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

[Wenjie Qi]

Hi Jaegeuk,

Thanks, v3 addresses the two error propagation issues I reported.

I have one question about successful rollback in a DIRSYNC directory.
f2fs_unlink() runs f2fs_sync_fs() after __do_unlink() succeeds, but
f2fs_symlink() calls __do_unlink() directly and returns without a
checkpoint.

Since f2fs_add_link() has already completed before symlink construction
fails, could a checkpoint between the addition and rollback make the
addition durable while the deletion remains dirty? Should the successful
rollback run f2fs_sync_fs() when IS_DIRSYNC(dir), while keeping the
original construction error as the return value?

Regards,
Wenjie