Re: 2.6.35+vfs-scale: INFO: suspicious rcu_dereference_check()usage (kernel/exit.c:1387)

[Paul E. McKenney]

On Mon, Aug 16, 2010 at 07:03:40PM +0200, Vegard Nossum wrote:
> On the third try it will hopefully work.
> 
> 
> Vegard
> 
> ---------- Forwarded message ----------
> From: Vegard Nossum <vegard.nossum@xxxxxxxxx>
> Date: 16 August 2010 19:01
> Subject: Re: 2.6.35+vfs-scale: INFO: suspicious
> rcu_dereference_check() usage (kernel/exit.c:1387)
> To: Nick Piggin <nickpiggin@xxxxxxxxxxxx>
> Cc: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>, David Howells
> <dhowells@xxxxxxxxxx>, "Paul E. McKenney"
> <paulmck@xxxxxxxxxxxxxxxxxx>, LKML <linux-kernel@xxxxxxxxxxxxxxx>
> 
> 
> Nick Piggin might update his git config because the commits in the
> vfs-scale tree appear to bear an invalid e-mail address. Please reply
> to this e-mail instead. Thanks.
> 
> Vegard
> 
> 
> ---------- Forwarded message ----------
> From: Vegard Nossum <vegard.nossum@xxxxxxxxx>
> Date: 16 August 2010 18:57
> Subject: 2.6.35+vfs-scale: INFO: suspicious rcu_dereference_check()
> usage (kernel/exit.c:1387)
> To: Nick Piggin <npiggin@xxxxxxx>
> Cc: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>, David Howells
> <dhowells@xxxxxxxxxx>, "Paul E. McKenney"
> <paulmck@xxxxxxxxxxxxxxxxxx>, LKML <linux-kernel@xxxxxxxxxxxxxxx>
> 
> 
> Hi,
> 
> I was testing your VFS scaling tree*, but I get the following in my dmesg:
> 
> ===================================================
> [ INFO: suspicious rcu_dereference_check() usage. ]
> ---------------------------------------------------
> kernel/exit.c:1387 invoked rcu_dereference_check() without protection!
> 
> other info that might help us debug this:
> 
> 
> rcu_scheduler_active = 1, debug_locks = 1
> 2 locks held by bash/2878:
>  #0:  (tasklist_lock){.+.+..}, at: [<ffffffff8103a275>] do_wait+0xb3/0x200
>  #1:  (&(&sighand->siglock)->rlock){-.....}, at: [<ffffffff810399e9>]
> wait_consider_task+0x2ca/0xaa3

Thank you for locating this!

OK, so we aren't holding rcu_read_lock(), nor is this task dead, so
something needs some help.

We get from wait_consider_task() as follows: wait_consider_task()
calls security_task_wait(), which calls either selinux_task_wait() or
smack_task_wait().  Let's start with selinux_task_wait(), which calls
task_has_perm(), which does a couple of __task_cred(), but does so
under rcu_read_lock() protection, so presumably we didn't come in
that way.

Dave, am I missing a code path somewhere here?

							Thanx, Paul

> stack backtrace:
> Pid: 2878, comm: bash Not tainted 2.6.35-00079-g4067b85 #7
> Call Trace:
>  [<ffffffff8105e9f2>] lockdep_rcu_dereference+0x9b/0xa6
>  [<ffffffff8103a036>] wait_consider_task+0x917/0xaa3
>  [<ffffffff8102cf36>] ? add_preempt_count+0xc2/0xc9
>  [<ffffffff8103a29d>] do_wait+0xdb/0x200
>  [<ffffffff8103a45b>] sys_wait4+0x99/0xd3
>  [<ffffffff810387e5>] ? child_wait_callback+0x0/0x5d
>  [<ffffffff8109be66>] ? might_fault+0x96/0xbc
>  [<ffffffff810737a6>] compat_sys_wait4+0x8a/0xc1
>  [<ffffffff8109be83>] ? might_fault+0xb3/0xbc
>  [<ffffffff8109be38>] ? might_fault+0x68/0xbc
>  [<ffffffff815d02c3>] ? trace_hardirqs_off_thunk+0x3a/0x3c
>  [<ffffffff8102862c>] sys32_waitpid+0xb/0xd
>  [<ffffffff81027bb4>] sysenter_dispatch+0x7/0x30
>  [<ffffffff815d0284>] ? trace_hardirqs_on_thunk+0x3a/0x3f
> 
> I don't think it's related to the VFS patches, however, so I added
> some other (maybe relevant) CCs too.
> 
> At the time I was simply invoking gcc from bash.
> 
> 
> Vegard
> 
> * git://git.kernel.org/pub/scm/linux/kernel/git/npiggin/linux-npiggin.git
> branch vfs-scale
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/