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:

> On Mon 11-05-26 23:32:45, OGAWA Hirofumi wrote:
>> Jan Kara <jack@suse.cz> writes:
>> 
>> > Use mmb inode buffer writeout infrastructure to reliably write out
>> > inode's buffer on fsync(2).
>> 
>> > Signed-off-by: Jan Kara <jack@suse.cz>
>> > ---
>> >  fs/fat/inode.c | 3 ++-
>> >  1 file changed, 2 insertions(+), 1 deletion(-)
>> >
>> > diff --git a/fs/fat/inode.c b/fs/fat/inode.c
>> > index 28f78df086ef..4ca00b7a618b 100644
>> > --- a/fs/fat/inode.c
>> > +++ b/fs/fat/inode.c
>> > @@ -907,6 +907,7 @@ static int __fat_write_inode(struct inode *inode, int wait)
>> >  	}
>> >  	spin_unlock(&sbi->inode_hash_lock);
>> >  	mark_buffer_dirty(bh);
>> > +	MSDOS_I(inode)->i_metadata_bhs.inode_blk = bh->b_blocknr;
>> 
>> When inode position was changed/removed, this will point the wrong
>> block. And maybe sync a unrelated block and wait.
>
> So I didn't realize that e.g. rename does change the backing inode block.
> But given we set i_metadata_bhs.inode_blk on each inode write, inode_blk
> should always contain the current position where the inode was written so
> fsync should be syncing the right block. Or am I still missing something?

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.
-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>