* [pbs-devel] [PATCH proxmox-backup stable-3] fix #6566: backup: api: conditionally drop group and snapshot locks
@ 2025-10-01 11:22 Christian Ebner
2025-10-02 8:00 ` [pbs-devel] applied: " Fabian Grünbichler
0 siblings, 1 reply; 2+ messages in thread
From: Christian Ebner @ 2025-10-01 11:22 UTC (permalink / raw)
To: pbs-devel
To guarantee consistency by possible concurrent operations, the
backup protocol locks the backup group, the previous backup
snapshot (if any) and holds a lock for the newly created backup
snapshot. All of these are currently stored in the backup worker
task, only released on its destruction.
The backup API however signals a successful backup via the return
status of the `finish` call, while still holding the locks.
Therefore, an immediate subsequent backup of the client to the same
group can fail because the locks cannot be acquired until the previous
backup task is completely destroyed, which can however outlive the
`finish` return for some time. This manifests in e.g. a push sync job
failing.
To fix this, store the lock guards inside the RPC environments shared
state instead, allowing to selectively drop the locks on successful
backup finish. On error, hold the locks until the cleanup was
successful.
Immediate verification of new snapshots already downgraded the lock
by dropping the exclusive lock and getting a shared lock. Since the
dropping is now already handled by the finish call, only gathering
the shared lock is required. While there is now a larger time window
for concurrent prunes, the underlying possible race between
verification and prune remains in place.
Backported from https://git.proxmox.com/?p=proxmox-backup.git;a=commit;h=b1ece6c70c7785191321525576ed6f53e9c4bc18
Fixes: https://bugzilla.proxmox.com/show_bug.cgi?id=6566
Signed-off-by: Christian Ebner <c.ebner@proxmox.com>
---
src/api2/backup/environment.rs | 39 +++++++++++++++++++++++++++++-----
src/api2/backup/mod.rs | 18 +++++++++-------
2 files changed, 44 insertions(+), 13 deletions(-)
diff --git a/src/api2/backup/environment.rs b/src/api2/backup/environment.rs
index 3d541b461..c4251c008 100644
--- a/src/api2/backup/environment.rs
+++ b/src/api2/backup/environment.rs
@@ -85,6 +85,27 @@ struct SharedBackupState {
known_chunks: KnownChunksMap,
backup_size: u64, // sums up size of all files
backup_stat: UploadStatistic,
+ backup_lock_guards: BackupLockGuards,
+}
+
+pub struct BackupLockGuards {
+ previous_snapshot: Option<BackupLockGuard>,
+ group: Option<BackupLockGuard>,
+ snapshot: Option<BackupLockGuard>,
+}
+
+impl BackupLockGuards {
+ pub(crate) fn new(
+ previous_snapshot: Option<BackupLockGuard>,
+ group: BackupLockGuard,
+ snapshot: BackupLockGuard,
+ ) -> Self {
+ Self {
+ previous_snapshot,
+ group: Some(group),
+ snapshot: Some(snapshot),
+ }
+ }
}
impl SharedBackupState {
@@ -125,6 +146,7 @@ impl BackupEnvironment {
worker: Arc<WorkerTask>,
datastore: Arc<DataStore>,
backup_dir: BackupDir,
+ backup_lock_guards: BackupLockGuards,
) -> Self {
let state = SharedBackupState {
finished: false,
@@ -135,6 +157,7 @@ impl BackupEnvironment {
known_chunks: HashMap::new(),
backup_size: 0,
backup_stat: UploadStatistic::new(),
+ backup_lock_guards,
};
Self {
@@ -607,6 +630,9 @@ impl BackupEnvironment {
bail!("backup does not contain valid files (file count == 0)");
}
+ // drop previous snapshot lock
+ state.backup_lock_guards.previous_snapshot.take();
+
// check for valid manifest and store stats
let stats = serde_json::to_value(state.backup_stat)?;
self.backup_dir
@@ -630,13 +656,17 @@ impl BackupEnvironment {
// marks the backup as successful
state.finished = true;
+ // drop snapshot and group lock only here so any error above will lead to
+ // the locks still being held in the env for the backup cleanup.
+ state.backup_lock_guards.snapshot.take();
+ state.backup_lock_guards.group.take();
+
Ok(())
}
/// If verify-new is set on the datastore, this will run a new verify task
- /// for the backup. If not, this will return and also drop the passed lock
- /// immediately.
- pub fn verify_after_complete(&self, excl_snap_lock: BackupLockGuard) -> Result<(), Error> {
+ /// for the backup. If not, this will return.
+ pub fn verify_after_complete(&self) -> Result<(), Error> {
self.ensure_finished()?;
if !self.datastore.verify_new() {
@@ -644,8 +674,7 @@ impl BackupEnvironment {
return Ok(());
}
- // Downgrade to shared lock, the backup itself is finished
- drop(excl_snap_lock);
+ // Get shared lock, the backup itself is finished
let snap_lock = self.backup_dir.lock_shared().with_context(|| {
format!(
"while trying to verify snapshot '{:?}' after completion",
diff --git a/src/api2/backup/mod.rs b/src/api2/backup/mod.rs
index 629df933e..2f3fa7527 100644
--- a/src/api2/backup/mod.rs
+++ b/src/api2/backup/mod.rs
@@ -140,7 +140,7 @@ fn upgrade_to_backup_protocol(
};
// lock backup group to only allow one backup per group at a time
- let (owner, _group_guard) = datastore.create_locked_backup_group(
+ let (owner, group_guard) = datastore.create_locked_backup_group(
backup_group.backup_ns(),
backup_group.as_ref(),
&auth_id,
@@ -179,7 +179,7 @@ fn upgrade_to_backup_protocol(
let backup_dir = backup_group.backup_dir(backup_dir_arg.time)?;
- let _last_guard = if let Some(last) = &last_backup {
+ let last_guard = if let Some(last) = &last_backup {
if backup_dir.backup_time() <= last.backup_dir.backup_time() {
bail!("backup timestamp is older than last backup.");
}
@@ -205,12 +205,19 @@ fn upgrade_to_backup_protocol(
auth_id.to_string(),
true,
move |worker| {
+ // Keep flock for the backup runtime by storing guards in backup env shared state.
+ // Drop them on successful backup finish or when dropping the env after cleanup in
+ // case of errors. The former is required for immediate subsequent backups (e.g.
+ // during a push sync) to be able to lock the group and snapshots.
+ let backup_lock_guards = BackupLockGuards::new(last_guard, group_guard, snap_guard);
+
let mut env = BackupEnvironment::new(
env_type,
auth_id,
worker.clone(),
datastore,
backup_dir,
+ backup_lock_guards,
);
env.debug = debug;
@@ -264,11 +271,6 @@ fn upgrade_to_backup_protocol(
let mut abort_future = abort_future.map(|_| Err(format_err!("task aborted")));
async move {
- // keep flock until task ends
- let _group_guard = _group_guard;
- let snap_guard = snap_guard;
- let _last_guard = _last_guard;
-
let res = select! {
req = req_fut => req,
abrt = abort_future => abrt,
@@ -280,7 +282,7 @@ fn upgrade_to_backup_protocol(
}
let verify = |env: BackupEnvironment| {
- if let Err(err) = env.verify_after_complete(snap_guard) {
+ if let Err(err) = env.verify_after_complete() {
env.log(format!(
"backup finished, but starting the requested verify task failed: {}",
err
--
2.39.5
_______________________________________________
pbs-devel mailing list
pbs-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pbs-devel
^ permalink raw reply [flat|nested] 2+ messages in thread
* [pbs-devel] applied: [PATCH proxmox-backup stable-3] fix #6566: backup: api: conditionally drop group and snapshot locks
2025-10-01 11:22 [pbs-devel] [PATCH proxmox-backup stable-3] fix #6566: backup: api: conditionally drop group and snapshot locks Christian Ebner
@ 2025-10-02 8:00 ` Fabian Grünbichler
0 siblings, 0 replies; 2+ messages in thread
From: Fabian Grünbichler @ 2025-10-02 8:00 UTC (permalink / raw)
To: pbs-devel, Christian Ebner
On Wed, 01 Oct 2025 13:22:51 +0200, Christian Ebner wrote:
> To guarantee consistency by possible concurrent operations, the
> backup protocol locks the backup group, the previous backup
> snapshot (if any) and holds a lock for the newly created backup
> snapshot. All of these are currently stored in the backup worker
> task, only released on its destruction.
>
> The backup API however signals a successful backup via the return
> status of the `finish` call, while still holding the locks.
> Therefore, an immediate subsequent backup of the client to the same
> group can fail because the locks cannot be acquired until the previous
> backup task is completely destroyed, which can however outlive the
> `finish` return for some time. This manifests in e.g. a push sync job
> failing.
>
> [...]
Applied, thanks!
[1/1] fix #6566: backup: api: conditionally drop group and snapshot locks
commit: c096925472a4f4dee4bd119d0d72ca66ee44654a
Best regards,
--
Fabian Grünbichler <f.gruenbichler@proxmox.com>
_______________________________________________
pbs-devel mailing list
pbs-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pbs-devel
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2025-10-02 8:01 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-10-01 11:22 [pbs-devel] [PATCH proxmox-backup stable-3] fix #6566: backup: api: conditionally drop group and snapshot locks Christian Ebner
2025-10-02 8:00 ` [pbs-devel] applied: " Fabian Grünbichler
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.