A remote change to a pinned file called `ContentCache::evict`, which does
not check pins, and only the pin request itself ever downloaded anything.
An edit from another device, or even a rename, left the file listed as kept
offline with no copy, so the next read downloaded it or failed without a
network. A file added to a pinned folder elsewhere was never downloaded,
and a folder pin whose downloads failed while offline never retried them.
A remote change no longer evicts a pinned file's copy. The copy is checked
against the revision's (mtime, size), so a newer revision does not read
it, and a rename keeps the mtime (B123), so the copy stays valid. A new
pdfs-pin-keeper thread checks pins again and downloads files that are
missing or stale:
- the pins a batch of remote events changed, after the batch's listings
are dropped, including a pinned folder that gets a new file;
- every pin at startup, when the link returns, and after the event feed
loses continuity;
- a file whose version is restored.
Failed downloads retry with a backoff of 30 s to 30 min. Pinning and the
keeper settle a provisional size first: a copy stored under the
ciphertext size never matched the plaintext it holds.
Fixes B282.
A daemon killed by the watchdog left its FUSE mounts dead. A dead mount
answers ENOTCONN, which create_dir_all took for an absent directory, so
every restart failed with EEXIST and retried forever without reaching the
code that clears stale mounts. mount_once now goes through
prepare_mountpoint, which clears a dead mount first.
Startup on a large account (70 s hydrating 48k nodes) plus the supervisor's
first ping one full period later outlasted WatchdogSec=120, so a restart
that did mount was aborted again. The daemon now asks for a 10-minute
watchdog before startup, the supervisor pings on its first tick, and the
unit's own timeout is restored once the control socket answers.
Records B279.
FUSE mounts over a non-empty folder without complaint, so files already
in the My files folder were hidden under the mount: not uploaded, and
back only once the service stopped, so they looked deleted (#67, B277).
- pdfs-core: new `mountpoint` module. `not_empty` lists what is in a
folder (hidden entries count); `mounted_fs_type` moves here from the
CLI so the daemon can use it. `MountFailure` gains an optional
`not_empty` field, which older readers ignore.
- pdfs-fuse: `mount_with` checks the folder before the root fetch and
before any worker starts. A dead mount is cleared first; one of our
own live mounts is left to `spawn_session`, as before.
- pdfs-cli: `cmd_daemon` records the folder in the failure record. The
refusal backs off and retries like any mount failure, so emptying the
folder is enough.
- pdfs-gui: the status pages explain the refusal, and a dialog offers
to open the folder or choose another one, once per run of failures.
The folder picker refuses a folder with files in it.
Translated in all catalogs. Docs: BUGS (B277), CHANGELOG, FAQ,
TROUBLESHOOTING, CONFIGURATION.
A mount that failed on every try retried every 5 s forever, and the
reason was only in the journal: the app and the tray said Connecting...
for days (B263, issue #40). The retry now doubles to a 2-minute ceiling,
and the last failure is kept in mount-failure.json for the app, the
tray, pdfs diagnose and systemctl status.
The early returns in mount left the workers they had started running
beside the next attempt (B264). They are now stopped on every error
path, without telling systemd the daemon is stopping.
One thread runs every synced folder's passes. A panic in a pass ended
it, and every later add, sync-now, file event and poll went nowhere
without a word until a restart, which fits issue #53 (B259).
Each pass now runs under catch_unwind; a panic sets the folder to error
and the next poll tries again. Diagnostics reports the engine's state,
the supervisor warns when it has ended or a pass has gone quiet, and
first passes log at info.
`pdfs unlock` resumes the daemon's session and may refresh it. The
daemon then holds a spent refresh token, and since B202 a refusal of it
read as an expired session although the keyring held a live one.
Before calling a refused session expired, compare the keyring with what
this process last stored. When it holds a newer rotation or another
login, end the mount with `MountOutcome::ResumeStored` and mount again
from the keyring. A rotation another process already wrote counts as
this session's, so the daemon keeps writing its later ones (B203).
A refresh token Proton refused, or a deleted or disabled account, was
treated as a lost connection: the mount came up "offline", the probe
retried forever and logged only at debug, and nothing told the user that
only signing in again helps (issue #35).
session_revoked() names those errors. The root fetch, probe, drain,
event poll, event seed and photos root now set session_expired, log one
warning naming `pdfs login`, and stop retrying. Cached files stay
readable and changes stay queued.
Response::Status carries session_expired (serde default false). The
tray shows "Session expired — sign in again" with a Sign In Again…
item that opens the app's sign-in page; the app's status and Sync card
say the same with a button. `pdfs status` and `pdfs diagnose` report it.
New strings are translated in every catalog.
docs/BUGS.md B202.
An unreadable subfolder, such as a container's root-owned cache, failed
every pass of its synced folder, so nothing else in the tree reached
Drive (#33, B196). The scan now leaves such a path out, together with an
entry that vanished mid-scan and a name that is not UTF-8, and the pass
drops everything under it from classification, the wipe guard and the
push pass's deletions. It is never read as deleted and never downloaded
into. An unreadable or absent root still fails the pass. The activity
feed names unreadable paths once each time the set changes.
Two related bugs on the way to online-only:
- The push pass read deletions from the raw baseline, so a synced path
that later became ignored was trashed on Drive (B197).
- The switch evicted everything once the folder was idle, including
ignored paths such as .git/, symlinks, unreadable folders and files
changed after the pass, none of which are on Drive (B198). It now
refuses and names the path.
Closes#33
66 comments cited `bugs.md B<n>`, the ledger's old name; they now say
`docs/BUGS.md B<n>`, and every cited number has an entry. About 90 more
cited offline.md, devices.md, features.md, mount-architecture.md,
plan.md and improvements.md, planning notes that were never in the
repository. Those citations are removed; the comments keep their text.
The known-bug confirmation run failed one_client_on_a_flaky_link seeds 6
and 11 on three bugs around a create whose answer was lost:
- B182: after a restart, a write over a file whose create landed but was
never read back forked a conflict copy. A revision of a file that has
no revision of Drive's now adopts the landed node first.
- B183: the SHA-1s a create sent were kept in memory, so after a restart
its trash or retry no longer knew the node Drive made. They are now on
the op's row (pending_op.sent_sha1, schema 42).
- B184: a rename rewrote the op's target, and the trash and the retry
looked for the node under the last name only. Each attempt's parent
and name are recorded (pending_op.sent_to); the trash looks under all
of them, and a retry adopts the node an earlier attempt made and moves
it to the current name.
An upload that landed let go of its blob before the tree had its node. A
write open in between took the base it sampled at its start, the revision
the upload replaced, found none of its bytes cached, and queued an
incomplete revision; the next write open failed with EIO.
The drain keeps the landing node aside until the tree has it, and a write
open with no queued revision takes its base from that node, or from the
tree as it is when the handle goes in. The fake Drive can hold the drain
in that window (`hold_next_landing`).
A stop joined every background thread for as long as each took. The sync
engine's mirror pass calls Drive with no deadline, and a stop that waited for
it was killed by systemd at TimeoutStopSec. The stop now waits up to 10 s for
the background threads and FUSE workers together, names any still running, and
the runtime shuts down within 2 s.
When the read-back after an upload went unanswered, the drain let go of
the staged blob without caching it. A partial write opened next had
nothing to fill its gaps from and queued an incomplete revision, so the
write open after it was refused with EIO.
The bytes are now cached under the size and mtime the tree still has,
which is the base the next write takes. Their SHA-1 stands in for the
revision id the read-back did not bring, so that write chains onto the
upload and does not fork a conflict copy of itself.
A new folder, a new file, a rename or a delete through the mount no longer
waits for Drive: it is recorded locally, the call returns at once, and the
drain sends it as it does offline. `"local_first": false` in config.json
brings back the old path for this release.
Fixes found by the simulation runs on the way:
- B129: a file deleted while its first upload is on the wire is trashed once
the upload lands.
- B132, B140: files made in a new folder survive a restart; a landed folder
hands its queued children over in the stored node too.
- B138: a create waits for the rename or trash that frees its name.
- B139: a file closed while its first upload lands keeps its last write.
- B141: a rename a worker holds stays queued, and the newer one waits behind
it instead of racing it.
- B142: a rename of a file whose create lands meanwhile goes to the uid Drive
gave it instead of answering EBUSY.
- B143: a create whose read-back fails as the link drops is adopted before
the next op for it is sent, so its next write is not a conflict copy.
On-demand sync folders kept their own State, which the drain, the
event applier and the sweeps had to visit one by one (B74). Every mount
now serves from the primary's tree. A session keeps only its root, which
the kernel knows as inode 1, and its notification channel; work that
changes the tree tells every mounted session. Size upgrades become
daemon-wide, and moves compare mounts by root instead of by state.
A node's inode is now derived from its row's local id instead of a
per-run counter, so the same node keeps the same inode across restarts
and across every place that interns it. A lid handed out before the
row is stored reads as no node.
Phase 1 of docs/MILESTONE-3.0.0.md, without a change in behaviour.
- Own sealed revisions live only in the own_sealed_rev table. The
in-memory map in front of it is gone, so revision_conflict gives the
same answer before and after a restart.
- A queued rename records the node's original name next to its original
parent, and keeps both across renames that replace the row. The drain
logs when Drive has the node neither where the rename found it nor
where it takes it.
The SDK's NodeUpdated event carries no revision id, so echoes of our own
changes stay recognised by the self-change timer.
mount() becomes a thin wrapper over mount_with(), which takes the
DriveApi to reach Drive through and a Host. The daemon's Host is
unchanged: stopped by SIGTERM or SIGINT, installed at the same point as
before, and indexing the user's home. A simulated daemon is stopped over
a channel and skips the home index, so several can share one test
process without walking the developer's home.
The account quota moves onto DriveApi. It was the only call outside the
trait that marks the link lost, so a simulated daemon would have gone
offline on it; the FakeDrive now also models a full account
(InsufficientQuota).
sim::daemon starts one daemon on its own state directory with a real
FUSE mount, and a smoke test (ignored; needs /dev/fuse) checks a local
write lands, a remote write arrives, and a write made offline lands once
the link is back.
`Core` held a concrete `ProtonDriveClient`, so no test could make Drive
answer late, answer stale or not answer. This is the first step of
phase 0 in docs/MILESTONE-3.0.0.md.
`DriveApi` (`crates/pdfs-fuse/src/drive.rs`) covers the calls the sync
core makes: listings, events, creates, uploads, downloads, renames,
moves, trashes and restores. `ProtonDriveClient` implements it by
forwarding. `Core::drive` holds it, and every call of those methods now
goes through it. Sharing, Photos, devices, invitations and public links
stay on `Core::client`.
The trait takes readers and writers as trait objects, and
`impl dyn DriveApi` puts the client's generic signatures back, so call
sites read the same. Block reads go through `RevisionRead` instead of
the SDK's `RevisionReader`, which only the SDK can construct.
No behaviour change.
The drain retired an upload's op first, and only then read the node back
and linked the staged blob into the content cache. For that one round
trip, the status reported an empty queue for a file that wasn't cached
yet, so anything that waited for an empty queue before reading the file
could still need the network. The account run of 2026-10-03 hit it over
a slow link: "regression B119" failed in the same second the upload
landed.
`Core::landing_uploads` now counts an upload from just before its op is
retired until its bytes are cached. The status adds it to
`pending_uploads`, and reads the queue before the count, so it always
sees the upload in one of them. A drop guard uncounts it on every exit
path.
The op still goes first. An op restored after a crash would re-send a
revision this daemon hasn't recorded as its own yet, and
`revision_conflict` would keep it as a conflict copy of the user's file.
The mount only ever went from offline to online: nothing marked it
offline again, so after the Wi-Fi dropped every create, mkdir, delete and
rename called Drive, waited out the SDK's retries for up to two minutes
and answered EIO. Background loops kept polling the dead link.
Connectivity now lives in pdfs-fuse's new link module and goes both
ways. A call that gets no answer (a transport error or a request
timeout) marks the mount offline; a probe thread checks every 2 s,
backing off to 30 s, and on recovery re-lists folders served from the
database, retries failed operations and wakes the drain. Interactive
calls get a 20 s deadline, so a dead link costs one slow call instead of
many.
While offline:
- create, mkdir, unlink, rmdir and rename fall back to the offline
queue instead of answering EIO;
- folders list from the database even when not marked listed, and are
re-listed once the link returns;
- reads of uncached data wait up to 30 s for the link, then retry once
within the 120 s read deadline;
- statfs serves the last quota instead of fetching, and quota refreshes
are single-flight;
- the event and photo event loops wait for the link instead of polling.
A create whose request timed out may have landed anyway. When the drain
replays it and finds a same-named twin of the same kind, created within
the last two minutes and still empty, it adopts that node instead of
making a conflict copy. The drain also leaves alone a name still held
by a queued trash.
The app and the tray stop polling. pdfs-core adds Request::Subscribe: the
connection stays open and carries Event lines (status, transfers, and
Changed with a Topic), with a heartbeat while quiet. In pdfs-fuse, an
EventHub gives each subscriber a mailbox, code that changes something
publishes its topic, and a sampler publishes the status, transfers and sync
progress when they differ. The app follows the feed on one thread, paints
status and transfers as they arrive, and reloads the page on screen when its
topic changes, including after changes made with the pdfs CLI. A hidden
window keeps the changes and catches up when focused. Front-ends fall back
to polling a daemon that predates subscriptions. Refresh buttons leave the
page headers; F5 and Ctrl+R still reload from the server.
Photos gets a scrubber overlaid on the right edge of the timeline in place
of the slider and scrollbar. It shows the years, appears when the pointer
comes near or the timeline scrolls, names the month under the pointer and
jumps to it on drag. Upload becomes a split button whose menu holds the
Google Photos import. Tiles are pooled per row and rebound instead of
rebuilt, hover scales the thumbnail to 1.02 without a shadow, and tiles
lose their tooltip.
On libadwaita 1.8 or newer the app uses AdwSpinner, AdwToggleGroup,
AdwButtonRow and AdwShortcutsDialog, and keeps look-alike stand-ins on
older releases (src/compat.rs, widgets/compat.rs). It still builds against
libadwaita 1.5 and picks at run time, creating the newer widgets by class
name through GtkBuilder, so every package gets them. PDFS_STAND_IN_WIDGETS
forces the stand-ins.
The Sync and share-link icons draw in full again: GTK drops group
transforms from symbolic icons, so their coordinates are baked into the
paths. All catalogs are updated and complete.
The duplicate finder gains an Exact / Similar switch. Similar asks the
daemon for PhotoSimilar, which starts a background pass that hashes the
thumbnail of every shot not hashed yet and replies at once with the sets
found so far and how far hashing has got. The dialog shows a progress bar
and asks again until the pass is done, then lists the sets.
The pass takes thumbnails from the server only: a photo without one is
skipped for the run instead of downloaded whole, and the dialog says how
many photos could not be compared. Decoding and hashing are spread over
the machine's cores.
Every similar photo starts ticked Keep, since photos that look alike can
be different shots. A similar photo stands for its whole shot, so one not
kept goes to Trash with the rest of its shot.
The first timeline refresh after the Places update reads the metadata of
the whole library, and until it is done the Places view has nothing to show
and no sign that anything is happening.
The refresh now records how many photos it has resolved out of how many,
counting photos read by an earlier pass as done. Response::Places carries
it as mapping (absent from older daemons, which parse as None). The Places
view shows "Mapping photos: 120 of 480" with a bar while it is set, asks
again every two seconds, and loads the new places once the pass ends. An
empty listing during the pass says the photos are being mapped rather than
that there are no places.
Every transfer charges the bytes it moves to its direction's rate limit,
a virtual clock shared by all transfers with a one-second burst, and sleeps
off any overshoot. On runtime workers the sleep goes through block_in_place
so FUSE reads keep their threads.
The caps are set live with SetBandwidthLimits, persisted to config and
applied at mount start. Preferences gains a Network group with debounced
MiB/s editors, and the CLI gains `pdfs sync limit --up/--down`.
Backend (docs/UI_UX_PLAN.md B-1, B-2):
- Request::SetSyncPaused { paused, until } pauses the drain worker and the
mirror reconcile. FUSE reads, write staging, the watcher and the poll keep
running. The pause is persisted in the `state` table (`sync_paused_until`),
so it survives a restart; a timed pause ends by itself through a timer
thread, re-armed on mount. Resuming wakes the drain and runs a full
reconcile. Response::Status gains `paused` and `paused_until`.
- Request::ListPendingOps lists the queue with paths, attempts, last error,
next attempt and parked/failing flags. Request::RetryPendingOp { id } makes
one backed-off op (or every failed one) due now. Parked ops are never
un-parked and nothing is dropped from the queue.
Front ends:
- Sync page: a status card with Pause/Resume (split button with 1 h, 8 h and
24 h pauses) and a "Waiting to Upload" list with Retry now and Retry All.
- The sidebar strip and the tray show the paused state; the tray gains Pause
Sync / Resume Sync and names failing changes.
- CLI: `pdfs sync pause [--for 30m|1h|2d]`, `pdfs sync resume`,
`pdfs sync queue`, `pdfs sync retry [id]`; `pdfs status` reports a pause.
Not in this phase: merging Activity into the Sync page, discarding a queued
op (needs a recovery copy first), and per-folder pause.
The gallery could show photos and nothing else: a photo deleted on the phone
stayed on the page for minutes, and a photo you wanted gone had to be deleted
on another device. Both are now handled here.
Deleting needs a uid-addressed request, because `Request::Delete` takes a
mountpoint path and the photos volume is not in the FUSE mount. `TrashNodes`
trashes through the SDK and answers per-uid outcomes, so a partial failure is
reportable; the daemon then drops those rows with `Db::photos_delete`, album
membership included, rather than leaving the gallery to wait for the next
timeline refresh. The GUI removes the tiles immediately with an Undo toast and
puts back anything the server refused — the photos are in Proton's trash and
stay recoverable there either way. Selection is Ctrl/Shift+click, the Select
toggle or Delete in the lightbox.
Remote deletions arrive on their own now. The Drive volume and the photos
volume have separate event streams, so the photos volume gets its own poller
and its own cursor. Trash and delete events remove rows; anything else only
marks the timeline stale, because there is no inode space to converge on this
volume. With the timeline TTL down to 60 s and an explicit Refresh that waits
for the server instead of only clearing the freshness stamp, a photo deleted
elsewhere leaves the grid in about ten seconds.
Teardown flagged the FUSE worker pool closed but never joined it, so the
process could drop the tokio runtime while a worker was still inside a
job — which is where 2026-09-18's "A Tokio 1.x context was found, but it
is being shutdown" panic came from. `Workers::stop_and_join` now waits
for the jobs in flight, bounded at 10 s so a worker wedged on the network
cannot hold the stop the user asked for; what is still stuck is named in
the log and shutdown continues around it.
The queues stay unbounded — the only caller that could feel backpressure
is fuser's dispatch loop, and blocking that is the stall the pool exists
to avoid — but a lane now logs its depth every 256 jobs, so a queue that
keeps climbing is visible before it is reported as a slow mount.
Documents the hang as B91, with the journal evidence, what made it
undiagnosable, and the candidates the evidence does not yet separate.
A supervisor thread ticks every 15 s and does three things: it WARNs
about a worker that has held one job for two minutes, a control request
that never returned, or a lane that has not completed anything while its
queue is non-empty; it samples resident size and the pending-op count
into the log every five minutes; and it answers the systemd watchdog.
The watchdog ping is deliberately not a heartbeat from a thread that
does nothing else — that would keep the unit "healthy" through exactly
the hang it is meant to catch. It goes out only after a real round trip
over the control socket, so it means "a client can still talk to this
daemon", which is the property that failed on 2026-09-18.
`sd_notify` is spoken directly rather than through libsystemd (READY,
WATCHDOG, STOPPING), including the abstract-namespace address; off
systemd every call is a no-op.
The unit gains `NotifyAccess=all` (rather than `Type=notify`, which
would fail the unit on the start timeout when no login is stored yet),
`WatchdogSec=120`, `Restart=always` — a hang is not a failing exit code
— and `MemoryHigh`/`MemoryMax` as a backstop against the 5.2 G growth
the hung run showed.
On 2026-09-18 the daemon hung: 43 minutes of CPU over 20 hours, 5.2 GiB
resident, silent from 17:10 until a manual restart the next day, and no
panic. Nothing in the journal said what it had been waiting on, and
nothing in the daemon could be asked.
- The worker pool publishes per-thread state: whether the thread is busy,
how long it has been in that state, and the label of the job in hand.
Every FUSE handler that hands work to the pool now names its job
(`read`, `lookup`, `rename`), so a stuck worker is named rather than
merely counted. Queue depths and per-lane completion counts are
mirrored outside the queue lock.
- `Request::Diagnostics` / `pdfs diagnostics` reports those, plus
resident size, the control handler count against its limit, the
in-flight control requests with their ages, and the pending-op count.
- Built to answer while the daemon is wedged: atomics and `try_lock`
only, and the one field that needs the database reads it on a throwaway
thread with a 500 ms deadline.
nvrl
emptyname