[BUG] 2.6.24-rc2-mm1 - kernel bug on nfs v4

[Torsten Kaiser]

On Nov 18, 2007 12:05 AM, Peter Zijlstra <a.p.zijlstra at chello.nl> wrote:
>
> On Sat, Nov 17, 2007 at 08:40:22PM +0100, Torsten Kaiser wrote:
>
> > Lockdep triggers immedetly before the freeze, but the result is still
> > not helpful:
> >
> > [  221.565011] INFO: trying to register non-static key.
> > [  221.566999] the code is fine but needs lockdep annotation.
> > [  221.569206] turning off the locking correctness validator.
> > [  221.571404]
> > [  221.571405] Call Trace:
> > [  221.572996]  [<ffffffff8025a1b4>] __lock_acquire+0x4c4/0x1140
> > [  221.575298]  [<ffffffff8025ae85>] lock_acquire+0x55/0x70
> > [  221.577429]  [<ffffffff8022d6fd>] __wake_up+0x2d/0x70
> > [  221.579457]  [<ffffffff805c5f04>] _spin_lock_irqsave+0x34/0x50
> > [  221.581800]  [<ffffffff805c5e45>] _spin_unlock_irqrestore+0x55/0x70
> > [  221.584317]  [<ffffffff8022d6fd>] __wake_up+0x2d/0x70
> > [  221.586344]  [<ffffffff805a88b0>] rpc_async_schedule+0x0/0x10
> > [  221.588648]  [<ffffffff802fface>] nfs_free_unlinkdata+0x1e/0x50
> > [  221.591023]  [<ffffffff805a7e96>] rpc_release_calldata+0x26/0x50
> > [  221.593428]  [<ffffffff8024778f>] run_workqueue+0x16f/0x210
> > [  221.595662]  [<ffffffff80259731>] trace_hardirqs_on+0xc1/0x160
> > [  221.598004]  [<ffffffff802483d0>] worker_thread+0x0/0xb0
> > [  221.600130]  [<ffffffff802483d0>] worker_thread+0x0/0xb0
> > [  221.602265]  [<ffffffff8024843d>] worker_thread+0x6d/0xb0
> > [  221.604431]  [<ffffffff8024bfc0>] autoremove_wake_function+0x0/0x30
> > [  221.606939]  [<ffffffff802483d0>] worker_thread+0x0/0xb0
> > [  221.609067]  [<ffffffff802483d0>] worker_thread+0x0/0xb0
> > [  221.611199]  [<ffffffff8024bbeb>] kthread+0x4b/0x80
> > [  221.613156]  [<ffffffff8020cb98>] child_rip+0xa/0x12
> > [  221.615151]  [<ffffffff8020c2af>] restore_args+0x0/0x30
> > [  221.617247]  [<ffffffff8024bba0>] kthread+0x0/0x80
> > [  221.619162]  [<ffffffff8020cb8e>] child_rip+0x0/0x12
> > [  221.621147]
> > [  221.621749] INFO: lockdep is turned off.
>
> I've been staring at this NFS code for a while an can't make any sense
> out of it. It seems to correctly initialize the waitqueue. So this would
> indicate corruption of some sort.

Not sure if this is helpful, but after looking into the code, the
above stacktrace looks somewhat damaged.
Might be my fault: # CONFIG_FRAME_POINTER is not set
On the other hand the stacktrace from the run with the SLUB lockdep
fix shows the same function names.

That trace contains this line:
 [<ffffffff8030167e>] nfs_free_unlinkdata+0x1e/0x50
(gdb) list *0xffffffff8030167e
0xffffffff8030167e is in nfs_free_unlinkdata (fs/nfs/unlink.c:33).
28       */
29      static void
30      nfs_free_unlinkdata(struct nfs_unlinkdata *data)
31      {
32              nfs_sb_deactive(NFS_SERVER(data->dir));
33              iput(data->dir);
34              put_rpccred(data->cred);
35              kfree(data->args.name.name);
36              kfree(data);
37      }

Is some inode lock guilty?
Please ask, if you need more information.

Torsten