Eelco Chaudron <echau...@redhat.com> writes:
> On 16 Nov 2023, at 11:07, Joseph Zhong wrote:
>
>> This patch is to avoid generating incorrect conntrack entry
>> In a certain use case of conntrack flow that if flow included
>> ct(commit, nat) action, but no detail action/direction specified,
>> CT will generate incorrect conntrack entry.
>> For example, add below flow:
>> ip,priority=500,in_port=1,ct_state=-trk actions=ct(table=1,nat)'
>> ip,priority=500,in_port=2,ct_state=-trk actions=ct(table=1,nat)'
>> table=1,in_port=1,ip,ct_state=+trk+new actions=ct*(commit,nat)*,2
>> table=1,in_port=1,ip,ct_state=-new+trk+est actions=2
>> table=1,in_port=2,ip,ct_state=-new+trk+est actions=1
>> start traffic from 192.168.2.2 to 192.168.2.7
>> ovs dpdk datpath generate CT entry as below:
>> icmp,orig=(src=192.168.2.2,dst=192.168.2.7,id=17038,type=8,code=0),
>> reply=(src=*0.0.0.0*,dst=192.168.2.2,id=17038,type=0,code=0)
>> reply key src 0.0.0.0 is generated not correct by "nat_get_unique_tuple".
>> but ovs kernel datapath will generate correct ct entry as below:
>> icmp,orig=(src=192.168.2.2,dst=192.168.2.7,id=17038,type=8,code=0),
>> reply=(src=192.168.2.7,dst=192.168.2.2,id=17038,type=0,code=0)
>>
>> To compatible with this use case of flow, and also be consistent with
>> kernel datapath's behavior, this patch treat this nat without action
>> specified as not nat, and don't do "nat_get_unique_tuple" and malloc
>> a nat_conn that is attached to nc.
>
> Hi Joseph,
>
> Thanks for the patch, I’m not a conntrack expert so I’ll let Aaron, or
> Paolo review it. But would it be possible to have a test case for
> this?
It should for sure be possible to have a test case for this.
> Cheers,
>
> Eelco
>
>
>> Signed-off-by: Zhong Zhong <zhongzh...@gmail.com>
>>
>> ---
>> lib/conntrack.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>> diff --git a/lib/conntrack.c b/lib/conntrack.c
>> index 47a443f..581b62b 100644
>> --- a/lib/conntrack.c
>> +++ b/lib/conntrack.c
>> @@ -942,7 +942,7 @@ conn_not_found(struct conntrack *ct, struct dp_packet
>> *pkt,
Something happened with your patch here and that caused issues with the
robot. Please fix.
>> nc->parent_key = alg_exp->parent_key;
>> }
>> - if (nat_action_info) {
>> + if (nat_action_info && nat_action_info->nat_action) {
>> nc->nat_action = nat_action_info->nat_action;
>> if (alg_exp) {
>> --
I think this will break expectations case with NAT.
We need a more comprehensive solution here, because there are cases to
legitimately flow into a ct(commit,nat) pipeline with a new connection
and expect that things should "just work."
Note that the kernel side has other considerations as well - because the
NAT table is shared with other subsystems. That said, I think the
expectation case is important and should be retained, so to get a more
consistent behavior, we probably need to have multiple checks for the
nat_action value.
And that showcases a deficiency in the design of the userspace NAT as
well, because it assumes:
1) either SRC or DST but not both
2) SRC or DST is set when calling these routines.
>> --
>> Best Regards
>>
>> Zhong, Zhong
>> Email: zhongzh...@gmail.com
>> _______________________________________________
>> dev mailing list
>> d...@openvswitch.org
>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
