nodatacsum broken?

nodatacsum broken?

[Steven Pratt]

Did we break the ability to run with -o nodatacsum somewhere along the 
way? I had not run just this option in a while (usually doing nodatacow 
which implies nodatacsum).  Just did a full run with just nodatacsum and 
it looks like we still did csums.

from proc/mounts:

/dev/ffsbdev1 on /mnt/ffsb1 type btrfs (rw,nodatacsum)

but profile:

samples % app name symbol name

1664460  15.3103  vmlinux-2.6.31-autokern1 copy_user_generic_unrolled
1514349  13.9295  vmlinux-2.6.31-autokern1 crc32c
606147    5.5756  vmlinux-2.6.31-autokern1 rb_get_reader_page
562936    5.1781  vmlinux-2.6.31-autokern1 mutex_spin_on_owner
433405    3.9866  vmlinux-2.6.31-autokern1 ring_buffer_consume


This was for a sequential read test on the unstable tree from 11-12.

Steve

Re: nodatacsum broken?

[Chris Mason]

On Mon, Nov 16, 2009 at 10:22:59AM -0600, Steven Pratt wrote:
> Did we break the ability to run with -o nodatacsum somewhere along
> the way? I had not run just this option in a while (usually doing
> nodatacow which implies nodatacsum).  Just did a full run with just
> nodatacsum and it looks like we still did csums.

Were the files created without nodatasum?  The sum option actually ends
up stored in the file itself, so if you're reading from an old dataset
you'll still sum.

-chris

Re: nodatacsum broken?

[Steven Pratt]

Chris Mason wrote:
> On Mon, Nov 16, 2009 at 10:22:59AM -0600, Steven Pratt wrote:
>   
>> Did we break the ability to run with -o nodatacsum somewhere along
>> the way? I had not run just this option in a while (usually doing
>> nodatacow which implies nodatacsum).  Just did a full run with just
>> nodatacsum and it looks like we still did csums.
>>     
>
> Were the files created without nodatasum?  The sum option actually ends
> up stored in the file itself, so if you're reading from an old dataset
> you'll still sum.
>   
No, fresh mkfs and all mounts used nodatacsum.

Steve

> -chris
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>   

Re: nodatacsum broken?

[Chris Mason]

On Mon, Nov 16, 2009 at 10:22:59AM -0600, Steven Pratt wrote:
> Did we break the ability to run with -o nodatacsum somewhere along
> the way? I had not run just this option in a while (usually doing
> nodatacow which implies nodatacsum).  Just did a full run with just
> nodatacsum and it looks like we still did csums.
> 
> from proc/mounts:
> 
> /dev/ffsbdev1 on /mnt/ffsb1 type btrfs (rw,nodatacsum)
                                         ^^^^^^^^^^^^^^

This should be mount -o nodatasum.

2.6.33 will have a patch from Sage to fail mounting when we don't
recognize the option instead of silently pretending it worked.  I just
didn't want to introduce it this late in the 2.6.32-rc series.

-chris

Re: nodatacsum broken?

[Steven Pratt]

Chris Mason wrote:
> On Mon, Nov 16, 2009 at 10:22:59AM -0600, Steven Pratt wrote:
>   
>> Did we break the ability to run with -o nodatacsum somewhere along
>> the way? I had not run just this option in a while (usually doing
>> nodatacow which implies nodatacsum).  Just did a full run with just
>> nodatacsum and it looks like we still did csums.
>>
>> from proc/mounts:
>>
>> /dev/ffsbdev1 on /mnt/ffsb1 type btrfs (rw,nodatacsum)
>>     
>                                          ^^^^^^^^^^^^^^
>
> This should be mount -o nodatasum.
>
>   
Oops. My bad. Re-running.

> 2.6.33 will have a patch from Sage to fail mounting when we don't
> recognize the option instead of silently pretending it worked.  I just
> didn't want to introduce it this late in the 2.6.32-rc series.
>   
Good.

Steve


> -chris
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>