Re: [PATCH] check bh->b_blocknr only if BH_Mapped is set

[Nikanth Karthikesan <knikanth@suse.de>]

On Wednesday 11 February 2009 02:16:40 Andrew Morton wrote:
> On Mon, 9 Feb 2009 17:48:11 +0530
>
> Nikanth Karthikesan <knikanth@suse.de> wrote:
> > check bh->b_blocknr only if BH_Mapped is set
>
> Why?
>

b_blocknr will have a valid blocknr only when BH_Mapped is set.

> > diff --git a/fs/buffer.c b/fs/buffer.c
> > index 665d446..782c365 100644
> > --- a/fs/buffer.c
> > +++ b/fs/buffer.c
> > @@ -344,13 +344,13 @@ __find_get_block_slow(struct block_device *bdev,
> > sector_t block)
> >  	head = page_buffers(page);
> >  	bh = head;
> >  	do {
> > -		if (bh->b_blocknr == block) {
> > +		if (!buffer_mapped(bh))
> > +			all_mapped = 0;
> > +		else if (bh->b_blocknr == block) {
> >  			ret = bh;
> >  			get_bh(bh);
> >  			goto out_unlock;
> >  		}
> > -		if (!buffer_mapped(bh))
> > -			all_mapped = 0;
> >  		bh = bh->b_this_page;
> >  	} while (bh != head);
>
> Does this fix some user-visible misbehaviour?  If so, what was that
> behaviour?
>

No. I was looking at the "Bug 12593 - __find_get_block_slow while using 
parted." Even though this may not be the cause of that bug, I think this could 
be a potential problem elsewhere.

> Please write good changelogs.  This is not some pointless book-keeping
> exercise.  People will make decisions about which kernel versions
> patches should be merged into, and they will want to know if a
> particular patch addresses a particular problem which they are
> experiencing.  For this, they need information.
>

Sorry for the bad changelog. Thanks for pointing it out.

Thanks
Nikanth