Re: [PATCH 6/9] fat: Fix possibly missing inode write on fsync(2)

[OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>]

Jan Kara <jack@suse.cz> writes:

>> I didn't check the case of rename completely, just recalled it when I
>> saw this code, need confirm/check.  But at least, the case of remove
>> will leave it even after the block is reused.
>
> Right. fat_detach() should set i_metadata_bhs.inode_blk to INVALID_BLK,
> thanks for catching that. I was thinking whether we should set
> i_metadata_bhs.inode_blk in fat_attach() instead of during inode dirtying.
> It would be somewhat more obviously correct but it could lead to
> unnecessary flushing in case the directory block gets dirtied by some other
> entry in it while the inode we are fsyncing got never dirtied. IMHO that's
> a sensible tradeoff so I'd do that but what is your opinion?

IMO, the marker should be cleared like b_assoc_buffers or I_DIRTY_*
flags after each sync. Otherwise, because the block is shared with other
inodes, it would sync/wait the unrelated dirty easily.

[And more serious implementation, looks like it should be cleared at
similar points or such with b_assoc_buffers is cleared to minimize
unrelated sync/wait.]
-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>