[PATCH] fix type of "offset" in ext4_io_end

[PATCH] fix type of "offset" in ext4_io_end

[Eric Sandeen]

The "offset" member in ext4_io_end holds bytes, not
blocks, so ext4_lblk_t is wrong - and too small (u32)

This caused the testcase "Possible ext4 data corruption
with large files and async I/O" sent by Giel to fail when it
wrapped around to 0.

Also fix up the type of arguments to
ext4_convert_unwritten_extents(), it gets ssize_t from
ext4_end_aio_dio_nolock() and ext4_ext_direct_IO()

Reported-by: Giel de Nijs <giel@vectorwise.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
---
 
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index d0a2afb..4a825c1 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -139,8 +139,8 @@ typedef struct ext4_io_end {
 	struct inode		*inode;		/* file being written to */
 	unsigned int		flag;		/* unwritten or not */
 	int			error;		/* I/O error code */
-	ext4_lblk_t		offset;		/* offset in the file */
-	size_t			size;		/* size of the extent */
+	loff_t			offset;		/* offset in the file */
+	ssize_t			size;		/* size of the extent */
 	struct work_struct	work;		/* data work queue */
 } ext4_io_end_t;
 
@@ -1740,7 +1740,7 @@ extern void ext4_ext_release(struct super_block *);
 extern long ext4_fallocate(struct inode *inode, int mode, loff_t offset,
 			  loff_t len);
 extern int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
-			  loff_t len);
+			  ssize_t len);
 extern int ext4_get_blocks(handle_t *handle, struct inode *inode,
 			   sector_t block, unsigned int max_blocks,
 			   struct buffer_head *bh, int flags);
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 8b8bae4..9333dc9 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -3547,7 +3547,7 @@ retry:
  * Returns 0 on success.
  */
 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
-				    loff_t len)
+				    ssize_t len)
 {
 	handle_t *handle;
 	ext4_lblk_t block;

Re: [PATCH] fix type of "offset" in ext4_io_end

[Josef Bacik]

On Fri, Jan 29, 2010 at 12:24:04PM -0600, Eric Sandeen wrote:
> The "offset" member in ext4_io_end holds bytes, not
> blocks, so ext4_lblk_t is wrong - and too small (u32)
> 
> This caused the testcase "Possible ext4 data corruption
> with large files and async I/O" sent by Giel to fail when it
> wrapped around to 0.
> 
> Also fix up the type of arguments to
> ext4_convert_unwritten_extents(), it gets ssize_t from
> ext4_end_aio_dio_nolock() and ext4_ext_direct_IO()
> 
> Reported-by: Giel de Nijs <giel@vectorwise.com>
> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> ---
>  

Reviewed-by: Josef Bacik <josef@redhat.com>

[PATCH V2] fix type of "offset" in ext4_io_end

[Eric Sandeen]

The "offset" member in ext4_io_end holds bytes, not
blocks, so ext4_lblk_t is wrong - and too small (u32)

This caused the testcase "Possible ext4 data corruption
with large files and async I/O" sent by Giel to fail when it
wrapped around to 0.

Also fix up the type of arguments to
ext4_convert_unwritten_extents(), it gets ssize_t from
ext4_end_aio_dio_nolock() and ext4_ext_direct_IO()

Reported-by: Giel de Nijs <giel@vectorwise.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
---

V2: Post-review Josef spotted another size_t in ext4_end_aio_dio_nolock()

---diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index d0a2afb..4a825c1 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -139,8 +139,8 @@ typedef struct ext4_io_end {
 	struct inode		*inode;		/* file being written to */
 	unsigned int		flag;		/* unwritten or not */
 	int			error;		/* I/O error code */
-	ext4_lblk_t		offset;		/* offset in the file */
-	size_t			size;		/* size of the extent */
+	loff_t			offset;		/* offset in the file */
+	ssize_t			size;		/* size of the extent */
 	struct work_struct	work;		/* data work queue */
 } ext4_io_end_t;
 
@@ -1740,7 +1740,7 @@ extern void ext4_ext_release(struct super_block *);
 extern long ext4_fallocate(struct inode *inode, int mode, loff_t offset,
 			  loff_t len);
 extern int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
-			  loff_t len);
+			  ssize_t len);
 extern int ext4_get_blocks(handle_t *handle, struct inode *inode,
 			   sector_t block, unsigned int max_blocks,
 			   struct buffer_head *bh, int flags);
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 8b8bae4..9333dc9 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -3547,7 +3547,7 @@ retry:
  * Returns 0 on success.
  */
 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
-				    loff_t len)
+				    ssize_t len)
 {
 	handle_t *handle;
 	ext4_lblk_t block;
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index e233879..4f36f7d 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3540,7 +3540,7 @@ static int ext4_end_aio_dio_nolock(ext4_io_end_t *io)
 {
 	struct inode *inode = io->inode;
 	loff_t offset = io->offset;
-	size_t size = io->size;
+	ssize_t size = io->size;
 	int ret = 0;
 
 	ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"

Re: [PATCH V2] fix type of "offset" in ext4_io_end

[Josef Bacik]

On Fri, Jan 29, 2010 at 02:28:23PM -0600, Eric Sandeen wrote:
> The "offset" member in ext4_io_end holds bytes, not
> blocks, so ext4_lblk_t is wrong - and too small (u32)
> 
> This caused the testcase "Possible ext4 data corruption
> with large files and async I/O" sent by Giel to fail when it
> wrapped around to 0.
> 
> Also fix up the type of arguments to
> ext4_convert_unwritten_extents(), it gets ssize_t from
> ext4_end_aio_dio_nolock() and ext4_ext_direct_IO()
> 
> Reported-by: Giel de Nijs <giel@vectorwise.com>
> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> ---
> 
> V2: Post-review Josef spotted another size_t in ext4_end_aio_dio_nolock()
> 

Reviewed-by: Josef Back <josef@redhat.com>

probably should have said something before replying to the last email :).
Thanks,

Josef

Re: [PATCH V2] fix type of "offset" in ext4_io_end

[tytso]

On Fri, Jan 29, 2010 at 02:28:23PM -0600, Eric Sandeen wrote:
> The "offset" member in ext4_io_end holds bytes, not
> blocks, so ext4_lblk_t is wrong - and too small (u32)
> 
> This caused the testcase "Possible ext4 data corruption
> with large files and async I/O" sent by Giel to fail when it
> wrapped around to 0.
> 
> Also fix up the type of arguments to
> ext4_convert_unwritten_extents(), it gets ssize_t from
> ext4_end_aio_dio_nolock() and ext4_ext_direct_IO()
> 
> Reported-by: Giel de Nijs <giel@vectorwise.com>
> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> ---

This looks important enough (and low risk enough) to push to Linus
before the next merge window.   Comments?

						- Ted

Re: [PATCH V2] fix type of "offset" in ext4_io_end

[Eric Sandeen]

tytso@mit.edu wrote:
> On Fri, Jan 29, 2010 at 02:28:23PM -0600, Eric Sandeen wrote:
>> The "offset" member in ext4_io_end holds bytes, not
>> blocks, so ext4_lblk_t is wrong - and too small (u32)
>>
>> This caused the testcase "Possible ext4 data corruption
>> with large files and async I/O" sent by Giel to fail when it
>> wrapped around to 0.
>>
>> Also fix up the type of arguments to
>> ext4_convert_unwritten_extents(), it gets ssize_t from
>> ext4_end_aio_dio_nolock() and ext4_ext_direct_IO()
>>
>> Reported-by: Giel de Nijs <giel@vectorwise.com>
>> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
>> ---
> 
> This looks important enough (and low risk enough) to push to Linus
> before the next merge window.   Comments?
> 
> 						- Ted

yeah, I think so.

if there's any concern over the size_t type changes I could
split it up.

-Eric

Re: [PATCH V2] fix type of "offset" in ext4_io_end

[tytso]

On Fri, Jan 29, 2010 at 02:28:23PM -0600, Eric Sandeen wrote:
> The "offset" member in ext4_io_end holds bytes, not
> blocks, so ext4_lblk_t is wrong - and too small (u32)
> 
> This caused the testcase "Possible ext4 data corruption
> with large files and async I/O" sent by Giel to fail when it
> wrapped around to 0.
> 
> Also fix up the type of arguments to
> ext4_convert_unwritten_extents(), it gets ssize_t from
> ext4_end_aio_dio_nolock() and ext4_ext_direct_IO()
> 
> Reported-by: Giel de Nijs <giel@vectorwise.com>
> Signed-off-by: Eric Sandeen <sandeen@redhat.com>

So I was going to submit this patch to Linus, but the last two times
I've run xfsqa ("xfsqa -g quick" and "xfsqa -g auto"), test #126 has
failed.  If I run the test stand-alone, it passes.  It's a bit of a
head-scratcher.  I'm currently backing out the patch and trying to do
an xfsqa -g auto run to make sure this was something that had crept in
before applying this patch, so this may very well not be this patch
--- I certainly can't see anything wrong with it.  But I thought I
would give a heads up....

      	     	       	       	      	   	- Ted

Re: [PATCH V2] fix type of "offset" in ext4_io_end

[Eric Sandeen]

tytso@mit.edu wrote:
> On Fri, Jan 29, 2010 at 02:28:23PM -0600, Eric Sandeen wrote:
>> The "offset" member in ext4_io_end holds bytes, not
>> blocks, so ext4_lblk_t is wrong - and too small (u32)
>>
>> This caused the testcase "Possible ext4 data corruption
>> with large files and async I/O" sent by Giel to fail when it
>> wrapped around to 0.
>>
>> Also fix up the type of arguments to
>> ext4_convert_unwritten_extents(), it gets ssize_t from
>> ext4_end_aio_dio_nolock() and ext4_ext_direct_IO()
>>
>> Reported-by: Giel de Nijs <giel@vectorwise.com>
>> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> 
> So I was going to submit this patch to Linus, but the last two times
> I've run xfsqa ("xfsqa -g quick" and "xfsqa -g auto"), test #126 has
> failed.  If I run the test stand-alone, it passes.  It's a bit of a
> head-scratcher.  I'm currently backing out the patch and trying to do
> an xfsqa -g auto run to make sure this was something that had crept in
> before applying this patch, so this may very well not be this patch
> --- I certainly can't see anything wrong with it.  But I thought I
> would give a heads up....
> 
>       	     	       	       	      	   	- Ted

hrm 126 is a permissions test, I can't imagine it'd be related...

I see it fail on a sorta-stock .32 kernel like this:

--- 126.out	2009-08-05 20:19:40.380978986 -0500
+++ 126.out.bad	2010-02-05 09:01:13.088372667 -0600
@@ -5,7 +5,7 @@
 w a 002 file owned by (99/99) as user/group(12/100)  PASS
 w a 020 file owned by (99/99) as user/group(200/99)  PASS
 w a 200 file owned by (99/99) as user/group(99/500)  PASS
-r a 004 file owned by (99/99) as user/group(12/100)  PASS
+r a 004 file owned by (99/99) as user/group(12/100)  FAIL
 r a 040 file owned by (99/99) as user/group(200/99)  PASS
 r a 400 file owned by (99/99) as user/group(99/500)  PASS
 r a 000 file owned by (99/99) as user/group(99/99)  FAIL


What's your failure?

-Eric

Re: [PATCH V2] fix type of "offset" in ext4_io_end

[tytso]

On Fri, Feb 05, 2010 at 08:37:51AM -0500, tytso@MIT.EDU wrote:
> So I was going to submit this patch to Linus, but the last two times
> I've run xfsqa ("xfsqa -g quick" and "xfsqa -g auto"), test #126 has
> failed.  If I run the test stand-alone, it passes.  It's a bit of a
> head-scratcher.  I'm currently backing out the patch and trying to do
> an xfsqa -g auto run to make sure this was something that had crept in
> before applying this patch, so this may very well not be this patch
> --- I certainly can't see anything wrong with it.  But I thought I
> would give a heads up....

Eric and I did more looking at this, and looks like it's a busted
test.  So we're considering this a false alarm for now.

(Safety Pup says: "Null terminate your arguments to fopen!")

	    	  		       - Ted