Ocfs2 devel - Nov 2008 - [PATCH 0/2] 2 bug fix for xattr set transaction merge.

When we extend one xattr's value to a large size, in case
the old value size is smaller than the size of value root,
we still need to guess the metadata allocation.

Reported-by: Tiger Yang <tiger.yang at oracle.com>
Signed-off-by: Tao Ma <tao.ma at oracle.com>
---
 fs/ocfs2/xattr.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)

diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
index 8195450..1504c6b 100644
--- a/fs/ocfs2/xattr.c
+++ b/fs/ocfs2/xattr.c
@@ -2270,6 +2270,7 @@ static int ocfs2_calc_xattr_set_need(struct inode *inode,
 								 value_size);
 			xv = (struct ocfs2_xattr_value_root *)
 			     (base + name_offset + name_len);
+			value_size = OCFS2_XATTR_ROOT_SIZE;
 		} else
 			xv = &def_xv.xv;
 
@@ -2283,7 +2284,8 @@ static int ocfs2_calc_xattr_set_need(struct inode *inode,
 							     &xv->xr_list,
 							     new_clusters -
 							     old_clusters);
-			goto out;
+			if (value_size >= OCFS2_XATTR_ROOT_SIZE)
+				goto out;
 		}
 	} else {
 		/*
-- 
1.5.5

In xattr set, when we move a xattr which was stored in inode to
the outside bucket, we have to delete it and it will use the old
"xis->not_found", but it is removed by the
ocfs2_calc_xattr_set_need.
So restore it.

Signed-off-by: Tao Ma <tao.ma at oracle.com>
---
 fs/ocfs2/xattr.c |    6 +++++-
 1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
index 1504c6b..7d6307f 100644
--- a/fs/ocfs2/xattr.c
+++ b/fs/ocfs2/xattr.c
@@ -2414,7 +2414,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
 				    struct ocfs2_xattr_search *xbs,
 				    struct ocfs2_xattr_set_ctxt *ctxt)
 {
-	int ret = 0, credits;
+	int ret = 0, credits, old_found;
 
 	if (!xi->value) {
 		/* Remove existing extended attribute */
@@ -2433,6 +2433,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
 			xi->value = NULL;
 			xi->value_len = 0;
 
+			old_found = xis->not_found;
 			xis->not_found = -ENODATA;
 			ret = ocfs2_calc_xattr_set_need(inode,
 							di,
@@ -2442,6 +2443,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
 							NULL,
 							NULL,
 							&credits);
+			xis->not_found = old_found;
 			if (ret) {
 				mlog_errno(ret);
 				goto out;
@@ -2462,6 +2464,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
 				if (ret)
 					goto out;
 
+				old_found = xis->not_found;
 				xis->not_found = -ENODATA;
 				ret = ocfs2_calc_xattr_set_need(inode,
 								di,
@@ -2471,6 +2474,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
 								NULL,
 								NULL,
 								&credits);
+				xis->not_found = old_found;
 				if (ret) {
 					mlog_errno(ret);
 					goto out;
-- 
1.5.5

Hi Mark,
	The 2 patches are 2 bug fix for my xattr set transaction merge patches. 
I don't know why tristan's powerful xattr test can't find it, and I
will
ask him to include the test case in it. The good news is that the merge 
patch set is still in your merge_window, so the bug fixes are not too 
late. They are both straightforward, so please push them to your 
merge_window.

Regards,
Tao

how quickly you solve this problem! I also found this problem with my 
test script but forgot to tell you when you address the first problem.

thanks,
tiger

Tao Ma wrote:
> In xattr set, when we move a xattr which was stored in inode to
> the outside bucket, we have to delete it and it will use the old
> "xis->not_found", but it is removed by the
ocfs2_calc_xattr_set_need.
> So restore it.
> 
> Signed-off-by: Tao Ma <tao.ma at oracle.com>
> ---
>  fs/ocfs2/xattr.c |    6 +++++-
>  1 files changed, 5 insertions(+), 1 deletions(-)
> 
> diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
> index 1504c6b..7d6307f 100644
> --- a/fs/ocfs2/xattr.c
> +++ b/fs/ocfs2/xattr.c
> @@ -2414,7 +2414,7 @@ static int __ocfs2_xattr_set_handle(struct inode
*inode,
>  				    struct ocfs2_xattr_search *xbs,
>  				    struct ocfs2_xattr_set_ctxt *ctxt)
>  {
> -	int ret = 0, credits;
> +	int ret = 0, credits, old_found;
>  
>  	if (!xi->value) {
>  		/* Remove existing extended attribute */
> @@ -2433,6 +2433,7 @@ static int __ocfs2_xattr_set_handle(struct inode
*inode,
>  			xi->value = NULL;
>  			xi->value_len = 0;
>  
> +			old_found = xis->not_found;
>  			xis->not_found = -ENODATA;
>  			ret = ocfs2_calc_xattr_set_need(inode,
>  							di,
> @@ -2442,6 +2443,7 @@ static int __ocfs2_xattr_set_handle(struct inode
*inode,
>  							NULL,
>  							NULL,
>  							&credits);
> +			xis->not_found = old_found;
>  			if (ret) {
>  				mlog_errno(ret);
>  				goto out;
> @@ -2462,6 +2464,7 @@ static int __ocfs2_xattr_set_handle(struct inode
*inode,
>  				if (ret)
>  					goto out;
>  
> +				old_found = xis->not_found;
>  				xis->not_found = -ENODATA;
>  				ret = ocfs2_calc_xattr_set_need(inode,
>  								di,
> @@ -2471,6 +2474,7 @@ static int __ocfs2_xattr_set_handle(struct inode
*inode,
>  								NULL,
>  								NULL,
>  								&credits);
> +				xis->not_found = old_found;
>  				if (ret) {
>  					mlog_errno(ret);
>  					goto out;

On Wed, Nov 19, 2008 at 04:52:53PM +0800, Tao Ma wrote:
> Hi Mark,
> 	The 2 patches are 2 bug fix for my xattr set transaction merge 
> 	patches. I don't know why tristan's powerful xattr test can't
find it, and
> I will ask him to include the test case in it. The good news is that the 
> merge patch set is still in your merge_window, so the bug fixes are not too
> late. They are both straightforward, so please push them to your 
> merge_window.
Yep, looks good. They're in merge_window now.
	--Mark

--
Mark Fasheh