[PATCH] xfs: overflow in xfs_iomap_eof_align_last_fsb

[PATCH] xfs: overflow in xfs_iomap_eof_align_last_fsb

[Peter Watkins]

Someone else may have run into this already, if not please take
a look.

 Peter

If extsize is set and new_last_fsb is larger than 32 bits, the
roundup to extsize will overflow the align variable. Instead,
combine alignments by rounding extsize hint up to stripe size.

Signed-off-by: Peter Watkins <treestem@gmail.com>
Reviewed-by: Nathaniel W. Turner <nate@houseofnate.net>
---
 fs/xfs/xfs_iomap.c |   19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index afcf3c9..0c4abfe 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -52,7 +52,7 @@ xfs_iomap_eof_align_last_fsb(
 	xfs_extlen_t	extsize,
 	xfs_fileoff_t	*last_fsb)
 {
-	xfs_fileoff_t	new_last_fsb = 0;
+	xfs_fileoff_t	new_last_fsb;
 	xfs_extlen_t	align = 0;
 	int		eof, error;
 
@@ -70,23 +70,24 @@ xfs_iomap_eof_align_last_fsb(
 		else if (mp->m_dalign)
 			align = mp->m_dalign;
 
-		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
-			new_last_fsb = roundup_64(*last_fsb, align);
+		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
+			align = 0;
 	}
 
 	/*
-	 * Always round up the allocation request to an extent boundary
-	 * (when file on a real-time subvolume or has di_extsize hint).
+	 * Round up the allocation request to an extent boundary. If
+	 * already aligned to a stripe, round extsize up to a stripe
+	 * boundary.
 	 */
 	if (extsize) {
-		if (new_last_fsb)
-			align = roundup_64(new_last_fsb, extsize);
+		if (align)
+			align = roundup_64(extsize, align);
 		else
 			align = extsize;
-		new_last_fsb = roundup_64(*last_fsb, align);
 	}
 
-	if (new_last_fsb) {
+	if (align) {
+		new_last_fsb = roundup_64(*last_fsb, align);
 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 		if (error)
 			return error;
-- 
1.7.9.5

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: overflow in xfs_iomap_eof_align_last_fsb

[Brian Foster]

On Mon, Nov 24, 2014 at 02:06:13PM -0500, Peter Watkins wrote:
> Someone else may have run into this already, if not please take
> a look.
> 
>  Peter
> 
> If extsize is set and new_last_fsb is larger than 32 bits, the
> roundup to extsize will overflow the align variable. Instead,
> combine alignments by rounding extsize hint up to stripe size.
> 
> Signed-off-by: Peter Watkins <treestem@gmail.com>
> Reviewed-by: Nathaniel W. Turner <nate@houseofnate.net>
> ---

So we basically assign a file offset value to an extent length variable.
Nice spot...

>  fs/xfs/xfs_iomap.c |   19 ++++++++++---------
>  1 file changed, 10 insertions(+), 9 deletions(-)
> 
> diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
> index afcf3c9..0c4abfe 100644
> --- a/fs/xfs/xfs_iomap.c
> +++ b/fs/xfs/xfs_iomap.c
> @@ -52,7 +52,7 @@ xfs_iomap_eof_align_last_fsb(
>  	xfs_extlen_t	extsize,
>  	xfs_fileoff_t	*last_fsb)
>  {
> -	xfs_fileoff_t	new_last_fsb = 0;
> +	xfs_fileoff_t	new_last_fsb;
>  	xfs_extlen_t	align = 0;
>  	int		eof, error;
>  
> @@ -70,23 +70,24 @@ xfs_iomap_eof_align_last_fsb(
>  		else if (mp->m_dalign)
>  			align = mp->m_dalign;
>  
> -		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
> -			new_last_fsb = roundup_64(*last_fsb, align);
> +		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
> +			align = 0;
>  	}
>  
>  	/*
> -	 * Always round up the allocation request to an extent boundary
> -	 * (when file on a real-time subvolume or has di_extsize hint).
> +	 * Round up the allocation request to an extent boundary. If
> +	 * already aligned to a stripe, round extsize up to a stripe
> +	 * boundary.
>  	 */
>  	if (extsize) {
> -		if (new_last_fsb)
> -			align = roundup_64(new_last_fsb, extsize);
> +		if (align)
> +			align = roundup_64(extsize, align);

I think the previous behavior would swap this around and round up the
alignment to the extsize. Because we aligned from the actual fsb, we'd
effectively use the stripe alignment as a min. alloc and always align to
the hint. Here we simply round up extsize to the stripe alignment. IOW,
the alignment was previously always to the hint, even if smaller than
the stripe alignment, and now we align to the larger of the hint or
stripe.

That said, it's kind of a weird scenario and it's not clear to me
whether it's important or even intended. If not, the patch seems Ok to
me...

Brian

>  		else
>  			align = extsize;
> -		new_last_fsb = roundup_64(*last_fsb, align);
>  	}
>  
> -	if (new_last_fsb) {
> +	if (align) {
> +		new_last_fsb = roundup_64(*last_fsb, align);
>  		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
>  		if (error)
>  			return error;
> -- 
> 1.7.9.5
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: overflow in xfs_iomap_eof_align_last_fsb

[Dave Chinner]

On Mon, Nov 24, 2014 at 02:06:13PM -0500, Peter Watkins wrote:
> Someone else may have run into this already, if not please take
> a look.
> 
>  Peter
> 
> If extsize is set and new_last_fsb is larger than 32 bits, the
> roundup to extsize will overflow the align variable. Instead,
> combine alignments by rounding extsize hint up to stripe size.

Change of allocator behaviour that will cause significant problems
for applications that depend on deterministic behaviour from the
extent size hint.

i.e. The extsize hint is supposed to override all other alignments
that are made - extent size hints are effectively a "unit of
allocation".  Hence we round up to stripe unit/width, then apply the
extent size hint alignment so that extent sizes are always aligned
to the extent size hint....

....
> @@ -52,7 +52,7 @@ xfs_iomap_eof_align_last_fsb(
>  	xfs_extlen_t	extsize,
>  	xfs_fileoff_t	*last_fsb)
>  {
> -	xfs_fileoff_t	new_last_fsb = 0;
> +	xfs_fileoff_t	new_last_fsb;

That can be declared in the context that uses it now.

>  	xfs_extlen_t	align = 0;
>  	int		eof, error;
>  
> @@ -70,23 +70,24 @@ xfs_iomap_eof_align_last_fsb(
>  		else if (mp->m_dalign)
>  			align = mp->m_dalign;
>  
> -		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
> -			new_last_fsb = roundup_64(*last_fsb, align);
> +		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
> +			align = 0;
>  	}
>  
>  	/*
> -	 * Always round up the allocation request to an extent boundary
> -	 * (when file on a real-time subvolume or has di_extsize hint).
> +	 * Round up the allocation request to an extent boundary. If
> +	 * already aligned to a stripe, round extsize up to a stripe
> +	 * boundary.
>  	 */

Drop the comment change...

>  	if (extsize) {
> -		if (new_last_fsb)
> -			align = roundup_64(new_last_fsb, extsize);
> +		if (align)
> +			align = roundup_64(extsize, align);

			align = roundup(align, extsize);

>  		else
>  			align = extsize;
> -		new_last_fsb = roundup_64(*last_fsb, align);
>  	}
>  
> -	if (new_last_fsb) {
> +	if (align) {
> +		new_last_fsb = roundup_64(*last_fsb, align);

		xfs_fileoff_t	new_last_fsb = roundup_64(*last_fsb, align);

>  		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
>  		if (error)
>  			return error;

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

[PATCH v2] xfs: overflow in xfs_iomap_eof_align_last_fsb

[Peter Watkins]

If extsize is set and new_last_fsb is larger than 32 bits, the
roundup to extsize will overflow the align variable. Instead,
combine alignments by rounding stripe size up to extsize.

Signed-off-by: Peter Watkins <treestem@gmail.com>
Reviewed-by: Nathaniel W. Turner <nate@houseofnate.net>
---
 fs/xfs/xfs_iomap.c |   13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index afcf3c9..3fad071 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -52,7 +52,6 @@ xfs_iomap_eof_align_last_fsb(
 	xfs_extlen_t	extsize,
 	xfs_fileoff_t	*last_fsb)
 {
-	xfs_fileoff_t	new_last_fsb = 0;
 	xfs_extlen_t	align = 0;
 	int		eof, error;
 
@@ -70,8 +69,8 @@ xfs_iomap_eof_align_last_fsb(
 		else if (mp->m_dalign)
 			align = mp->m_dalign;
 
-		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
-			new_last_fsb = roundup_64(*last_fsb, align);
+		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
+			align = 0;
 	}
 
 	/*
@@ -79,14 +78,14 @@ xfs_iomap_eof_align_last_fsb(
 	 * (when file on a real-time subvolume or has di_extsize hint).
 	 */
 	if (extsize) {
-		if (new_last_fsb)
-			align = roundup_64(new_last_fsb, extsize);
+		if (align)
+			align = roundup_64(align, extsize);
 		else
 			align = extsize;
-		new_last_fsb = roundup_64(*last_fsb, align);
 	}
 
-	if (new_last_fsb) {
+	if (align) {
+		xfs_fileoff_t	new_last_fsb = roundup_64(*last_fsb, align);
 		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
 		if (error)
 			return error;
-- 
1.7.9.5

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH v2] xfs: overflow in xfs_iomap_eof_align_last_fsb

[Brian Foster]

On Mon, Dec 01, 2014 at 05:45:28PM -0500, Peter Watkins wrote:
> If extsize is set and new_last_fsb is larger than 32 bits, the
> roundup to extsize will overflow the align variable. Instead,
> combine alignments by rounding stripe size up to extsize.
> 
> Signed-off-by: Peter Watkins <treestem@gmail.com>
> Reviewed-by: Nathaniel W. Turner <nate@houseofnate.net>
> ---

Looks fine to me:

Reviewed-by: Brian Foster <bfoster@redhat.com>

>  fs/xfs/xfs_iomap.c |   13 ++++++-------
>  1 file changed, 6 insertions(+), 7 deletions(-)
> 
> diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
> index afcf3c9..3fad071 100644
> --- a/fs/xfs/xfs_iomap.c
> +++ b/fs/xfs/xfs_iomap.c
> @@ -52,7 +52,6 @@ xfs_iomap_eof_align_last_fsb(
>  	xfs_extlen_t	extsize,
>  	xfs_fileoff_t	*last_fsb)
>  {
> -	xfs_fileoff_t	new_last_fsb = 0;
>  	xfs_extlen_t	align = 0;
>  	int		eof, error;
>  
> @@ -70,8 +69,8 @@ xfs_iomap_eof_align_last_fsb(
>  		else if (mp->m_dalign)
>  			align = mp->m_dalign;
>  
> -		if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
> -			new_last_fsb = roundup_64(*last_fsb, align);
> +		if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
> +			align = 0;
>  	}
>  
>  	/*
> @@ -79,14 +78,14 @@ xfs_iomap_eof_align_last_fsb(
>  	 * (when file on a real-time subvolume or has di_extsize hint).
>  	 */
>  	if (extsize) {
> -		if (new_last_fsb)
> -			align = roundup_64(new_last_fsb, extsize);
> +		if (align)
> +			align = roundup_64(align, extsize);
>  		else
>  			align = extsize;
> -		new_last_fsb = roundup_64(*last_fsb, align);
>  	}
>  
> -	if (new_last_fsb) {
> +	if (align) {
> +		xfs_fileoff_t	new_last_fsb = roundup_64(*last_fsb, align);
>  		error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
>  		if (error)
>  			return error;
> -- 
> 1.7.9.5
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs