) {
-    let folders = match core.db.sync_folder_list() {
-        Ok(f) => f,
-        Err(e) => {
-            warn!(error = ?e, "sync: cannot list folders");
-            return;
-        }
-    };
-    for folder in folders {
-        core.reconcile_folder(&folder);
+/// The reconcile a watcher event asks for: the one mirror folder all its paths
+/// lie in, so an edit in one folder does not walk every other one too. An event
+/// that spans folders, or that no watched folder claims (a watch list a moment
+/// out of date), falls back to every folder.
+pub(super) fn event_target(watched: &[(PathBuf, i64)], paths: &[PathBuf]) -> SyncMsg {
+    let mut ids = paths.iter().map(|path| {
+        watched
+            .iter()
+            .filter(|(root, _)| path.starts_with(root))
+            // The deepest root, should one mirror folder sit inside another.
+            .max_by_key(|(root, _)| root.components().count())
+            .map(|(_, id)| *id)
+    });
+    match ids.next() {
+        Some(Some(first)) if ids.all(|id| id == Some(first)) => SyncMsg::Reconcile(first),
+        _ => SyncMsg::ReconcileAll,
     }
 }
 
 /// Bring the filesystem watches in line with the current mirror-folder set:
-/// watch newly-added folders, drop removed ones.
+/// watch newly-added folders, drop removed ones. Returns whether a folder is
+/// left unwatched because its watch failed.
+///
+/// The watch calls run on a copy, outside the lock: the watcher's event
+/// callback takes the same lock, and a backend that hands a watch request to
+/// its event thread and waits for the answer would deadlock against it.
 fn rewatch(
     core: &Core,
     watcher: &mut notify::RecommendedWatcher,
     watched: &Mutex<Vec<(PathBuf, i64)>>,
-) {
+) -> bool {
     let want: Vec<(PathBuf, i64)> = match core.db.sync_folder_list() {
         Ok(f) => f
             .into_iter()
@@ -259,29 +447,60 @@ fn rewatch(
             .collect(),
         Err(e) => {
             warn!(error = ?e, "sync: cannot list folders to watch");
-            return;
+            return false;
         }
     };
-    let mut have = watched.lock();
+    let mut have = watched.lock().clone();
     // Drop watches no longer wanted.
-    have.retain(|(path, _)| {
-        if want.iter().any(|(p, _)| p == path) {
-            true
-        } else {
-            let _ = watcher.unwatch(path);
-            false
-        }
-    });
+    for path in keep_wanted(&mut have, &want) {
+        let _ = watcher.unwatch(&path);
+    }
     // Add newly wanted watches.
+    let mut failed = false;
     for (path, id) in &want {
         if have.iter().any(|(p, _)| p == path) {
             continue;
         }
         match watcher.watch(path, RecursiveMode::Recursive) {
             Ok(()) => have.push((path.clone(), *id)),
-            Err(e) => warn!(path = %path.display(), error = %e, "sync: watch failed"),
+            Err(e) => {
+                // A recursive watch that fails partway keeps the watches it had
+                // added. Their events would match no folder and run a pass over
+                // every one, and they hold watches the next folder needs (B262).
+                let _ = watcher.unwatch(path);
+                failed = true;
+                warn!(
+                    path = %path.display(),
+                    error = %e,
+                    "sync: watch failed; the poll finds this folder's changes until a retry \
+                     works. A low fs.inotify.max_user_watches is the usual cause"
+                );
+            }
         }
     }
+    *watched.lock() = have;
+    failed
+}
+
+/// Keep the watches in `have` that `want` still asks for, and return the paths
+/// of the others, to unwatch.
+///
+/// A path is matched alone, but takes the id `want` gives it: a folder removed
+/// and added again in one burst keeps its watch, and its events go to the new
+/// row rather than to the removed one (`docs/BUGS.md` B272).
+pub(super) fn keep_wanted(have: &mut Vec<(PathBuf, i64)>, want: &[(PathBuf, i64)]) -> Vec<PathBuf> {
+    let mut dropped = Vec::new();
+    have.retain_mut(|(path, id)| match want.iter().find(|(p, _)| p == path) {
+        Some((_, wanted)) => {
+            *id = *wanted;
+            true
+        }
+        None => {
+            dropped.push(path.clone());
+            false
+        }
+    });
+    dropped
 }
 
 // ---- reconcile ------------------------------------------------------------

crates/pdfs-fuse/src/sync/model.rs

@@ -20,7 +20,7 @@ impl LocalSig {
     /// Nanoseconds decide it when both sides have them; otherwise the comparison
     /// falls back to whole seconds, which is what a baseline row written before
     /// schema 28 — or a filesystem that reports no sub-second time — can support
-    /// (bugs.md B25).
+    /// (`docs/BUGS.md` B25).
     pub(super) fn same_content(&self, other: &Self) -> bool {
         if self.size != other.size {
             return false;
@@ -83,6 +83,46 @@ pub(super) struct RemoteItem {
     pub(super) is_dir: bool,
     pub(super) mtime: i64,
     pub(super) size: i64,
+    /// The active revision's id. The light walk carries it for every file, so
+    /// it is the signal that tells another client's new revision from a link
+    /// merely re-stamped.
+    pub(super) revision: Option<String>,
+    /// The active revision's content SHA-1, read only for files the walk
+    /// enumerated in full (no baseline, or a moved mtime). `None` is never
+    /// proof of anything.
+    pub(super) sha1: Option<String>,
+}
+
+/// What a pass saw of a remote file: the signature its baseline records once
+/// the file is synced, and what an upload checks Drive still holds before it
+/// supersedes it.
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub(super) struct RemoteSeen {
+    pub(super) mtime: i64,
+    pub(super) size: i64,
+    pub(super) revision: Option<String>,
+}
+
+impl RemoteSeen {
+    /// Whether `node` is still the revision this was taken from. The revision
+    /// id decides it when both sides have one; otherwise the modification time
+    /// and size, which is all a baseline row from before schema 44 holds.
+    pub(super) fn still_current(&self, node: &Node) -> bool {
+        match (&self.revision, crate::node_revision_id(node)) {
+            (Some(seen), Some(now)) => *seen == now,
+            _ => (self.mtime, self.size) == remote_file_sig(node),
+        }
+    }
+}
+
+impl From<&RemoteItem> for RemoteSeen {
+    fn from(item: &RemoteItem) -> Self {
+        Self {
+            mtime: item.mtime,
+            size: item.size,
+            revision: item.revision.clone(),
+        }
+    }
 }
 
 /// The result of a reconcile pass: what it moved, how many paths were kept as
@@ -98,6 +138,11 @@ pub(super) struct Outcome {
     pub(super) errors: usize,
     /// Files skipped because another process held them open for writing.
     pub(super) deferred: usize,
+    /// Paths the local scan could not read, left out of the pass. Not part of
+    /// [`is_empty`](Self::is_empty) or the summary: an unreadable folder stays
+    /// unreadable from one poll to the next, so the pass reports it once, when
+    /// the set changes, rather than on every poll.
+    pub(super) unscanned: Vec<String>,
 }
 
 impl Outcome {
@@ -108,6 +153,7 @@ impl Outcome {
             Applied::Uploaded => self.uploaded += 1,
             Applied::Downloaded => self.downloaded += 1,
             Applied::Conflict => self.conflicts += 1,
+            Applied::Vanished | Applied::Adopted => {}
         }
     }
 
@@ -151,29 +197,45 @@ pub(super) enum Pending {
     CreateDir { rel: String, parent: NodeUid },
     /// Upload a brand-new local file into `parent`.
     UploadNew { rel: String, parent: NodeUid },
-    /// Upload a changed local file as a new revision of `uid`.
-    UploadRevision { rel: String, uid: NodeUid },
-    /// Download remote `uid` to the local path, stamping `mtime`. `size` is the
-    /// remote's reported size, used as the transfer's expected total.
+    /// Upload a changed local file as a new revision of `uid`, if Drive still
+    /// holds the revision `seen` describes (B249).
+    UploadRevision {
+        rel: String,
+        uid: NodeUid,
+        seen: RemoteSeen,
+    },
+    /// Download remote `uid` to the local path, stamping `seen.mtime`.
+    /// `seen.size` is the remote's reported size, used as the transfer's
+    /// expected total.
     ///
     /// `local` is what the planning walk saw at this path, or `None` if nothing
     /// was there. A download can run for minutes, so the apply step revalidates
     /// against it immediately before it renames over the destination: a writer
     /// that finished an edit in the meantime must not be overwritten without a
-    /// conflict copy (bugs.md B39).
+    /// conflict copy (`docs/BUGS.md` B39).
     Download {
         rel: String,
         uid: NodeUid,
-        mtime: i64,
-        size: i64,
+        seen: RemoteSeen,
         local: Option<LocalSig>,
     },
+    /// First sync of a path present on both sides at the same size, with a
+    /// digest on Drive: hash the local file, and record the baseline if it holds
+    /// the same bytes. A different file of the same size goes the way of
+    /// [`Pending::Conflict`]: two machines syncing one folder must not read as
+    /// in step while they hold different content.
+    Adopt {
+        rel: String,
+        uid: NodeUid,
+        seen: RemoteSeen,
+        sha1: String,
+        local: LocalSig,
+    },
     /// Both sides changed: set the local copy aside, then download remote `uid`.
     Conflict {
         rel: String,
         uid: NodeUid,
-        mtime: i64,
-        size: i64,
+        seen: RemoteSeen,
     },
     /// Both sides changed during a [`push pass`](Core::push_pass): upload the local
     /// copy into `parent` under a conflict name, leaving the remote file as it is.
@@ -188,6 +250,7 @@ impl Pending {
             | Pending::UploadNew { rel, .. }
             | Pending::UploadRevision { rel, .. }
             | Pending::Download { rel, .. }
+            | Pending::Adopt { rel, .. }
             | Pending::Conflict { rel, .. }
             | Pending::PushConflict { rel, .. } => rel,
         }
@@ -200,7 +263,9 @@ impl Pending {
             Pending::UploadNew { .. }
             | Pending::UploadRevision { .. }
             | Pending::PushConflict { .. } => ActivityKind::Upload,
-            Pending::Download { .. } | Pending::Conflict { .. } => ActivityKind::Download,
+            Pending::Download { .. } | Pending::Adopt { .. } | Pending::Conflict { .. } => {
+                ActivityKind::Download
+            }
         }
     }
 
@@ -211,6 +276,7 @@ impl Pending {
             Pending::UploadNew { .. } => "new file",
             Pending::UploadRevision { .. } => "new version",
             Pending::Download { .. } => "from Drive",
+            Pending::Adopt { .. } => "already the same on Drive",
             Pending::Conflict { .. } => "local changes kept as a conflict copy",
             Pending::PushConflict { .. } => "local changes uploaded as a conflict copy",
         }
@@ -226,6 +292,10 @@ pub(super) enum PassAbort {
     Interrupted,
     /// The pass could not establish its diff (a walk or the baseline load failed).
     Failed(String),
+    /// The daemon is stopping. What the pass sent before the stop has landed or
+    /// will be cancelled with the runtime; nothing more is sent, and the next
+    /// start's pass picks up from the baseline.
+    Stopped,
 }
 
 impl From<String> for PassAbort {

crates/pdfs-fuse/src/sync/planner.rs

@@ -90,6 +90,96 @@ pub(super) fn filter_baseline<'a, B>(
         .collect()
 }
 
+/// The paths a local scan left out because it could not read them: a folder
+/// it has no permission for, an entry that vanished between the listing and
+/// its `stat`, a name that is not UTF-8.
+///
+/// A left-out path is not a deleted one. A pass drops everything an entry
+/// covers from classification, exactly as it drops ignored paths, so a
+/// baseline row under an unreadable folder is neither read as a local deletion
+/// nor answered with a download into a folder the pass cannot write. Failing
+/// the whole pass instead is what B55 did, and it left one root-owned folder
+/// stopping every other file in the tree from syncing (B196).
+#[derive(Default)]
+pub(super) struct Unscanned {
+    covered: HashSet<String>,
+    /// The subset worth telling the user about, with why: an entry that only
+    /// vanished mid-scan is not, since the next pass sees it settled.
+    unreadable: Vec<String>,
+}
+
+impl Unscanned {
+    /// Leave out `rel`, which vanished while the scan looked at it.
+    pub(super) fn vanished(&mut self, rel: String) {
+        self.covered.insert(rel);
+    }
+
+    /// Leave out `rel`, which cannot be read for `reason`.
+    pub(super) fn unreadable(&mut self, rel: String, reason: &str) {
+        self.unreadable.push(format!("{rel} ({reason})"));
+        self.covered.insert(rel);
+    }
+
+    /// Whether `rel` is a left-out path or lies under one.
+    pub(super) fn covers(&self, rel: &str) -> bool {
+        if self.covered.is_empty() {
+            return false;
+        }
+        let mut at = rel;
+        loop {
+            if self.covered.contains(at) {
+                return true;
+            }
+            match at.rfind('/') {
+                Some(i) => at = &at[..i],
+                None => return false,
+            }
+        }
+    }
+
+    /// The unreadable paths with their reasons, sorted, for the activity feed.
+    pub(super) fn report(&self) -> Vec<String> {
+        let mut lines = self.unreadable.clone();
+        lines.sort_unstable();
+        lines
+    }
+}
+
+/// The baseline paths a pass can see: [`filter_baseline`], minus everything
+/// under a path the scan could not read. This is the set [`guard_local_wipe`]
+/// checks and the set a push pass reads deletions from, since a path outside
+/// it is absent from the local walk by rule or by failure, never by loss.
+pub(super) fn tracked_baseline<'a, B>(
+    baseline: &'a HashMap<String, B>,
+    rules: &IgnoreRules,
+    unscanned: &Unscanned,
+) -> HashMap<String, &'a B> {
+    let mut tracked = filter_baseline(baseline, rules);
+    tracked.retain(|rel, _| !unscanned.covers(rel));
+    tracked
+}
+
+/// The baseline paths a push pass has to trash on Drive: tracked, and gone
+/// from the local walk. Shallowest first, so a trashed folder takes its
+/// children with it.
+///
+/// Built from [`tracked_baseline`] rather than from the raw baseline. Reading
+/// it raw trashed every synced path that later became ignored, because the walk
+/// skips those, and the switch to on-demand that the push pass runs for then
+/// deleted the local copy too (B197).
+pub(super) fn local_deletions<B, L>(
+    tracked: &HashMap<String, &B>,
+    local: &HashMap<String, L>,
+) -> Vec<String> {
+    let mut missing: Vec<String> = tracked
+        .keys()
+        .filter(|rel| !local.contains_key(*rel))
+        .cloned()
+        .collect();
+    missing.sort_by_key(|p| p.matches('/').count());
+    missing
+}
+
 /// Refuse to run a pass whose local side has vanished in its entirety.
 ///
 /// When every baseline path is absent locally, the likely cause is an unavailable
@@ -102,21 +192,180 @@ pub(super) fn guard_local_wipe<B, L>(
     if !baseline.is_empty() && baseline.keys().all(|rel| !local.contains_key(rel)) {
         return Err(format!(
             "every one of the {} synced paths is missing locally; refusing to trash \
-             them on Drive. Check that the folder is mounted and readable.",
+             them on Drive. Check that the folder is mounted and readable; to download \
+             them again, sync the folder from scratch.",
             baseline.len()
         ));
     }
     Ok(())
 }
 
+/// Refuse to run a pass whose remote side has vanished in its entirety: the
+/// remote twin of [`guard_local_wipe`].
+///
+/// An empty or truncated remote walk reads to the planner as "Drive deleted
+/// everything", which a pass answers by deleting the local copies. When not
+/// one synced path is left on Drive, a failed listing is far likelier than a
+/// deliberate wipe of the whole folder, so the pass stops instead.
+pub(super) fn guard_remote_wipe<B, R>(
+    baseline: &HashMap<String, B>,
+    remote: &HashMap<String, R>,
+) -> Result<(), String> {
+    if !baseline.is_empty() && baseline.keys().all(|rel| !remote.contains_key(rel)) {
+        return Err(format!(
+            "every one of the {} synced paths is missing on Proton Drive; refusing to \
+             delete them here. If they were deleted on purpose, remove the folder from sync; \
+             to upload them again, sync the folder from scratch.",
+            baseline.len()
+        ));
+    }
+    Ok(())
+}
+
+/// What a remote walk saw but could not read: a node the SDK left out because
+/// it would not decrypt, or a file whose size could not be read.
+///
+/// A path missing from the walk is read as deleted on Drive, so a pass must
+/// not classify one it only failed to read. Since a node that failed to decrypt
+/// has no name, all a walk can say is that a folder's listing came back short;
+/// every path under that folder that the walk did not return is then left out,
+/// as [`Unscanned`] leaves out what the local scan could not read.
+#[derive(Default)]
+pub(super) struct RemoteGaps {
+    /// Folders whose listing came back with fewer nodes than it named.
+    short: HashSet<String>,
+    /// Paths seen, but whose entry could not be read.
+    unread: HashSet<String>,
+    /// Folders whose listing failed outright, after the SDK's retries. Nothing
+    /// under one is known, so all of it is left out of the pass (B271).
+    unlisted: HashSet<String>,
+}
+
+impl RemoteGaps {
+    /// The listing of `folder` named nodes that did not come back.
+    pub(super) fn short_listing(&mut self, folder: &str) {
+        self.short.insert(folder.to_string());
+    }
+
+    /// `rel` is on Drive, but the walk could not read what it needs of it.
+    pub(super) fn unread(&mut self, rel: String) {
+        self.unread.insert(rel);
+    }
+
+    /// The listing of `folder` failed, so nothing under it was walked.
+    pub(super) fn unlisted(&mut self, folder: &str) {
+        self.unlisted.insert(folder.to_string());
+    }
+
+    pub(super) fn is_empty(&self) -> bool {
+        self.short.is_empty() && self.unread.is_empty() && self.unlisted.is_empty()
+    }
+
+    /// Whether `rel` may be missing from `remote` only because the walk could
+    /// not read it. A path the walk returned is never hidden.
+    pub(super) fn hides<R>(
+        &self,
+        rel: &str,
+        remote: &HashMap<String, R>,
+        remote_dirs: &HashMap<String, NodeUid>,
+    ) -> bool {
+        if self.is_empty() || remote.contains_key(rel) {
+            return false;
+        }
+        if self.unread.contains(rel) {
+            return true;
+        }
+        // The deepest folder the walk did list decides: under a short listing,
+        // anything not returned may be one of the nodes it could not read.
+        let mut at = parent_rel(rel);
+        loop {
+            if remote_dirs.contains_key(at) {
+                return self.short.contains(at) || self.unlisted.contains(at);
+            }
+            if at.is_empty() {
+                return false;
+            }
+            at = parent_rel(at);
+        }
+    }
+
+    /// The folders to name to the user, sorted.
+    pub(super) fn report(&self) -> Vec<String> {
+        let mut lines: Vec<String> = self
+            .short
+            .iter()
+            .map(|rel| {
+                let shown = if rel.is_empty() { "the folder" } else { rel };
+                format!("{shown} (items on Proton Drive that cannot be read)")
+            })
+            .chain(
+                self.unread
+                    .iter()
+                    .map(|rel| format!("{rel} (cannot be read on Proton Drive)")),
+            )
+            .chain(self.unlisted.iter().map(|rel| {
+                format!("{rel} (could not be listed on Proton Drive; tried again next pass)")
+            }))
+            .collect();
+        lines.sort_unstable();
+        lines
+    }
+}
+
+/// Why the remote folder at `rel` must not be trashed although it was deleted
+/// here: something under it is new or changed on Drive since the last sync,
+/// so it is another device's work that this computer has never had.
+///
+/// Trashing the folder would take that work with it, and the next pass would
+/// then delete the copy this pass just downloaded. `None` when everything under
+/// it is exactly what the baseline recorded.
+pub(super) fn remote_subtree_changed(
+    rel: &str,
+    remote: &HashMap<String, RemoteItem>,
+    baseline: &HashMap<String, StoredSyncEntry>,
+) -> Option<String> {
+    let prefix = format!("{rel}/");
+    let mut changed: Vec<&String> = remote
+        .iter()
+        .filter(|(path, _)| path.starts_with(&prefix))
+        .filter(|(path, item)| match baseline.get(*path) {
+            None => true,
+            Some(_) if item.is_dir => false,
+            Some(base) => {
+                remote_sig(base) != Some((item.mtime, item.size))
+                    || revision_moved(Some(base), item.revision.as_deref())
+            }
+        })
+        .map(|(path, _)| path)
+        .collect();
+    changed.sort_unstable();
+    changed
+        .first()
+        .map(|path| format!("{path} is new or changed on Proton Drive"))
+}
+
 /// The size the baseline recorded for `rel`, if its remote signature's mtime
-/// still matches `mtime`.
+/// still matches `mtime` and its revision is still `revision`.
+///
+/// A row without a revision id, while Drive reports one, cannot vouch for the
+/// revision, so the size is read for real. Such a row comes from before
+/// schema 44, or from an upload Drive did not list yet when it was read back.
+/// Another computer's revision in the same second as that upload kept the
+/// modification time, and only the size could show it (B254). The pass then
+/// records the id on the row, so this costs one read per row.
 pub(super) fn unchanged_remote_size(
     baseline: &HashMap<String, StoredSyncEntry>,
     rel: &str,
     mtime: i64,
+    revision: Option<&str>,
 ) -> Option<i64> {
-    match baseline.get(rel).and_then(remote_sig) {
+    let base = baseline.get(rel)?;
+    if revision_moved(Some(base), revision)
+        || (base.remote_revision_id.is_none() && revision.is_some())
+    {
+        return None;
+    }
+    match remote_sig(base) {
         Some((recorded, size)) if recorded == mtime => Some(size),
         _ => None,
     }
@@ -131,6 +380,17 @@ pub(super) fn join_rel(prefix: &str, name: &str) -> String {
     }
 }
 
+/// Whether Drive's active revision `now` is not the one the baseline row was
+/// synced at. Only a revision id on both sides can say so; a row from before
+/// schema 44, or a node read without one, leaves it to the modification time
+/// and size.
+pub(super) fn revision_moved(base: Option<&StoredSyncEntry>, now: Option<&str>) -> bool {
+    match (base.and_then(|b| b.remote_revision_id.as_deref()), now) {
+        (Some(recorded), Some(now)) => recorded != now,
+        _ => false,
+    }
+}
+
 /// The stored remote signature of a baseline row, if it has one.
 pub(super) fn remote_sig(e: &StoredSyncEntry) -> Option<(i64, i64)> {
     match (&e.remote_rev, &e.remote_hash) {
@@ -161,7 +421,8 @@ pub(super) fn rel_to_path(rel: &str) -> PathBuf {
     rel.split('/').collect()
 }
 
-/// The name for a conflict copy of `path`, e.g. `notes (sync-conflict 1700000000).txt`.
+/// The name for a conflict copy of `path`, e.g.
+/// `notes (sync-conflict laptop 1700000000).txt`.
 #[cfg(test)]
 pub(super) fn conflict_path(path: &Path, stamp: i64) -> PathBuf {
     conflict_path_with_suffix(path, stamp, 0)
@@ -170,16 +431,17 @@ pub(super) fn conflict_path(path: &Path, stamp: i64) -> PathBuf {
 /// As [`conflict_path`], with a deterministic suffix used when that name already
 /// exists. This keeps conflict preservation from ever replacing an older copy.
 pub(super) fn conflict_path_with_suffix(path: &Path, stamp: i64, suffix: u32) -> PathBuf {
+    conflict_path_as(path, crate::conflict_label(), stamp, suffix)
+}
+
+/// [`conflict_path_with_suffix`] with the machine's label given.
+pub(super) fn conflict_path_as(path: &Path, label: &str, stamp: i64, suffix: u32) -> PathBuf {
     let stem = path.file_stem().and_then(|s| s.to_str()).unwrap_or("file");
     let ext = path.extension().and_then(|s| s.to_str());
-    let suffix = if suffix == 0 {
-        String::new()
-    } else {
-        format!("-{suffix}")
-    };
+    let tag = crate::conflict_tag(label, stamp, suffix);
     let name = match ext {
-        Some(ext) => format!("{stem} (sync-conflict {stamp}{suffix}).{ext}"),
-        None => format!("{stem} (sync-conflict {stamp}{suffix})"),
+        Some(ext) => format!("{stem} (sync-conflict {tag}).{ext}"),
+        None => format!("{stem} (sync-conflict {tag})"),
     };
     match path.parent() {
         Some(dir) => dir.join(name),
@@ -200,11 +462,14 @@ pub(super) enum FilePlan {
     DeleteLocal,
     DeleteRemote,
     ForgetBaseline,
-    /// First sync of a file that already exists on both sides with identical
-    /// content: record the baseline without transferring anything. Without this,
+    /// First sync of a file that exists on both sides at the same size, with a
+    /// digest on Drive: record the baseline without transferring anything if
+    /// the local file hashes the same, else keep a conflict copy. Without this,
     /// a folder synced for the first time cuts a `(sync-conflict)` copy of every
     /// pre-existing file, because a missing baseline makes both sides read as
-    /// "changed" (see the mass first-sync conflict storm in docs/BUGS.md).
+    /// "changed" (see the mass first-sync conflict storm in docs/BUGS.md). Size
+    /// alone is not enough: a second machine joining the folder would read a
+    /// different file of the same size as in step (B248).
     AdoptBaseline,
 }
 
@@ -218,13 +483,18 @@ pub(super) fn plan_file(
     }
 
     let local_changed = local.is_some_and(|item| {
-        // Nanoseconds where both sides have them (bugs.md B25): whole seconds
+        // Nanoseconds where both sides have them (`docs/BUGS.md` B25): whole seconds
         // could not see an edit made in the same second as the last sync that
         // left the file the same length, so it was never uploaded.
         baseline.is_none_or(|base| !LocalSig::from(item).same_content(&LocalSig::from(base)))
     });
+    // Another client's new revision shows in its id even when the link's
+    // modification time and the size come out the same.
     let remote_changed = remote.is_some_and(|item| {
-        baseline.is_none_or(|base| remote_sig(base) != Some((item.mtime, item.size)))
+        baseline.is_none_or(|base| {
+            remote_sig(base) != Some((item.mtime, item.size))
+                || revision_moved(Some(base), item.revision.as_deref())
+        })
     });
 
     match (local, remote) {
@@ -234,10 +504,13 @@ pub(super) fn plan_file(
             (false, true) => FilePlan::Download,
             // With a baseline, both-changed is a genuine concurrent edit and must
             // be preserved as a conflict. Without one, "both changed" only means
-            // this is the first sync and there was nothing to compare to: adopt
-            // the baseline when the sizes already match rather than conflict-copy
-            // an already-identical file. Divergent sizes stay a real conflict.
-            (true, true) if baseline.is_none() && l.size == r.size => FilePlan::AdoptBaseline,
+            // this is the first sync and there was nothing to compare to: when the
+            // sizes match and Drive has a digest to compare against, the apply
+            // step hashes the local file before adopting it. Divergent sizes, or
+            // no digest, stay a real conflict: a missing digest proves nothing.
+            (true, true) if baseline.is_none() && l.size == r.size && r.sha1.is_some() => {
+                FilePlan::AdoptBaseline
+            }
             (true, true) => FilePlan::Conflict,
         },
         (Some(_), None) if baseline.is_none() || local_changed => FilePlan::UploadNew,
@@ -269,6 +542,8 @@ mod file_plan_tests {
             is_dir: false,
             mtime,
             size,
+            revision: None,
+            sha1: None,
         }
     }
 
@@ -286,12 +561,13 @@ mod file_plan_tests {
             local_size,
             remote_rev: Some(remote_mtime.to_string()),
             remote_hash: Some(remote_size.to_string()),
+            remote_revision_id: None,
         }
     }
 
     /// Whole seconds cannot see an edit made inside the same second that leaves the
     /// file the same length, so a mirror silently stopped uploading it. With a
-    /// nanosecond time on both sides it is an ordinary local change (bugs.md B25).
+    /// nanosecond time on both sides it is an ordinary local change (`docs/BUGS.md` B25).
     #[test]
     fn a_same_second_same_size_edit_is_still_a_local_change() {
         let mut base = baseline(10, 20, 30, 40);
@@ -407,12 +683,28 @@ mod file_plan_tests {
     fn first_sync_of_an_identical_file_adopts_the_baseline_not_a_conflict() {
         // No baseline yet (first sync). Same size on both sides is a pre-existing
         // identical file, not a concurrent edit: adopt it rather than cut a
-        // conflict copy of every file in the folder.
+        // conflict copy of every file in the folder. The apply step still
+        // hashes the local file before it records anything.
+        let local = local(10, 40);
+        let remote = RemoteItem {
+            sha1: Some("da39a3ee".into()),
+            ..remote(30, 40)
+        };
+        assert_eq!(
+            plan_file(Some(&local), Some(&remote), None),
+            FilePlan::AdoptBaseline
+        );
+    }
+
+    /// A file on Drive without a digest cannot be compared without a download,
+    /// and the same size is not the same content (B248).
+    #[test]
+    fn first_sync_without_a_remote_digest_is_a_conflict() {
         let local = local(10, 40);
         let remote = remote(30, 40);
         assert_eq!(
             plan_file(Some(&local), Some(&remote), None),
-            FilePlan::AdoptBaseline
+            FilePlan::Conflict
         );
     }
 
@@ -428,6 +720,73 @@ mod file_plan_tests {
         );
     }
 
+    /// Another computer's new revision can leave the modification time and
+    /// the size as they were. The revision id still tells it apart (B249).
+    #[test]
+    fn a_new_remote_revision_is_a_change_even_at_the_same_mtime_and_size() {
+        let mut base = baseline(10, 20, 30, 40);
+        base.remote_revision_id = Some("r1".into());
+        let unedited = local(10, 20);
+        let same = RemoteItem {
+            revision: Some("r1".into()),
+            ..remote(30, 40)
+        };
+        let newer = RemoteItem {
+            revision: Some("r2".into()),
+            ..remote(30, 40)
+        };
+        assert_eq!(
+            plan_file(Some(&unedited), Some(&same), Some(&base)),
+            FilePlan::Unchanged
+        );
+        assert_eq!(
+            plan_file(Some(&unedited), Some(&newer), Some(&base)),
+            FilePlan::Download
+        );
+        let edited = local(11, 20);
+        assert_eq!(
+            plan_file(Some(&edited), Some(&newer), Some(&base)),
+            FilePlan::Conflict
+        );
+
+        // A row from before schema 44 has no id to compare, and a node read
+        // without one says nothing: both fall back to the signature.
+        let old = baseline(10, 20, 30, 40);
+        assert_eq!(
+            plan_file(Some(&unedited), Some(&newer), Some(&old)),
+            FilePlan::Unchanged
+        );
+        assert_eq!(
+            plan_file(Some(&unedited), Some(&remote(30, 40)), Some(&base)),
+            FilePlan::Unchanged
+        );
+    }
+
+    /// The walk reuses the recorded size only while the revision is the one
+    /// the baseline holds.
+    #[test]
+    fn a_moved_revision_makes_the_walk_read_the_size_again() {
+        let mut base = baseline(10, 20, 30, 40);
+        base.remote_revision_id = Some("r1".into());
+        let rows = HashMap::from([("file".to_string(), base)]);
+        assert_eq!(
+            unchanged_remote_size(&rows, "file", 30, Some("r1")),
+            Some(40)
+        );
+        assert_eq!(unchanged_remote_size(&rows, "file", 30, None), Some(40));
+        assert_eq!(unchanged_remote_size(&rows, "file", 30, Some("r2")), None);
+        assert_eq!(unchanged_remote_size(&rows, "file", 31, Some("r1")), None);
+    }
+
+    /// A row without a revision id reuses its size only while Drive gives no
+    /// id either (B254).
+    #[test]
+    fn a_row_without_a_revision_makes_the_walk_read_the_size_again() {
+        let rows = HashMap::from([("file".to_string(), baseline(10, 20, 30, 40))]);
+        assert_eq!(unchanged_remote_size(&rows, "file", 30, Some("r1")), None);
+        assert_eq!(unchanged_remote_size(&rows, "file", 30, None), Some(40));
+    }
+
     #[test]
     fn a_missing_remote_signature_is_treated_as_changed() {
         let local = local(10, 20);
@@ -440,4 +799,122 @@ mod file_plan_tests {
             FilePlan::Download
         );
     }
+
+    fn uid(id: &str) -> NodeUid {
+        NodeUid::new(VolumeId::from("v"), LinkId::from(id))
+    }
+
+    fn folder() -> RemoteItem {
+        RemoteItem {
+            is_dir: true,
+            mtime: 0,
+            size: 0,
+            ..remote(0, 0)
+        }
+    }
+
+    /// An empty or failed remote listing must not read as "Drive deleted
+    /// everything" and delete the local copies.
+    #[test]
+    fn the_remote_wipe_guard_blocks_only_a_total_disappearance() {
+        let base = HashMap::from([("a".to_string(), ()), ("b".to_string(), ())]);
+        let none: HashMap<String, RemoteItem> = HashMap::new();
+        assert!(guard_remote_wipe(&base, &none).is_err());
+        let one = HashMap::from([("b".to_string(), remote(1, 1))]);
+        assert!(guard_remote_wipe(&base, &one).is_ok());
+        let empty: HashMap<String, ()> = HashMap::new();
+        assert!(guard_remote_wipe(&empty, &none).is_ok());
+    }
+
+    /// A node that failed to decrypt has no name, so everything a short
+    /// listing did not return is left out, and nothing else.
+    #[test]
+    fn a_short_remote_listing_hides_what_it_did_not_return() {
+        let remote = HashMap::from([
+            ("docs".to_string(), folder()),
+            ("docs/a.txt".to_string(), remote(1, 1)),
+            ("pics".to_string(), folder()),
+        ]);
+        let dirs = HashMap::from([
+            (String::new(), uid("root")),
+            ("docs".to_string(), uid("docs")),
+            ("pics".to_string(), uid("pics")),
+        ]);
+        let mut gaps = RemoteGaps::default();
+        assert!(!gaps.hides("docs/b.txt", &remote, &dirs));
+        gaps.short_listing("docs");
+        // Returned paths stand; unreturned ones under the short folder hide,
+        // and so does anything under a subfolder it failed to return.
+        assert!(!gaps.hides("docs/a.txt", &remote, &dirs));
+        assert!(gaps.hides("docs/b.txt", &remote, &dirs));
+        assert!(gaps.hides("docs/sub/deep.txt", &remote, &dirs));
+        // A complete sibling listing still reads deletions as deletions.
+        assert!(!gaps.hides("pics/gone.jpg", &remote, &dirs));
+        assert!(!gaps.hides("top.txt", &remote, &dirs));
+        gaps.short_listing("");
+        assert!(gaps.hides("top.txt", &remote, &dirs));
+    }
+
+    /// A folder whose listing failed was walked no further: nothing under it
+    /// reads as deleted on Drive, and a sibling that listed still does
+    /// (`docs/BUGS.md` B271).
+    #[test]
+    fn a_folder_that_failed_to_list_hides_everything_under_it() {
+        let remote = HashMap::from([
+            ("docs".to_string(), folder()),
+            ("pics".to_string(), folder()),
+        ]);
+        let dirs = HashMap::from([
+            (String::new(), uid("root")),
+            ("docs".to_string(), uid("docs")),
+            ("pics".to_string(), uid("pics")),
+        ]);
+        let mut gaps = RemoteGaps::default();
+        gaps.unlisted("docs");
+        assert!(gaps.hides("docs/a.txt", &remote, &dirs));
+        assert!(gaps.hides("docs/sub/deep.txt", &remote, &dirs));
+        assert!(!gaps.hides("docs", &remote, &dirs));
+        assert!(!gaps.hides("pics/gone.jpg", &remote, &dirs));
+        assert_eq!(
+            gaps.report(),
+            ["docs (could not be listed on Proton Drive; tried again next pass)"]
+        );
+    }
+
+    #[test]
+    fn a_file_whose_size_could_not_be_read_is_hidden() {
+        let remote: HashMap<String, RemoteItem> = HashMap::new();
+        let dirs = HashMap::from([(String::new(), uid("root"))]);
+        let mut gaps = RemoteGaps::default();
+        gaps.unread("a.txt".to_string());
+        assert!(gaps.hides("a.txt", &remote, &dirs));
+        assert!(!gaps.hides("b.txt", &remote, &dirs));
+        assert_eq!(gaps.report(), ["a.txt (cannot be read on Proton Drive)"]);
+    }
+
+    /// A folder deleted here holds another device's new work on Drive, so
+    /// trashing it would lose that work.
+    #[test]
+    fn a_remote_folder_with_new_or_changed_files_is_not_trashed() {
+        let mut base = HashMap::from([
+            ("x".to_string(), baseline(0, 0, 0, 0)),
+            ("x/old.txt".to_string(), baseline(1, 1, 5, 6)),
+        ]);
+        let mut remote_tree = HashMap::from([
+            ("x".to_string(), folder()),
+            ("x/old.txt".to_string(), remote(5, 6)),
+            ("xy/new.txt".to_string(), remote(1, 1)),
+        ]);
+        // Unchanged under x; a name-prefixed sibling does not count.
+        assert_eq!(remote_subtree_changed("x", &remote_tree, &base), None);
+
+        remote_tree.insert("x/old.txt".to_string(), remote(9, 6));
+        assert!(remote_subtree_changed("x", &remote_tree, &base).is_some());
+
+        remote_tree.insert("x/old.txt".to_string(), remote(5, 6));
+        remote_tree.insert("x/sub".to_string(), folder());
+        assert!(remote_subtree_changed("x", &remote_tree, &base).is_some());
+        base.insert("x/sub".to_string(), baseline(0, 0, 0, 0));
+        assert_eq!(remote_subtree_changed("x", &remote_tree, &base), None);
+    }
 }

crates/pdfs-fuse/src/systemd.rs

@@ -64,16 +64,44 @@ fn send_abstract(socket: &UnixDatagram, name: &str, message: &str) -> bool {
 /// so `systemctl --user start` returns when the mount actually works rather than
 /// when the process exists.
 pub(crate) fn ready() {
-    if notify("READY=1\n") {
+    if notify("READY=1\nSTATUS=Mounted\n") {
         debug!("notified systemd: ready");
     }
 }
 
+/// Set the one-line status `systemctl --user status` shows for the unit, such
+/// as why the mount keeps failing (`docs/BUGS.md` B263).
+pub fn status(text: &str) {
+    // A newline would end the `STATUS=` line and start another variable.
+    let line = text.replace('\n', " ");
+    notify(&format!("STATUS={line}\n"));
+}
+
 /// Tell the manager the daemon is still healthy.
 pub(crate) fn watchdog_ping() {
     let _ = notify("WATCHDOG=1\n");
 }
 
+/// Sleep for `duration`, pinging the watchdog on the way.
+///
+/// For the stretches where the daemon is alive but has no mount, and so no
+/// supervisor: waiting for a login, or backing off after a failed mount. The
+/// manager arms the watchdog when the process starts, so without these pings a
+/// daemon that is merely retrying is aborted after `WatchdogSec` and leaves a
+/// coredump instead of the error it was logging. Never use this while mounted:
+/// there the ping must prove the control socket answers.
+pub fn idle_sleep(duration: Duration) {
+    const CHUNK: Duration = Duration::from_secs(10);
+    let mut left = duration;
+    watchdog_ping();
+    while !left.is_zero() {
+        let step = left.min(CHUNK);
+        std::thread::sleep(step);
+        left -= step;
+        watchdog_ping();
+    }
+}
+
 /// Tell the manager the daemon is stopping, so a slow teardown is not mistaken
 /// for a hang by the watchdog it is racing.
 pub(crate) fn stopping() {