xfs_io "BUG: lock held when returning to user space!" on suspend

xfs_io "BUG: lock held when returning to user space!" on suspend

[Jeremy Fitzhardinge]

I'm getting this:

================================================
[ BUG: lock held when returning to user space! ]
------------------------------------------------
xfs_io/18796 is leaving the kernel with locks still held!
1 lock held by xfs_io/18796:
 #0:  (&type->s_umount_key#19){----}, at: [<c048c9ac>] get_super+0x42/0x87

when I suspend, possibly during an xfs-freeze.  There don't seem to be any ill-effects.

Thanks,
	J

Re: xfs_io "BUG: lock held when returning to user space!" on suspend

[David Chinner]

On Wed, Apr 02, 2008 at 11:58:54AM -0700, Jeremy Fitzhardinge wrote:
> I'm getting this:
> 
> ================================================
> [ BUG: lock held when returning to user space! ]
> ------------------------------------------------
> xfs_io/18796 is leaving the kernel with locks still held!
> 1 lock held by xfs_io/18796:
> #0:  (&type->s_umount_key#19){----}, at: [<c048c9ac>] get_super+0x42/0x87
> 
> when I suspend, possibly during an xfs-freeze.  There don't seem to be any 
> ill-effects.

Yup, both the sb->s_umount and bdev->bd_mount_sem seaphores are held
across freeze_bdev()/thaw_bdev(), and they are issued via separate
ioctls generally from separate processes. Not great design, but not
a bug....

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group

Re: xfs_io "BUG: lock held when returning to user space!" on suspend

[Peter Zijlstra]

On Thu, 2008-04-03 at 07:57 +1000, David Chinner wrote:
> On Wed, Apr 02, 2008 at 11:58:54AM -0700, Jeremy Fitzhardinge wrote:
> > I'm getting this:
> > 
> > ================================================
> > [ BUG: lock held when returning to user space! ]
> > ------------------------------------------------
> > xfs_io/18796 is leaving the kernel with locks still held!
> > 1 lock held by xfs_io/18796:
> > #0:  (&type->s_umount_key#19){----}, at: [<c048c9ac>] get_super+0x42/0x87
> > 
> > when I suspend, possibly during an xfs-freeze.  There don't seem to be any 
> > ill-effects.
> 
> Yup, both the sb->s_umount and bdev->bd_mount_sem seaphores are held
> across freeze_bdev()/thaw_bdev(), and they are issued via separate
> ioctls generally from separate processes. Not great design, but not
> a bug....

Actually... we do consider that bugs.

Re: xfs_io "BUG: lock held when returning to user space!" on suspend

[Peter Zijlstra]

On Thu, 2008-04-03 at 15:53 +0200, Peter Zijlstra wrote:
> On Thu, 2008-04-03 at 07:57 +1000, David Chinner wrote:
> > On Wed, Apr 02, 2008 at 11:58:54AM -0700, Jeremy Fitzhardinge wrote:
> > > I'm getting this:
> > > 
> > > ================================================
> > > [ BUG: lock held when returning to user space! ]
> > > ------------------------------------------------
> > > xfs_io/18796 is leaving the kernel with locks still held!
> > > 1 lock held by xfs_io/18796:
> > > #0:  (&type->s_umount_key#19){----}, at: [<c048c9ac>] get_super+0x42/0x87
> > > 
> > > when I suspend, possibly during an xfs-freeze.  There don't seem to be any 
> > > ill-effects.
> > 
> > Yup, both the sb->s_umount and bdev->bd_mount_sem seaphores are held
> > across freeze_bdev()/thaw_bdev(), and they are issued via separate
> > ioctls generally from separate processes. Not great design, but not
> > a bug....
> 
> Actually... we do consider that bugs.

http://lkml.org/lkml/2007/10/27/135

Where Linus says:

 Definitely not a sane thing to do. It should use ref-counting and/or a 
single bit to say "busy". 

Re: xfs_io "BUG: lock held when returning to user space!" on suspend

[David Chinner]

On Thu, Apr 03, 2008 at 03:53:14PM +0200, Peter Zijlstra wrote:
> On Thu, 2008-04-03 at 07:57 +1000, David Chinner wrote:
> > On Wed, Apr 02, 2008 at 11:58:54AM -0700, Jeremy Fitzhardinge wrote:
> > > I'm getting this:
> > > 
> > > ================================================
> > > [ BUG: lock held when returning to user space! ]
> > > ------------------------------------------------
> > > xfs_io/18796 is leaving the kernel with locks still held!
> > > 1 lock held by xfs_io/18796:
> > > #0:  (&type->s_umount_key#19){----}, at: [<c048c9ac>] get_super+0x42/0x87
> > > 
> > > when I suspend, possibly during an xfs-freeze.  There don't seem to be any 
> > > ill-effects.
> > 
> > Yup, both the sb->s_umount and bdev->bd_mount_sem seaphores are held
> > across freeze_bdev()/thaw_bdev(), and they are issued via separate
> > ioctls generally from separate processes. Not great design, but not
> > a bug....
> 
> Actually... we do consider that bugs.

Ok, you're welcome to fix it then. It's way below my care factor
right now....

FWIW, I did suggest that this needed fixing before FIFREEZE/FITHAW
are introduced, but that hasn't happened yet.

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group