[strongSwan] Occasional kernel crash at __xfrm_state_lookup

[Jiri Horky]

Hi Martin,

thx for the response.

On 09/19/2014 09:53 AM, Martin Willi wrote:
> Hi Jiri,
>
>> We experience relatively frequent kernel crash (~2 a day out of 102
>> nodes). I enclose a stack trace:
>> [17243.492885]  [<ffffffff81601c06>] xfrm_state_lookup+0x66/0x90
>> [17243.492907]  [<ffffffff8160796e>] xfrm_user_state_lookup+0x6e/0xe0
>> [17243.492930]  [<ffffffff81255078>] ? security_capable+0x18/0x20
>> [17243.492952]  [<ffffffff81608912>] xfrm_get_sa+0x42/0xc0
>> [17243.492971]  [<ffffffff816078db>] xfrm_user_rcv_msg+0x12b/0x150
> I haven't seen such crashes on any somewhat recent kernel.
>
>> The failing offset is 0x3168, which is on line 676:
>>    675          hlist_for_each_entry(x, net->xfrm.state_byspi+h, byspi) {
>>    676                  if (x->props.family != family ||
> To me this looks like a list corruption or something, hard to say what
> causes this. Probably a bug lingering somewhere in the kernel.
That's what I think as well, I suspect it is some kind of race condition
because of number of tunnels we have.
>
> I don't know what non-vanilla changes your kernel has. Probably you
> should check that issue with your kernel distributor.
Actually, it is pure vanilla kernel v3.14.13.

I will try to post the problem on kernel.org mailing list and get back
to you if it get resolved.

Jirka H.