On 2021-05-04 16:50 +02, Sebastien Marie <[email protected]> wrote:
> On Tue, May 04, 2021 at 02:15:05PM +0200, Alexandr Nedvedicky wrote:
>> Hello Sebastien,
>>
>> thank you for additional info about previously working kernel.
>>
>> it looks like your older kernel, which works, might be running without
>> my commit
>>
>> revision 1.1116
>> date: 2021/04/27 09:38:29; author: sashan; state: Exp;\
>> lines: +14 -6; commitid: 3W1fRTkLb3ZlUanF;
>> pf_state_key_link_reverse() is prone to race on parallel forwarding
>>
>> we need to adjust assertions. at time we call pf_state_key_link_reverse()
>> is state_key either linked to correct reverse peer or not linked at all.
>> The pf_state_key_link_reverse() is being called as a reader ons
>> tate_lock.
>> There might be more packets, which try to update the state key.
>>
>> OK bluhm@
>>
>> > # cat /etc/pf.conf
>> > # See pf.conf(5) and /etc/examples/pf.conf
>> > #
>> > # avoid skip on lo to get tryton redir
>> > #set skip on lo
>> >
>> > block return # block stateless traffic
>> > pass # establish keep-state
>> >
>> > # redir 80 -> 8000
>> > pass in proto tcp to (self) port 80 rdr-to
>> > 2001:41d0:fe39:c05c:56b8:d15b:2e0a:8775 port 8000
>> >
>>
>> the rule above is also interesting bit. this may trigger NAT-64.
>> I'm not using IPv6 on my hosts/routers. I'll try to use similar rule
>> to try to reproduce the crash.
>
> I will first try with another rule:
>
> pass in inet6 proto tcp to (self) port 80 rdr-to
> 2001:41d0:fe39:c05c:56b8:d15b:2e0a:8775 port 8000
>
> to avoid the possible NAT-64.
fwiw I intentionally run NAT64 and I don't see this.
But I only have one client, my android phone.
kern.version=OpenBSD 6.9-current (GENERIC.MP) #559: Fri Apr 30 03:31:48 MDT 2021
set skip on lo
block return # block stateless traffic
pass # establish keep-state
pass in log on vlan64 inet6 from any to 64:ff9b::/96 af-to inet \
from 192.168.178.23
>
> And next (if I still trigger the assert) I will rebuild a kernel and
> backout this specific commit.
>
> It should permit to see if the commit is the problem or not.
>
> Thanks.
--
I'm not entirely sure you are real.
