Skip to contentSkip to conversation

btrfs: don't force read-only after error in drop snapshot

PATCH In review David Sterba 3 participants 4 messages Feb 25, 2020 – Apr 10, 2020 Btrfs file system

lore
Share & tools
lore

David Sterba

author maintainer
#0 lore raw

PATCH 1/1 btrfs: don't force read-only after error in drop snapshot

Deleting a subvolume on a full filesystem leads to ENOSPC followed by a forced read-only. This is not a transaction abort and the filesystem is otherwise ok, so the error should be just propagated.

This is caused by unnecessary call to btrfs_handle_fs_error for almost all errors, except EAGAIN. This does not make sense as the standard transaction abort mechanism is in btrfs_drop_snapshot so all relevant failures are handled.

Originally in commit cb1b69f4508a ("Btrfs: forced readonly when btrfs_drop_snapshot() fails") there was no return value at all, so the btrfs_std_error made some sense but once the error handling and propagation has been we don't need it.

---

The use of btrfs_handle_fs_error in other places looks fishy, it makes sense only in case there's a real error and transaction abort is not possible, ~40 calls sound too much.

 fs/btrfs/extent-tree.c | 2 --
 1 file changed, 2 deletions(-)
fs/btrfs/extent-tree.c +0 -2
index 161274118853..b18db1b3a412 100644
@@ -5426,8 +5426,6 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
 	 */
 	if (!for_reloc && !root_dropped)
 		btrfs_add_dead_root(root);
-	if (err && err != -EAGAIN)
-		btrfs_handle_fs_error(fs_info, err, NULL);
 	return err;
 }
 
Signature · 1 line
2.25.0
#1 lore raw
Quoted 3 lines · David Sterba

Deleting a subvolume on a full filesystem leads to ENOSPC followed by a forced read-only. This is not a transaction abort and the filesystem is otherwise ok, so the error should be just propagated.

yep.

Quoted 25 lines

This is caused by unnecessary call to btrfs_handle_fs_error for almost all errors, except EAGAIN. This does not make sense as the standard transaction abort mechanism is in btrfs_drop_snapshot so all relevant failures are handled.

Originally in commit cb1b69f4508a ("Btrfs: forced readonly when btrfs_drop_snapshot() fails") there was no return value at all, so the btrfs_std_error made some sense but once the error handling and propagation has been we don't need it.

---

The use of btrfs_handle_fs_error in other places looks fishy, it makes sense only in case there's a real error and transaction abort is not possible, ~40 calls sound too much.

  fs/btrfs/extent-tree.c | 2 --
  1 file changed, 2 deletions(-)
fs/btrfs/extent-tree.c +0 -0
index 161274118853..b18db1b3a412 100644

fs/btrfs/extent-tree.c int btrfs_drop_snapshot(struct btrfs_root *root, · line 5430

Quoted 7 lines
fs/btrfs/extent-tree.c +0 -2
@@ -5426,8 +5426,6 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
  	 */
  	if (!for_reloc && !root_dropped)
  		btrfs_add_dead_root(root);
-	if (err && err != -EAGAIN)
-		btrfs_handle_fs_error(fs_info, err, NULL);
  	return err;
  }

However can we confirm that the error returned here are logged by its parents (relocation thread and the cleaner thread) ?

Thanks, Anand

#2 lore raw
Quoted 27 lines · Anand Jain
Quoted 3 lines · David Sterba

Deleting a subvolume on a full filesystem leads to ENOSPC followed by a forced read-only. This is not a transaction abort and the filesystem is otherwise ok, so the error should be just propagated.

yep.

Quoted 20 lines

This is caused by unnecessary call to btrfs_handle_fs_error for almost all errors, except EAGAIN. This does not make sense as the standard transaction abort mechanism is in btrfs_drop_snapshot so all relevant failures are handled.

Originally in commit cb1b69f4508a ("Btrfs: forced readonly when btrfs_drop_snapshot() fails") there was no return value at all, so the btrfs_std_error made some sense but once the error handling and propagation has been we don't need it.

---

The use of btrfs_handle_fs_error in other places looks fishy, it makes sense only in case there's a real error and transaction abort is not possible, ~40 calls sound too much.

  fs/btrfs/extent-tree.c | 2 --
  1 file changed, 2 deletions(-)
Quoted 15 lines

fs/btrfs/extent-tree.c int btrfs_drop_snapshot(struct btrfs_root *root, · line 5430

Quoted 12 lines
fs/btrfs/extent-tree.c +0 -2
index 161274118853..b18db1b3a412 100644
@@ -5426,8 +5426,6 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
  	 */
  	if (!for_reloc && !root_dropped)
  		btrfs_add_dead_root(root);
-	if (err && err != -EAGAIN)
-		btrfs_handle_fs_error(fs_info, err, NULL);
  	return err;
  }

However can we confirm that the error returned here are logged by its parents (relocation thread and the cleaner thread) ?

What do you mean logged? The error is propagated to the callers.

In btrfs_clean_one_deleted_snapshot it will result in not looping again to clean futher subvolumes, silent error should be ok here.

In clean_dirty_subvols the last error is propagated as the loop continues until all subvolumes from the list are processed, but reloc dropping snapshot can handle EAGAIN and ENOSPC will be the same.

#3 lore raw
In reply to David Sterba
Quoted 9 lines · David Sterba

Deleting a subvolume on a full filesystem leads to ENOSPC followed by a forced read-only. This is not a transaction abort and the filesystem is otherwise ok, so the error should be just propagated.

This is caused by unnecessary call to btrfs_handle_fs_error for almost all errors, except EAGAIN. This does not make sense as the standard transaction abort mechanism is in btrfs_drop_snapshot so all relevant failures are handled.

Quoted 6 lines

Originally in commit cb1b69f4508a ("Btrfs: forced readonly when btrfs_drop_snapshot() fails") there was no return value at all, so the btrfs_std_error made some sense but once the error handling and propagation has been we don't need it.

I'm hitting the ENOSPC and now also EINTR after the fast balance cancel patches so this is needed. If anybody has still concerns, please let me know, patch goes to misc-next.