Re: inet6_dump_fib lock held when returning to user space

[David Miller <davem@davemloft.net>]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 19 Jan 2012 13:33:19 +0100

> Le jeudi 19 janvier 2012 à 11:48 +0100, Hans Schillstrom a écrit :
>> Hello
>> Have this one been seen before ?
>> It happens during a container start i.e. in a netns,
>> 
>> [   19.232941] ================================================
>> [   19.268785] [ BUG: lock held when returning to user space! ]
>> [   19.270146] 3.2.0+ #14 Tainted: G        W   
>> [   19.271217] ------------------------------------------------
>> [   19.272968] nsm/1042 is leaving the kernel with locks still held!
>> [   19.274603] 1 lock held by nsm/1042:
>> [   19.275524]  #0:  (rcu_read_lock){.+.+..}, at: [<ffffffff8136a9b2>] inet6_dump_fib+0xbd/0x34a
>> 
> 
> Yep, Dave Jones reported this some days ago
> 
> "Re: recvmsg sleeping from invalid context"
> 
> I have no idea of what happens yet.

The only way I can this happening in inet6_dump_fib() is if we somehow
sleep during fib6_dump_table(), and then return to userspace in
another task.

But that makes no sense, since in such a case we should see the
warning instead when the task schedules out, not when the
inet6_dump_fib calling task goes back to userspace.

I wonder if Dave Jones's report is even related to this one.