Hi all,

On Thu, Jul 31, 2025 at 11:49 AM Tiago Pires <[email protected]> wrote:
>
> Hi all,
>
> I have a potential issue to report where the use case is using
> dnat_and_snat and it is keeping many SYN_SENT with UNREPLIED status on
> the conntrack table.
> The TCP handshake is working fine, but when a user establishes a
> connection, there is always a SYN_SENT left on the conntrack table.
>
> I was able to figure out that this patch [0] (Fix direct access to
> SNAT network) introduced this behavior.
>
> [0] 
> https://github.com/ovn-org/ovn/commit/40136a2f2c84dca48f7b67c7dc63759e82a44668
>
> When we have many connections on the chassis gateways, more than 50%
> of the conntrack entries stay in this UNREPLIED state until the entry
> is expired according with the nf_conntrack_tcp_timeout_syn_sent
> setting.
>
> How to reproduce:
>
> #Setup
> OS_BASE=ubuntu OS_IMAGE=docker.io/library/ubuntu:jammy ENABLE_SSL=no
> CENTRAL_COUNT=1 CHASSIS_COUNT=2 GW_COUNT=1 ./ovn_cluster.sh start
> podman exec -it ovn-gw-1 ip link set dev br-ex up
> podman exec -it ovn-central-az1-1 ovn-nbctl lr-add lr11
> podman exec -it ovn-central-az1-1 ovn-nbctl ls-add bar11
> podman exec -it ovn-central-az1-1 ovn-nbctl lrp-add lr11 lr11-public1
> 00:00:00:00:aa:02 172.16.1.100/24
> podman exec -it ovn-central-az1-1 ovn-nbctl lrp-set-gateway-chassis
> lr11-public1 ovn-gw-1
> podman exec -it ovn-central-az1-1 ovn-nbctl lsp-add public1 lr11-lsp-public1
> podman exec -it ovn-central-az1-1 ovn-nbctl lsp-set-type lr11-lsp-public1 
> router
> podman exec -it ovn-central-az1-1 ovn-nbctl lsp-set-addresses
> lr11-lsp-public1 router
> podman exec -it ovn-central-az1-1 ovn-nbctl lsp-set-options
> lr11-lsp-public1 router-port=lr11-public1
> podman exec -it ovn-central-az1-1 ovn-nbctl lr-route-add lr11
> 0.0.0.0/0 172.16.1.1
> podman exec -it ovn-central-az1-1 ovn-nbctl lr-nat-add lr11
> dnat_and_snat 172.16.1.101 192.168.10.10
> podman exec -it ovn-central-az1-1 ovn-nbctl lr-nat-add lr11 snat
> 172.16.1.100 192.168.10.0/24
> podman exec -it ovn-central-az1-1 ovn-nbctl lrp-add lr11 rp11-bar
> 00:01:ff:22:00:01 192.168.10.1/24
> podman exec -it ovn-central-az1-1 ovn-nbctl lsp-add bar11 haproxy1 \
>     -- lsp-set-addresses haproxy1 "00:00:ff:ff:ff:21 192.168.10.10"
> podman exec -it ovn-central-az1-1 ovn-nbctl lsp-add bar11 bar-rp11 --
> set Logical_Switch_Port bar-rp11     type=router
> options:router-port=rp11-bar     -- lsp-set-addresses bar-rp11 router
> podman exec -it ovn-chassis-1 ip netns add haproxy1
> podman exec -it ovn-chassis-1 ovs-vsctl add-port br-int haproxy1 --
> set interface haproxy1 type=internal
> podman exec -it ovn-chassis-1 ip link set haproxy1 netns haproxy1
> podman exec -it ovn-chassis-1 ip netns exec haproxy1 ip addr add
> 192.168.10.10/24 dev haproxy1
> podman exec -it ovn-chassis-1 ip netns exec haproxy1 ip link set dev
> haproxy1 mtu 1400
> podman exec -it ovn-chassis-1 ip netns exec haproxy1 ip link set
> haproxy1 address 00:00:ff:ff:ff:21
> podman exec -it ovn-chassis-1 ip netns exec haproxy1 ip link set haproxy1 up
> podman exec -it ovn-chassis-1 ip netns exec haproxy1 ip route add
> default via 192.168.10.1
> podman exec -it ovn-chassis-1 ovs-vsctl set Interface haproxy1
> external_ids:iface-id=haproxy1
> podman exec ovn-chassis-1 ip netns exec haproxy1 python3 -m
> http.server 80 -d /tmp/ >/dev/null 2>&1 &
>
> #Test (client IP - 172.16.1.50)
> ip netns exec ovnfake-ext1 curl http://172.16.1.101
>
> #tcpdump on ovn-gw-1 (eth2)
>
> 13:50:31.703106 IP 172.16.1.50.44742 > 172.16.1.101.80: Flags [S], seq
> 2055144661, win 64240, options [mss 1460,sackOK,TS val 2795500537 ecr
> 0,nop,wscale 7], length 0
> 13:50:31.705836 IP 172.16.1.101.80 > 172.16.1.50.44742: Flags [S.],
> seq 3211627047, ack 2055144662, win 64704, options [mss 1360,sackOK,TS
> val 875814075 ecr 2795500537,nop,wscale 7], length 0
> 13:50:31.705973 IP 172.16.1.50.44742 > 172.16.1.101.80: Flags [.], ack
> 1, win 502, options [nop,nop,TS val 2795500540 ecr 875814075], length
> 0
> 13:50:31.706065 IP 172.16.1.50.44742 > 172.16.1.101.80: Flags [P.],
> seq 1:77, ack 1, win 502, options [nop,nop,TS val 2795500540 ecr
> 875814075], length 76: HTTP: GET / HTTP/1.1
> 13:50:31.706323 IP 172.16.1.101.80 > 172.16.1.50.44742: Flags [.], ack
> 77, win 505, options [nop,nop,TS val 875814077 ecr 2795500540], length
> 0
> 13:50:31.708484 IP 172.16.1.101.80 > 172.16.1.50.44742: Flags [P.],
> seq 1:187, ack 77, win 505, options [nop,nop,TS val 875814079 ecr
> 2795500540], length 186: HTTP: HTTP/1.0 200 OK
> 13:50:31.708494 IP 172.16.1.50.44742 > 172.16.1.101.80: Flags [.], ack
> 187, win 501, options [nop,nop,TS val 2795500542 ecr 875814079],
> length 0
> 13:50:31.708567 IP 172.16.1.101.80 > 172.16.1.50.44742: Flags [P.],
> seq 187:248, ack 77, win 505, options [nop,nop,TS val 875814080 ecr
> 2795500542], length 61: HTTP
> 13:50:31.708571 IP 172.16.1.50.44742 > 172.16.1.101.80: Flags [.], ack
> 248, win 501, options [nop,nop,TS val 2795500543 ecr 875814080],
> length 0
> 13:50:31.708605 IP 172.16.1.101.80 > 172.16.1.50.44742: Flags [F.],
> seq 248, ack 77, win 505, options [nop,nop,TS val 875814080 ecr
> 2795500543], length 0
> 13:50:31.708673 IP 172.16.1.50.44742 > 172.16.1.101.80: Flags [F.],
> seq 77, ack 249, win 501, options [nop,nop,TS val 2795500543 ecr
> 875814080], length 0
> 13:50:31.708759 IP 172.16.1.101.80 > 172.16.1.50.44742: Flags [.], ack
> 78, win 505, options [nop,nop,TS val 875814080 ecr 2795500543], length
> 0
>
>
> #conntrack UNREPLIED entries on ovn-gw-1
> # conntrack -L | grep -v 6081 | grep -v 6642 | grep UNREP
> conntrack v1.4.6 (conntrack-tools): 4 flow entries have been shown.
> tcp      6 118 SYN_SENT src=172.16.1.50 dst=192.168.10.10 sport=44742
> dport=80 [UNREPLIED] src=192.168.10.10 dst=172.16.1.50 sport=80
> dport=44742 mark=0 zone=13 use=1
>
> Even though the packets are replied and NATed, it seems there is
> something wrong that is keeping this conntrack entry there.
> I was able to check there was a later patch [1] that fixed some issues
> with the patch [0] but it did not address this specific behavior.
>
> [1] 
> https://github.com/ovn-org/ovn/commit/8114a5c606eb5d0b3f04cf3e8219e094f1cc1b37
>
> The patch is creating the way for an external IP to communicate with a
> private IP on the overlay without NATing when having DNAT/SNAT
> configured on the LR.
>
> Could you give a hint how we could work to solve this issue?
>
> If Martin or Ales could check, I would appreciate it.
>

The following line is creating the behavior when using true:

https://github.com/ovn-org/ovn/blob/main/northd/northd.c#L16516

So, changing it to false solves my problem but I believe it would
create an issue for the main purpose of the patch.
Checking the logical flows I was able to verify that there is a
difference in the direction of the flow on the lr_out_snat stage:

#true
table=3 (lr_out_snat        ), priority=153  , match=(ip && ip4.dst ==
192.168.10.0/24 && inport == "lr11-public1" &&
is_chassis_resident("cr-lr11-public1")), action=(ct_snat;)
table=3 (lr_out_snat        ), priority=153  , match=(ip && ip4.src ==
192.168.10.0/24 && outport == "lr11-public1" &&
is_chassis_resident("cr-lr11-public1") && (!ct.trk || !ct.rpl)),
action=(ct_snat(172.16.1.100);)

#false
table=3 (lr_out_snat        ), priority=153  , match=(ip && ip4.src ==
192.168.10.0/24 && outport == "lr11-public1" &&
is_chassis_resident("cr-lr11-public1") && (!ct.trk || !ct.rpl)),
action=(ct_snat(172.16.1.100);)
table=3 (lr_out_snat        ), priority=153  , match=(ip && ip4.src ==
192.168.10.0/24 && outport == "lr11-public1" &&
is_chassis_resident("cr-lr11-public1") && (!ct.trk || !ct.rpl)),
action=(ct_snat;)

Regards,

Tiago Pires


> Thanks.
>
> Regards,
>
> Tiago Pires

-- 




_‘Esta mensagem é direcionada apenas para os endereços constantes no 
cabeçalho inicial. Se você não está listado nos endereços constantes no 
cabeçalho, pedimos-lhe que desconsidere completamente o conteúdo dessa 
mensagem e cuja cópia, encaminhamento e/ou execução das ações citadas estão 
imediatamente anuladas e proibidas’._


* **‘Apesar do Magazine Luiza tomar 
todas as precauções razoáveis para assegurar que nenhum vírus esteja 
presente nesse e-mail, a empresa não poderá aceitar a responsabilidade por 
quaisquer perdas ou danos causados por esse e-mail ou por seus anexos’.*



_______________________________________________
discuss mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-discuss

Reply via email to