Bug 32972

Reply-To: xiaoqiangnk@gmail.com

Ah, it is a wrong fix.

original code:
	map_bh(bh, inode->i_sb, map.m_pblk);
	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;

	if (buffer_unwritten(bh)) {
		/* A delayed write to unwritten bh should be marked
		 * new and mapped.  Mapped ensures that we don't do
		 * get_block multiple times when we write to the same
		 * offset and new ensures that we do proper zero out
		 * for partial write.
		 */
		set_buffer_new(bh);
		set_buffer_mapped(bh);
	}

after the patch:
	map_bh(bh, inode->i_sb, map.m_pblk);
	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;

	if (buffer_unwritten(bh)) {
		/* A delayed write to unwritten bh should be marked
		 * new and mapped.  Mapped ensures that we don't do
		 * get_block multiple times when we write to the same
		 * offset and new ensures that we do proper zero out
		 * for partial write.
		 */
		set_buffer_new(bh);
	}

Actually, mapped flag is cleared by statement:
	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;

So the right code should be：

	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
	map_bh(bh, inode->i_sb, map.m_pblk);
	if (buffer_unwritten(bh)) {
		/* A delayed write to unwritten bh should be marked
		 * new and mapped.  Mapped ensures that we don't do
		 * get_block multiple times when we write to the same
		 * offset and new ensures that we do proper zero out
		 * for partial write.
		 */
		set_buffer_new(bh);
	}



On Sun, Apr 10, 2011 at 9:44 PM,  <bugzilla-daemon@bugzilla.kernel.org> wrote:
> https://bugzilla.kernel.org/show_bug.cgi?id=32972
>
>           Summary: EXT4 causes corrupt BitTorrent downloads
>           Product: File System
>           Version: 2.5
>    Kernel Version: 2.6.39-rc2+
>          Platform: All
>        OS/Version: Linux
>              Tree: Mainline
>            Status: NEW
>          Severity: high
>          Priority: P1
>         Component: ext4
>        AssignedTo: fs_ext4@kernel-bugs.osdl.org
>        ReportedBy: damien@grassart.com
>                CC: feng.tang@intel.com
>        Regression: Yes
>
>
> Using the Transmission BitTorrent client (version 2.11) with the most recent
> kernel (2.6.39-rc2+), any torrent I try to download is corrupt. By using
> Transmission's "Verify Local Data" functionality, I am able to reproduce this
> consistently. During verification, the percent complete will reverse itself
> as
> the verification process invalidates pieces that were just downloaded.
>
> I was able to bisect the issue and track it down to this commit:
>
> commit 6de9843dab3f2a1d4d66d80aa9e5782f80977d20
> Author: Feng Tang <feng.tang@intel.com>
> Date:   Wed Mar 23 14:05:03 2011 -0400
>
>    ext4: remove redundant set_buffer_mapped() in ext4_da_get_block_prep()
>
>    The map_bh() call will have already set the buffer_head to mapped.
>
>    Signed-off-by: Feng Tang <feng.tang@intel.com>
>    Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
>
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index f44307a..dec10e2 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -2502,7 +2502,6 @@ static int ext4_da_get_block_prep(struct inode *inode,
> sector_t iblock,
>                 * for partial write.
>                 */
>                set_buffer_new(bh);
> -               set_buffer_mapped(bh);
>        }
>        return 0;
>  }
>
>
> I confirmed that reverting this commit makes the problem go away.
>
> Thanks,
> -Damien
>
> --
> Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
> ------- You are receiving this mail because: -------
> You are watching the assignee of the bug.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Hi, thanks for the quick response. I've just tested and confirmed that your fix works.

Also, should that comment inside the conditional be updated (or moved up) as well? Otherwise it seems to be out of sync with the code.

Thanks,
-Damien

On Sun, Apr 10, 2011 at 10:30:13PM +0800, Yongqiang Yang wrote:
> So the right code should be：
> 
>       bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
>       map_bh(bh, inode->i_sb, map.m_pblk);
>       if (buffer_unwritten(bh)) {
>               /* A delayed write to unwritten bh should be marked
>                * new and mapped.  Mapped ensures that we don't do
>                * get_block multiple times when we write to the same
>                * offset and new ensures that we do proper zero out
>                * for partial write.
>                */
>               set_buffer_new(bh);
>       }

Actually, I'm much more comfortable backing out commit 6de9843da
entirely.  The above is *not* equivalent to what we had before ---
consider the case where ext4_map_blocks returns !EXT4_MAP_MAPPED && 
!EXT4_MAP_UNWRITTEN.

I don't *think* this should happen in the case where ext4_map_blocks
returns a value > 0, but the fact that it's not obvious, means I'd
much rather keep things the way that they are.  It's not like dropping
the set_buffer_mapped(bh) was saving anything measurable anyway....

						- Ted

On Mon, Apr 11, 2011 at 9:46 AM, Ted Ts'o <tytso@mit.edu> wrote:
> On Sun, Apr 10, 2011 at 10:30:13PM +0800, Yongqiang Yang wrote:
>> So the right code should be：
>>
>>       bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
>>       map_bh(bh, inode->i_sb, map.m_pblk);
>>       if (buffer_unwritten(bh)) {
>>               /* A delayed write to unwritten bh should be marked
>>                * new and mapped.  Mapped ensures that we don't do
>>                * get_block multiple times when we write to the same
>>                * offset and new ensures that we do proper zero out
>>                * for partial write.
>>                */
>>               set_buffer_new(bh);
>>       }
>
> Actually, I'm much more comfortable backing out commit 6de9843da
> entirely.  The above is *not* equivalent to what we had before ---
> consider the case where ext4_map_blocks returns !EXT4_MAP_MAPPED &&
> !EXT4_MAP_UNWRITTEN.
>
> I don't *think* this should happen in the case where ext4_map_blocks
> returns a value > 0, but the fact that it's not obvious, means I'd
> much rather keep things the way that they are.  It's not like dropping
> the set_buffer_mapped(bh) was saving anything measurable anyway....
Agree.  this way, the comment for unwritten case is also much clearer.

Yongqiang
>
>                                                - Ted
>

> 
> Actually, I'm much more comfortable backing out commit 6de9843da
> entirely.  The above is *not* equivalent to what we had before ---
> consider the case where ext4_map_blocks returns !EXT4_MAP_MAPPED && 
> !EXT4_MAP_UNWRITTEN.
> 
> I don't *think* this should happen in the case where ext4_map_blocks
> returns a value > 0, but the fact that it's not obvious, means I'd
> much rather keep things the way that they are.  It's not like dropping
> the set_buffer_mapped(bh) was saving anything measurable anyway....
> 
>                         - Ted

Agree for the revert and sorry for the inconvenience brought by my patch. I walked across the code when debugging a problem, and thought the patch can
clean the code a little :( 

Thanks,
Feng

First-Bad-Commit : 6de9843dab3f2a1d4d66d80aa9e5782f80977d20

A patch referencing this bug report has been merged in v2.6.39-rc3:

commit c8205636029fc869278c55b7336053b3e7ae3ef4
Author: Theodore Ts'o <tytso@mit.edu>
Date:   Sun Apr 10 22:30:07 2011 -0400

    ext4: fix data corruption regression by reverting commit 6de9843dab3f

*** Bug 33112 has been marked as a duplicate of this bug. ***