btrfs: zoned: mark relocation as writing
authorNaohiro Aota <naohiro.aota@wdc.com>
Fri, 18 Feb 2022 04:14:19 +0000 (13:14 +0900)
committerDavid Sterba <dsterba@suse.com>
Mon, 28 Feb 2022 16:01:58 +0000 (17:01 +0100)
commit6e3068172e9e3e54d754c809c5418314b3f0be36
tree334a97e190f09d5268d64bed97f56181d7b0c527
parent1e19d3e20329189a96d91fa6d88fbec8ef637174
btrfs: zoned: mark relocation as writing

There is a hung_task issue with running generic/068 on an SMR
device. The hang occurs while a process is trying to thaw the
filesystem. The process is trying to take sb->s_umount to thaw the
FS. The lock is held by fsstress, which calls btrfs_sync_fs() and is
waiting for an ordered extent to finish. However, as the FS is frozen,
the ordered extents never finish.

Having an ordered extent while the FS is frozen is the root cause of
the hang. The ordered extent is initiated from btrfs_relocate_chunk()
which is called from btrfs_reclaim_bgs_work().

This commit adds sb_*_write() around btrfs_relocate_chunk() call
site. For the usual "btrfs balance" command, we already call it with
mnt_want_file() in btrfs_ioctl_balance().

Additionally, add an ASSERT in btrfs_relocate_chunk() to check it is
properly called.

Fixes: 18bb8bbf13c1 ("btrfs: zoned: automatically reclaim zones")
CC: stable@vger.kernel.org # 5.13+
Link: https://github.com/naota/linux/issues/56
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/block-group.c
fs/btrfs/volumes.c