[PATCH] xfs: don't return 0 if generic_segment_checks() find nothing to write

[PATCH] xfs: don't return 0 if generic_segment_checks() find nothing to write

[Jeff Liu]

From: Jie Liu <jeff.liu@oracle.com>

At xfs_file_aio_write(), we just return 0 which means a successfully operation if the ocount
is evaluated to 0 at generic_segment_checks().  Generally, it'll occurred if the specified
write count is 0 from the user space.  However, for example, if the given offset is over the
maximum file size limitation although nothing would be written into the file, in this case,
the underlying file system should detect this issue and return a proper errno instead.
Hence, the beginning pre-checking up against ocount would conceal some real cause and confuse
the user space program maybe.

To fix it, we can delay the write count verification until xfs_file_aio_write_checks() is done.

This test is performed on 32-bit system without CONFIG_LBADF is enabled.
Before patching:
xfs_io -f -c "pwrite 17592186040320 0" /storage/test_file
wrote 0/0 bytes at offset 17592186040320
0.000000 bytes, 0 ops; 0.0000 sec (0.000000 bytes/sec and 0.0000 ops/sec)

After patching:
xfs_io -f -c "pwrite 17592186040320 0" /storage/test_file
pwrite64: File too large

Signed-off-by: Jie Liu <jeff.liu@oracle.com>

---
 fs/xfs/xfs_file.c |    9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index f03bf1a..f8a933e 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -671,6 +671,9 @@ xfs_file_dio_aio_write(
 	if (ret)
 		goto out;
 
+	if (ocount == 0)
+		goto out;
+
 	if (mapping->nrpages) {
 		ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
 						    pos, -1);
@@ -725,6 +728,9 @@ xfs_file_buffered_aio_write(
 	if (ret)
 		goto out;
 
+	if (ocount == 0)
+		goto out;
+
 	/* We can write back this queue in page reclaim */
 	current->backing_dev_info = mapping->backing_dev_info;
 
@@ -772,9 +778,6 @@ xfs_file_aio_write(
 	if (ret)
 		return ret;
 
-	if (ocount == 0)
-		return 0;
-
 	sb_start_write(inode->i_sb);
 
 	if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
-- 
1.7.9.5

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

Re: [PATCH] xfs: don't return 0 if generic_segment_checks() find nothing to write

[Ben Myers]

Hey Jeff,

On Fri, Apr 12, 2013 at 11:48:28PM +0800, Jeff Liu wrote:
> From: Jie Liu <jeff.liu@oracle.com>
> 
> At xfs_file_aio_write(), we just return 0 which means a successfully operation if the ocount
> is evaluated to 0 at generic_segment_checks().  Generally, it'll occurred if the specified
> write count is 0 from the user space.  However, for example, if the given offset is over the
> maximum file size limitation although nothing would be written into the file, in this case,
> the underlying file system should detect this issue and return a proper errno instead.
> Hence, the beginning pre-checking up against ocount would conceal some real cause and confuse
> the user space program maybe.
> 
> To fix it, we can delay the write count verification until xfs_file_aio_write_checks() is done.

Could you wrap these around 72 columns or so in the future?  Don't worry
about reposting this one unless something technical comes up.

Thanks,
	Ben

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

Re: [PATCH] xfs: don't return 0 if generic_segment_checks() find nothing to write

[Jeff Liu]

Hi Ben,

On 04/17/2013 02:26 AM, Ben Myers wrote:
> Hey Jeff,
> 
> On Fri, Apr 12, 2013 at 11:48:28PM +0800, Jeff Liu wrote:
>> From: Jie Liu <jeff.liu@oracle.com>
>>
>> At xfs_file_aio_write(), we just return 0 which means a successfully operation if the ocount
>> is evaluated to 0 at generic_segment_checks().  Generally, it'll occurred if the specified
>> write count is 0 from the user space.  However, for example, if the given offset is over the
>> maximum file size limitation although nothing would be written into the file, in this case,
>> the underlying file system should detect this issue and return a proper errno instead.
>> Hence, the beginning pre-checking up against ocount would conceal some real cause and confuse
>> the user space program maybe.
>>
>> To fix it, we can delay the write count verification until xfs_file_aio_write_checks() is done.
> 
> Could you wrap these around 72 columns or so in the future?  Don't worry
> about reposting this one unless something technical comes up.
Sure, I'll take care of the patch format in the future.

Thanks,
-Jeff

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