Re: [GIT PULL] Fix for btrfs/070 checksum error

[Chris Mason <clm@fb.com>]

On Wed, Jul 29, 2015 at 04:21:33PM +0800, Zhao Lei wrote:
> Hi, Chris
> 
> > -----Original Message-----
> > From: linux-btrfs-owner@vger.kernel.org
> > [mailto:linux-btrfs-owner@vger.kernel.org] On Behalf Of Qu Wenruo
> > Sent: Tuesday, July 28, 2015 3:11 PM
> > To: Chris Mason; btrfs
> > Subject: Re: [GIT PULL] Fix for btrfs/070 checksum error
> > 
> > Chris Mason wrote on 2015/07/23 21:57 -0400:
> > > On Fri, Jul 24, 2015 at 08:29:05AM +0800, Qu Wenruo wrote:
> > >
> > > [ deadlock with the 070 patches ]
> > >
> > >> Thanks Chris
> > >>
> > >> We will investigate it with highest priority.
> > >>
> > >> Thanks,
> > >> Qu
> > >>
> > >
> > > Thanks!  I'm doing a few more runs to make sure the lockup is new with
> > > these patches.
> > >
> > > -chris
> > >
> > Hi Chris,
> > 
> > I'm very sorry that we are unable to fix the lockup in a short time, so it may not
> > fit in the v4.2 merge window.
> > 
> > Please ignore this patchset for now.
> > 
> 
> Sorry for taking quite a long time for investigate because it is
> randomly happened.
> 
> We got reason of process blocking:
> 1: In some case, this patch caused __btrfs_cow_block()->btrfs_reloc_cow_block()
>   failed from btrfs_balance operation.(need more investigation)
> 
> 2: __btrfs_cow_block()'s error handle code hadn't unlock/free
>   new_allocated tree block before return error.
> 
> 3: do_relocation(), which is caller of __btrfs_cow_block(), have error handle
>   code, but also can't work in this case, because new_allocated eb is not
>   returned.
> 
> 4: subsequent code in do_relocation() try to lock above eb again,
>   and caused dead lock.

Excellent, thanks for tracking this down.  I agree investigating #1 is
the top priority, since it's possible the patches are just making it
happen more often.

-chris