Re: [PATCH v3] f2fs: don't leave the hashed inode while it's unlinked

[Wenjie Qi]

Hi Jaegeuk,

Yes, that's true for the normal path.

What I meant is the rollback case: err is non-zero, but __do_unlink()
may still return ret == 0. If the preceding f2fs_add_link() has already
been checkpointed, a crash before the rollback deletion is checkpointed
could expose the incomplete symlink again. Should we run f2fs_sync_fs()
for !ret && IS_DIRSYNC(dir), while preserving err?

Regards,
Wenjie