Re: [PATCH 1/3] btrfs: simplify counting number of eb pages

[David Sterba <dsterba@suse.cz>]

On Tue, Apr 24, 2018 at 02:22:15PM +0800, Qu Wenruo wrote:
> 
> 
> On 2018年04月24日 13:59, Nikolay Borisov wrote:
> > 
> > 
> > On 24.04.2018 02:03, David Sterba wrote:
> >> The eb length is nodesize, as initialized in __alloc_extent_buffer.
> >> Regardless of start, we should always get the same number of pages, so
> >> use that fact.
> >>
> >> Signed-off-by: David Sterba <dsterba@suse.com>
> >> ---
> >>  fs/btrfs/extent_io.h | 3 +--
> >>  1 file changed, 1 insertion(+), 2 deletions(-)
> >>
> >> diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
> >> index a53009694b16..ee92c1289edd 100644
> >> --- a/fs/btrfs/extent_io.h
> >> +++ b/fs/btrfs/extent_io.h
> >> @@ -454,8 +454,7 @@ void wait_on_extent_buffer_writeback(struct extent_buffer *eb);
> >>  
> >>  static inline unsigned long num_extent_pages(u64 start, u64 len)
> >>  {
> >> -	return ((start + len + PAGE_SIZE - 1) >> PAGE_SHIFT) -
> >> -		(start >> PAGE_SHIFT);
> >> +	return len >> PAGE_SHIFT;
> > 
> > Shouldn't this really be len + PAGE_SIZE -1 or in fact DIV_ROUND_DOWN
> > (len, PAGE_SIZE). Because with a nodesize of 4k (and basically less than
> > a page size) we can get into a situation where we do:
> > 
> > 4096 >> 13 = 0
> > 
> > On powerpc for example we have:
> > 
> > arch/powerpc/include/asm/page.h:#define PAGE_SHIFT              18
> > arch/powerpc/include/asm/page.h:#define PAGE_SHIFT              16
> > arch/powerpc/include/asm/page.h:#define PAGE_SHIFT              14
> 
> For such case, the fs won't be mounted as we don't have sub-pagesized
> nodesize support yet.
> So won't hit the problem.
> 
> Although a WARN_ON(len < PAGE_SIZE || IS_ALIGNED(start, PAGE_SIZE))
> would do no harm here.

Such check is fine, but would be better placed in __alloc_extent_buffer,
not each time we access the eb.