Message ID | 20240723194049.96305-1-roberto.acosta@luizalabs.com |
---|---|
State | New |
Headers | show |
Series | [ovs-dev,v4] northd: Fix logical router load-balancer nat rules when using DGP. | expand |
Context | Check | Description |
---|---|---|
ovsrobot/apply-robot | success | apply and check: success |
ovsrobot/github-robot-_Build_and_Test | success | github build: passed |
ovsrobot/github-robot-_ovn-kubernetes | success | github build: passed |
Hi Roberto, Overall, I think this approach doesn't work except for the most simple configurations. Reading the code, the approach appears to be the following: 1) On each chassis with a DGP, add a flow that will load balance using ct_lb_mark() 2) On each chassis with a DGP, avoid the undnat operation by replacing the typical undat action with a "next;" action. 3) On each chassis with a DGP, install a stateless SNAT flow so that the source IP address is changed back into load balancer VIP address. The idea is that the traffic can be load balanced on any chassis with a DGP, then the reply traffic can be statelessly SNATted on any other chassis since the conntrack state is not needed to statelessly SNAT the reply traffic back to the load balancer VIP. There are several issues with this approach * The ct_lb_mark() flows have several ip4.dst=X actions added to them. However, these actions don't actually accomplish anything, because the ct_lb_mark() action at the end is what ultimately will select a new destination for the traffic. * The changes to the undnat flows ignore the lb_force_snat_ip and lb_skip_snat_ip options. * The stateless SNATs will not handle related traffic properly. For instance, if there is a TCP session that is load balanced, and the backend sends an ICMP message, the ICMP message will not be SNATted properly. In order to handle load balancing with DGPs bound to multiple chassis, the approach either has to be 1) Ensure that the traffic will always be redirected to the chassis where the load balancing occurred. 2) Distribute conntrack state. Currently (2) is not something that I know is supported, so (1) is the better option. I suppose you or someone else could write a patch that works this way, but honestly I feel like restricting load balancing to a single DGP makes this more easy and reliable. (2) may eventually become a reality. I know that there is currently research being done to determine the feasibility and scalability of distributed conntrack. If that comes to fruition, then it'll be much easier to support load balancing on logical routers with multiple DGPs. On 7/23/24 15:40, Roberto Bartzen Acosta via dev wrote: > This commit fixes the build_distr_lrouter_nat_flows_for_lb function to > include one NAT stateless flow entry for each DGP in use. Since we have > added support to create multiple gateway ports per logical router, it's > necessary to include in the LR nat rules pipeline a specific entry for each > attached DGP. Otherwise, the ingress traffic is only redirected when the > incoming LRP matches the chassis_resident field. > > Considering that DNAT rules for DGPs were implemented with the need to > configure the DGP-related gateway-port column, the load-balancer NAT rule > configuration can use a similar idea. In this case, we don't know the LRP > responsible for the incoming traffic, and therefore we need to automatically > apply a stateless NAT rule to the load-balancer on all DGPs to allow inbound > traffic. > > After applying this patch, the incoming and/or outgoing traffic can pass > through any chassis where the DGP resides without having problems with CT > state. > > Reported-at: https://bugs.launchpad.net/ubuntu/+source/ovn/+bug/2054322 > Fixes: 15348b7b806f ("ovn-northd: Multiple distributed gateway port support.") > Signed-off-by: Roberto Bartzen Acosta <roberto.acosta@luizalabs.com> > --- > northd/en-lr-stateful.c | 12 --- > northd/northd.c | 96 ++++++++++++----- > tests/ovn-northd.at | 222 ++++++++++++++++++++++++++++++++++++++++ > 3 files changed, 293 insertions(+), 37 deletions(-) > > diff --git a/northd/en-lr-stateful.c b/northd/en-lr-stateful.c > index baf1bd2f8..f09691af6 100644 > --- a/northd/en-lr-stateful.c > +++ b/northd/en-lr-stateful.c > @@ -516,18 +516,6 @@ lr_stateful_record_create(struct lr_stateful_table *table, > > table->array[od->index] = lr_stateful_rec; > > - /* Load balancers are not supported (yet) if a logical router has multiple > - * distributed gateway port. Log a warning. */ > - if (lr_stateful_rec->has_lb_vip && lr_has_multiple_gw_ports(od)) { > - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); > - VLOG_WARN_RL(&rl, "Load-balancers are configured on logical " > - "router %s, which has %"PRIuSIZE" distributed " > - "gateway ports. Load-balancer is not supported " > - "yet when there is more than one distributed " > - "gateway port on the router.", > - od->nbr->name, od->n_l3dgw_ports); > - } > - > return lr_stateful_rec; > } > > diff --git a/northd/northd.c b/northd/northd.c > index 6898daa00..853d58f29 100644 > --- a/northd/northd.c > +++ b/northd/northd.c > @@ -11026,31 +11026,30 @@ static void > build_distr_lrouter_nat_flows_for_lb(struct lrouter_nat_lb_flows_ctx *ctx, > enum lrouter_nat_lb_flow_type type, > struct ovn_datapath *od, > - struct lflow_ref *lflow_ref) > + struct lflow_ref *lflow_ref, > + struct ovn_port *dgp) > { > - struct ovn_port *dgp = od->l3dgw_ports[0]; > - > - const char *undnat_action; > - > - switch (type) { > - case LROUTER_NAT_LB_FLOW_FORCE_SNAT: > - undnat_action = "flags.force_snat_for_lb = 1; next;"; > - break; > - case LROUTER_NAT_LB_FLOW_SKIP_SNAT: > - undnat_action = "flags.skip_snat_for_lb = 1; next;"; > - break; > - case LROUTER_NAT_LB_FLOW_NORMAL: > - case LROUTER_NAT_LB_FLOW_MAX: > - undnat_action = lrouter_use_common_zone(od) > - ? "ct_dnat_in_czone;" > - : "ct_dnat;"; > - break; > - } > + struct ds dnat_action = DS_EMPTY_INITIALIZER; > > /* Store the match lengths, so we can reuse the ds buffer. */ > size_t new_match_len = ctx->new_match->length; > size_t undnat_match_len = ctx->undnat_match->length; > > + /* Create stateless LB NAT rules when using DGPs. > + * dnat_action: Add the LB backend IPs as a destination action of the > + * lr_in_dnat NAT rule with cumulative effect because any > + * backend dst IP used in the action list will redirect the > + * packet to the ct_lb pipeline. > + */ > + if (od->n_l3dgw_ports > 1) { > + for (size_t i = 0; i < ctx->lb_vip->n_backends; i++) { > + struct ovn_lb_backend *backend = &ctx->lb_vip->backends[i]; > + bool ipv6 = !IN6_IS_ADDR_V4MAPPED(&backend->ip); > + ds_put_format(&dnat_action, "%s.dst=%s;", ipv6 ? "ip6" : "ip4", > + backend->ip_str); > + } > + } > + ds_put_format(&dnat_action, "%s", ctx->new_action[type]); > > const char *meter = NULL; > > @@ -11060,20 +11059,47 @@ build_distr_lrouter_nat_flows_for_lb(struct lrouter_nat_lb_flows_ctx *ctx, > > if (ctx->lb_vip->n_backends || !ctx->lb_vip->empty_backend_rej) { > ds_put_format(ctx->new_match, " && is_chassis_resident(%s)", > - od->l3dgw_ports[0]->cr_port->json_key); > + dgp->cr_port->json_key); > } > > ovn_lflow_add_with_hint__(ctx->lflows, od, S_ROUTER_IN_DNAT, ctx->prio, > - ds_cstr(ctx->new_match), ctx->new_action[type], > + ds_cstr(ctx->new_match), ds_cstr(&dnat_action), > NULL, meter, &ctx->lb->nlb->header_, > lflow_ref); > > ds_truncate(ctx->new_match, new_match_len); > > + ds_destroy(&dnat_action); > if (!ctx->lb_vip->n_backends) { > return; > } > > + struct ds undnat_action = DS_EMPTY_INITIALIZER; > + struct ds snat_action = DS_EMPTY_INITIALIZER; > + > + switch (type) { > + case LROUTER_NAT_LB_FLOW_FORCE_SNAT: > + ds_put_format(&undnat_action, "flags.force_snat_for_lb = 1; next;"); > + break; > + case LROUTER_NAT_LB_FLOW_SKIP_SNAT: > + ds_put_format(&undnat_action, "flags.skip_snat_for_lb = 1; next;"); > + break; > + case LROUTER_NAT_LB_FLOW_NORMAL: > + case LROUTER_NAT_LB_FLOW_MAX: > + ds_put_format(&undnat_action, "%s", > + lrouter_use_common_zone(od) ? "ct_dnat_in_czone;" > + : "ct_dnat;"); > + break; > + } > + > + /* Create stateless LB NAT rules when using DGPs. > + * undnat_action: Remove the ct action from the lr_out_undenat NAT rule. > + */ > + if (od->n_l3dgw_ports > 1) { > + ds_clear(&undnat_action); > + ds_put_format(&undnat_action, "next;"); > + } > + > /* We need to centralize the LB traffic to properly perform > * the undnat stage. > */ > @@ -11093,10 +11119,27 @@ build_distr_lrouter_nat_flows_for_lb(struct lrouter_nat_lb_flows_ctx *ctx, > " && is_chassis_resident(%s)", dgp->json_key, dgp->json_key, > dgp->cr_port->json_key); > ovn_lflow_add_with_hint(ctx->lflows, od, S_ROUTER_OUT_UNDNAT, 120, > - ds_cstr(ctx->undnat_match), undnat_action, > - &ctx->lb->nlb->header_, > + ds_cstr(ctx->undnat_match), > + ds_cstr(&undnat_action), &ctx->lb->nlb->header_, > lflow_ref); > + > + /* Create stateless LB NAT rules when using DGPs. > + * snat_action: Add a new lr_out_snat rule with the LB VIP as source IP > + * action to perform the NAT stateless pipeline completely. > + */ > + if (od->n_l3dgw_ports > 1) { > + ds_put_format(&snat_action, "%s.src=%s; next;", > + ctx->lb_vip->address_family == AF_INET6 ? "ip6" : "ip4", > + ctx->lb_vip->vip_str); > + ovn_lflow_add_with_hint(ctx->lflows, od, S_ROUTER_OUT_SNAT, 120, > + ds_cstr(ctx->undnat_match), > + ds_cstr(&snat_action), &ctx->lb->nlb->header_, > + lflow_ref); > + } > + > ds_truncate(ctx->undnat_match, undnat_match_len); > + ds_destroy(&undnat_action); > + ds_destroy(&snat_action); > } > > static void > @@ -11263,8 +11306,11 @@ build_lrouter_nat_flows_for_lb( > if (!od->n_l3dgw_ports) { > bitmap_set1(gw_dp_bitmap[type], index); > } else { > - build_distr_lrouter_nat_flows_for_lb(&ctx, type, od, > - lb_dps->lflow_ref); > + for (size_t i = 0; i < od->n_l3dgw_ports; i++) { > + struct ovn_port *dgp = od->l3dgw_ports[i]; > + build_distr_lrouter_nat_flows_for_lb(&ctx, type, od, > + lb_dps->lflow_ref, dgp); > + } > } > > if (lb->affinity_timeout) { > diff --git a/tests/ovn-northd.at b/tests/ovn-northd.at > index a389d1988..e1b26b774 100644 > --- a/tests/ovn-northd.at > +++ b/tests/ovn-northd.at > @@ -12721,3 +12721,225 @@ AT_CHECK([ovn-sbctl dump-flows lr | grep lr_in_dnat | ovn_strip_lflows], [0], [d > > AT_CLEANUP > ]) > + > +OVN_FOR_EACH_NORTHD_NO_HV_PARALLELIZATION([ > +AT_SETUP([Load balancer with Distributed Gateway Ports (LB + DGP + NAT Stateless)]) > +ovn_start > + > +check ovn-nbctl ls-add public > +check ovn-nbctl lr-add lr1 > + > +# lr1 DGP ts1 > +check ovn-nbctl ls-add ts1 > +check ovn-nbctl lrp-add lr1 lr1-ts1 00:00:01:02:03:04 172.16.10.1/24 > +check ovn-nbctl lrp-set-gateway-chassis lr1-ts1 chassis-2 > + > +# lr1 DGP ts2 > +check ovn-nbctl ls-add ts2 > +check ovn-nbctl lrp-add lr1 lr1-ts2 00:00:01:02:03:05 172.16.20.1/24 > +check ovn-nbctl lrp-set-gateway-chassis lr1-ts2 chassis-3 > + > +# lr1 DGP public > +check ovn-nbctl lrp-add lr1 lr1_public 00:de:ad:ff:00:01 173.16.0.1/16 > +check ovn-nbctl lrp-add lr1 lr1_s1 00:de:ad:fe:00:02 172.16.0.1/24 > +check ovn-nbctl lrp-set-gateway-chassis lr1_public chassis-1 > + > +check ovn-nbctl ls-add s1 > +# s1 - lr1 > +check ovn-nbctl lsp-add s1 s1_lr1 > +check ovn-nbctl lsp-set-type s1_lr1 router > +check ovn-nbctl lsp-set-addresses s1_lr1 "00:de:ad:fe:00:02 172.16.0.1" > +check ovn-nbctl lsp-set-options s1_lr1 router-port=lr1_s1 > + > +# s1 - backend vm1 > +check ovn-nbctl lsp-add s1 vm1 > +check ovn-nbctl lsp-set-addresses vm1 "00:de:ad:01:00:01 172.16.0.101" > + > +# s1 - backend vm2 > +check ovn-nbctl lsp-add s1 vm2 > +check ovn-nbctl lsp-set-addresses vm2 "00:de:ad:01:00:02 172.16.0.102" > + > +# s1 - backend vm3 > +check ovn-nbctl lsp-add s1 vm3 > +check ovn-nbctl lsp-set-addresses vm3 "00:de:ad:01:00:03 172.16.0.103" > + > +# Add the lr1 DGP ts1 to the public switch > +check ovn-nbctl lsp-add public public_lr1_ts1 > +check ovn-nbctl lsp-set-type public_lr1_ts1 router > +check ovn-nbctl lsp-set-addresses public_lr1_ts1 router > +check ovn-nbctl lsp-set-options public_lr1_ts1 router-port=lr1-ts1 nat-addresses=router > + > +# Add the lr1 DGP ts2 to the public switch > +check ovn-nbctl lsp-add public public_lr1_ts2 > +check ovn-nbctl lsp-set-type public_lr1_ts2 router > +check ovn-nbctl lsp-set-addresses public_lr1_ts2 router > +check ovn-nbctl lsp-set-options public_lr1_ts2 router-port=lr1-ts2 nat-addresses=router > + > +# Add the lr1 DGP public to the public switch > +check ovn-nbctl lsp-add public public_lr1 > +check ovn-nbctl lsp-set-type public_lr1 router > +check ovn-nbctl lsp-set-addresses public_lr1 router > +check ovn-nbctl lsp-set-options public_lr1 router-port=lr1_public nat-addresses=router > + > +# Create the Load Balancer lb1 > +check ovn-nbctl --wait=sb lb-add lb1 "30.0.0.1" "172.16.0.103,172.16.0.102,172.16.0.101" > + > +# Associate load balancer to s1 > +check ovn-nbctl ls-lb-add s1 lb1 > +check ovn-nbctl --wait=sb sync > + > +ovn-sbctl dump-flows s1 > s1flows > +AT_CAPTURE_FILE([s1flows]) > + > +AT_CHECK([grep "ls_in_pre_stateful" s1flows | ovn_strip_lflows | grep "30.0.0.1"], [0], [dnl > + table=??(ls_in_pre_stateful ), priority=120 , match=(reg0[[2]] == 1 && ip4.dst == 30.0.0.1), action=(reg1 = 30.0.0.1; ct_lb_mark;) > +]) > +AT_CHECK([grep "ls_in_lb" s1flows | ovn_strip_lflows | grep "30.0.0.1"], [0], [dnl > + table=??(ls_in_lb ), priority=110 , match=(ct.new && ip4.dst == 30.0.0.1), action=(reg0[[1]] = 0; ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) > +]) > + > +# Associate load balancer to lr1 with DGP > +check ovn-nbctl lr-lb-add lr1 lb1 > +check ovn-nbctl --wait=sb sync > + > +ovn-sbctl dump-flows lr1 > lr1flows > +AT_CAPTURE_FILE([lr1flows]) > + > +# Check stateless NAT rules for load balancer with multiple DGP > +# 1. Check if the backend IPs are in the ipX.dst action > +AT_CHECK([grep "lr_in_dnat" lr1flows | ovn_strip_lflows | grep "30.0.0.1"], [0], [dnl > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip4 && ip4.dst == 30.0.0.1 && is_chassis_resident("cr-lr1-ts1")), action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip4 && ip4.dst == 30.0.0.1 && is_chassis_resident("cr-lr1-ts2")), action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip4 && ip4.dst == 30.0.0.1 && is_chassis_resident("cr-lr1_public")), action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) > +]) > + > +# 2. Check if the DGP ports are in the match with action next > +AT_CHECK([grep "lr_out_undnat" lr1flows | ovn_strip_lflows], [0], [dnl > + table=??(lr_out_undnat ), priority=0 , match=(1), action=(next;) > + table=??(lr_out_undnat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(next;) > + table=??(lr_out_undnat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(next;) > + table=??(lr_out_undnat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1_public" || outport == "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(next;) > +]) > + > +# 3. Check if the VIP IP is in the ipX.src action > +AT_CHECK([grep "lr_out_snat" lr1flows | ovn_strip_lflows], [0], [dnl > + table=??(lr_out_snat ), priority=0 , match=(1), action=(next;) > + table=??(lr_out_snat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(ip4.src=30.0.0.1; next;) > + table=??(lr_out_snat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(ip4.src=30.0.0.1; next;) > + table=??(lr_out_snat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1_public" || outport == "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(ip4.src=30.0.0.1; next;) > + table=??(lr_out_snat ), priority=120 , match=(nd_ns), action=(next;) > +]) > + > +AT_CLEANUP > +]) > + > +OVN_FOR_EACH_NORTHD_NO_HV_PARALLELIZATION([ > +AT_SETUP([Load balancer with Distributed Gateway Ports (LB + DGP + NAT Stateless) - IPv6]) > +ovn_start > + > +check ovn-nbctl ls-add public > +check ovn-nbctl lr-add lr1 > + > +# lr1 DGP ts1 > +check ovn-nbctl ls-add ts1 > +check ovn-nbctl lrp-add lr1 lr1-ts1 00:00:01:02:03:04 2001:db8:aaaa:1::1/64 > +check ovn-nbctl lrp-set-gateway-chassis lr1-ts1 chassis-2 > + > +# lr1 DGP ts2 > +check ovn-nbctl ls-add ts2 > +check ovn-nbctl lrp-add lr1 lr1-ts2 00:00:01:02:03:05 2001:db8:aaaa:2::1/64 > +check ovn-nbctl lrp-set-gateway-chassis lr1-ts2 chassis-3 > + > +# lr1 DGP public > +check ovn-nbctl lrp-add lr1 lr1_public 00:de:ad:ff:00:01 2001:db8:bbbb::1/64 > +check ovn-nbctl lrp-add lr1 lr1_s1 00:de:ad:fe:00:02 2001:db8:aaaa:3::1/64 > +check ovn-nbctl lrp-set-gateway-chassis lr1_public chassis-1 > + > +check ovn-nbctl ls-add s1 > +# s1 - lr1 > +check ovn-nbctl lsp-add s1 s1_lr1 > +check ovn-nbctl lsp-set-type s1_lr1 router > +check ovn-nbctl lsp-set-addresses s1_lr1 "00:de:ad:fe:00:02 2001:db8:aaaa:3::1" > +check ovn-nbctl lsp-set-options s1_lr1 router-port=lr1_s1 > + > +# s1 - backend vm1 > +check ovn-nbctl lsp-add s1 vm1 > +check ovn-nbctl lsp-set-addresses vm1 "00:de:ad:01:00:01 2001:db8:aaaa:3::101" > + > +# s1 - backend vm2 > +check ovn-nbctl lsp-add s1 vm2 > +check ovn-nbctl lsp-set-addresses vm2 "00:de:ad:01:00:02 2001:db8:aaaa:3::102" > + > +# s1 - backend vm3 > +check ovn-nbctl lsp-add s1 vm3 > +check ovn-nbctl lsp-set-addresses vm3 "00:de:ad:01:00:03 2001:db8:aaaa:3::103" > + > +# Add the lr1 DGP ts1 to the public switch > +check ovn-nbctl lsp-add public public_lr1_ts1 > +check ovn-nbctl lsp-set-type public_lr1_ts1 router > +check ovn-nbctl lsp-set-addresses public_lr1_ts1 router > +check ovn-nbctl lsp-set-options public_lr1_ts1 router-port=lr1-ts1 nat-addresses=router > + > +# Add the lr1 DGP ts2 to the public switch > +check ovn-nbctl lsp-add public public_lr1_ts2 > +check ovn-nbctl lsp-set-type public_lr1_ts2 router > +check ovn-nbctl lsp-set-addresses public_lr1_ts2 router > +check ovn-nbctl lsp-set-options public_lr1_ts2 router-port=lr1-ts2 nat-addresses=router > + > +# Add the lr1 DGP public to the public switch > +check ovn-nbctl lsp-add public public_lr1 > +check ovn-nbctl lsp-set-type public_lr1 router > +check ovn-nbctl lsp-set-addresses public_lr1 router > +check ovn-nbctl lsp-set-options public_lr1 router-port=lr1_public nat-addresses=router > + > +# Create the Load Balancer lb1 > +check ovn-nbctl --wait=sb lb-add lb1 "2001:db8:cccc::1" "2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101" > + > +# Associate load balancer to s1 > +check ovn-nbctl ls-lb-add s1 lb1 > +check ovn-nbctl --wait=sb sync > + > +ovn-sbctl dump-flows s1 > s1flows > +AT_CAPTURE_FILE([s1flows]) > + > +AT_CHECK([grep "ls_in_pre_stateful" s1flows | ovn_strip_lflows | grep "2001:db8:cccc::1"], [0], [dnl > + table=??(ls_in_pre_stateful ), priority=120 , match=(reg0[[2]] == 1 && ip6.dst == 2001:db8:cccc::1), action=(xxreg1 = 2001:db8:cccc::1; ct_lb_mark;) > +]) > +AT_CHECK([grep "ls_in_lb" s1flows | ovn_strip_lflows | grep "2001:db8:cccc::1"], [0], [dnl > + table=??(ls_in_lb ), priority=110 , match=(ct.new && ip6.dst == 2001:db8:cccc::1), action=(reg0[[1]] = 0; ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) > +]) > + > +# Associate load balancer to lr1 with DGP > +check ovn-nbctl lr-lb-add lr1 lb1 > +check ovn-nbctl --wait=sb sync > + > +ovn-sbctl dump-flows lr1 > lr1flows > +AT_CAPTURE_FILE([lr1flows]) > + > +# Check stateless NAT rules for load balancer with multiple DGP > +# 1. Check if the backend IPs are in the ipX.dst action > +AT_CHECK([grep "lr_in_dnat" lr1flows | ovn_strip_lflows | grep "2001:db8:cccc::1"], [0], [dnl > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && is_chassis_resident("cr-lr1-ts1")), action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && is_chassis_resident("cr-lr1-ts2")), action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && is_chassis_resident("cr-lr1_public")), action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) > +]) > + > +# 2. Check if the DGP ports are in the match with action next > +AT_CHECK([grep "lr_out_undnat" lr1flows | ovn_strip_lflows], [0], [dnl > + table=??(lr_out_undnat ), priority=0 , match=(1), action=(next;) > + table=??(lr_out_undnat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(next;) > + table=??(lr_out_undnat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(next;) > + table=??(lr_out_undnat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1_public" || outport == "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(next;) > +]) > + > +# 3. Check if the VIP IP is in the ipX.src action > +AT_CHECK([grep "lr_out_snat" lr1flows | ovn_strip_lflows], [0], [dnl > + table=??(lr_out_snat ), priority=0 , match=(1), action=(next;) > + table=??(lr_out_snat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(ip6.src=2001:db8:cccc::1; next;) > + table=??(lr_out_snat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(ip6.src=2001:db8:cccc::1; next;) > + table=??(lr_out_snat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1_public" || outport == "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(ip6.src=2001:db8:cccc::1; next;) > + table=??(lr_out_snat ), priority=120 , match=(nd_ns), action=(next;) > +]) > + > +AT_CLEANUP > +])
Hi Mark, thanks for your feedback. Em qui., 8 de ago. de 2024 às 17:41, Mark Michelson <mmichels@redhat.com> escreveu: > Hi Roberto, > > Overall, I think this approach doesn't work except for the most simple > configurations. > Could you provide a concrete scenario where my approach of using stateless NAT doesn't work? We can think of something to solve simple or complex cases that are not covered. The proposed patch passed all current OVN tests and e2e/kubernetes integrated tests. In addition, I am performing integrated tests with ovn-interconnect and with simple LB + DGPs setups (with or without ECMP). I haven't seen any problems in any topology so far, it works with one chassis or with multiple chassis. It even solves the Priyankar case(from Nutanix) that was mentioned in the other discussion. > > Reading the code, the approach appears to be the following: > > 1) On each chassis with a DGP, add a flow that will load balance using > ct_lb_mark() > > 2) On each chassis with a DGP, avoid the undnat operation by replacing > the typical undat action with a "next;" action. > > 3) On each chassis with a DGP, install a stateless SNAT flow so that the > source IP address is changed back into load balancer VIP address. > > The idea is that the traffic can be load balanced on any chassis with a > DGP, then the reply traffic can be statelessly SNATted on any other > chassis since the conntrack state is not needed to statelessly SNAT the > reply traffic back to the load balancer VIP. > > There are several issues with this approach > > * The ct_lb_mark() flows have several ip4.dst=X actions added to them. > However, these actions don't actually accomplish anything, because the > ct_lb_mark() action at the end is what ultimately will select a new > destination for the traffic. > This is not true as a whole right? Correct me if I'm wrong but from what I saw in the ovs actions documentation[1], the actions are cumulative only when they modify the same field. I agree that ip.dst modifies the same field and the last one will be applied but this is irrelevant because the destination IP needs to be in the context of LB backends. So, selecting one or applying all IPs would result in the same case (I can modify the logic to use only one of the backends - but I thought it made sense to put all of them as it is done in ct_lb_mark). Furthermore, the ct_lb_mark action modifies other fields (group) and the result is that the 2 openflow flows actions are added to DNAT. For instance: Using my proposed approach of including the ip.dst + ct_lb_mark table=7 (lr_in_dnat ), priority=120 , match=(ct.new && !ct.rel && ip4 && ip4.dst == 30.0.0.1 && tcp && tcp.dst == 80 && is_chassis_resident("cr-r1-ts1")), action=(ip4.dst=192.168.2.101;ip4.dst=192.168.2.102;ip4.dst=192.168.2.103;ct_lb_mark(backends= 192.168.2.101:80,192.168.2.102:80,192.168.2.103:80);) table=7 (lr_in_dnat ), priority=120 , match=(ct.new && !ct.rel && ip4 && ip4.dst == 30.0.0.1 && tcp && tcp.dst == 80 && is_chassis_resident("cr-r1-ts2")), action=(ip4.dst=192.168.2.101;ip4.dst=192.168.2.102;ip4.dst=192.168.2.103;ct_lb_mark(backends= 192.168.2.101:80,192.168.2.102:80,192.168.2.103:80);) This results is as additional nw_dst field for table 15: cookie=0xab99f240, duration=1086.022s, table=15, n_packets=610, n_bytes=48979, idle_age=825, priority=120,ct_state=+new-rel+trk,tcp,metadata=0x2,nw_dst=30.0.0.1,tp_dst=80 actions=mod_nw_dst:192.168.2.101,mod_nw_dst:192.168.2.102,mod_nw_dst:192.168.2.103,group:1 This proposal does not remove the use of ct_lb_mark usage. So, the "group:1" continues to be added to the created action and this makes the traffic to the LB work, regardless of the DGP it enters. Note: Because of the need to use gateway-port in the NAT rules for DGP, I noticed that the priority applied to these rules was 154 for lr_out_snat. So, I changed the priority of the S_ROUTER_OUT_SNAT rule to 160 to ensure that the traffic will always be SNATed to the LB VIP without conflicting with the router's default SNAT. table=3 (lr_out_snat ), priority=160 , match=(ip4 && ((ip4.src == 192.168.2.101 && tcp.src == 80) || (ip4.src == 192.168.2.102 && tcp.src == 80) || (ip4.src == 192.168.2.103 && tcp.src == 80)) && outport == "r1-ts1" && is_chassis_resident("cr-r1-ts1")), action=(ip4.src=30.0.0.1; next;) table=3 (lr_out_snat ), priority=160 , match=(ip4 && ((ip4.src == 192.168.2.101 && tcp.src == 80) || (ip4.src == 192.168.2.102 && tcp.src == 80) || (ip4.src == 192.168.2.103 && tcp.src == 80)) && outport == "r1-ts2" && is_chassis_resident("cr-r1-ts2")), action=(ip4.src=30.0.0.1; next;) table=3 (lr_out_snat ), priority=154 , match=(ip && ip4.src == 192.168.2.0/24 && outport == "r1-ts1" && is_chassis_resident("cr-r1-ts1") && reg9[4] == 1), action=(reg9[4] = 0; ct_snat(10.0.0.3);) table=3 (lr_out_snat ), priority=154 , match=(ip && ip4.src == 192.168.2.0/24 && outport == "r1-ts2" && is_chassis_resident("cr-r1-ts2") && reg9[4] == 1), action=(reg9[4] = 0; ct_snat(20.0.0.3);) > * The changes to the undnat flows ignore the lb_force_snat_ip and > lb_skip_snat_ip options. > These options don't make sense with the use of multiple DGPs IMO, wouldn't it be a case of documenting that the use with multiple DGPs disables these options? > > * The stateless SNATs will not handle related traffic properly. For > instance, if there is a TCP session that is load balanced, and the > backend sends an ICMP message, the ICMP message will not be SNATted > properly. > I didn't understand why not! The lr_out_snat rule created for stateless NAT is specific to the backend IP:PORT, the ICMP traffic and any other traffic with other source ports will continue to be SNATed via the router's outbound IP. By the way, for this to work with multiple DGPs we need specific SNAT rules, using the gateway port for that (same case of FIPs for ovn-interconnect scenarios in OpenStack). Please take a look at the end of this content: https://pastebin.ubuntu.com/p/Xdb2KsfnqZ/ > > In order to handle load balancing with DGPs bound to multiple chassis, > the approach either has to be > > 1) Ensure that the traffic will always be redirected to the chassis > where the load balancing occurred. > The use of NAT stateless guarantees this because it's done in conjunction with ct_lb_mark, so regardless of which DPG the packet comes from, the LB will forward the traffic to the same backend that is related to the end-to-end connection. Please take a look at the tests here: https://pastebin.ubuntu.com/p/Xdb2KsfnqZ/ > > 2) Distribute conntrack state. > > Currently (2) is not something that I know is supported, so (1) is the > better option. I suppose you or someone else could write a patch that > works this way, but honestly I feel like restricting load balancing to a > single DGP makes this more easy and reliable. > > (2) may eventually become a reality. I know that there is currently > research being done to determine the feasibility and scalability of > distributed conntrack. If that comes to fruition, then it'll be much > easier to support load balancing on logical routers with multiple DGPs. > I don't imagine this is feasible in the short term, of course, a conntrack sync solution would be ideal, such as Maglev for large scale LB solutions, for example. However, at the moment, the use of OVN LBs with multiple DGPs is broken in the upstream code and my proposal seems very reasonable to me. wdyt Mark? [1] https://docs.openvswitch.org/en/latest/ref/ovs-actions.7 > On 7/23/24 15:40, Roberto Bartzen Acosta via dev wrote: > > This commit fixes the build_distr_lrouter_nat_flows_for_lb function to > > include one NAT stateless flow entry for each DGP in use. Since we have > > added support to create multiple gateway ports per logical router, it's > > necessary to include in the LR nat rules pipeline a specific entry for > each > > attached DGP. Otherwise, the ingress traffic is only redirected when the > > incoming LRP matches the chassis_resident field. > > > > Considering that DNAT rules for DGPs were implemented with the need to > > configure the DGP-related gateway-port column, the load-balancer NAT rule > > configuration can use a similar idea. In this case, we don't know the LRP > > responsible for the incoming traffic, and therefore we need to > automatically > > apply a stateless NAT rule to the load-balancer on all DGPs to allow > inbound > > traffic. > > > > After applying this patch, the incoming and/or outgoing traffic can pass > > through any chassis where the DGP resides without having problems with CT > > state. > > > > Reported-at: https://bugs.launchpad.net/ubuntu/+source/ovn/+bug/2054322 > > Fixes: 15348b7b806f ("ovn-northd: Multiple distributed gateway port > support.") > > Signed-off-by: Roberto Bartzen Acosta <roberto.acosta@luizalabs.com> > > --- > > northd/en-lr-stateful.c | 12 --- > > northd/northd.c | 96 ++++++++++++----- > > tests/ovn-northd.at | 222 ++++++++++++++++++++++++++++++++++++++++ > > 3 files changed, 293 insertions(+), 37 deletions(-) > > > > diff --git a/northd/en-lr-stateful.c b/northd/en-lr-stateful.c > > index baf1bd2f8..f09691af6 100644 > > --- a/northd/en-lr-stateful.c > > +++ b/northd/en-lr-stateful.c > > @@ -516,18 +516,6 @@ lr_stateful_record_create(struct lr_stateful_table > *table, > > > > table->array[od->index] = lr_stateful_rec; > > > > - /* Load balancers are not supported (yet) if a logical router has > multiple > > - * distributed gateway port. Log a warning. */ > > - if (lr_stateful_rec->has_lb_vip && lr_has_multiple_gw_ports(od)) { > > - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); > > - VLOG_WARN_RL(&rl, "Load-balancers are configured on logical " > > - "router %s, which has %"PRIuSIZE" distributed " > > - "gateway ports. Load-balancer is not supported " > > - "yet when there is more than one distributed " > > - "gateway port on the router.", > > - od->nbr->name, od->n_l3dgw_ports); > > - } > > - > > return lr_stateful_rec; > > } > > > > diff --git a/northd/northd.c b/northd/northd.c > > index 6898daa00..853d58f29 100644 > > --- a/northd/northd.c > > +++ b/northd/northd.c > > @@ -11026,31 +11026,30 @@ static void > > build_distr_lrouter_nat_flows_for_lb(struct lrouter_nat_lb_flows_ctx > *ctx, > > enum lrouter_nat_lb_flow_type > type, > > struct ovn_datapath *od, > > - struct lflow_ref *lflow_ref) > > + struct lflow_ref *lflow_ref, > > + struct ovn_port *dgp) > > { > > - struct ovn_port *dgp = od->l3dgw_ports[0]; > > - > > - const char *undnat_action; > > - > > - switch (type) { > > - case LROUTER_NAT_LB_FLOW_FORCE_SNAT: > > - undnat_action = "flags.force_snat_for_lb = 1; next;"; > > - break; > > - case LROUTER_NAT_LB_FLOW_SKIP_SNAT: > > - undnat_action = "flags.skip_snat_for_lb = 1; next;"; > > - break; > > - case LROUTER_NAT_LB_FLOW_NORMAL: > > - case LROUTER_NAT_LB_FLOW_MAX: > > - undnat_action = lrouter_use_common_zone(od) > > - ? "ct_dnat_in_czone;" > > - : "ct_dnat;"; > > - break; > > - } > > + struct ds dnat_action = DS_EMPTY_INITIALIZER; > > > > /* Store the match lengths, so we can reuse the ds buffer. */ > > size_t new_match_len = ctx->new_match->length; > > size_t undnat_match_len = ctx->undnat_match->length; > > > > + /* Create stateless LB NAT rules when using DGPs. > > + * dnat_action: Add the LB backend IPs as a destination action of > the > > + * lr_in_dnat NAT rule with cumulative effect because > any > > + * backend dst IP used in the action list will > redirect the > > + * packet to the ct_lb pipeline. > > + */ > > + if (od->n_l3dgw_ports > 1) { > > + for (size_t i = 0; i < ctx->lb_vip->n_backends; i++) { > > + struct ovn_lb_backend *backend = &ctx->lb_vip->backends[i]; > > + bool ipv6 = !IN6_IS_ADDR_V4MAPPED(&backend->ip); > > + ds_put_format(&dnat_action, "%s.dst=%s;", ipv6 ? "ip6" : > "ip4", > > + backend->ip_str); > > + } > > + } > > + ds_put_format(&dnat_action, "%s", ctx->new_action[type]); > > > > const char *meter = NULL; > > > > @@ -11060,20 +11059,47 @@ build_distr_lrouter_nat_flows_for_lb(struct > lrouter_nat_lb_flows_ctx *ctx, > > > > if (ctx->lb_vip->n_backends || !ctx->lb_vip->empty_backend_rej) { > > ds_put_format(ctx->new_match, " && is_chassis_resident(%s)", > > - od->l3dgw_ports[0]->cr_port->json_key); > > + dgp->cr_port->json_key); > > } > > > > ovn_lflow_add_with_hint__(ctx->lflows, od, S_ROUTER_IN_DNAT, > ctx->prio, > > - ds_cstr(ctx->new_match), > ctx->new_action[type], > > + ds_cstr(ctx->new_match), > ds_cstr(&dnat_action), > > NULL, meter, &ctx->lb->nlb->header_, > > lflow_ref); > > > > ds_truncate(ctx->new_match, new_match_len); > > > > + ds_destroy(&dnat_action); > > if (!ctx->lb_vip->n_backends) { > > return; > > } > > > > + struct ds undnat_action = DS_EMPTY_INITIALIZER; > > + struct ds snat_action = DS_EMPTY_INITIALIZER; > > + > > + switch (type) { > > + case LROUTER_NAT_LB_FLOW_FORCE_SNAT: > > + ds_put_format(&undnat_action, "flags.force_snat_for_lb = 1; > next;"); > > + break; > > + case LROUTER_NAT_LB_FLOW_SKIP_SNAT: > > + ds_put_format(&undnat_action, "flags.skip_snat_for_lb = 1; > next;"); > > + break; > > + case LROUTER_NAT_LB_FLOW_NORMAL: > > + case LROUTER_NAT_LB_FLOW_MAX: > > + ds_put_format(&undnat_action, "%s", > > + lrouter_use_common_zone(od) ? "ct_dnat_in_czone;" > > + : "ct_dnat;"); > > + break; > > + } > > + > > + /* Create stateless LB NAT rules when using DGPs. > > + * undnat_action: Remove the ct action from the lr_out_undenat NAT > rule. > > + */ > > + if (od->n_l3dgw_ports > 1) { > > + ds_clear(&undnat_action); > > + ds_put_format(&undnat_action, "next;"); > > + } > > + > > /* We need to centralize the LB traffic to properly perform > > * the undnat stage. > > */ > > @@ -11093,10 +11119,27 @@ build_distr_lrouter_nat_flows_for_lb(struct > lrouter_nat_lb_flows_ctx *ctx, > > " && is_chassis_resident(%s)", dgp->json_key, > dgp->json_key, > > dgp->cr_port->json_key); > > ovn_lflow_add_with_hint(ctx->lflows, od, S_ROUTER_OUT_UNDNAT, 120, > > - ds_cstr(ctx->undnat_match), undnat_action, > > - &ctx->lb->nlb->header_, > > + ds_cstr(ctx->undnat_match), > > + ds_cstr(&undnat_action), > &ctx->lb->nlb->header_, > > lflow_ref); > > + > > + /* Create stateless LB NAT rules when using DGPs. > > + * snat_action: Add a new lr_out_snat rule with the LB VIP as > source IP > > + * action to perform the NAT stateless pipeline > completely. > > + */ > > + if (od->n_l3dgw_ports > 1) { > > + ds_put_format(&snat_action, "%s.src=%s; next;", > > + ctx->lb_vip->address_family == AF_INET6 ? "ip6" : > "ip4", > > + ctx->lb_vip->vip_str); > > + ovn_lflow_add_with_hint(ctx->lflows, od, S_ROUTER_OUT_SNAT, 120, > > + ds_cstr(ctx->undnat_match), > > + ds_cstr(&snat_action), > &ctx->lb->nlb->header_, > > + lflow_ref); > > + } > > + > > ds_truncate(ctx->undnat_match, undnat_match_len); > > + ds_destroy(&undnat_action); > > + ds_destroy(&snat_action); > > } > > > > static void > > @@ -11263,8 +11306,11 @@ build_lrouter_nat_flows_for_lb( > > if (!od->n_l3dgw_ports) { > > bitmap_set1(gw_dp_bitmap[type], index); > > } else { > > - build_distr_lrouter_nat_flows_for_lb(&ctx, type, od, > > - lb_dps->lflow_ref); > > + for (size_t i = 0; i < od->n_l3dgw_ports; i++) { > > + struct ovn_port *dgp = od->l3dgw_ports[i]; > > + build_distr_lrouter_nat_flows_for_lb(&ctx, type, od, > > + lb_dps->lflow_ref, > dgp); > > + } > > } > > > > if (lb->affinity_timeout) { > > diff --git a/tests/ovn-northd.at b/tests/ovn-northd.at > > index a389d1988..e1b26b774 100644 > > --- a/tests/ovn-northd.at > > +++ b/tests/ovn-northd.at > > @@ -12721,3 +12721,225 @@ AT_CHECK([ovn-sbctl dump-flows lr | grep > lr_in_dnat | ovn_strip_lflows], [0], [d > > > > AT_CLEANUP > > ]) > > + > > +OVN_FOR_EACH_NORTHD_NO_HV_PARALLELIZATION([ > > +AT_SETUP([Load balancer with Distributed Gateway Ports (LB + DGP + NAT > Stateless)]) > > +ovn_start > > + > > +check ovn-nbctl ls-add public > > +check ovn-nbctl lr-add lr1 > > + > > +# lr1 DGP ts1 > > +check ovn-nbctl ls-add ts1 > > +check ovn-nbctl lrp-add lr1 lr1-ts1 00:00:01:02:03:04 172.16.10.1/24 > > +check ovn-nbctl lrp-set-gateway-chassis lr1-ts1 chassis-2 > > + > > +# lr1 DGP ts2 > > +check ovn-nbctl ls-add ts2 > > +check ovn-nbctl lrp-add lr1 lr1-ts2 00:00:01:02:03:05 172.16.20.1/24 > > +check ovn-nbctl lrp-set-gateway-chassis lr1-ts2 chassis-3 > > + > > +# lr1 DGP public > > +check ovn-nbctl lrp-add lr1 lr1_public 00:de:ad:ff:00:01 173.16.0.1/16 > > +check ovn-nbctl lrp-add lr1 lr1_s1 00:de:ad:fe:00:02 172.16.0.1/24 > > +check ovn-nbctl lrp-set-gateway-chassis lr1_public chassis-1 > > + > > +check ovn-nbctl ls-add s1 > > +# s1 - lr1 > > +check ovn-nbctl lsp-add s1 s1_lr1 > > +check ovn-nbctl lsp-set-type s1_lr1 router > > +check ovn-nbctl lsp-set-addresses s1_lr1 "00:de:ad:fe:00:02 172.16.0.1" > > +check ovn-nbctl lsp-set-options s1_lr1 router-port=lr1_s1 > > + > > +# s1 - backend vm1 > > +check ovn-nbctl lsp-add s1 vm1 > > +check ovn-nbctl lsp-set-addresses vm1 "00:de:ad:01:00:01 172.16.0.101" > > + > > +# s1 - backend vm2 > > +check ovn-nbctl lsp-add s1 vm2 > > +check ovn-nbctl lsp-set-addresses vm2 "00:de:ad:01:00:02 172.16.0.102" > > + > > +# s1 - backend vm3 > > +check ovn-nbctl lsp-add s1 vm3 > > +check ovn-nbctl lsp-set-addresses vm3 "00:de:ad:01:00:03 172.16.0.103" > > + > > +# Add the lr1 DGP ts1 to the public switch > > +check ovn-nbctl lsp-add public public_lr1_ts1 > > +check ovn-nbctl lsp-set-type public_lr1_ts1 router > > +check ovn-nbctl lsp-set-addresses public_lr1_ts1 router > > +check ovn-nbctl lsp-set-options public_lr1_ts1 router-port=lr1-ts1 > nat-addresses=router > > + > > +# Add the lr1 DGP ts2 to the public switch > > +check ovn-nbctl lsp-add public public_lr1_ts2 > > +check ovn-nbctl lsp-set-type public_lr1_ts2 router > > +check ovn-nbctl lsp-set-addresses public_lr1_ts2 router > > +check ovn-nbctl lsp-set-options public_lr1_ts2 router-port=lr1-ts2 > nat-addresses=router > > + > > +# Add the lr1 DGP public to the public switch > > +check ovn-nbctl lsp-add public public_lr1 > > +check ovn-nbctl lsp-set-type public_lr1 router > > +check ovn-nbctl lsp-set-addresses public_lr1 router > > +check ovn-nbctl lsp-set-options public_lr1 router-port=lr1_public > nat-addresses=router > > + > > +# Create the Load Balancer lb1 > > +check ovn-nbctl --wait=sb lb-add lb1 "30.0.0.1" > "172.16.0.103,172.16.0.102,172.16.0.101" > > + > > +# Associate load balancer to s1 > > +check ovn-nbctl ls-lb-add s1 lb1 > > +check ovn-nbctl --wait=sb sync > > + > > +ovn-sbctl dump-flows s1 > s1flows > > +AT_CAPTURE_FILE([s1flows]) > > + > > +AT_CHECK([grep "ls_in_pre_stateful" s1flows | ovn_strip_lflows | grep > "30.0.0.1"], [0], [dnl > > + table=??(ls_in_pre_stateful ), priority=120 , match=(reg0[[2]] == 1 > && ip4.dst == 30.0.0.1), action=(reg1 = 30.0.0.1; ct_lb_mark;) > > +]) > > +AT_CHECK([grep "ls_in_lb" s1flows | ovn_strip_lflows | grep > "30.0.0.1"], [0], [dnl > > + table=??(ls_in_lb ), priority=110 , match=(ct.new && > ip4.dst == 30.0.0.1), action=(reg0[[1]] = 0; > ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) > > +]) > > + > > +# Associate load balancer to lr1 with DGP > > +check ovn-nbctl lr-lb-add lr1 lb1 > > +check ovn-nbctl --wait=sb sync > > + > > +ovn-sbctl dump-flows lr1 > lr1flows > > +AT_CAPTURE_FILE([lr1flows]) > > + > > +# Check stateless NAT rules for load balancer with multiple DGP > > +# 1. Check if the backend IPs are in the ipX.dst action > > +AT_CHECK([grep "lr_in_dnat" lr1flows | ovn_strip_lflows | grep > "30.0.0.1"], [0], [dnl > > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && > !ct.rel && ip4 && ip4.dst == 30.0.0.1 && > is_chassis_resident("cr-lr1-ts1")), > action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) > > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && > !ct.rel && ip4 && ip4.dst == 30.0.0.1 && > is_chassis_resident("cr-lr1-ts2")), > action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) > > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && > !ct.rel && ip4 && ip4.dst == 30.0.0.1 && > is_chassis_resident("cr-lr1_public")), > action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) > > +]) > > + > > +# 2. Check if the DGP ports are in the match with action next > > +AT_CHECK([grep "lr_out_undnat" lr1flows | ovn_strip_lflows], [0], [dnl > > + table=??(lr_out_undnat ), priority=0 , match=(1), > action=(next;) > > + table=??(lr_out_undnat ), priority=120 , match=(ip4 && > ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == > 172.16.0.101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && > is_chassis_resident("cr-lr1-ts1")), action=(next;) > > + table=??(lr_out_undnat ), priority=120 , match=(ip4 && > ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == > 172.16.0.101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && > is_chassis_resident("cr-lr1-ts2")), action=(next;) > > + table=??(lr_out_undnat ), priority=120 , match=(ip4 && > ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == > 172.16.0.101)) && (inport == "lr1_public" || outport == "lr1_public") && > is_chassis_resident("cr-lr1_public")), action=(next;) > > +]) > > + > > +# 3. Check if the VIP IP is in the ipX.src action > > +AT_CHECK([grep "lr_out_snat" lr1flows | ovn_strip_lflows], [0], [dnl > > + table=??(lr_out_snat ), priority=0 , match=(1), > action=(next;) > > + table=??(lr_out_snat ), priority=120 , match=(ip4 && > ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == > 172.16.0.101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && > is_chassis_resident("cr-lr1-ts1")), action=(ip4.src=30.0.0.1; next;) > > + table=??(lr_out_snat ), priority=120 , match=(ip4 && > ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == > 172.16.0.101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && > is_chassis_resident("cr-lr1-ts2")), action=(ip4.src=30.0.0.1; next;) > > + table=??(lr_out_snat ), priority=120 , match=(ip4 && > ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == > 172.16.0.101)) && (inport == "lr1_public" || outport == "lr1_public") && > is_chassis_resident("cr-lr1_public")), action=(ip4.src=30.0.0.1; next;) > > + table=??(lr_out_snat ), priority=120 , match=(nd_ns), > action=(next;) > > +]) > > + > > +AT_CLEANUP > > +]) > > + > > +OVN_FOR_EACH_NORTHD_NO_HV_PARALLELIZATION([ > > +AT_SETUP([Load balancer with Distributed Gateway Ports (LB + DGP + NAT > Stateless) - IPv6]) > > +ovn_start > > + > > +check ovn-nbctl ls-add public > > +check ovn-nbctl lr-add lr1 > > + > > +# lr1 DGP ts1 > > +check ovn-nbctl ls-add ts1 > > +check ovn-nbctl lrp-add lr1 lr1-ts1 00:00:01:02:03:04 > 2001:db8:aaaa:1::1/64 > > +check ovn-nbctl lrp-set-gateway-chassis lr1-ts1 chassis-2 > > + > > +# lr1 DGP ts2 > > +check ovn-nbctl ls-add ts2 > > +check ovn-nbctl lrp-add lr1 lr1-ts2 00:00:01:02:03:05 > 2001:db8:aaaa:2::1/64 > > +check ovn-nbctl lrp-set-gateway-chassis lr1-ts2 chassis-3 > > + > > +# lr1 DGP public > > +check ovn-nbctl lrp-add lr1 lr1_public 00:de:ad:ff:00:01 > 2001:db8:bbbb::1/64 > > +check ovn-nbctl lrp-add lr1 lr1_s1 00:de:ad:fe:00:02 > 2001:db8:aaaa:3::1/64 > > +check ovn-nbctl lrp-set-gateway-chassis lr1_public chassis-1 > > + > > +check ovn-nbctl ls-add s1 > > +# s1 - lr1 > > +check ovn-nbctl lsp-add s1 s1_lr1 > > +check ovn-nbctl lsp-set-type s1_lr1 router > > +check ovn-nbctl lsp-set-addresses s1_lr1 "00:de:ad:fe:00:02 > 2001:db8:aaaa:3::1" > > +check ovn-nbctl lsp-set-options s1_lr1 router-port=lr1_s1 > > + > > +# s1 - backend vm1 > > +check ovn-nbctl lsp-add s1 vm1 > > +check ovn-nbctl lsp-set-addresses vm1 "00:de:ad:01:00:01 > 2001:db8:aaaa:3::101" > > + > > +# s1 - backend vm2 > > +check ovn-nbctl lsp-add s1 vm2 > > +check ovn-nbctl lsp-set-addresses vm2 "00:de:ad:01:00:02 > 2001:db8:aaaa:3::102" > > + > > +# s1 - backend vm3 > > +check ovn-nbctl lsp-add s1 vm3 > > +check ovn-nbctl lsp-set-addresses vm3 "00:de:ad:01:00:03 > 2001:db8:aaaa:3::103" > > + > > +# Add the lr1 DGP ts1 to the public switch > > +check ovn-nbctl lsp-add public public_lr1_ts1 > > +check ovn-nbctl lsp-set-type public_lr1_ts1 router > > +check ovn-nbctl lsp-set-addresses public_lr1_ts1 router > > +check ovn-nbctl lsp-set-options public_lr1_ts1 router-port=lr1-ts1 > nat-addresses=router > > + > > +# Add the lr1 DGP ts2 to the public switch > > +check ovn-nbctl lsp-add public public_lr1_ts2 > > +check ovn-nbctl lsp-set-type public_lr1_ts2 router > > +check ovn-nbctl lsp-set-addresses public_lr1_ts2 router > > +check ovn-nbctl lsp-set-options public_lr1_ts2 router-port=lr1-ts2 > nat-addresses=router > > + > > +# Add the lr1 DGP public to the public switch > > +check ovn-nbctl lsp-add public public_lr1 > > +check ovn-nbctl lsp-set-type public_lr1 router > > +check ovn-nbctl lsp-set-addresses public_lr1 router > > +check ovn-nbctl lsp-set-options public_lr1 router-port=lr1_public > nat-addresses=router > > + > > +# Create the Load Balancer lb1 > > +check ovn-nbctl --wait=sb lb-add lb1 "2001:db8:cccc::1" > "2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101" > > + > > +# Associate load balancer to s1 > > +check ovn-nbctl ls-lb-add s1 lb1 > > +check ovn-nbctl --wait=sb sync > > + > > +ovn-sbctl dump-flows s1 > s1flows > > +AT_CAPTURE_FILE([s1flows]) > > + > > +AT_CHECK([grep "ls_in_pre_stateful" s1flows | ovn_strip_lflows | grep > "2001:db8:cccc::1"], [0], [dnl > > + table=??(ls_in_pre_stateful ), priority=120 , match=(reg0[[2]] == 1 > && ip6.dst == 2001:db8:cccc::1), action=(xxreg1 = 2001:db8:cccc::1; > ct_lb_mark;) > > +]) > > +AT_CHECK([grep "ls_in_lb" s1flows | ovn_strip_lflows | grep > "2001:db8:cccc::1"], [0], [dnl > > + table=??(ls_in_lb ), priority=110 , match=(ct.new && > ip6.dst == 2001:db8:cccc::1), action=(reg0[[1]] = 0; > ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) > > +]) > > + > > +# Associate load balancer to lr1 with DGP > > +check ovn-nbctl lr-lb-add lr1 lb1 > > +check ovn-nbctl --wait=sb sync > > + > > +ovn-sbctl dump-flows lr1 > lr1flows > > +AT_CAPTURE_FILE([lr1flows]) > > + > > +# Check stateless NAT rules for load balancer with multiple DGP > > +# 1. Check if the backend IPs are in the ipX.dst action > > +AT_CHECK([grep "lr_in_dnat" lr1flows | ovn_strip_lflows | grep > "2001:db8:cccc::1"], [0], [dnl > > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && > !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && > is_chassis_resident("cr-lr1-ts1")), > action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) > > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && > !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && > is_chassis_resident("cr-lr1-ts2")), > action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) > > + table=??(lr_in_dnat ), priority=110 , match=(ct.new && > !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && > is_chassis_resident("cr-lr1_public")), > action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) > > +]) > > + > > +# 2. Check if the DGP ports are in the match with action next > > +AT_CHECK([grep "lr_out_undnat" lr1flows | ovn_strip_lflows], [0], [dnl > > + table=??(lr_out_undnat ), priority=0 , match=(1), > action=(next;) > > + table=??(lr_out_undnat ), priority=120 , match=(ip6 && > ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || > (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts1" || outport == > "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(next;) > > + table=??(lr_out_undnat ), priority=120 , match=(ip6 && > ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || > (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts2" || outport == > "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(next;) > > + table=??(lr_out_undnat ), priority=120 , match=(ip6 && > ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || > (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1_public" || outport == > "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(next;) > > +]) > > + > > +# 3. Check if the VIP IP is in the ipX.src action > > +AT_CHECK([grep "lr_out_snat" lr1flows | ovn_strip_lflows], [0], [dnl > > + table=??(lr_out_snat ), priority=0 , match=(1), > action=(next;) > > + table=??(lr_out_snat ), priority=120 , match=(ip6 && > ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || > (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts1" || outport == > "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), > action=(ip6.src=2001:db8:cccc::1; next;) > > + table=??(lr_out_snat ), priority=120 , match=(ip6 && > ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || > (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts2" || outport == > "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), > action=(ip6.src=2001:db8:cccc::1; next;) > > + table=??(lr_out_snat ), priority=120 , match=(ip6 && > ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || > (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1_public" || outport == > "lr1_public") && is_chassis_resident("cr-lr1_public")), > action=(ip6.src=2001:db8:cccc::1; next;) > > + table=??(lr_out_snat ), priority=120 , match=(nd_ns), > action=(next;) > > +]) > > + > > +AT_CLEANUP > > +]) > >
diff --git a/northd/en-lr-stateful.c b/northd/en-lr-stateful.c index baf1bd2f8..f09691af6 100644 --- a/northd/en-lr-stateful.c +++ b/northd/en-lr-stateful.c @@ -516,18 +516,6 @@ lr_stateful_record_create(struct lr_stateful_table *table, table->array[od->index] = lr_stateful_rec; - /* Load balancers are not supported (yet) if a logical router has multiple - * distributed gateway port. Log a warning. */ - if (lr_stateful_rec->has_lb_vip && lr_has_multiple_gw_ports(od)) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); - VLOG_WARN_RL(&rl, "Load-balancers are configured on logical " - "router %s, which has %"PRIuSIZE" distributed " - "gateway ports. Load-balancer is not supported " - "yet when there is more than one distributed " - "gateway port on the router.", - od->nbr->name, od->n_l3dgw_ports); - } - return lr_stateful_rec; } diff --git a/northd/northd.c b/northd/northd.c index 6898daa00..853d58f29 100644 --- a/northd/northd.c +++ b/northd/northd.c @@ -11026,31 +11026,30 @@ static void build_distr_lrouter_nat_flows_for_lb(struct lrouter_nat_lb_flows_ctx *ctx, enum lrouter_nat_lb_flow_type type, struct ovn_datapath *od, - struct lflow_ref *lflow_ref) + struct lflow_ref *lflow_ref, + struct ovn_port *dgp) { - struct ovn_port *dgp = od->l3dgw_ports[0]; - - const char *undnat_action; - - switch (type) { - case LROUTER_NAT_LB_FLOW_FORCE_SNAT: - undnat_action = "flags.force_snat_for_lb = 1; next;"; - break; - case LROUTER_NAT_LB_FLOW_SKIP_SNAT: - undnat_action = "flags.skip_snat_for_lb = 1; next;"; - break; - case LROUTER_NAT_LB_FLOW_NORMAL: - case LROUTER_NAT_LB_FLOW_MAX: - undnat_action = lrouter_use_common_zone(od) - ? "ct_dnat_in_czone;" - : "ct_dnat;"; - break; - } + struct ds dnat_action = DS_EMPTY_INITIALIZER; /* Store the match lengths, so we can reuse the ds buffer. */ size_t new_match_len = ctx->new_match->length; size_t undnat_match_len = ctx->undnat_match->length; + /* Create stateless LB NAT rules when using DGPs. + * dnat_action: Add the LB backend IPs as a destination action of the + * lr_in_dnat NAT rule with cumulative effect because any + * backend dst IP used in the action list will redirect the + * packet to the ct_lb pipeline. + */ + if (od->n_l3dgw_ports > 1) { + for (size_t i = 0; i < ctx->lb_vip->n_backends; i++) { + struct ovn_lb_backend *backend = &ctx->lb_vip->backends[i]; + bool ipv6 = !IN6_IS_ADDR_V4MAPPED(&backend->ip); + ds_put_format(&dnat_action, "%s.dst=%s;", ipv6 ? "ip6" : "ip4", + backend->ip_str); + } + } + ds_put_format(&dnat_action, "%s", ctx->new_action[type]); const char *meter = NULL; @@ -11060,20 +11059,47 @@ build_distr_lrouter_nat_flows_for_lb(struct lrouter_nat_lb_flows_ctx *ctx, if (ctx->lb_vip->n_backends || !ctx->lb_vip->empty_backend_rej) { ds_put_format(ctx->new_match, " && is_chassis_resident(%s)", - od->l3dgw_ports[0]->cr_port->json_key); + dgp->cr_port->json_key); } ovn_lflow_add_with_hint__(ctx->lflows, od, S_ROUTER_IN_DNAT, ctx->prio, - ds_cstr(ctx->new_match), ctx->new_action[type], + ds_cstr(ctx->new_match), ds_cstr(&dnat_action), NULL, meter, &ctx->lb->nlb->header_, lflow_ref); ds_truncate(ctx->new_match, new_match_len); + ds_destroy(&dnat_action); if (!ctx->lb_vip->n_backends) { return; } + struct ds undnat_action = DS_EMPTY_INITIALIZER; + struct ds snat_action = DS_EMPTY_INITIALIZER; + + switch (type) { + case LROUTER_NAT_LB_FLOW_FORCE_SNAT: + ds_put_format(&undnat_action, "flags.force_snat_for_lb = 1; next;"); + break; + case LROUTER_NAT_LB_FLOW_SKIP_SNAT: + ds_put_format(&undnat_action, "flags.skip_snat_for_lb = 1; next;"); + break; + case LROUTER_NAT_LB_FLOW_NORMAL: + case LROUTER_NAT_LB_FLOW_MAX: + ds_put_format(&undnat_action, "%s", + lrouter_use_common_zone(od) ? "ct_dnat_in_czone;" + : "ct_dnat;"); + break; + } + + /* Create stateless LB NAT rules when using DGPs. + * undnat_action: Remove the ct action from the lr_out_undenat NAT rule. + */ + if (od->n_l3dgw_ports > 1) { + ds_clear(&undnat_action); + ds_put_format(&undnat_action, "next;"); + } + /* We need to centralize the LB traffic to properly perform * the undnat stage. */ @@ -11093,10 +11119,27 @@ build_distr_lrouter_nat_flows_for_lb(struct lrouter_nat_lb_flows_ctx *ctx, " && is_chassis_resident(%s)", dgp->json_key, dgp->json_key, dgp->cr_port->json_key); ovn_lflow_add_with_hint(ctx->lflows, od, S_ROUTER_OUT_UNDNAT, 120, - ds_cstr(ctx->undnat_match), undnat_action, - &ctx->lb->nlb->header_, + ds_cstr(ctx->undnat_match), + ds_cstr(&undnat_action), &ctx->lb->nlb->header_, lflow_ref); + + /* Create stateless LB NAT rules when using DGPs. + * snat_action: Add a new lr_out_snat rule with the LB VIP as source IP + * action to perform the NAT stateless pipeline completely. + */ + if (od->n_l3dgw_ports > 1) { + ds_put_format(&snat_action, "%s.src=%s; next;", + ctx->lb_vip->address_family == AF_INET6 ? "ip6" : "ip4", + ctx->lb_vip->vip_str); + ovn_lflow_add_with_hint(ctx->lflows, od, S_ROUTER_OUT_SNAT, 120, + ds_cstr(ctx->undnat_match), + ds_cstr(&snat_action), &ctx->lb->nlb->header_, + lflow_ref); + } + ds_truncate(ctx->undnat_match, undnat_match_len); + ds_destroy(&undnat_action); + ds_destroy(&snat_action); } static void @@ -11263,8 +11306,11 @@ build_lrouter_nat_flows_for_lb( if (!od->n_l3dgw_ports) { bitmap_set1(gw_dp_bitmap[type], index); } else { - build_distr_lrouter_nat_flows_for_lb(&ctx, type, od, - lb_dps->lflow_ref); + for (size_t i = 0; i < od->n_l3dgw_ports; i++) { + struct ovn_port *dgp = od->l3dgw_ports[i]; + build_distr_lrouter_nat_flows_for_lb(&ctx, type, od, + lb_dps->lflow_ref, dgp); + } } if (lb->affinity_timeout) { diff --git a/tests/ovn-northd.at b/tests/ovn-northd.at index a389d1988..e1b26b774 100644 --- a/tests/ovn-northd.at +++ b/tests/ovn-northd.at @@ -12721,3 +12721,225 @@ AT_CHECK([ovn-sbctl dump-flows lr | grep lr_in_dnat | ovn_strip_lflows], [0], [d AT_CLEANUP ]) + +OVN_FOR_EACH_NORTHD_NO_HV_PARALLELIZATION([ +AT_SETUP([Load balancer with Distributed Gateway Ports (LB + DGP + NAT Stateless)]) +ovn_start + +check ovn-nbctl ls-add public +check ovn-nbctl lr-add lr1 + +# lr1 DGP ts1 +check ovn-nbctl ls-add ts1 +check ovn-nbctl lrp-add lr1 lr1-ts1 00:00:01:02:03:04 172.16.10.1/24 +check ovn-nbctl lrp-set-gateway-chassis lr1-ts1 chassis-2 + +# lr1 DGP ts2 +check ovn-nbctl ls-add ts2 +check ovn-nbctl lrp-add lr1 lr1-ts2 00:00:01:02:03:05 172.16.20.1/24 +check ovn-nbctl lrp-set-gateway-chassis lr1-ts2 chassis-3 + +# lr1 DGP public +check ovn-nbctl lrp-add lr1 lr1_public 00:de:ad:ff:00:01 173.16.0.1/16 +check ovn-nbctl lrp-add lr1 lr1_s1 00:de:ad:fe:00:02 172.16.0.1/24 +check ovn-nbctl lrp-set-gateway-chassis lr1_public chassis-1 + +check ovn-nbctl ls-add s1 +# s1 - lr1 +check ovn-nbctl lsp-add s1 s1_lr1 +check ovn-nbctl lsp-set-type s1_lr1 router +check ovn-nbctl lsp-set-addresses s1_lr1 "00:de:ad:fe:00:02 172.16.0.1" +check ovn-nbctl lsp-set-options s1_lr1 router-port=lr1_s1 + +# s1 - backend vm1 +check ovn-nbctl lsp-add s1 vm1 +check ovn-nbctl lsp-set-addresses vm1 "00:de:ad:01:00:01 172.16.0.101" + +# s1 - backend vm2 +check ovn-nbctl lsp-add s1 vm2 +check ovn-nbctl lsp-set-addresses vm2 "00:de:ad:01:00:02 172.16.0.102" + +# s1 - backend vm3 +check ovn-nbctl lsp-add s1 vm3 +check ovn-nbctl lsp-set-addresses vm3 "00:de:ad:01:00:03 172.16.0.103" + +# Add the lr1 DGP ts1 to the public switch +check ovn-nbctl lsp-add public public_lr1_ts1 +check ovn-nbctl lsp-set-type public_lr1_ts1 router +check ovn-nbctl lsp-set-addresses public_lr1_ts1 router +check ovn-nbctl lsp-set-options public_lr1_ts1 router-port=lr1-ts1 nat-addresses=router + +# Add the lr1 DGP ts2 to the public switch +check ovn-nbctl lsp-add public public_lr1_ts2 +check ovn-nbctl lsp-set-type public_lr1_ts2 router +check ovn-nbctl lsp-set-addresses public_lr1_ts2 router +check ovn-nbctl lsp-set-options public_lr1_ts2 router-port=lr1-ts2 nat-addresses=router + +# Add the lr1 DGP public to the public switch +check ovn-nbctl lsp-add public public_lr1 +check ovn-nbctl lsp-set-type public_lr1 router +check ovn-nbctl lsp-set-addresses public_lr1 router +check ovn-nbctl lsp-set-options public_lr1 router-port=lr1_public nat-addresses=router + +# Create the Load Balancer lb1 +check ovn-nbctl --wait=sb lb-add lb1 "30.0.0.1" "172.16.0.103,172.16.0.102,172.16.0.101" + +# Associate load balancer to s1 +check ovn-nbctl ls-lb-add s1 lb1 +check ovn-nbctl --wait=sb sync + +ovn-sbctl dump-flows s1 > s1flows +AT_CAPTURE_FILE([s1flows]) + +AT_CHECK([grep "ls_in_pre_stateful" s1flows | ovn_strip_lflows | grep "30.0.0.1"], [0], [dnl + table=??(ls_in_pre_stateful ), priority=120 , match=(reg0[[2]] == 1 && ip4.dst == 30.0.0.1), action=(reg1 = 30.0.0.1; ct_lb_mark;) +]) +AT_CHECK([grep "ls_in_lb" s1flows | ovn_strip_lflows | grep "30.0.0.1"], [0], [dnl + table=??(ls_in_lb ), priority=110 , match=(ct.new && ip4.dst == 30.0.0.1), action=(reg0[[1]] = 0; ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) +]) + +# Associate load balancer to lr1 with DGP +check ovn-nbctl lr-lb-add lr1 lb1 +check ovn-nbctl --wait=sb sync + +ovn-sbctl dump-flows lr1 > lr1flows +AT_CAPTURE_FILE([lr1flows]) + +# Check stateless NAT rules for load balancer with multiple DGP +# 1. Check if the backend IPs are in the ipX.dst action +AT_CHECK([grep "lr_in_dnat" lr1flows | ovn_strip_lflows | grep "30.0.0.1"], [0], [dnl + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip4 && ip4.dst == 30.0.0.1 && is_chassis_resident("cr-lr1-ts1")), action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip4 && ip4.dst == 30.0.0.1 && is_chassis_resident("cr-lr1-ts2")), action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip4 && ip4.dst == 30.0.0.1 && is_chassis_resident("cr-lr1_public")), action=(ip4.dst=172.16.0.103;ip4.dst=172.16.0.102;ip4.dst=172.16.0.101;ct_lb_mark(backends=172.16.0.103,172.16.0.102,172.16.0.101);) +]) + +# 2. Check if the DGP ports are in the match with action next +AT_CHECK([grep "lr_out_undnat" lr1flows | ovn_strip_lflows], [0], [dnl + table=??(lr_out_undnat ), priority=0 , match=(1), action=(next;) + table=??(lr_out_undnat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(next;) + table=??(lr_out_undnat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(next;) + table=??(lr_out_undnat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1_public" || outport == "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(next;) +]) + +# 3. Check if the VIP IP is in the ipX.src action +AT_CHECK([grep "lr_out_snat" lr1flows | ovn_strip_lflows], [0], [dnl + table=??(lr_out_snat ), priority=0 , match=(1), action=(next;) + table=??(lr_out_snat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(ip4.src=30.0.0.1; next;) + table=??(lr_out_snat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(ip4.src=30.0.0.1; next;) + table=??(lr_out_snat ), priority=120 , match=(ip4 && ((ip4.src == 172.16.0.103) || (ip4.src == 172.16.0.102) || (ip4.src == 172.16.0.101)) && (inport == "lr1_public" || outport == "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(ip4.src=30.0.0.1; next;) + table=??(lr_out_snat ), priority=120 , match=(nd_ns), action=(next;) +]) + +AT_CLEANUP +]) + +OVN_FOR_EACH_NORTHD_NO_HV_PARALLELIZATION([ +AT_SETUP([Load balancer with Distributed Gateway Ports (LB + DGP + NAT Stateless) - IPv6]) +ovn_start + +check ovn-nbctl ls-add public +check ovn-nbctl lr-add lr1 + +# lr1 DGP ts1 +check ovn-nbctl ls-add ts1 +check ovn-nbctl lrp-add lr1 lr1-ts1 00:00:01:02:03:04 2001:db8:aaaa:1::1/64 +check ovn-nbctl lrp-set-gateway-chassis lr1-ts1 chassis-2 + +# lr1 DGP ts2 +check ovn-nbctl ls-add ts2 +check ovn-nbctl lrp-add lr1 lr1-ts2 00:00:01:02:03:05 2001:db8:aaaa:2::1/64 +check ovn-nbctl lrp-set-gateway-chassis lr1-ts2 chassis-3 + +# lr1 DGP public +check ovn-nbctl lrp-add lr1 lr1_public 00:de:ad:ff:00:01 2001:db8:bbbb::1/64 +check ovn-nbctl lrp-add lr1 lr1_s1 00:de:ad:fe:00:02 2001:db8:aaaa:3::1/64 +check ovn-nbctl lrp-set-gateway-chassis lr1_public chassis-1 + +check ovn-nbctl ls-add s1 +# s1 - lr1 +check ovn-nbctl lsp-add s1 s1_lr1 +check ovn-nbctl lsp-set-type s1_lr1 router +check ovn-nbctl lsp-set-addresses s1_lr1 "00:de:ad:fe:00:02 2001:db8:aaaa:3::1" +check ovn-nbctl lsp-set-options s1_lr1 router-port=lr1_s1 + +# s1 - backend vm1 +check ovn-nbctl lsp-add s1 vm1 +check ovn-nbctl lsp-set-addresses vm1 "00:de:ad:01:00:01 2001:db8:aaaa:3::101" + +# s1 - backend vm2 +check ovn-nbctl lsp-add s1 vm2 +check ovn-nbctl lsp-set-addresses vm2 "00:de:ad:01:00:02 2001:db8:aaaa:3::102" + +# s1 - backend vm3 +check ovn-nbctl lsp-add s1 vm3 +check ovn-nbctl lsp-set-addresses vm3 "00:de:ad:01:00:03 2001:db8:aaaa:3::103" + +# Add the lr1 DGP ts1 to the public switch +check ovn-nbctl lsp-add public public_lr1_ts1 +check ovn-nbctl lsp-set-type public_lr1_ts1 router +check ovn-nbctl lsp-set-addresses public_lr1_ts1 router +check ovn-nbctl lsp-set-options public_lr1_ts1 router-port=lr1-ts1 nat-addresses=router + +# Add the lr1 DGP ts2 to the public switch +check ovn-nbctl lsp-add public public_lr1_ts2 +check ovn-nbctl lsp-set-type public_lr1_ts2 router +check ovn-nbctl lsp-set-addresses public_lr1_ts2 router +check ovn-nbctl lsp-set-options public_lr1_ts2 router-port=lr1-ts2 nat-addresses=router + +# Add the lr1 DGP public to the public switch +check ovn-nbctl lsp-add public public_lr1 +check ovn-nbctl lsp-set-type public_lr1 router +check ovn-nbctl lsp-set-addresses public_lr1 router +check ovn-nbctl lsp-set-options public_lr1 router-port=lr1_public nat-addresses=router + +# Create the Load Balancer lb1 +check ovn-nbctl --wait=sb lb-add lb1 "2001:db8:cccc::1" "2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101" + +# Associate load balancer to s1 +check ovn-nbctl ls-lb-add s1 lb1 +check ovn-nbctl --wait=sb sync + +ovn-sbctl dump-flows s1 > s1flows +AT_CAPTURE_FILE([s1flows]) + +AT_CHECK([grep "ls_in_pre_stateful" s1flows | ovn_strip_lflows | grep "2001:db8:cccc::1"], [0], [dnl + table=??(ls_in_pre_stateful ), priority=120 , match=(reg0[[2]] == 1 && ip6.dst == 2001:db8:cccc::1), action=(xxreg1 = 2001:db8:cccc::1; ct_lb_mark;) +]) +AT_CHECK([grep "ls_in_lb" s1flows | ovn_strip_lflows | grep "2001:db8:cccc::1"], [0], [dnl + table=??(ls_in_lb ), priority=110 , match=(ct.new && ip6.dst == 2001:db8:cccc::1), action=(reg0[[1]] = 0; ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) +]) + +# Associate load balancer to lr1 with DGP +check ovn-nbctl lr-lb-add lr1 lb1 +check ovn-nbctl --wait=sb sync + +ovn-sbctl dump-flows lr1 > lr1flows +AT_CAPTURE_FILE([lr1flows]) + +# Check stateless NAT rules for load balancer with multiple DGP +# 1. Check if the backend IPs are in the ipX.dst action +AT_CHECK([grep "lr_in_dnat" lr1flows | ovn_strip_lflows | grep "2001:db8:cccc::1"], [0], [dnl + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && is_chassis_resident("cr-lr1-ts1")), action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && is_chassis_resident("cr-lr1-ts2")), action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) + table=??(lr_in_dnat ), priority=110 , match=(ct.new && !ct.rel && ip6 && ip6.dst == 2001:db8:cccc::1 && is_chassis_resident("cr-lr1_public")), action=(ip6.dst=2001:db8:aaaa:3::103;ip6.dst=2001:db8:aaaa:3::102;ip6.dst=2001:db8:aaaa:3::101;ct_lb_mark(backends=2001:db8:aaaa:3::103,2001:db8:aaaa:3::102,2001:db8:aaaa:3::101);) +]) + +# 2. Check if the DGP ports are in the match with action next +AT_CHECK([grep "lr_out_undnat" lr1flows | ovn_strip_lflows], [0], [dnl + table=??(lr_out_undnat ), priority=0 , match=(1), action=(next;) + table=??(lr_out_undnat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(next;) + table=??(lr_out_undnat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(next;) + table=??(lr_out_undnat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1_public" || outport == "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(next;) +]) + +# 3. Check if the VIP IP is in the ipX.src action +AT_CHECK([grep "lr_out_snat" lr1flows | ovn_strip_lflows], [0], [dnl + table=??(lr_out_snat ), priority=0 , match=(1), action=(next;) + table=??(lr_out_snat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts1" || outport == "lr1-ts1") && is_chassis_resident("cr-lr1-ts1")), action=(ip6.src=2001:db8:cccc::1; next;) + table=??(lr_out_snat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1-ts2" || outport == "lr1-ts2") && is_chassis_resident("cr-lr1-ts2")), action=(ip6.src=2001:db8:cccc::1; next;) + table=??(lr_out_snat ), priority=120 , match=(ip6 && ((ip6.src == 2001:db8:aaaa:3::103) || (ip6.src == 2001:db8:aaaa:3::102) || (ip6.src == 2001:db8:aaaa:3::101)) && (inport == "lr1_public" || outport == "lr1_public") && is_chassis_resident("cr-lr1_public")), action=(ip6.src=2001:db8:cccc::1; next;) + table=??(lr_out_snat ), priority=120 , match=(nd_ns), action=(next;) +]) + +AT_CLEANUP +])
This commit fixes the build_distr_lrouter_nat_flows_for_lb function to include one NAT stateless flow entry for each DGP in use. Since we have added support to create multiple gateway ports per logical router, it's necessary to include in the LR nat rules pipeline a specific entry for each attached DGP. Otherwise, the ingress traffic is only redirected when the incoming LRP matches the chassis_resident field. Considering that DNAT rules for DGPs were implemented with the need to configure the DGP-related gateway-port column, the load-balancer NAT rule configuration can use a similar idea. In this case, we don't know the LRP responsible for the incoming traffic, and therefore we need to automatically apply a stateless NAT rule to the load-balancer on all DGPs to allow inbound traffic. After applying this patch, the incoming and/or outgoing traffic can pass through any chassis where the DGP resides without having problems with CT state. Reported-at: https://bugs.launchpad.net/ubuntu/+source/ovn/+bug/2054322 Fixes: 15348b7b806f ("ovn-northd: Multiple distributed gateway port support.") Signed-off-by: Roberto Bartzen Acosta <roberto.acosta@luizalabs.com> --- northd/en-lr-stateful.c | 12 --- northd/northd.c | 96 ++++++++++++----- tests/ovn-northd.at | 222 ++++++++++++++++++++++++++++++++++++++++ 3 files changed, 293 insertions(+), 37 deletions(-)