[PATCH] xfs: fix node forward in xfs_node_toosmall

[PATCH] xfs: fix node forward in xfs_node_toosmall

[Ben Myers]

From: Mark Tinguely <tinguely@sgi.com>

Commit f5ea1100 cleans up the disk to host conversions for
node directory entries, but because a variable is reused in
xfs_node_toosmall() the next node is not correctly found.
If the original node is small enough (<= 3/8 of the node size),
this change may incorrectly cause a node collapse when it should
not. That will cause an assert in xfstest generic/319:

   Assertion failed: first <= last && last < BBTOB(bp->b_length),
   file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569

Keep the original node header to get the correct forward node.

(When a node is considered for a merge with a sibling, it overwrites the
 sibling pointers of the original incore nodehdr with the sibling's
 pointers.  This leads to loop considering the original node as a merge
 candidate with itself in the second pass, and so it incorrectly
 determines a merge should occur.)

upstream commit 997def25e4b9cee3b01609e18a52f926bca8bd2b

Signed-off-by: Mark Tinguely <tinguely@sgi.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>

[v3: added Dave Chinner's (slightly modified) suggestion to the commit header,
	cleaned up whitespace.  -bpm]
---

Hi Stable Folk,

   Please take this commit for stable branches back to 3.10.  This fixes a bug
   in XFS directory code.

Thanks,
	Ben

 fs/xfs/xfs_da_btree.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/fs/xfs/xfs_da_btree.c b/fs/xfs/xfs_da_btree.c
index 069537c..20bf8e8 100644
--- a/fs/xfs/xfs_da_btree.c
+++ b/fs/xfs/xfs_da_btree.c
@@ -1224,6 +1224,7 @@ xfs_da3_node_toosmall(
 	/* start with smaller blk num */
 	forward = nodehdr.forw < nodehdr.back;
 	for (i = 0; i < 2; forward = !forward, i++) {
+		struct xfs_da3_icnode_hdr thdr;
 		if (forward)
 			blkno = nodehdr.forw;
 		else
@@ -1236,10 +1237,10 @@ xfs_da3_node_toosmall(
 			return(error);
 
 		node = bp->b_addr;
-		xfs_da3_node_hdr_from_disk(&nodehdr, node);
+		xfs_da3_node_hdr_from_disk(&thdr, node);
 		xfs_trans_brelse(state->args->trans, bp);
 
-		if (count - nodehdr.count >= 0)
+		if (count - thdr.count >= 0)
 			break;	/* fits with at least 25% to spare */
 	}
 	if (i >= 2) {
-- 
1.7.12.4

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

Re: [PATCH] xfs: fix node forward in xfs_node_toosmall

[Greg KH]

On Tue, Oct 08, 2013 at 07:09:26PM -0500, Ben Myers wrote:
> From: Mark Tinguely <tinguely@sgi.com>
> 
> Commit f5ea1100 cleans up the disk to host conversions for
> node directory entries, but because a variable is reused in
> xfs_node_toosmall() the next node is not correctly found.
> If the original node is small enough (<= 3/8 of the node size),
> this change may incorrectly cause a node collapse when it should
> not. That will cause an assert in xfstest generic/319:
> 
>    Assertion failed: first <= last && last < BBTOB(bp->b_length),
>    file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569
> 
> Keep the original node header to get the correct forward node.
> 
> (When a node is considered for a merge with a sibling, it overwrites the
>  sibling pointers of the original incore nodehdr with the sibling's
>  pointers.  This leads to loop considering the original node as a merge
>  candidate with itself in the second pass, and so it incorrectly
>  determines a merge should occur.)
> 
> upstream commit 997def25e4b9cee3b01609e18a52f926bca8bd2b

Applied to the 3.10 and 3.11 stable queues, thanks.

greg k-h

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

Re: [PATCH] xfs: fix node forward in xfs_node_toosmall

[Mark Tinguely]

On 09/23/13 16:34, Michael L. Semon wrote:
> On 09/20/2013 06:05 PM, Mark Tinguely wrote:
>> Commit f5ea1100 cleans up the disk to host conversions for
>> node directory entries, but because a variable is reused in
>> xfs_node_toosmall() the next node is not correctly found.
>> If the original node is small enough (<= 3/8 of the node size),
>> this change may incorrectly cause a node collapse when it should
>> not. That will cause an assert in xfstest generic/319:
>>
>>     Assertion failed: first<= last&&  last<  BBTOB(bp->b_length),
>>     file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569
>>
>> Keep the original node header to get the correct forward node.
>
> This works for xfstests generic/319 on 32-bit x86.  The xfstests
> run was okay, with a lockdep in the middle and a crash in xfs/300.
> Neither the lockdep nor the crash could reproduced by running
> individual tests one by one.
>
> generic/319 itself fails only because it couldn't find its
> _cleanup routine.  I'm not sure if I'm using v1 or v2 of
> generic/319.
>
> The kernel is set to the commit just before the latest aio-next
> merge, which is causing severe problems here.  xfsprogs was the
> latest production xfsprogs because my copy of xfstests is having
> trouble checking v4 XFS filesystems otherwise.
>
> Will test xfs_node_toosmall patch v2 once I get home.
>
> Thanks!
>
> Michael
>

319 v1 had the cleanup and removed in v2 because of feedback.
IMO, it should be there even basically empty.

--Mark.

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

Re: [PATCH] xfs: fix node forward in xfs_node_toosmall

[Michael L. Semon]

On 09/20/2013 06:05 PM, Mark Tinguely wrote:
> Commit f5ea1100 cleans up the disk to host conversions for
> node directory entries, but because a variable is reused in
> xfs_node_toosmall() the next node is not correctly found.
> If the original node is small enough (<= 3/8 of the node size),
> this change may incorrectly cause a node collapse when it should
> not. That will cause an assert in xfstest generic/319:
> 
>    Assertion failed: first <= last && last < BBTOB(bp->b_length),
>    file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569
> 
> Keep the original node header to get the correct forward node.

This works for xfstests generic/319 on 32-bit x86.  The xfstests 
run was okay, with a lockdep in the middle and a crash in xfs/300. 
Neither the lockdep nor the crash could reproduced by running 
individual tests one by one.

generic/319 itself fails only because it couldn't find its 
_cleanup routine.  I'm not sure if I'm using v1 or v2 of 
generic/319.

The kernel is set to the commit just before the latest aio-next 
merge, which is causing severe problems here.  xfsprogs was the 
latest production xfsprogs because my copy of xfstests is having 
trouble checking v4 XFS filesystems otherwise.

Will test xfs_node_toosmall patch v2 once I get home.

Thanks!

Michael

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

Re: [PATCH] xfs: fix node forward in xfs_node_toosmall

[Mark Tinguely]

On 09/22/13 19:08, Dave Chinner wrote:
> On Fri, Sep 20, 2013 at 05:05:08PM -0500, Mark Tinguely wrote:
>> Commit f5ea1100 cleans up the disk to host conversions for
>> node directory entries, but because a variable is reused in
>> xfs_node_toosmall() the next node is not correctly found.
>> If the original node is small enough (<= 3/8 of the node size),
>> this change may incorrectly cause a node collapse when it should
>> not. That will cause an assert in xfstest generic/319:
>>
>>     Assertion failed: first<= last&&  last<  BBTOB(bp->b_length),
>>     file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569
>>
>> Keep the original node header to get the correct forward node.
>>
>> Signed-off-by: Mark Tinguely<tinguely@sgi.com>
>> ---

...

> Yes, that definitely a bug, but I think that the change doesn't
> scope correctly. The original node header doesn't need to be saved
> like this - the node header decoded in the loop needs a loop-scope
> variable. i.e.:
>
> 	/* start with smaller blk num */
> 	forward = nodehdr.forw<  nodehdr.back;
> 	for (i = 0; i<  2; forward = !forward, i++) {
> +		struct xfs_da3_icnode_hdr thdr;
> +
> 		if (forward)
> 			blkno = nodehdr.forw;
> 		else
> 			blkno = nodehdr.back;
> 		if (blkno == 0)
> 			continue;
> 		error = xfs_da3_node_read(state->args->trans, state->args->dp,
> 					blkno, -1,&bp, state->args->whichfork);
> 		if (error)
> 			return(error);
>
> 		node = bp->b_addr;
> -		xfs_da3_node_hdr_from_disk(&nodehdr, node);
> +		xfs_da3_node_hdr_from_disk(&thdr, node);
> 		xfs_trans_brelse(state->args->trans, bp);
>
> -		if (count - nodehdr.count>= 0)
> +		if (count - thdr.count>= 0)
> 			break;  /* fits with at least 25% to spare */
> 	}
>
> Cheers,
>
> Dave.

Okay thanks.

--Mark.

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

Re: [PATCH] xfs: fix node forward in xfs_node_toosmall

[Dave Chinner]

On Fri, Sep 20, 2013 at 05:05:08PM -0500, Mark Tinguely wrote:
> Commit f5ea1100 cleans up the disk to host conversions for
> node directory entries, but because a variable is reused in
> xfs_node_toosmall() the next node is not correctly found.
> If the original node is small enough (<= 3/8 of the node size),
> this change may incorrectly cause a node collapse when it should
> not. That will cause an assert in xfstest generic/319:
> 
>    Assertion failed: first <= last && last < BBTOB(bp->b_length),
>    file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569
> 
> Keep the original node header to get the correct forward node.
> 
> Signed-off-by: Mark Tinguely <tinguely@sgi.com>
> ---
>  fs/xfs/xfs_da_btree.c |    6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> Index: b/fs/xfs/xfs_da_btree.c
> ===================================================================
> --- a/fs/xfs/xfs_da_btree.c
> +++ b/fs/xfs/xfs_da_btree.c
> @@ -1163,6 +1163,7 @@ xfs_da3_node_toosmall(
>  	xfs_dablk_t		blkno;
>  	struct xfs_buf		*bp;
>  	struct xfs_da3_icnode_hdr nodehdr;
> +	struct xfs_da3_icnode_hdr firstnhdr;
>  	int			count;
>  	int			forward;
>  	int			error;
> @@ -1221,13 +1222,14 @@ xfs_da3_node_toosmall(
>  	count -= state->node_ents >> 2;
>  	count -= nodehdr.count;
>  
> +	firstnhdr = nodehdr;
>  	/* start with smaller blk num */
>  	forward = nodehdr.forw < nodehdr.back;
>  	for (i = 0; i < 2; forward = !forward, i++) {
>  		if (forward)
> -			blkno = nodehdr.forw;
> +			blkno = firstnhdr.forw;
>  		else
> -			blkno = nodehdr.back;
> +			blkno = firstnhdr.back;
>  		if (blkno == 0)
>  			continue;
>  		error = xfs_da3_node_read(state->args->trans, state->args->dp,

Yes, that definitely a bug, but I think that the change doesn't
scope correctly. The original node header doesn't need to be saved
like this - the node header decoded in the loop needs a loop-scope
variable. i.e.:

	/* start with smaller blk num */
	forward = nodehdr.forw < nodehdr.back;
	for (i = 0; i < 2; forward = !forward, i++) {
+		struct xfs_da3_icnode_hdr thdr;
+
		if (forward)
			blkno = nodehdr.forw;
		else
			blkno = nodehdr.back;
		if (blkno == 0)
			continue;
		error = xfs_da3_node_read(state->args->trans, state->args->dp,
					blkno, -1, &bp, state->args->whichfork);
		if (error)
			return(error);

		node = bp->b_addr;
-		xfs_da3_node_hdr_from_disk(&nodehdr, node);
+		xfs_da3_node_hdr_from_disk(&thdr, node);
		xfs_trans_brelse(state->args->trans, bp);

-		if (count - nodehdr.count >= 0)
+		if (count - thdr.count >= 0)
			break;  /* fits with at least 25% to spare */
	}

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

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

[PATCH] xfs: fix node forward in xfs_node_toosmall

[Mark Tinguely]

[-- Attachment #1: xfs-fix-xfs_da3_node_toosmall-test.patch --]
[-- Type: text/plain, Size: 1670 bytes --]

Commit f5ea1100 cleans up the disk to host conversions for
node directory entries, but because a variable is reused in
xfs_node_toosmall() the next node is not correctly found.
If the original node is small enough (<= 3/8 of the node size),
this change may incorrectly cause a node collapse when it should
not. That will cause an assert in xfstest generic/319:

   Assertion failed: first <= last && last < BBTOB(bp->b_length),
   file: /root/newest/xfs/fs/xfs/xfs_trans_buf.c, line: 569

Keep the original node header to get the correct forward node.

Signed-off-by: Mark Tinguely <tinguely@sgi.com>
---
 fs/xfs/xfs_da_btree.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

Index: b/fs/xfs/xfs_da_btree.c
===================================================================
--- a/fs/xfs/xfs_da_btree.c
+++ b/fs/xfs/xfs_da_btree.c
@@ -1163,6 +1163,7 @@ xfs_da3_node_toosmall(
 	xfs_dablk_t		blkno;
 	struct xfs_buf		*bp;
 	struct xfs_da3_icnode_hdr nodehdr;
+	struct xfs_da3_icnode_hdr firstnhdr;
 	int			count;
 	int			forward;
 	int			error;
@@ -1221,13 +1222,14 @@ xfs_da3_node_toosmall(
 	count -= state->node_ents >> 2;
 	count -= nodehdr.count;
 
+	firstnhdr = nodehdr;
 	/* start with smaller blk num */
 	forward = nodehdr.forw < nodehdr.back;
 	for (i = 0; i < 2; forward = !forward, i++) {
 		if (forward)
-			blkno = nodehdr.forw;
+			blkno = firstnhdr.forw;
 		else
-			blkno = nodehdr.back;
+			blkno = firstnhdr.back;
 		if (blkno == 0)
 			continue;
 		error = xfs_da3_node_read(state->args->trans, state->args->dp,


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