Re: Workaroundable bug with 2.6.0-test3 xfs & xfs_growfs

[Tupshin Harper <tupshin@tupshin.com>]

Nathan Scott wrote:

>On Tue, Aug 19, 2003 at 12:34:00PM +0300, Erkki Seppala wrote:
>  
>
>>Growing filesystem twice after mount fails. Reproducing:
>>
>>1) mount /dev/store/media2 - 100G volume
>>2) lvextend -L 200G /dev/store/media2
>>3) xfs_growfs -d /dev/store/media2
>>  -> works ok
>>4) lvextend -L 350G /dev/store/media2
>>5) xfs_growfs -d /dev/store/media2
>>  -> says there is nothing to be done
>>
>>Unmounting and mounting the system will allow xfs_growfs to work
>>again.
>>
>>    
>>
>
>This sounds like the device is not properly updating its size.
>Can you cat /proc/partitions after each device size extension,
>and check that it is the expected size.
>  
>
/proc/partitions only contains regular hard drive partitions. The 
problem is occuring on lvm partitions. Is there some equivalent check 
you would like me to make for lvm2?

>xfs_growfs is issuing an ioctl querying the device size before
>it tells the XFS kernel code what size to extend to - for some
>reason it seems that your second growfs there is getting the
>same size as before... if so, this would point to a problem in
>the device driver and not XFS.
>
>cheers.
>  
>
I'd believe this.

>Can you try the latest XFS 1.3-pre6 code and see if the problem
>is resolved there?  There was a growfs fix went in there in the
>last few days.  Make sure you start from a known not-corrupt fs
>of course.  Seth, I think that FAQ entry can be removed/updated
>now to say the problem is resolved in 1.3.
>
As I mentioned, I'm using kernel 2.6.x where X is newer than test3. I'm 
not sure if you are referring to userland tools, or the kernel code. If 
you mean userland stuff, than I can try it out, but if you mean kernel 
code, I'm not aware of a patch against recent 2.6 kernels.

-Tupshin