Re: [PATCH] btrfs: don't force the filesystem read-only on EDQUOT/ENOSPC verity rollback

[Daniel Linjama]

On 2026/9/14 18:45, Qu Wenruo wrote:
> On 2026/9/14 16:02, Qu Wenruo wrote:
>> Please also submit a fstests case for it.

Sent to fstests as btrfs/354 (qgroup limit) and btrfs/355 (full
filesystem):
https://lore.kernel.org/fstests/20260915053815.307674-1-daniel@xxxxxxxxxxxxxxx/

>> EDQUOT is fine, but I'm not sure if ENOSPC is really recoverable.
>>
>> Even if we ignore ENOSPC for now, we may have really exhausted the
>> metadata, and the next transaction may be aborted anyway.
>
> Sashiko also mentioned even if we ignored EDQUOT for now, the next mount
> we still fallback to call btrfs_drop_verity_items() on the inode cleanup.
>
> And it will return -EDQUOT again, so the inode will not be cleanred up.
>
> Considering we're only delete items, I think the better solution would
> be calling btrfs_start_transaction_fallback_global_rsv(), which will not
> enforce qgroup, thus should not return -EDQUOT in the first place.

You are both right. With v1 the next mount fails the orphan cleanup with
-EDQUOT again, every lookup of the subvolume fails, and mounting it with
-o subvol fails too, so the subvolume is unreachable as long as the limit
is in place.

v2 does what you suggest: btrfs_start_transaction_fallback_global_rsv()
in drop_verity_items(), for the second transaction in rollback_verity()
and for the orphan item deletion in btrfs_orphan_cleanup(). All three are
needed, the reservations of the dropped items are only released at
commit, so the following transactions hit the limit as well.
https://lore.kernel.org/linux-btrfs/20260915055109.311839-1-daniel@xxxxxxxxxxxxxxx/

>> I believe your test case is only utilizing qgroup limit, or did you hit
>> a case where btrfs_drop_verity_items() returns -ENOSPC and the fs still
>> works fine?

Yes, on a full mixed block group filesystem the rollback fails with
-ENOSPC the same way. Without the fix the orphan cleanup then fails the
next read-write mount (open_ctree failed: -ENOSPC), and the filesystem
can only be mounted read-only after that. With v2 the mount succeeds and
the leftover items are dropped. btrfs/355 covers this.

Thanks,
Daniel