Kernel 2.6.30.4 loop(..?) regression (& with/2.6.31-rc6)

[Justin Piszcz]

On Wed, 26 Aug 2009, Christoph Hellwig wrote:

> On Wed, Aug 26, 2009 at 05:19:11PM -0400, Justin Piszcz wrote:
>> For the root filesystem, on / type xfs (rw,noatime)
>> For larger partitions (per the cryptoloop doc) the partition itself is
>> setup via cryptoloop and then XFS ontop of that.  Nothing special.
>
> Just to make sure we have all data here - you see the problem only
> with loop devices backed by XFS, only with loop devices backed by
> raw partitions or with both?
Both.

>
>>> Any way to trgiger it with just loop but not crypto, etc.
>> Unfortunately not that I am aware of..  Would trying the kernel with the
>> patch removed and mounted with -o nobarrier help to show us anything, or?
>
> Actually, yes - seeing what happens if you run plain 2.6.30 or 2.6.31-rc
> without the backout patch, but with -o nobarrier would be very
> interesting.
>
Will give it a try with 2.6.31-rc6 and mount all loop devices (both those 
backed by XFS and raw partitions) with -o nobarrier to see if the problem 
recurs.  So far though with that patch up until this next test, there were 
no problems.

Justin.