Eric Dumazet wrote:
> Le dimanche 02 mai 2010 à 13:46 -0400, Valdis.Kletnieks@vt.edu a écrit :
>> On Wed, 28 Apr 2010 16:53:32 PDT, akpm@linux-foundation.org said:
>>> The mm-of-the-moment snapshot 2010-04-28-16-53 has been uploaded to
>>>
>>>    http://userweb.kernel.org/~akpm/mmotm/
>> I thought we swatted all these, hit another one...
>>
>> [    9.131490] ctnetlink v0.93: registering with nfnetlink.
>> [    9.131535]
>> [    9.131535] ===================================================
>> [    9.131704] [ INFO: suspicious rcu_dereference_check() usage. ]
>> [    9.131794] ---------------------------------------------------
>> [    9.131883] net/netfilter/nf_conntrack_ecache.c:88 invoked rcu_dereference_check() without protection!
>> [    9.131977]
>> [    9.131977] other info that might help us debug this:
>> [    9.131978]
>> [    9.132218]
>> [    9.132219] rcu_scheduler_active = 1, debug_locks = 0
>> [    9.132434] 1 lock held by swapper/1:
>> [    9.132519]  #0:  (nf_ct_ecache_mutex){+.+...}, at: [<ffffffff8148922d>] nf_conntrack_register_notifier+0x1a/0x75
>> [    9.132938]
>> [    9.132939] stack backtrace:
>> [    9.133129] Pid: 1, comm: swapper Tainted: G        W   2.6.34-rc5-mmotm0428 #1
>> [    9.133220] Call Trace:
>> [    9.133319]  [<ffffffff81064832>] lockdep_rcu_dereference+0xaa/0xb2
>> [    9.133410]  [<ffffffff81489250>] nf_conntrack_register_notifier+0x3d/0x75
>> [    9.133521]  [<ffffffff81b5a157>] ctnetlink_init+0x71/0xd5
>> [    9.133627]  [<ffffffff81b5a0e6>] ? ctnetlink_init+0x0/0xd5
>> [    9.133735]  [<ffffffff810001ef>] do_one_initcall+0x59/0x14e
>> [    9.133843]  [<ffffffff81b2e68a>] kernel_init+0x144/0x1ce
>> [    9.133949]  [<ffffffff81003414>] kernel_thread_helper+0x4/0x10
>> [    9.134060]  [<ffffffff81598a40>] ? restore_args+0x0/0x30
>> [    9.134196]  [<ffffffff81b2e546>] ? kernel_init+0x0/0x1ce
>> [    9.134328]  [<ffffffff81003410>] ? kernel_thread_helper+0x0/0x10
>> [    9.134530] ip_tables: (C) 2000-2006 Netfilter Core Team
>> [    9.134655] TCP bic registered
>>
> 
> Thanks for the report !
> 
> We can use rcu_dereference_protected() in those cases.
> 
> [PATCH] net: Use rcu_dereference_protected in nf_conntrack_ecache
> 
> Writers own nf_ct_ecache_mutex.

I've committed this patch to my tree, which also fixes up the nf_log
changes I already had queued.

I've also figured out how to prevent the false commits from showing
up using the '^' notation, I'll submit everything after some final
testing.