XFS Direct IO, unaligned block zeroing at 128K boundery

XFS Direct IO, unaligned block zeroing at 128K boundery

[Ajeet Yadav]

[-- Attachment #1.1: Type: text/plain, Size: 3114 bytes --]

Dear all
I made a simple script that creates a 2MB file using directio write
operation,
The objective was to test with different buffer sizes from 4K-5K-6K..8K, the
buffer was filled with 5,
Than take hexdump of file to check data integrity.

With 4K the file is fine, but with 5K, It can be noticed that

there is zeroing is at 128K boundry further this shift 1K on successive 128K
chunk.
I guess this is a bug.
------------------------------------------------------------------------------------
*****************************
BUFFER SIZE : 4096
*****************************
0000000 0505 0505 0505 0505 0505 0505 0505 0505
*
0200000
*****************************
BUFFER SIZE : 5120
*****************************
0000000 0505 0505 0505 0505 0505 0505 0505 0505
*
001f400 0000 0000 0000 0000 0000 0000 0000 0000
*
0020000 0505 0505 0505 0505 0505 0505 0505 0505
*
002f800 0000 0000 0000 0000 0000 0000 0000 0000
*
0030000 0505 0505 0505 0505 0505 0505 0505 0505
*
003fc00 0000 0000 0000 0000 0000 0000 0000 0000
*
0040000 0505 0505 0505 0505 0505 0505 0505 0505
*
006f400 0000 0000 0000 0000 0000 0000 0000 0000
*
0070000 0505 0505 0505 0505 0505 0505 0505 0505
*
007f800 0000 0000 0000 0000 0000 0000 0000 0000
*
0080000 0505 0505 0505 0505 0505 0505 0505 0505
*
008fc00 0000 0000 0000 0000 0000 0000 0000 0000
*
0090000 0505 0505 0505 0505 0505 0505 0505 0505
*
00bf400 0000 0000 0000 0000 0000 0000 0000 0000
*
00c0000 0505 0505 0505 0505 0505 0505 0505 0505
*
00cf800 0000 0000 0000 0000 0000 0000 0000 0000
*
00d0000 0505 0505 0505 0505 0505 0505 0505 0505
*
00dfc00 0000 0000 0000 0000 0000 0000 0000 0000
*
00e0000 0505 0505 0505 0505 0505 0505 0505 0505
*
010f400 0000 0000 0000 0000 0000 0000 0000 0000
*
0110000 0505 0505 0505 0505 0505 0505 0505 0505
*
011f800 0000 0000 0000 0000 0000 0000 0000 0000
*
0120000 0505 0505 0505 0505 0505 0505 0505 0505
*
012fc00 0000 0000 0000 0000 0000 0000 0000 0000
*
0130000 0505 0505 0505 0505 0505 0505 0505 0505
*
015f400 0000 0000 0000 0000 0000 0000 0000 0000
*
0160000 0505 0505 0505 0505 0505 0505 0505 0505
*
016f800 0000 0000 0000 0000 0000 0000 0000 0000
*
0170000 0505 0505 0505 0505 0505 0505 0505 0505
*
017fc00 0000 0000 0000 0000 0000 0000 0000 0000
*
0180000 0505 0505 0505 0505 0505 0505 0505 0505
*
01af400 0000 0000 0000 0000 0000 0000 0000 0000
*
01b0000 0505 0505 0505 0505 0505 0505 0505 0505
*
01bf800 0000 0000 0000 0000 0000 0000 0000 0000
*
01c0000 0505 0505 0505 0505 0505 0505 0505 0505
*
01cfc00 0000 0000 0000 0000 0000 0000 0000 0000
*
01d0000 0505 0505 0505 0505 0505 0505 0505 0505
*
01ff400

On Wed, Nov 24, 2010 at 10:10 AM, Ajeet Yadav <ajeet.yadav.77@gmail.com>wrote:

> We have backported complete xfs from 2.6.34 to 2.6.30.9, because there were
> many fixes since 2.6.30.9
>   On Tue, Nov 23, 2010 at 6:51 PM, Christoph Hellwig <hch@infradead.org>wrote:
>
>> On Tue, Nov 23, 2010 at 03:50:37PM +0530, Ajeet Yadav wrote:
>> > I have the following backtrace on my ARM target, please have a look
>>
>> Are sure you're on 2.6.34?  Something similar was fixed by commit
>> f1f724e4b523d444c5a598d74505aefa3d6844d2 in 2.6.34.
>>
>>
>

[-- Attachment #1.2: Type: text/html, Size: 4099 bytes --]

[-- Attachment #2: Type: text/plain, Size: 121 bytes --]

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

Re: XFS Direct IO, unaligned block zeroing at 128K boundery

[Dave Chinner]

On Thu, Nov 25, 2010 at 12:23:11PM +0530, Ajeet Yadav wrote:
> Dear all
> I made a simple script that creates a 2MB file using directio write
> operation,
> The objective was to test with different buffer sizes from 4K-5K-6K..8K, the
> buffer was filled with 5,
> Than take hexdump of file to check data integrity.

what kernel? Can you reproduce it on a mainline kernel? If so,
can you apply the three patches in this thread:

http://marc.info/?l=linux-fsdevel&m=128920213904443&w=2

and see if that fixes the problem?

Also, if you are writing new test scripts that demonstrate problems,
it's handy to attach them to your bug reports, or better yet, send
patches to add them to the xfstests suite....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

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

Re: XFS Direct IO, unaligned block zeroing at 128K boundery

[Ajeet Yadav]

[-- Attachment #1.1: Type: text/plain, Size: 1138 bytes --]

Hello,
Its my fault,
Our kernel 2.6.30.9, however we migrated XFS to higher version 2.6.34

We undo the XFS changes from commit 5fe878ae7f82fbf0830dbfaee4c5ca18f3aee442
because our fs/direct_io.c belong to version 2.6.30.9, and we do not want to
change it.


On Thu, Nov 25, 2010 at 3:27 PM, Dave Chinner <david@fromorbit.com> wrote:

> On Thu, Nov 25, 2010 at 12:23:11PM +0530, Ajeet Yadav wrote:
> > Dear all
> > I made a simple script that creates a 2MB file using directio write
> > operation,
> > The objective was to test with different buffer sizes from 4K-5K-6K..8K,
> the
> > buffer was filled with 5,
> > Than take hexdump of file to check data integrity.
>
> what kernel? Can you reproduce it on a mainline kernel? If so,
> can you apply the three patches in this thread:
>
> http://marc.info/?l=linux-fsdevel&m=128920213904443&w=2
>
> and see if that fixes the problem?
>
> Also, if you are writing new test scripts that demonstrate problems,
> it's handy to attach them to your bug reports, or better yet, send
> patches to add them to the xfstests suite....
>
> Cheers,
>
> Dave.
> --
> Dave Chinner
> david@fromorbit.com
>

[-- Attachment #1.2: Type: text/html, Size: 1718 bytes --]

[-- Attachment #2: Type: text/plain, Size: 121 bytes --]

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