f2fs: quota: fix quota flush failure during filesystem freeze

Quota flush retries can be exhausted during filesystem freeze.

freeze_super() and the quota checkpoint may race as below.

freeze_super()                       f2fs_ckpt
- down_write(&sb->s_umount)
- sb->s_writers.frozen = SB_FREEZE_PAGEFAULT
- sync_filesystem()
 - f2fs_sync_fs()
  - f2fs_issue_checkpoint()
   - queue CP_SYNC -----------------> - block_operations()
                                      - down_read_trylock(s_umount)
                                        : fails; freeze holds write lock
                                      - quota flush retries exhausted
                                      - set CP_QUOTA_NEED_FSCK_FLAG

During the PAGEFAULT sync pass, freeze_super() holds s_umount for
write, but umount_lock_holder is not set until f2fs_freeze().  Thus
quota writeback tries to acquire s_umount again and can exhaust retries.

With checkpoint merge enabled, f2fs_issue_checkpoint() may dispatch
CP_SYNC to f2fs_ckpt. That thread also cannot acquire the freeze lock.

Record current as umount_lock_holder around the VFS PAGEFAULT sync so
checkpoint runs in the freeze caller and quota flush uses its existing
s_umount write lock.

Fixes: eb85c2410d6f ("f2fs: quota: fix to avoid warning in dquot_writeback_dquots()")
Signed-off-by: Jianan Huang <jnhuang95@gmail.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

0 files changed, 0 insertions, 0 deletions