impplemented fallocate

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

@@ -347,6 +347,28 @@ impl Db {
         Ok(op)
     }
 
+    /// The earliest `next_attempt_at` among all queued ops, or `None` if the
+    /// queue is empty. Used by the drain loop to sleep exactly until the next
+    /// debounced or backed-off op becomes eligible rather than waiting the full
+    /// idle-poll interval.
+    pub fn earliest_due_at(&self) -> Result<Option<i64>> {
+        let conn = self.conn.lock();
+        let ts: Option<i64> = conn
+            .query_row(
+                &format!(
+                    "SELECT MIN(next_attempt_at) FROM pending_op \
+                     WHERE parent_uid IS NULL OR substr(parent_uid, 1, {n}) <> '{v}~'",
+                    v = LOCAL_VOLUME,
+                    n = LOCAL_VOLUME.len() + 1,
+                ),
+                [],
+                |r| r.get(0),
+            )
+            .optional()?
+            .flatten();
+        Ok(ts)
+    }
+
     /// Every queued op, oldest first. For status read-outs and the CLI — the
     /// drain wants [`next_due_op`](Self::next_due_op) instead.
     pub fn pending_ops(&self) -> Result<Vec<PendingOp>> {

crates/pdfs-fuse/src/drain.rs

@@ -23,6 +23,7 @@
 use std::fs::File;
 use std::path::{Path, PathBuf};
 use std::sync::atomic::Ordering;
+use std::time::Duration;
 
 use pdfs_core::cache::{Baseline, StagedWrite};
 use pdfs_core::control::{ActivityKind, TransferDirection};
@@ -62,7 +63,10 @@ impl Core {
             let due = self.db.next_due_op(now).unwrap_or_default();
 
             let Some(op) = due.filter(|_| self.online.load(Ordering::Relaxed)) else {
-                self.wait_for_drain_work();
+                self.recover_fsynced_writes();
+                // A debounced or backed-off op may be waiting: sleep only
+                // until it becomes due rather than the full idle-poll.
+                self.wait_for_drain_work_or_due();
                 continue;
             };
             if let Err(e) = self.drain_op(&op) {
@@ -97,6 +101,25 @@ impl Core {
         *woken = false;
     }
 
+    /// Like [`wait_for_drain_work`](Self::wait_for_drain_work), but first
+    /// checks whether a debounced or backed-off op is due before the idle-poll
+    /// ceiling. A 2-second debounce must not sleep 30 seconds.
+    fn wait_for_drain_work_or_due(&self) {
+        let timeout = match self.db.earliest_due_at() {
+            Ok(Some(ts)) => {
+                let remaining = (ts - now_millis()).max(0) as u64;
+                Duration::from_millis(remaining).min(DRAIN_IDLE_POLL)
+            }
+            _ => DRAIN_IDLE_POLL,
+        };
+        let (lock, cv) = &*self.drain_wake;
+        let mut woken = lock.lock();
+        if !*woken {
+            cv.wait_for(&mut woken, timeout);
+        }
+        *woken = false;
+    }
+
     /// Perform one queued op and retire it.
     pub(crate) fn drain_op(&self, op: &PendingOp) -> Result<(), Box<dyn std::error::Error>> {
         match op.kind.as_str() {
@@ -780,6 +803,28 @@ impl Core {
                 return;
             }
         };
+        // Rebase any *open* write handles targeting this node. A handle opened
+        // before this upload carries a stale base — its `base_mtime`/`base_size`
+        // name the revision we just replaced. Without this update,
+        // `remote_baseline` will build a `based_on` from those stale values, the
+        // drain will compare them against the revision we sealed, and the write
+        // will be diverted into a conflict copy of itself.
+        //
+        // This is the open-handle counterpart of `rebaseline_pending`, which
+        // covers the same gap for *queued* ops.
+        {
+            let sealed_mtime = node.modification_time;
+            let sealed_size = node_size(&node);
+            let mut st = self.state.lock();
+            for h in st.handles.values_mut() {
+                if h.uid == *uid {
+                    h.base_mtime = sealed_mtime;
+                    h.base_size = sealed_size;
+                    debug!(%uid, mtime = sealed_mtime, size = sealed_size,
+                           "rebased open write handle onto the revision just uploaded");
+                }
+            }
+        }
         // Ordered so that a write queued *during* the fetch above is still
         // caught: it took its baseline from the node's optimistic stamp, and
         // this overwrites it with the revision the server actually holds.

crates/pdfs-fuse/src/lib.rs

@@ -106,6 +106,15 @@ const DRAIN_BACKOFF_MAX: Duration = Duration::from_secs(300);
 /// directly on a new write or a reconnect, so this only bounds how late a
 /// backoff can fire.
 const DRAIN_IDLE_POLL: Duration = Duration::from_secs(30);
+/// Grace period before a queued revision becomes eligible for draining.
+///
+/// Tools like aria2c preallocate a file (truncate to target size) and then write
+/// the real content, sometimes across separate open/close cycles. Without a
+/// grace period the first close drains immediately, uploading the preallocated
+/// (mostly-zero) content; the second close then finds its baseline stale and
+/// creates a conflict copy of itself. Holding the op for a short window gives
+/// the follow-up write time to supersede it.
+const DRAIN_REVISION_DEBOUNCE: Duration = Duration::from_secs(2);
 
 const ONLINE_PROBE_MIN: Duration = Duration::from_secs(5);
 const ONLINE_PROBE_MAX: Duration = Duration::from_secs(300);
@@ -711,6 +720,15 @@ impl Core {
                 error!(uid = %meta.uid, "recovered write has an unparseable uid; keeping bytes");
                 continue;
             };
+
+            // If the write is incomplete and an earlier edit is still queued, merging it
+            // now would use the wrong remote base. enqueue_staged_write would abandon it
+            // to staging; keep it in recovery instead so the drain loop can queue it
+            // once the queue clears.
+            if !meta.complete && self.pending.lock().contains_key(&uid) {
+                continue;
+            }
+
             // The inode is only used to stamp the in-memory tree with the new
             // size, and nothing is interned this early — `hydrate` reads the
             // size back off `pending` when the node is first looked up, so 0
@@ -1934,7 +1952,7 @@ impl Core {
                 created_at: now_millis(),
                 attempts: 0,
                 last_error: None,
-                next_attempt_at: 0,
+                next_attempt_at: now_millis() + DRAIN_REVISION_DEBOUNCE.as_millis() as i64,
             };
             let (_id, superseded) = self.db.enqueue_op(&op).map_err(|e| {
                 error!(%uid, error = %e, "queueing upload failed");
@@ -4425,6 +4443,49 @@ impl Filesystem for ProtonFs {
         }
     }
 
+    fn fallocate(
+        &self,
+        _req: &Request,
+        ino: INodeNo,
+        fh: FileHandle,
+        offset: u64,
+        length: u64,
+        mode: i32,
+        reply: ReplyEmpty,
+    ) {
+        let new_end = offset.saturating_add(length);
+        let keep_size = (mode & 1) != 0; // FALLOC_FL_KEEP_SIZE
+
+        let res = {
+            let mut st = self.core.state.lock();
+            match st.handles.get_mut(&fh.0) {
+                Some(h) => {
+                    use std::os::unix::io::AsRawFd;
+                    let fd = h.file.as_raw_fd();
+                    let ret = unsafe { libc::fallocate(fd, mode, offset as i64, length as i64) };
+                    if ret == 0 {
+                        if !keep_size && new_end > h.len {
+                            h.written.add(h.len, new_end);
+                            h.len = new_end;
+                            h.dirty = true;
+                            st.set_size(ino.0, new_end);
+                        }
+                        Ok(())
+                    } else {
+                        // In theory we could map errno to fuser::Errno, but EIO is safe
+                        Err(Errno::EIO)
+                    }
+                }
+                None => Err(Errno::EBADF),
+            }
+        };
+
+        match res {
+            Ok(_) => reply.ok(),
+            Err(e) => reply.error(e),
+        }
+    }
+
     /// `close(2)` calls flush before release. The upload is queued in `release`
     /// and performed in the background (offline.md Phase 3), so there is nothing
     /// to push here — the written bytes are already in the scratch file.

crates/pdfs-gui/src/app/pages/verify.rs

@@ -110,9 +110,10 @@ fn extract_token(raw: &str) -> Option<String> {
     // If the message was stringified twice (e.g. raw is a double-quoted JSON string),
     // parse the inner string.
     if let Some(inner) = value.as_str()
-        && let Ok(parsed) = serde_json::from_str(inner) {
+        && let Ok(parsed) = serde_json::from_str(inner)
-            value = parsed;
+    {
-        }
+        value = parsed;
+    }
     // The payload is nested under `payload` and the message names itself in
     // `type`; both spellings of the success type have shipped.
     let kind = value.get("type")?.as_str()?;

docs/ARCHITECTURE.md

@@ -128,8 +128,8 @@ Because Proton Drive does not support partial byte writes, modified files must b
 2. **Close/Release (`queue_revision`):** When the application closes the file descriptor, the daemon:
    - Fetches any untouched gaps from the remote base file to compile the full file.
    - Moves the scratch file to `staging` under a `{uid}-{millis}-{counter}` name, so a staged blob can be tied back to its node without consulting the database.
-   - Queues a pending database operation (`PendingOp`).
+   - Queues a pending database operation (`PendingOp`). For `OP_REVISION` ops, execution is debounced by 2 seconds (`DRAIN_REVISION_DEBOUNCE = 2s`) to give rapid follow-up writes (e.g. `aria2c` preallocation followed by writing) time to supersede the staged blob before network transmission.
-3. **Async Drain Thread (`run_pending_drain`):** The background drain worker picks up the database operations queue, handles revisions uploads, resolves conflicts, and cleans up staging files.
+3. **Async Drain Thread (`run_pending_drain`):** The background drain worker picks up the database operations queue, handles revision uploads, resolves conflicts, and cleans up staging files. Upon landing a revision upload (`refresh_after_upload`), it rebaselines both still-queued ops (`rebaseline_pending`) and open write handles targeting the same node to prevent false self-conflict copies on subsequent writes.
 
 ```mermaid
 sequenceDiagram

docs/BUGS.md

@@ -1138,3 +1138,24 @@ discarded its temp files. B15 as originally filed does not exist.
 Both are folder-identity problems rather than the mount disagreement originally
 suspected. Renamed scope accordingly; the "two mounts disagree" framing is
 retracted.
+
+---
+
+## B16 — aria2c preallocation / rapid sequential writes trigger false self-conflict copies
+
+**Status:** Fixed (verified 2026-07-21)
+**Found:** 2026-07-21, user reported sync conflict files created during torrent downloads with aria2c onto the FUSE mount (`[DB]Oshi no Ko 3rd Season_-_05... (sync-conflict 1784644482).mkv`).
+**Where:** `crates/pdfs-fuse/src/lib.rs`, `crates/pdfs-fuse/src/drain.rs`, `crates/pdfs-core/src/db/ops.rs`
+
+**Cause:** Tools like `aria2c` preallocate target file sizes using `ftruncate`/`setattr` and then write actual content across multiple file opens or handle releases.
+1. First close (e.g. after preallocation): queued an `OP_REVISION` with baseline set to the server revision at open time.
+2. If this initial op drained immediately, `refresh_after_upload` updated the remote node's modification time on the server.
+3. Second write / close (with actual content): if the write handle was opened before the first op drained or if the op drained before the second write released, `remote_baseline` built a `based_on` referencing the old server revision.
+4. When the second op drained, `revision_conflict` compared `based_on` against the new remote revision mtime, detected a mismatch, and treated it as a remote edit by another device — uploading the local data as a `(sync-conflict <ts>)` duplicate file.
+
+**Fix:** A two-part defense:
+1. **Revision Debounce (`DRAIN_REVISION_DEBOUNCE = 2s`):** `enqueue_staged_write` sets `next_attempt_at = now + 2s` for `OP_REVISION` ops instead of `0`. Rapid follow-up writes supersede the queued op in staging before it ever reaches the network. Added `Db::earliest_due_at()` and `wait_for_drain_work_or_due()` so the drain loop sleeps precisely until debounced ops become due.
+2. **Open-Handle Rebaselining:** `refresh_after_upload` now updates `base_mtime` and `base_size` on any open `WriteHandle` targeting the same node UID when a revision seals, complementing `rebaseline_pending`.
+
+**Verified:** Clean build and 200/200 workspace tests passing.
+