On Thu, Feb 24, 2022 at 02:15:58AM +0000, Naohiro Aota wrote:
> On Wed, Feb 23, 2022 at 11:31:07AM +0100, David Sterba wrote:
> > On Fri, Feb 18, 2022 at 01:14:19PM +0900, Naohiro Aota wrote:
> > [ 2927.114871] BTRFS warning (device vdc: state EX): csum failed root 5 ino 258 off 131072 csum 0x7e797e3e expected csum 0x8941f998 mirror 2
> > [ 2927.115469] btrfs_balance+0x4ed/0x7e0 [btrfs]
> > [ 2927.118802] BTRFS warning (device vdc: state EX): csum failed root 5 ino 258 off 139264 csum 0x27df6522 expected csum 0x8941f998 mirror 2
> > [ 2927.119691] ? btrfs_balance+0x7e0/0x7e0 [btrfs]
> > [ 2927.123158] BTRFS warning (device vdc: state EX): csum failed root 5 ino 258 off 143360 csum 0x9f144c35 expected csum 0x8941f998 mirror 2
> > [ 2927.123965] balance_kthread+0x37/0x50 [btrfs]
>
> It looks like this occurs when the balance is resumed. We also need
> sb_{start,end}_write around btrfs_balance() in balance_kthred().
Sounds plausible.
> I guess we can cause a hang if we resume the balance and freeze the FS
> at the same time.
The background balance starts only when the filesystem is mounted for
write, so right after the sb_rdonly check in open_ctree, but I think
you're right that freeze during that can lead to a hang.
