2 * Licensed under the Apache License, Version 2.0 (the "License");
3 * you may not use this file except in compliance with the License.
4 * You may obtain a copy of the License at:
6 * http://www.apache.org/licenses/LICENSE-2.0
8 * Unless required by applicable law or agreed to in writing, software
9 * distributed under the License is distributed on an "AS IS" BASIS,
10 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
11 * See the License for the specific language governing permissions and
12 * limitations under the License.
21 #include "command-line.h"
24 #include "openvswitch/dynamic-string.h"
25 #include "fatal-signal.h"
29 #include "ovn/lib/lex.h"
30 #include "ovn/lib/ovn-nb-idl.h"
31 #include "ovn/lib/ovn-sb-idl.h"
32 #include "ovn/lib/ovn-util.h"
34 #include "poll-loop.h"
38 #include "stream-ssl.h"
42 #include "openvswitch/vlog.h"
44 VLOG_DEFINE_THIS_MODULE(ovn_northd);
46 static unixctl_cb_func ovn_northd_exit;
48 struct northd_context {
49 struct ovsdb_idl *ovnnb_idl;
50 struct ovsdb_idl *ovnsb_idl;
51 struct ovsdb_idl_txn *ovnnb_txn;
52 struct ovsdb_idl_txn *ovnsb_txn;
55 static const char *ovnnb_db;
56 static const char *ovnsb_db;
58 static const char *default_nb_db(void);
59 static const char *default_sb_db(void);
61 /* Pipeline stages. */
63 /* The two pipelines in an OVN logical flow table. */
65 P_IN, /* Ingress pipeline. */
66 P_OUT /* Egress pipeline. */
69 /* The two purposes for which ovn-northd uses OVN logical datapaths. */
70 enum ovn_datapath_type {
71 DP_SWITCH, /* OVN logical switch. */
72 DP_ROUTER /* OVN logical router. */
75 /* Returns an "enum ovn_stage" built from the arguments.
77 * (It's better to use ovn_stage_build() for type-safety reasons, but inline
78 * functions can't be used in enums or switch cases.) */
79 #define OVN_STAGE_BUILD(DP_TYPE, PIPELINE, TABLE) \
80 (((DP_TYPE) << 9) | ((PIPELINE) << 8) | (TABLE))
82 /* A stage within an OVN logical switch or router.
84 * An "enum ovn_stage" indicates whether the stage is part of a logical switch
85 * or router, whether the stage is part of the ingress or egress pipeline, and
86 * the table within that pipeline. The first three components are combined to
87 * form the stage's full name, e.g. S_SWITCH_IN_PORT_SEC_L2,
88 * S_ROUTER_OUT_DELIVERY. */
90 #define PIPELINE_STAGES \
91 /* Logical switch ingress stages. */ \
92 PIPELINE_STAGE(SWITCH, IN, PORT_SEC_L2, 0, "ls_in_port_sec_l2") \
93 PIPELINE_STAGE(SWITCH, IN, PORT_SEC_IP, 1, "ls_in_port_sec_ip") \
94 PIPELINE_STAGE(SWITCH, IN, PORT_SEC_ND, 2, "ls_in_port_sec_nd") \
95 PIPELINE_STAGE(SWITCH, IN, PRE_ACL, 3, "ls_in_pre_acl") \
96 PIPELINE_STAGE(SWITCH, IN, PRE_LB, 4, "ls_in_pre_lb") \
97 PIPELINE_STAGE(SWITCH, IN, PRE_STATEFUL, 5, "ls_in_pre_stateful") \
98 PIPELINE_STAGE(SWITCH, IN, ACL, 6, "ls_in_acl") \
99 PIPELINE_STAGE(SWITCH, IN, LB, 7, "ls_in_lb") \
100 PIPELINE_STAGE(SWITCH, IN, STATEFUL, 8, "ls_in_stateful") \
101 PIPELINE_STAGE(SWITCH, IN, ARP_ND_RSP, 9, "ls_in_arp_rsp") \
102 PIPELINE_STAGE(SWITCH, IN, L2_LKUP, 10, "ls_in_l2_lkup") \
104 /* Logical switch egress stages. */ \
105 PIPELINE_STAGE(SWITCH, OUT, PRE_LB, 0, "ls_out_pre_lb") \
106 PIPELINE_STAGE(SWITCH, OUT, PRE_ACL, 1, "ls_out_pre_acl") \
107 PIPELINE_STAGE(SWITCH, OUT, PRE_STATEFUL, 2, "ls_out_pre_stateful") \
108 PIPELINE_STAGE(SWITCH, OUT, LB, 3, "ls_out_lb") \
109 PIPELINE_STAGE(SWITCH, OUT, ACL, 4, "ls_out_acl") \
110 PIPELINE_STAGE(SWITCH, OUT, STATEFUL, 5, "ls_out_stateful") \
111 PIPELINE_STAGE(SWITCH, OUT, PORT_SEC_IP, 6, "ls_out_port_sec_ip") \
112 PIPELINE_STAGE(SWITCH, OUT, PORT_SEC_L2, 7, "ls_out_port_sec_l2") \
114 /* Logical router ingress stages. */ \
115 PIPELINE_STAGE(ROUTER, IN, ADMISSION, 0, "lr_in_admission") \
116 PIPELINE_STAGE(ROUTER, IN, IP_INPUT, 1, "lr_in_ip_input") \
117 PIPELINE_STAGE(ROUTER, IN, UNSNAT, 2, "lr_in_unsnat") \
118 PIPELINE_STAGE(ROUTER, IN, DNAT, 3, "lr_in_dnat") \
119 PIPELINE_STAGE(ROUTER, IN, IP_ROUTING, 4, "lr_in_ip_routing") \
120 PIPELINE_STAGE(ROUTER, IN, ARP_RESOLVE, 5, "lr_in_arp_resolve") \
121 PIPELINE_STAGE(ROUTER, IN, ARP_REQUEST, 6, "lr_in_arp_request") \
123 /* Logical router egress stages. */ \
124 PIPELINE_STAGE(ROUTER, OUT, SNAT, 0, "lr_out_snat") \
125 PIPELINE_STAGE(ROUTER, OUT, DELIVERY, 1, "lr_out_delivery")
127 #define PIPELINE_STAGE(DP_TYPE, PIPELINE, STAGE, TABLE, NAME) \
128 S_##DP_TYPE##_##PIPELINE##_##STAGE \
129 = OVN_STAGE_BUILD(DP_##DP_TYPE, P_##PIPELINE, TABLE),
131 #undef PIPELINE_STAGE
134 /* Due to various hard-coded priorities need to implement ACLs, the
135 * northbound database supports a smaller range of ACL priorities than
136 * are available to logical flows. This value is added to an ACL
137 * priority to determine the ACL's logical flow priority. */
138 #define OVN_ACL_PRI_OFFSET 1000
140 #define REGBIT_CONNTRACK_DEFRAG "reg0[0]"
141 #define REGBIT_CONNTRACK_COMMIT "reg0[1]"
142 #define REGBIT_CONNTRACK_NAT "reg0[2]"
144 /* Returns an "enum ovn_stage" built from the arguments. */
145 static enum ovn_stage
146 ovn_stage_build(enum ovn_datapath_type dp_type, enum ovn_pipeline pipeline,
149 return OVN_STAGE_BUILD(dp_type, pipeline, table);
152 /* Returns the pipeline to which 'stage' belongs. */
153 static enum ovn_pipeline
154 ovn_stage_get_pipeline(enum ovn_stage stage)
156 return (stage >> 8) & 1;
159 /* Returns the table to which 'stage' belongs. */
161 ovn_stage_get_table(enum ovn_stage stage)
166 /* Returns a string name for 'stage'. */
168 ovn_stage_to_str(enum ovn_stage stage)
171 #define PIPELINE_STAGE(DP_TYPE, PIPELINE, STAGE, TABLE, NAME) \
172 case S_##DP_TYPE##_##PIPELINE##_##STAGE: return NAME;
174 #undef PIPELINE_STAGE
175 default: return "<unknown>";
179 /* Returns the type of the datapath to which a flow with the given 'stage' may
181 static enum ovn_datapath_type
182 ovn_stage_to_datapath_type(enum ovn_stage stage)
185 #define PIPELINE_STAGE(DP_TYPE, PIPELINE, STAGE, TABLE, NAME) \
186 case S_##DP_TYPE##_##PIPELINE##_##STAGE: return DP_##DP_TYPE;
188 #undef PIPELINE_STAGE
189 default: OVS_NOT_REACHED();
197 %s: OVN northbound management daemon\n\
198 usage: %s [OPTIONS]\n\
201 --ovnnb-db=DATABASE connect to ovn-nb database at DATABASE\n\
203 --ovnsb-db=DATABASE connect to ovn-sb database at DATABASE\n\
205 -h, --help display this help message\n\
206 -o, --options list available options\n\
207 -V, --version display version information\n\
208 ", program_name, program_name, default_nb_db(), default_sb_db());
211 stream_usage("database", true, true, false);
215 struct hmap_node hmap_node;
220 destroy_tnlids(struct hmap *tnlids)
222 struct tnlid_node *node;
223 HMAP_FOR_EACH_POP (node, hmap_node, tnlids) {
226 hmap_destroy(tnlids);
230 add_tnlid(struct hmap *set, uint32_t tnlid)
232 struct tnlid_node *node = xmalloc(sizeof *node);
233 hmap_insert(set, &node->hmap_node, hash_int(tnlid, 0));
238 tnlid_in_use(const struct hmap *set, uint32_t tnlid)
240 const struct tnlid_node *node;
241 HMAP_FOR_EACH_IN_BUCKET (node, hmap_node, hash_int(tnlid, 0), set) {
242 if (node->tnlid == tnlid) {
250 allocate_tnlid(struct hmap *set, const char *name, uint32_t max,
253 for (uint32_t tnlid = *hint + 1; tnlid != *hint;
254 tnlid = tnlid + 1 <= max ? tnlid + 1 : 1) {
255 if (!tnlid_in_use(set, tnlid)) {
256 add_tnlid(set, tnlid);
262 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
263 VLOG_WARN_RL(&rl, "all %s tunnel ids exhausted", name);
267 /* The 'key' comes from nbs->header_.uuid or nbr->header_.uuid or
268 * sb->external_ids:logical-switch. */
269 struct ovn_datapath {
270 struct hmap_node key_node; /* Index on 'key'. */
271 struct uuid key; /* (nbs/nbr)->header_.uuid. */
273 const struct nbrec_logical_switch *nbs; /* May be NULL. */
274 const struct nbrec_logical_router *nbr; /* May be NULL. */
275 const struct sbrec_datapath_binding *sb; /* May be NULL. */
277 struct ovs_list list; /* In list of similar records. */
279 /* Logical switch data. */
280 struct ovn_port **router_ports;
281 size_t n_router_ports;
283 struct hmap port_tnlids;
284 uint32_t port_key_hint;
289 static struct ovn_datapath *
290 ovn_datapath_create(struct hmap *datapaths, const struct uuid *key,
291 const struct nbrec_logical_switch *nbs,
292 const struct nbrec_logical_router *nbr,
293 const struct sbrec_datapath_binding *sb)
295 struct ovn_datapath *od = xzalloc(sizeof *od);
300 hmap_init(&od->port_tnlids);
301 od->port_key_hint = 0;
302 hmap_insert(datapaths, &od->key_node, uuid_hash(&od->key));
307 ovn_datapath_destroy(struct hmap *datapaths, struct ovn_datapath *od)
310 /* Don't remove od->list. It is used within build_datapaths() as a
311 * private list and once we've exited that function it is not safe to
313 hmap_remove(datapaths, &od->key_node);
314 destroy_tnlids(&od->port_tnlids);
315 free(od->router_ports);
320 /* Returns 'od''s datapath type. */
321 static enum ovn_datapath_type
322 ovn_datapath_get_type(const struct ovn_datapath *od)
324 return od->nbs ? DP_SWITCH : DP_ROUTER;
327 static struct ovn_datapath *
328 ovn_datapath_find(struct hmap *datapaths, const struct uuid *uuid)
330 struct ovn_datapath *od;
332 HMAP_FOR_EACH_WITH_HASH (od, key_node, uuid_hash(uuid), datapaths) {
333 if (uuid_equals(uuid, &od->key)) {
340 static struct ovn_datapath *
341 ovn_datapath_from_sbrec(struct hmap *datapaths,
342 const struct sbrec_datapath_binding *sb)
346 if (!smap_get_uuid(&sb->external_ids, "logical-switch", &key) &&
347 !smap_get_uuid(&sb->external_ids, "logical-router", &key)) {
350 return ovn_datapath_find(datapaths, &key);
354 lrouter_is_enabled(const struct nbrec_logical_router *lrouter)
356 return !lrouter->enabled || *lrouter->enabled;
360 join_datapaths(struct northd_context *ctx, struct hmap *datapaths,
361 struct ovs_list *sb_only, struct ovs_list *nb_only,
362 struct ovs_list *both)
364 hmap_init(datapaths);
365 ovs_list_init(sb_only);
366 ovs_list_init(nb_only);
369 const struct sbrec_datapath_binding *sb, *sb_next;
370 SBREC_DATAPATH_BINDING_FOR_EACH_SAFE (sb, sb_next, ctx->ovnsb_idl) {
372 if (!smap_get_uuid(&sb->external_ids, "logical-switch", &key) &&
373 !smap_get_uuid(&sb->external_ids, "logical-router", &key)) {
374 ovsdb_idl_txn_add_comment(
376 "deleting Datapath_Binding "UUID_FMT" that lacks "
377 "external-ids:logical-switch and "
378 "external-ids:logical-router",
379 UUID_ARGS(&sb->header_.uuid));
380 sbrec_datapath_binding_delete(sb);
384 if (ovn_datapath_find(datapaths, &key)) {
385 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
387 &rl, "deleting Datapath_Binding "UUID_FMT" with "
388 "duplicate external-ids:logical-switch/router "UUID_FMT,
389 UUID_ARGS(&sb->header_.uuid), UUID_ARGS(&key));
390 sbrec_datapath_binding_delete(sb);
394 struct ovn_datapath *od = ovn_datapath_create(datapaths, &key,
396 ovs_list_push_back(sb_only, &od->list);
399 const struct nbrec_logical_switch *nbs;
400 NBREC_LOGICAL_SWITCH_FOR_EACH (nbs, ctx->ovnnb_idl) {
401 struct ovn_datapath *od = ovn_datapath_find(datapaths,
405 ovs_list_remove(&od->list);
406 ovs_list_push_back(both, &od->list);
408 od = ovn_datapath_create(datapaths, &nbs->header_.uuid,
410 ovs_list_push_back(nb_only, &od->list);
414 const struct nbrec_logical_router *nbr;
415 NBREC_LOGICAL_ROUTER_FOR_EACH (nbr, ctx->ovnnb_idl) {
416 if (!lrouter_is_enabled(nbr)) {
420 struct ovn_datapath *od = ovn_datapath_find(datapaths,
425 ovs_list_remove(&od->list);
426 ovs_list_push_back(both, &od->list);
429 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
431 "duplicate UUID "UUID_FMT" in OVN_Northbound",
432 UUID_ARGS(&nbr->header_.uuid));
436 od = ovn_datapath_create(datapaths, &nbr->header_.uuid,
438 ovs_list_push_back(nb_only, &od->list);
444 ovn_datapath_allocate_key(struct hmap *dp_tnlids)
446 static uint32_t hint;
447 return allocate_tnlid(dp_tnlids, "datapath", (1u << 24) - 1, &hint);
450 /* Updates the southbound Datapath_Binding table so that it contains the
451 * logical switches and routers specified by the northbound database.
453 * Initializes 'datapaths' to contain a "struct ovn_datapath" for every logical
454 * switch and router. */
456 build_datapaths(struct northd_context *ctx, struct hmap *datapaths)
458 struct ovs_list sb_only, nb_only, both;
460 join_datapaths(ctx, datapaths, &sb_only, &nb_only, &both);
462 if (!ovs_list_is_empty(&nb_only)) {
463 /* First index the in-use datapath tunnel IDs. */
464 struct hmap dp_tnlids = HMAP_INITIALIZER(&dp_tnlids);
465 struct ovn_datapath *od;
466 LIST_FOR_EACH (od, list, &both) {
467 add_tnlid(&dp_tnlids, od->sb->tunnel_key);
470 /* Add southbound record for each unmatched northbound record. */
471 LIST_FOR_EACH (od, list, &nb_only) {
472 uint16_t tunnel_key = ovn_datapath_allocate_key(&dp_tnlids);
477 od->sb = sbrec_datapath_binding_insert(ctx->ovnsb_txn);
479 char uuid_s[UUID_LEN + 1];
480 sprintf(uuid_s, UUID_FMT, UUID_ARGS(&od->key));
481 const char *key = od->nbs ? "logical-switch" : "logical-router";
482 const struct smap id = SMAP_CONST1(&id, key, uuid_s);
483 sbrec_datapath_binding_set_external_ids(od->sb, &id);
485 sbrec_datapath_binding_set_tunnel_key(od->sb, tunnel_key);
487 destroy_tnlids(&dp_tnlids);
490 /* Delete southbound records without northbound matches. */
491 struct ovn_datapath *od, *next;
492 LIST_FOR_EACH_SAFE (od, next, list, &sb_only) {
493 ovs_list_remove(&od->list);
494 sbrec_datapath_binding_delete(od->sb);
495 ovn_datapath_destroy(datapaths, od);
500 struct hmap_node key_node; /* Index on 'key'. */
501 char *key; /* nbs->name, nbr->name, sb->logical_port. */
502 char *json_key; /* 'key', quoted for use in JSON. */
504 const struct sbrec_port_binding *sb; /* May be NULL. */
506 /* Logical switch port data. */
507 const struct nbrec_logical_switch_port *nbsp; /* May be NULL. */
509 struct lport_addresses *lsp_addrs; /* Logical switch port addresses. */
510 unsigned int n_lsp_addrs;
512 struct lport_addresses *ps_addrs; /* Port security addresses. */
513 unsigned int n_ps_addrs;
515 /* Logical router port data. */
516 const struct nbrec_logical_router_port *nbrp; /* May be NULL. */
518 struct lport_addresses lrp_networks;
520 struct ovn_port *peer;
522 struct ovn_datapath *od;
524 struct ovs_list list; /* In list of similar records. */
527 static struct ovn_port *
528 ovn_port_create(struct hmap *ports, const char *key,
529 const struct nbrec_logical_switch_port *nbsp,
530 const struct nbrec_logical_router_port *nbrp,
531 const struct sbrec_port_binding *sb)
533 struct ovn_port *op = xzalloc(sizeof *op);
535 struct ds json_key = DS_EMPTY_INITIALIZER;
536 json_string_escape(key, &json_key);
537 op->json_key = ds_steal_cstr(&json_key);
539 op->key = xstrdup(key);
543 hmap_insert(ports, &op->key_node, hash_string(op->key, 0));
548 ovn_port_destroy(struct hmap *ports, struct ovn_port *port)
551 /* Don't remove port->list. It is used within build_ports() as a
552 * private list and once we've exited that function it is not safe to
554 hmap_remove(ports, &port->key_node);
556 for (int i = 0; i < port->n_lsp_addrs; i++) {
557 destroy_lport_addresses(&port->lsp_addrs[i]);
559 free(port->lsp_addrs);
561 for (int i = 0; i < port->n_ps_addrs; i++) {
562 destroy_lport_addresses(&port->ps_addrs[i]);
564 free(port->ps_addrs);
566 destroy_lport_addresses(&port->lrp_networks);
567 free(port->json_key);
573 static struct ovn_port *
574 ovn_port_find(struct hmap *ports, const char *name)
578 HMAP_FOR_EACH_WITH_HASH (op, key_node, hash_string(name, 0), ports) {
579 if (!strcmp(op->key, name)) {
587 ovn_port_allocate_key(struct ovn_datapath *od)
589 return allocate_tnlid(&od->port_tnlids, "port",
590 (1u << 15) - 1, &od->port_key_hint);
594 join_logical_ports(struct northd_context *ctx,
595 struct hmap *datapaths, struct hmap *ports,
596 struct ovs_list *sb_only, struct ovs_list *nb_only,
597 struct ovs_list *both)
600 ovs_list_init(sb_only);
601 ovs_list_init(nb_only);
604 const struct sbrec_port_binding *sb;
605 SBREC_PORT_BINDING_FOR_EACH (sb, ctx->ovnsb_idl) {
606 struct ovn_port *op = ovn_port_create(ports, sb->logical_port,
608 ovs_list_push_back(sb_only, &op->list);
611 struct ovn_datapath *od;
612 HMAP_FOR_EACH (od, key_node, datapaths) {
614 for (size_t i = 0; i < od->nbs->n_ports; i++) {
615 const struct nbrec_logical_switch_port *nbsp
617 struct ovn_port *op = ovn_port_find(ports, nbsp->name);
619 if (op->nbsp || op->nbrp) {
620 static struct vlog_rate_limit rl
621 = VLOG_RATE_LIMIT_INIT(5, 1);
622 VLOG_WARN_RL(&rl, "duplicate logical port %s",
627 ovs_list_remove(&op->list);
628 ovs_list_push_back(both, &op->list);
630 /* This port exists due to a SB binding, but should
631 * not have been initialized fully. */
632 ovs_assert(!op->n_lsp_addrs && !op->n_ps_addrs);
634 op = ovn_port_create(ports, nbsp->name, nbsp, NULL, NULL);
635 ovs_list_push_back(nb_only, &op->list);
639 = xmalloc(sizeof *op->lsp_addrs * nbsp->n_addresses);
640 for (size_t j = 0; j < nbsp->n_addresses; j++) {
641 if (!strcmp(nbsp->addresses[j], "unknown")) {
644 if (!extract_lsp_addresses(nbsp->addresses[j],
645 &op->lsp_addrs[op->n_lsp_addrs])) {
646 static struct vlog_rate_limit rl
647 = VLOG_RATE_LIMIT_INIT(1, 1);
648 VLOG_INFO_RL(&rl, "invalid syntax '%s' in logical "
649 "switch port addresses. No MAC "
651 op->nbsp->addresses[j]);
658 = xmalloc(sizeof *op->ps_addrs * nbsp->n_port_security);
659 for (size_t j = 0; j < nbsp->n_port_security; j++) {
660 if (!extract_lsp_addresses(nbsp->port_security[j],
661 &op->ps_addrs[op->n_ps_addrs])) {
662 static struct vlog_rate_limit rl
663 = VLOG_RATE_LIMIT_INIT(1, 1);
664 VLOG_INFO_RL(&rl, "invalid syntax '%s' in port "
665 "security. No MAC address found",
666 op->nbsp->port_security[j]);
675 for (size_t i = 0; i < od->nbr->n_ports; i++) {
676 const struct nbrec_logical_router_port *nbrp
679 struct lport_addresses lrp_networks;
680 if (!extract_lrp_networks(nbrp, &lrp_networks)) {
681 static struct vlog_rate_limit rl
682 = VLOG_RATE_LIMIT_INIT(5, 1);
683 VLOG_WARN_RL(&rl, "bad 'mac' %s", nbrp->mac);
687 if (!lrp_networks.n_ipv4_addrs && !lrp_networks.n_ipv6_addrs) {
691 struct ovn_port *op = ovn_port_find(ports, nbrp->name);
693 if (op->nbsp || op->nbrp) {
694 static struct vlog_rate_limit rl
695 = VLOG_RATE_LIMIT_INIT(5, 1);
696 VLOG_WARN_RL(&rl, "duplicate logical router port %s",
701 ovs_list_remove(&op->list);
702 ovs_list_push_back(both, &op->list);
704 /* This port exists but should not have been
705 * initialized fully. */
706 ovs_assert(!op->lrp_networks.n_ipv4_addrs
707 && !op->lrp_networks.n_ipv6_addrs);
709 op = ovn_port_create(ports, nbrp->name, NULL, nbrp, NULL);
710 ovs_list_push_back(nb_only, &op->list);
713 op->lrp_networks = lrp_networks;
719 /* Connect logical router ports, and logical switch ports of type "router",
722 HMAP_FOR_EACH (op, key_node, ports) {
723 if (op->nbsp && !strcmp(op->nbsp->type, "router")) {
724 const char *peer_name = smap_get(&op->nbsp->options, "router-port");
729 struct ovn_port *peer = ovn_port_find(ports, peer_name);
730 if (!peer || !peer->nbrp) {
736 op->od->router_ports = xrealloc(
737 op->od->router_ports,
738 sizeof *op->od->router_ports * (op->od->n_router_ports + 1));
739 op->od->router_ports[op->od->n_router_ports++] = op;
740 } else if (op->nbrp && op->nbrp->peer) {
741 op->peer = ovn_port_find(ports, op->nbrp->peer);
747 ovn_port_update_sbrec(const struct ovn_port *op)
749 sbrec_port_binding_set_datapath(op->sb, op->od->sb);
751 /* If the router is for l3 gateway, it resides on a chassis
752 * and its port type is "gateway". */
753 const char *chassis = smap_get(&op->od->nbr->options, "chassis");
755 sbrec_port_binding_set_type(op->sb, "gateway");
757 sbrec_port_binding_set_type(op->sb, "patch");
760 const char *peer = op->peer ? op->peer->key : "<error>";
763 smap_add(&new, "peer", peer);
765 smap_add(&new, "gateway-chassis", chassis);
767 sbrec_port_binding_set_options(op->sb, &new);
770 sbrec_port_binding_set_parent_port(op->sb, NULL);
771 sbrec_port_binding_set_tag(op->sb, NULL, 0);
772 sbrec_port_binding_set_mac(op->sb, NULL, 0);
774 if (strcmp(op->nbsp->type, "router")) {
775 sbrec_port_binding_set_type(op->sb, op->nbsp->type);
776 sbrec_port_binding_set_options(op->sb, &op->nbsp->options);
778 const char *chassis = NULL;
779 if (op->peer && op->peer->od && op->peer->od->nbr) {
780 chassis = smap_get(&op->peer->od->nbr->options, "chassis");
783 /* A switch port connected to a gateway router is also of
786 sbrec_port_binding_set_type(op->sb, "gateway");
788 sbrec_port_binding_set_type(op->sb, "patch");
791 const char *router_port = smap_get(&op->nbsp->options,
794 router_port = "<error>";
798 smap_add(&new, "peer", router_port);
800 smap_add(&new, "gateway-chassis", chassis);
802 sbrec_port_binding_set_options(op->sb, &new);
805 sbrec_port_binding_set_parent_port(op->sb, op->nbsp->parent_name);
806 sbrec_port_binding_set_tag(op->sb, op->nbsp->tag, op->nbsp->n_tag);
807 sbrec_port_binding_set_mac(op->sb, (const char **) op->nbsp->addresses,
808 op->nbsp->n_addresses);
812 /* Updates the southbound Port_Binding table so that it contains the logical
813 * switch ports specified by the northbound database.
815 * Initializes 'ports' to contain a "struct ovn_port" for every logical port,
816 * using the "struct ovn_datapath"s in 'datapaths' to look up logical
819 build_ports(struct northd_context *ctx, struct hmap *datapaths,
822 struct ovs_list sb_only, nb_only, both;
824 join_logical_ports(ctx, datapaths, ports, &sb_only, &nb_only, &both);
826 /* For logical ports that are in both databases, update the southbound
827 * record based on northbound data. Also index the in-use tunnel_keys. */
828 struct ovn_port *op, *next;
829 LIST_FOR_EACH_SAFE (op, next, list, &both) {
830 ovn_port_update_sbrec(op);
832 add_tnlid(&op->od->port_tnlids, op->sb->tunnel_key);
833 if (op->sb->tunnel_key > op->od->port_key_hint) {
834 op->od->port_key_hint = op->sb->tunnel_key;
838 /* Add southbound record for each unmatched northbound record. */
839 LIST_FOR_EACH_SAFE (op, next, list, &nb_only) {
840 uint16_t tunnel_key = ovn_port_allocate_key(op->od);
845 op->sb = sbrec_port_binding_insert(ctx->ovnsb_txn);
846 ovn_port_update_sbrec(op);
848 sbrec_port_binding_set_logical_port(op->sb, op->key);
849 sbrec_port_binding_set_tunnel_key(op->sb, tunnel_key);
852 /* Delete southbound records without northbound matches. */
853 LIST_FOR_EACH_SAFE(op, next, list, &sb_only) {
854 ovs_list_remove(&op->list);
855 sbrec_port_binding_delete(op->sb);
856 ovn_port_destroy(ports, op);
860 #define OVN_MIN_MULTICAST 32768
861 #define OVN_MAX_MULTICAST 65535
863 struct multicast_group {
865 uint16_t key; /* OVN_MIN_MULTICAST...OVN_MAX_MULTICAST. */
868 #define MC_FLOOD "_MC_flood"
869 static const struct multicast_group mc_flood = { MC_FLOOD, 65535 };
871 #define MC_UNKNOWN "_MC_unknown"
872 static const struct multicast_group mc_unknown = { MC_UNKNOWN, 65534 };
875 multicast_group_equal(const struct multicast_group *a,
876 const struct multicast_group *b)
878 return !strcmp(a->name, b->name) && a->key == b->key;
881 /* Multicast group entry. */
882 struct ovn_multicast {
883 struct hmap_node hmap_node; /* Index on 'datapath' and 'key'. */
884 struct ovn_datapath *datapath;
885 const struct multicast_group *group;
887 struct ovn_port **ports;
888 size_t n_ports, allocated_ports;
892 ovn_multicast_hash(const struct ovn_datapath *datapath,
893 const struct multicast_group *group)
895 return hash_pointer(datapath, group->key);
898 static struct ovn_multicast *
899 ovn_multicast_find(struct hmap *mcgroups, struct ovn_datapath *datapath,
900 const struct multicast_group *group)
902 struct ovn_multicast *mc;
904 HMAP_FOR_EACH_WITH_HASH (mc, hmap_node,
905 ovn_multicast_hash(datapath, group), mcgroups) {
906 if (mc->datapath == datapath
907 && multicast_group_equal(mc->group, group)) {
915 ovn_multicast_add(struct hmap *mcgroups, const struct multicast_group *group,
916 struct ovn_port *port)
918 struct ovn_datapath *od = port->od;
919 struct ovn_multicast *mc = ovn_multicast_find(mcgroups, od, group);
921 mc = xmalloc(sizeof *mc);
922 hmap_insert(mcgroups, &mc->hmap_node, ovn_multicast_hash(od, group));
926 mc->allocated_ports = 4;
927 mc->ports = xmalloc(mc->allocated_ports * sizeof *mc->ports);
929 if (mc->n_ports >= mc->allocated_ports) {
930 mc->ports = x2nrealloc(mc->ports, &mc->allocated_ports,
933 mc->ports[mc->n_ports++] = port;
937 ovn_multicast_destroy(struct hmap *mcgroups, struct ovn_multicast *mc)
940 hmap_remove(mcgroups, &mc->hmap_node);
947 ovn_multicast_update_sbrec(const struct ovn_multicast *mc,
948 const struct sbrec_multicast_group *sb)
950 struct sbrec_port_binding **ports = xmalloc(mc->n_ports * sizeof *ports);
951 for (size_t i = 0; i < mc->n_ports; i++) {
952 ports[i] = CONST_CAST(struct sbrec_port_binding *, mc->ports[i]->sb);
954 sbrec_multicast_group_set_ports(sb, ports, mc->n_ports);
958 /* Logical flow generation.
960 * This code generates the Logical_Flow table in the southbound database, as a
961 * function of most of the northbound database.
965 struct hmap_node hmap_node;
967 struct ovn_datapath *od;
968 enum ovn_stage stage;
975 ovn_lflow_hash(const struct ovn_lflow *lflow)
977 size_t hash = uuid_hash(&lflow->od->key);
978 hash = hash_2words((lflow->stage << 16) | lflow->priority, hash);
979 hash = hash_string(lflow->match, hash);
980 return hash_string(lflow->actions, hash);
984 ovn_lflow_equal(const struct ovn_lflow *a, const struct ovn_lflow *b)
986 return (a->od == b->od
987 && a->stage == b->stage
988 && a->priority == b->priority
989 && !strcmp(a->match, b->match)
990 && !strcmp(a->actions, b->actions));
994 ovn_lflow_init(struct ovn_lflow *lflow, struct ovn_datapath *od,
995 enum ovn_stage stage, uint16_t priority,
996 char *match, char *actions)
999 lflow->stage = stage;
1000 lflow->priority = priority;
1001 lflow->match = match;
1002 lflow->actions = actions;
1005 /* Adds a row with the specified contents to the Logical_Flow table. */
1007 ovn_lflow_add(struct hmap *lflow_map, struct ovn_datapath *od,
1008 enum ovn_stage stage, uint16_t priority,
1009 const char *match, const char *actions)
1011 ovs_assert(ovn_stage_to_datapath_type(stage) == ovn_datapath_get_type(od));
1013 struct ovn_lflow *lflow = xmalloc(sizeof *lflow);
1014 ovn_lflow_init(lflow, od, stage, priority,
1015 xstrdup(match), xstrdup(actions));
1016 hmap_insert(lflow_map, &lflow->hmap_node, ovn_lflow_hash(lflow));
1019 static struct ovn_lflow *
1020 ovn_lflow_find(struct hmap *lflows, struct ovn_datapath *od,
1021 enum ovn_stage stage, uint16_t priority,
1022 const char *match, const char *actions)
1024 struct ovn_lflow target;
1025 ovn_lflow_init(&target, od, stage, priority,
1026 CONST_CAST(char *, match), CONST_CAST(char *, actions));
1028 struct ovn_lflow *lflow;
1029 HMAP_FOR_EACH_WITH_HASH (lflow, hmap_node, ovn_lflow_hash(&target),
1031 if (ovn_lflow_equal(lflow, &target)) {
1039 ovn_lflow_destroy(struct hmap *lflows, struct ovn_lflow *lflow)
1042 hmap_remove(lflows, &lflow->hmap_node);
1044 free(lflow->actions);
1049 /* Appends port security constraints on L2 address field 'eth_addr_field'
1050 * (e.g. "eth.src" or "eth.dst") to 'match'. 'ps_addrs', with 'n_ps_addrs'
1051 * elements, is the collection of port_security constraints from an
1052 * OVN_NB Logical_Switch_Port row generated by extract_lsp_addresses(). */
1054 build_port_security_l2(const char *eth_addr_field,
1055 struct lport_addresses *ps_addrs,
1056 unsigned int n_ps_addrs,
1063 ds_put_format(match, " && %s == {", eth_addr_field);
1065 for (size_t i = 0; i < n_ps_addrs; i++) {
1066 ds_put_format(match, "%s ", ps_addrs[i].ea_s);
1068 ds_chomp(match, ' ');
1069 ds_put_cstr(match, "}");
1073 build_port_security_ipv6_nd_flow(
1074 struct ds *match, struct eth_addr ea, struct ipv6_netaddr *ipv6_addrs,
1077 ds_put_format(match, " && ip6 && nd && ((nd.sll == "ETH_ADDR_FMT" || "
1078 "nd.sll == "ETH_ADDR_FMT") || ((nd.tll == "ETH_ADDR_FMT" || "
1079 "nd.tll == "ETH_ADDR_FMT")", ETH_ADDR_ARGS(eth_addr_zero),
1080 ETH_ADDR_ARGS(ea), ETH_ADDR_ARGS(eth_addr_zero),
1082 if (!n_ipv6_addrs) {
1083 ds_put_cstr(match, "))");
1087 char ip6_str[INET6_ADDRSTRLEN + 1];
1088 struct in6_addr lla;
1089 in6_generate_lla(ea, &lla);
1090 memset(ip6_str, 0, sizeof(ip6_str));
1091 ipv6_string_mapped(ip6_str, &lla);
1092 ds_put_format(match, " && (nd.target == %s", ip6_str);
1094 for(int i = 0; i < n_ipv6_addrs; i++) {
1095 memset(ip6_str, 0, sizeof(ip6_str));
1096 ipv6_string_mapped(ip6_str, &ipv6_addrs[i].addr);
1097 ds_put_format(match, " || nd.target == %s", ip6_str);
1100 ds_put_format(match, ")))");
1104 build_port_security_ipv6_flow(
1105 enum ovn_pipeline pipeline, struct ds *match, struct eth_addr ea,
1106 struct ipv6_netaddr *ipv6_addrs, int n_ipv6_addrs)
1108 char ip6_str[INET6_ADDRSTRLEN + 1];
1110 ds_put_format(match, " && %s == {",
1111 pipeline == P_IN ? "ip6.src" : "ip6.dst");
1113 /* Allow link-local address. */
1114 struct in6_addr lla;
1115 in6_generate_lla(ea, &lla);
1116 ipv6_string_mapped(ip6_str, &lla);
1117 ds_put_format(match, "%s, ", ip6_str);
1119 /* Allow ip6.dst=ff00::/8 for multicast packets */
1120 if (pipeline == P_OUT) {
1121 ds_put_cstr(match, "ff00::/8, ");
1123 for(int i = 0; i < n_ipv6_addrs; i++) {
1124 ipv6_string_mapped(ip6_str, &ipv6_addrs[i].addr);
1125 ds_put_format(match, "%s, ", ip6_str);
1127 /* Replace ", " by "}". */
1128 ds_chomp(match, ' ');
1129 ds_chomp(match, ',');
1130 ds_put_cstr(match, "}");
1134 * Build port security constraints on ARP and IPv6 ND fields
1135 * and add logical flows to S_SWITCH_IN_PORT_SEC_ND stage.
1137 * For each port security of the logical port, following
1138 * logical flows are added
1139 * - If the port security has no IP (both IPv4 and IPv6) or
1140 * if it has IPv4 address(es)
1141 * - Priority 90 flow to allow ARP packets for known MAC addresses
1142 * in the eth.src and arp.spa fields. If the port security
1143 * has IPv4 addresses, allow known IPv4 addresses in the arp.tpa field.
1145 * - If the port security has no IP (both IPv4 and IPv6) or
1146 * if it has IPv6 address(es)
1147 * - Priority 90 flow to allow IPv6 ND packets for known MAC addresses
1148 * in the eth.src and nd.sll/nd.tll fields. If the port security
1149 * has IPv6 addresses, allow known IPv6 addresses in the nd.target field
1150 * for IPv6 Neighbor Advertisement packet.
1152 * - Priority 80 flow to drop ARP and IPv6 ND packets.
1155 build_port_security_nd(struct ovn_port *op, struct hmap *lflows)
1157 struct ds match = DS_EMPTY_INITIALIZER;
1159 for (size_t i = 0; i < op->n_ps_addrs; i++) {
1160 struct lport_addresses *ps = &op->ps_addrs[i];
1162 bool no_ip = !(ps->n_ipv4_addrs || ps->n_ipv6_addrs);
1165 if (ps->n_ipv4_addrs || no_ip) {
1166 ds_put_format(&match,
1167 "inport == %s && eth.src == %s && arp.sha == %s",
1168 op->json_key, ps->ea_s, ps->ea_s);
1170 if (ps->n_ipv4_addrs) {
1171 ds_put_cstr(&match, " && arp.spa == {");
1172 for (size_t j = 0; j < ps->n_ipv4_addrs; j++) {
1173 /* When the netmask is applied, if the host portion is
1174 * non-zero, the host can only use the specified
1175 * address in the arp.spa. If zero, the host is allowed
1176 * to use any address in the subnet. */
1177 if (ps->ipv4_addrs[j].plen == 32
1178 || ps->ipv4_addrs[j].addr & ~ps->ipv4_addrs[j].mask) {
1179 ds_put_cstr(&match, ps->ipv4_addrs[j].addr_s);
1181 ds_put_format(&match, "%s/%d",
1182 ps->ipv4_addrs[j].network_s,
1183 ps->ipv4_addrs[j].plen);
1185 ds_put_cstr(&match, ", ");
1187 ds_chomp(&match, ' ');
1188 ds_chomp(&match, ',');
1189 ds_put_cstr(&match, "}");
1191 ovn_lflow_add(lflows, op->od, S_SWITCH_IN_PORT_SEC_ND, 90,
1192 ds_cstr(&match), "next;");
1195 if (ps->n_ipv6_addrs || no_ip) {
1197 ds_put_format(&match, "inport == %s && eth.src == %s",
1198 op->json_key, ps->ea_s);
1199 build_port_security_ipv6_nd_flow(&match, ps->ea, ps->ipv6_addrs,
1201 ovn_lflow_add(lflows, op->od, S_SWITCH_IN_PORT_SEC_ND, 90,
1202 ds_cstr(&match), "next;");
1207 ds_put_format(&match, "inport == %s && (arp || nd)", op->json_key);
1208 ovn_lflow_add(lflows, op->od, S_SWITCH_IN_PORT_SEC_ND, 80,
1209 ds_cstr(&match), "drop;");
1214 * Build port security constraints on IPv4 and IPv6 src and dst fields
1215 * and add logical flows to S_SWITCH_(IN/OUT)_PORT_SEC_IP stage.
1217 * For each port security of the logical port, following
1218 * logical flows are added
1219 * - If the port security has IPv4 addresses,
1220 * - Priority 90 flow to allow IPv4 packets for known IPv4 addresses
1222 * - If the port security has IPv6 addresses,
1223 * - Priority 90 flow to allow IPv6 packets for known IPv6 addresses
1225 * - If the port security has IPv4 addresses or IPv6 addresses or both
1226 * - Priority 80 flow to drop all IPv4 and IPv6 traffic
1229 build_port_security_ip(enum ovn_pipeline pipeline, struct ovn_port *op,
1230 struct hmap *lflows)
1232 char *port_direction;
1233 enum ovn_stage stage;
1234 if (pipeline == P_IN) {
1235 port_direction = "inport";
1236 stage = S_SWITCH_IN_PORT_SEC_IP;
1238 port_direction = "outport";
1239 stage = S_SWITCH_OUT_PORT_SEC_IP;
1242 for (size_t i = 0; i < op->n_ps_addrs; i++) {
1243 struct lport_addresses *ps = &op->ps_addrs[i];
1245 if (!(ps->n_ipv4_addrs || ps->n_ipv6_addrs)) {
1249 if (ps->n_ipv4_addrs) {
1250 struct ds match = DS_EMPTY_INITIALIZER;
1251 if (pipeline == P_IN) {
1252 /* Permit use of the unspecified address for DHCP discovery */
1253 struct ds dhcp_match = DS_EMPTY_INITIALIZER;
1254 ds_put_format(&dhcp_match, "inport == %s"
1256 " && ip4.src == 0.0.0.0"
1257 " && ip4.dst == 255.255.255.255"
1258 " && udp.src == 68 && udp.dst == 67",
1259 op->json_key, ps->ea_s);
1260 ovn_lflow_add(lflows, op->od, stage, 90,
1261 ds_cstr(&dhcp_match), "next;");
1262 ds_destroy(&dhcp_match);
1263 ds_put_format(&match, "inport == %s && eth.src == %s"
1264 " && ip4.src == {", op->json_key,
1267 ds_put_format(&match, "outport == %s && eth.dst == %s"
1268 " && ip4.dst == {255.255.255.255, 224.0.0.0/4, ",
1269 op->json_key, ps->ea_s);
1272 for (int j = 0; j < ps->n_ipv4_addrs; j++) {
1273 ovs_be32 mask = ps->ipv4_addrs[j].mask;
1274 /* When the netmask is applied, if the host portion is
1275 * non-zero, the host can only use the specified
1276 * address. If zero, the host is allowed to use any
1277 * address in the subnet.
1279 if (ps->ipv4_addrs[j].plen == 32
1280 || ps->ipv4_addrs[j].addr & ~mask) {
1281 ds_put_format(&match, "%s", ps->ipv4_addrs[j].addr_s);
1282 if (pipeline == P_OUT && ps->ipv4_addrs[j].plen != 32) {
1283 /* Host is also allowed to receive packets to the
1284 * broadcast address in the specified subnet. */
1285 ds_put_format(&match, ", %s",
1286 ps->ipv4_addrs[j].bcast_s);
1289 /* host portion is zero */
1290 ds_put_format(&match, "%s/%d", ps->ipv4_addrs[j].network_s,
1291 ps->ipv4_addrs[j].plen);
1293 ds_put_cstr(&match, ", ");
1296 /* Replace ", " by "}". */
1297 ds_chomp(&match, ' ');
1298 ds_chomp(&match, ',');
1299 ds_put_cstr(&match, "}");
1300 ovn_lflow_add(lflows, op->od, stage, 90, ds_cstr(&match), "next;");
1304 if (ps->n_ipv6_addrs) {
1305 struct ds match = DS_EMPTY_INITIALIZER;
1306 if (pipeline == P_IN) {
1307 /* Permit use of unspecified address for duplicate address
1309 struct ds dad_match = DS_EMPTY_INITIALIZER;
1310 ds_put_format(&dad_match, "inport == %s"
1313 " && ip6.dst == ff02::/16"
1314 " && icmp6.type == {131, 135, 143}", op->json_key,
1316 ovn_lflow_add(lflows, op->od, stage, 90,
1317 ds_cstr(&dad_match), "next;");
1318 ds_destroy(&dad_match);
1320 ds_put_format(&match, "%s == %s && %s == %s",
1321 port_direction, op->json_key,
1322 pipeline == P_IN ? "eth.src" : "eth.dst", ps->ea_s);
1323 build_port_security_ipv6_flow(pipeline, &match, ps->ea,
1324 ps->ipv6_addrs, ps->n_ipv6_addrs);
1325 ovn_lflow_add(lflows, op->od, stage, 90,
1326 ds_cstr(&match), "next;");
1330 char *match = xasprintf("%s == %s && %s == %s && ip",
1331 port_direction, op->json_key,
1332 pipeline == P_IN ? "eth.src" : "eth.dst",
1334 ovn_lflow_add(lflows, op->od, stage, 80, match, "drop;");
1341 lsp_is_enabled(const struct nbrec_logical_switch_port *lsp)
1343 return !lsp->enabled || *lsp->enabled;
1347 lsp_is_up(const struct nbrec_logical_switch_port *lsp)
1349 return !lsp->up || *lsp->up;
1353 has_stateful_acl(struct ovn_datapath *od)
1355 for (size_t i = 0; i < od->nbs->n_acls; i++) {
1356 struct nbrec_acl *acl = od->nbs->acls[i];
1357 if (!strcmp(acl->action, "allow-related")) {
1366 build_pre_acls(struct ovn_datapath *od, struct hmap *lflows,
1369 bool has_stateful = has_stateful_acl(od);
1370 struct ovn_port *op;
1372 /* Ingress and Egress Pre-ACL Table (Priority 0): Packets are
1373 * allowed by default. */
1374 ovn_lflow_add(lflows, od, S_SWITCH_IN_PRE_ACL, 0, "1", "next;");
1375 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PRE_ACL, 0, "1", "next;");
1377 /* If there are any stateful ACL rules in this dapapath, we must
1378 * send all IP packets through the conntrack action, which handles
1379 * defragmentation, in order to match L4 headers. */
1381 HMAP_FOR_EACH (op, key_node, ports) {
1382 if (op->od == od && !strcmp(op->nbsp->type, "router")) {
1383 /* Can't use ct() for router ports. Consider the
1384 * following configuration: lp1(10.0.0.2) on
1385 * hostA--ls1--lr0--ls2--lp2(10.0.1.2) on hostB, For a
1386 * ping from lp1 to lp2, First, the response will go
1387 * through ct() with a zone for lp2 in the ls2 ingress
1388 * pipeline on hostB. That ct zone knows about this
1389 * connection. Next, it goes through ct() with the zone
1390 * for the router port in the egress pipeline of ls2 on
1391 * hostB. This zone does not know about the connection,
1392 * as the icmp request went through the logical router
1393 * on hostA, not hostB. This would only work with
1394 * distributed conntrack state across all chassis. */
1395 struct ds match_in = DS_EMPTY_INITIALIZER;
1396 struct ds match_out = DS_EMPTY_INITIALIZER;
1398 ds_put_format(&match_in, "ip && inport == %s", op->json_key);
1399 ds_put_format(&match_out, "ip && outport == %s", op->json_key);
1400 ovn_lflow_add(lflows, od, S_SWITCH_IN_PRE_ACL, 110,
1401 ds_cstr(&match_in), "next;");
1402 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PRE_ACL, 110,
1403 ds_cstr(&match_out), "next;");
1405 ds_destroy(&match_in);
1406 ds_destroy(&match_out);
1409 /* Ingress and Egress Pre-ACL Table (Priority 110).
1411 * Not to do conntrack on ND packets. */
1412 ovn_lflow_add(lflows, od, S_SWITCH_IN_PRE_ACL, 110, "nd", "next;");
1413 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PRE_ACL, 110, "nd", "next;");
1415 /* Ingress and Egress Pre-ACL Table (Priority 100).
1417 * Regardless of whether the ACL is "from-lport" or "to-lport",
1418 * we need rules in both the ingress and egress table, because
1419 * the return traffic needs to be followed.
1421 * 'REGBIT_CONNTRACK_DEFRAG' is set to let the pre-stateful table send
1422 * it to conntrack for tracking and defragmentation. */
1423 ovn_lflow_add(lflows, od, S_SWITCH_IN_PRE_ACL, 100, "ip",
1424 REGBIT_CONNTRACK_DEFRAG" = 1; next;");
1425 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PRE_ACL, 100, "ip",
1426 REGBIT_CONNTRACK_DEFRAG" = 1; next;");
1430 /* For a 'key' of the form "IP:port" or just "IP", sets 'port' and
1431 * 'ip_address'. The caller must free() the memory allocated for
1434 ip_address_and_port_from_lb_key(const char *key, char **ip_address,
1437 char *ip_str, *start, *next;
1441 next = start = xstrdup(key);
1442 ip_str = strsep(&next, ":");
1443 if (!ip_str || !ip_str[0]) {
1444 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1445 VLOG_WARN_RL(&rl, "bad ip address for load balancer key %s", key);
1451 char *error = ip_parse_masked(ip_str, &ip, &mask);
1452 if (error || mask != OVS_BE32_MAX) {
1453 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1454 VLOG_WARN_RL(&rl, "bad ip address for load balancer key %s", key);
1461 if (next && next[0]) {
1462 if (!str_to_int(next, 0, &l4_port) || l4_port < 0 || l4_port > 65535) {
1463 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1464 VLOG_WARN_RL(&rl, "bad ip port for load balancer key %s", key);
1471 *ip_address = strdup(ip_str);
1476 build_pre_lb(struct ovn_datapath *od, struct hmap *lflows)
1478 /* Allow all packets to go to next tables by default. */
1479 ovn_lflow_add(lflows, od, S_SWITCH_IN_PRE_LB, 0, "1", "next;");
1480 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PRE_LB, 0, "1", "next;");
1482 struct sset all_ips = SSET_INITIALIZER(&all_ips);
1483 if (od->nbs->load_balancer) {
1484 struct nbrec_load_balancer *lb = od->nbs->load_balancer;
1485 struct smap *vips = &lb->vips;
1486 struct smap_node *node;
1487 bool vip_configured = false;
1489 SMAP_FOR_EACH (node, vips) {
1490 vip_configured = true;
1492 /* node->key contains IP:port or just IP. */
1493 char *ip_address = NULL;
1495 ip_address_and_port_from_lb_key(node->key, &ip_address, &port);
1500 if (!sset_contains(&all_ips, ip_address)) {
1501 sset_add(&all_ips, ip_address);
1506 /* Ignore L4 port information in the key because fragmented packets
1507 * may not have L4 information. The pre-stateful table will send
1508 * the packet through ct() action to de-fragment. In stateful
1509 * table, we will eventually look at L4 information. */
1512 /* 'REGBIT_CONNTRACK_DEFRAG' is set to let the pre-stateful table send
1513 * packet to conntrack for defragmentation. */
1514 const char *ip_address;
1515 SSET_FOR_EACH(ip_address, &all_ips) {
1516 char *match = xasprintf("ip && ip4.dst == %s", ip_address);
1517 ovn_lflow_add(lflows, od, S_SWITCH_IN_PRE_LB,
1518 100, match, REGBIT_CONNTRACK_DEFRAG" = 1; next;");
1522 sset_destroy(&all_ips);
1524 if (vip_configured) {
1525 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PRE_LB,
1526 100, "ip", REGBIT_CONNTRACK_DEFRAG" = 1; next;");
1532 build_pre_stateful(struct ovn_datapath *od, struct hmap *lflows)
1534 /* Ingress and Egress pre-stateful Table (Priority 0): Packets are
1535 * allowed by default. */
1536 ovn_lflow_add(lflows, od, S_SWITCH_IN_PRE_STATEFUL, 0, "1", "next;");
1537 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PRE_STATEFUL, 0, "1", "next;");
1539 /* If REGBIT_CONNTRACK_DEFRAG is set as 1, then the packets should be
1540 * sent to conntrack for tracking and defragmentation. */
1541 ovn_lflow_add(lflows, od, S_SWITCH_IN_PRE_STATEFUL, 100,
1542 REGBIT_CONNTRACK_DEFRAG" == 1", "ct_next;");
1543 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PRE_STATEFUL, 100,
1544 REGBIT_CONNTRACK_DEFRAG" == 1", "ct_next;");
1548 build_acls(struct ovn_datapath *od, struct hmap *lflows)
1550 bool has_stateful = has_stateful_acl(od);
1552 /* Ingress and Egress ACL Table (Priority 0): Packets are allowed by
1553 * default. A related rule at priority 1 is added below if there
1554 * are any stateful ACLs in this datapath. */
1555 ovn_lflow_add(lflows, od, S_SWITCH_IN_ACL, 0, "1", "next;");
1556 ovn_lflow_add(lflows, od, S_SWITCH_OUT_ACL, 0, "1", "next;");
1559 /* Ingress and Egress ACL Table (Priority 1).
1561 * By default, traffic is allowed. This is partially handled by
1562 * the Priority 0 ACL flows added earlier, but we also need to
1563 * commit IP flows. This is because, while the initiater's
1564 * direction may not have any stateful rules, the server's may
1565 * and then its return traffic would not have an associated
1566 * conntrack entry and would return "+invalid".
1568 * We use "ct_commit" for a connection that is not already known
1569 * by the connection tracker. Once a connection is committed,
1570 * subsequent packets will hit the flow at priority 0 that just
1573 * We also check for established connections that have ct_label[0]
1574 * set on them. That's a connection that was disallowed, but is
1575 * now allowed by policy again since it hit this default-allow flow.
1576 * We need to set ct_label[0]=0 to let the connection continue,
1577 * which will be done by ct_commit() in the "stateful" stage.
1578 * Subsequent packets will hit the flow at priority 0 that just
1580 ovn_lflow_add(lflows, od, S_SWITCH_IN_ACL, 1,
1581 "ip && (!ct.est || (ct.est && ct_label[0] == 1))",
1582 REGBIT_CONNTRACK_COMMIT" = 1; next;");
1583 ovn_lflow_add(lflows, od, S_SWITCH_OUT_ACL, 1,
1584 "ip && (!ct.est || (ct.est && ct_label[0] == 1))",
1585 REGBIT_CONNTRACK_COMMIT" = 1; next;");
1587 /* Ingress and Egress ACL Table (Priority 65535).
1589 * Always drop traffic that's in an invalid state. Also drop
1590 * reply direction packets for connections that have been marked
1591 * for deletion (bit 0 of ct_label is set).
1593 * This is enforced at a higher priority than ACLs can be defined. */
1594 ovn_lflow_add(lflows, od, S_SWITCH_IN_ACL, UINT16_MAX,
1595 "ct.inv || (ct.est && ct.rpl && ct_label[0] == 1)",
1597 ovn_lflow_add(lflows, od, S_SWITCH_OUT_ACL, UINT16_MAX,
1598 "ct.inv || (ct.est && ct.rpl && ct_label[0] == 1)",
1601 /* Ingress and Egress ACL Table (Priority 65535).
1603 * Allow reply traffic that is part of an established
1604 * conntrack entry that has not been marked for deletion
1605 * (bit 0 of ct_label). We only match traffic in the
1606 * reply direction because we want traffic in the request
1607 * direction to hit the currently defined policy from ACLs.
1609 * This is enforced at a higher priority than ACLs can be defined. */
1610 ovn_lflow_add(lflows, od, S_SWITCH_IN_ACL, UINT16_MAX,
1611 "ct.est && !ct.rel && !ct.new && !ct.inv "
1612 "&& ct.rpl && ct_label[0] == 0",
1614 ovn_lflow_add(lflows, od, S_SWITCH_OUT_ACL, UINT16_MAX,
1615 "ct.est && !ct.rel && !ct.new && !ct.inv "
1616 "&& ct.rpl && ct_label[0] == 0",
1619 /* Ingress and Egress ACL Table (Priority 65535).
1621 * Allow traffic that is related to an existing conntrack entry that
1622 * has not been marked for deletion (bit 0 of ct_label).
1624 * This is enforced at a higher priority than ACLs can be defined.
1626 * NOTE: This does not support related data sessions (eg,
1627 * a dynamically negotiated FTP data channel), but will allow
1628 * related traffic such as an ICMP Port Unreachable through
1629 * that's generated from a non-listening UDP port. */
1630 ovn_lflow_add(lflows, od, S_SWITCH_IN_ACL, UINT16_MAX,
1631 "!ct.est && ct.rel && !ct.new && !ct.inv "
1632 "&& ct_label[0] == 0",
1634 ovn_lflow_add(lflows, od, S_SWITCH_OUT_ACL, UINT16_MAX,
1635 "!ct.est && ct.rel && !ct.new && !ct.inv "
1636 "&& ct_label[0] == 0",
1639 /* Ingress and Egress ACL Table (Priority 65535).
1641 * Not to do conntrack on ND packets. */
1642 ovn_lflow_add(lflows, od, S_SWITCH_IN_ACL, UINT16_MAX, "nd", "next;");
1643 ovn_lflow_add(lflows, od, S_SWITCH_OUT_ACL, UINT16_MAX, "nd", "next;");
1646 /* Ingress or Egress ACL Table (Various priorities). */
1647 for (size_t i = 0; i < od->nbs->n_acls; i++) {
1648 struct nbrec_acl *acl = od->nbs->acls[i];
1649 bool ingress = !strcmp(acl->direction, "from-lport") ? true :false;
1650 enum ovn_stage stage = ingress ? S_SWITCH_IN_ACL : S_SWITCH_OUT_ACL;
1652 if (!strcmp(acl->action, "allow")
1653 || !strcmp(acl->action, "allow-related")) {
1654 /* If there are any stateful flows, we must even commit "allow"
1655 * actions. This is because, while the initiater's
1656 * direction may not have any stateful rules, the server's
1657 * may and then its return traffic would not have an
1658 * associated conntrack entry and would return "+invalid". */
1659 if (!has_stateful) {
1660 ovn_lflow_add(lflows, od, stage,
1661 acl->priority + OVN_ACL_PRI_OFFSET,
1662 acl->match, "next;");
1664 struct ds match = DS_EMPTY_INITIALIZER;
1666 /* Commit the connection tracking entry if it's a new
1667 * connection that matches this ACL. After this commit,
1668 * the reply traffic is allowed by a flow we create at
1669 * priority 65535, defined earlier.
1671 * It's also possible that a known connection was marked for
1672 * deletion after a policy was deleted, but the policy was
1673 * re-added while that connection is still known. We catch
1674 * that case here and un-set ct_label[0] (which will be done
1675 * by ct_commit in the "stateful" stage) to indicate that the
1676 * connection should be allowed to resume.
1678 ds_put_format(&match, "((ct.new && !ct.est)"
1679 " || (!ct.new && ct.est && !ct.rpl "
1680 "&& ct_label[0] == 1)) "
1681 "&& (%s)", acl->match);
1682 ovn_lflow_add(lflows, od, stage,
1683 acl->priority + OVN_ACL_PRI_OFFSET,
1685 REGBIT_CONNTRACK_COMMIT" = 1; next;");
1687 /* Match on traffic in the request direction for an established
1688 * connection tracking entry that has not been marked for
1689 * deletion. There is no need to commit here, so we can just
1690 * proceed to the next table. We use this to ensure that this
1691 * connection is still allowed by the currently defined
1694 ds_put_format(&match,
1695 "!ct.new && ct.est && !ct.rpl"
1696 " && ct_label[0] == 0 && (%s)",
1698 ovn_lflow_add(lflows, od, stage,
1699 acl->priority + OVN_ACL_PRI_OFFSET,
1700 ds_cstr(&match), "next;");
1704 } else if (!strcmp(acl->action, "drop")
1705 || !strcmp(acl->action, "reject")) {
1706 struct ds match = DS_EMPTY_INITIALIZER;
1708 /* XXX Need to support "reject", treat it as "drop;" for now. */
1709 if (!strcmp(acl->action, "reject")) {
1710 VLOG_INFO("reject is not a supported action");
1713 /* The implementation of "drop" differs if stateful ACLs are in
1714 * use for this datapath. In that case, the actions differ
1715 * depending on whether the connection was previously committed
1716 * to the connection tracker with ct_commit. */
1718 /* If the packet is not part of an established connection, then
1719 * we can simply drop it. */
1720 ds_put_format(&match,
1721 "(!ct.est || (ct.est && ct_label[0] == 1)) "
1724 ovn_lflow_add(lflows, od, stage, acl->priority +
1725 OVN_ACL_PRI_OFFSET, ds_cstr(&match), "drop;");
1727 /* For an existing connection without ct_label set, we've
1728 * encountered a policy change. ACLs previously allowed
1729 * this connection and we committed the connection tracking
1730 * entry. Current policy says that we should drop this
1731 * connection. First, we set bit 0 of ct_label to indicate
1732 * that this connection is set for deletion. By not
1733 * specifying "next;", we implicitly drop the packet after
1734 * updating conntrack state. We would normally defer
1735 * ct_commit() to the "stateful" stage, but since we're
1736 * dropping the packet, we go ahead and do it here. */
1738 ds_put_format(&match,
1739 "ct.est && ct_label[0] == 0 && (%s)",
1741 ovn_lflow_add(lflows, od, stage,
1742 acl->priority + OVN_ACL_PRI_OFFSET,
1743 ds_cstr(&match), "ct_commit(ct_label=1/1);");
1747 /* There are no stateful ACLs in use on this datapath,
1748 * so a "drop" ACL is simply the "drop" logical flow action
1750 ovn_lflow_add(lflows, od, stage,
1751 acl->priority + OVN_ACL_PRI_OFFSET,
1752 acl->match, "drop;");
1759 build_lb(struct ovn_datapath *od, struct hmap *lflows)
1761 /* Ingress and Egress LB Table (Priority 0): Packets are allowed by
1763 ovn_lflow_add(lflows, od, S_SWITCH_IN_LB, 0, "1", "next;");
1764 ovn_lflow_add(lflows, od, S_SWITCH_OUT_LB, 0, "1", "next;");
1766 if (od->nbs->load_balancer) {
1767 /* Ingress and Egress LB Table (Priority 65535).
1769 * Send established traffic through conntrack for just NAT. */
1770 ovn_lflow_add(lflows, od, S_SWITCH_IN_LB, UINT16_MAX,
1771 "ct.est && !ct.rel && !ct.new && !ct.inv",
1772 REGBIT_CONNTRACK_NAT" = 1; next;");
1773 ovn_lflow_add(lflows, od, S_SWITCH_OUT_LB, UINT16_MAX,
1774 "ct.est && !ct.rel && !ct.new && !ct.inv",
1775 REGBIT_CONNTRACK_NAT" = 1; next;");
1780 build_stateful(struct ovn_datapath *od, struct hmap *lflows)
1782 /* Ingress and Egress stateful Table (Priority 0): Packets are
1783 * allowed by default. */
1784 ovn_lflow_add(lflows, od, S_SWITCH_IN_STATEFUL, 0, "1", "next;");
1785 ovn_lflow_add(lflows, od, S_SWITCH_OUT_STATEFUL, 0, "1", "next;");
1787 /* If REGBIT_CONNTRACK_COMMIT is set as 1, then the packets should be
1788 * committed to conntrack. We always set ct_label[0] to 0 here as
1789 * any packet that makes it this far is part of a connection we
1790 * want to allow to continue. */
1791 ovn_lflow_add(lflows, od, S_SWITCH_IN_STATEFUL, 100,
1792 REGBIT_CONNTRACK_COMMIT" == 1", "ct_commit(ct_label=0/1); next;");
1793 ovn_lflow_add(lflows, od, S_SWITCH_OUT_STATEFUL, 100,
1794 REGBIT_CONNTRACK_COMMIT" == 1", "ct_commit(ct_label=0/1); next;");
1796 /* If REGBIT_CONNTRACK_NAT is set as 1, then packets should just be sent
1797 * through nat (without committing).
1799 * REGBIT_CONNTRACK_COMMIT is set for new connections and
1800 * REGBIT_CONNTRACK_NAT is set for established connections. So they
1803 ovn_lflow_add(lflows, od, S_SWITCH_IN_STATEFUL, 100,
1804 REGBIT_CONNTRACK_NAT" == 1", "ct_lb;");
1805 ovn_lflow_add(lflows, od, S_SWITCH_OUT_STATEFUL, 100,
1806 REGBIT_CONNTRACK_NAT" == 1", "ct_lb;");
1808 /* Load balancing rules for new connections get committed to conntrack
1809 * table. So even if REGBIT_CONNTRACK_COMMIT is set in a previous table
1810 * a higher priority rule for load balancing below also commits the
1811 * connection, so it is okay if we do not hit the above match on
1812 * REGBIT_CONNTRACK_COMMIT. */
1813 if (od->nbs->load_balancer) {
1814 struct nbrec_load_balancer *lb = od->nbs->load_balancer;
1815 struct smap *vips = &lb->vips;
1816 struct smap_node *node;
1818 SMAP_FOR_EACH (node, vips) {
1821 /* node->key contains IP:port or just IP. */
1822 char *ip_address = NULL;
1823 ip_address_and_port_from_lb_key(node->key, &ip_address, &port);
1828 /* New connections in Ingress table. */
1829 char *action = xasprintf("ct_lb(%s);", node->value);
1830 struct ds match = DS_EMPTY_INITIALIZER;
1831 ds_put_format(&match, "ct.new && ip && ip4.dst == %s", ip_address);
1833 if (lb->protocol && !strcmp(lb->protocol, "udp")) {
1834 ds_put_format(&match, "&& udp && udp.dst == %d", port);
1836 ds_put_format(&match, "&& tcp && tcp.dst == %d", port);
1838 ovn_lflow_add(lflows, od, S_SWITCH_IN_STATEFUL,
1839 120, ds_cstr(&match), action);
1841 ovn_lflow_add(lflows, od, S_SWITCH_IN_STATEFUL,
1842 110, ds_cstr(&match), action);
1852 build_lswitch_flows(struct hmap *datapaths, struct hmap *ports,
1853 struct hmap *lflows, struct hmap *mcgroups)
1855 /* This flow table structure is documented in ovn-northd(8), so please
1856 * update ovn-northd.8.xml if you change anything. */
1858 struct ds match = DS_EMPTY_INITIALIZER;
1859 struct ds actions = DS_EMPTY_INITIALIZER;
1861 /* Build pre-ACL and ACL tables for both ingress and egress.
1862 * Ingress tables 3 and 4. Egress tables 0 and 1. */
1863 struct ovn_datapath *od;
1864 HMAP_FOR_EACH (od, key_node, datapaths) {
1869 build_pre_acls(od, lflows, ports);
1870 build_pre_lb(od, lflows);
1871 build_pre_stateful(od, lflows);
1872 build_acls(od, lflows);
1873 build_lb(od, lflows);
1874 build_stateful(od, lflows);
1877 /* Logical switch ingress table 0: Admission control framework (priority
1879 HMAP_FOR_EACH (od, key_node, datapaths) {
1884 /* Logical VLANs not supported. */
1885 ovn_lflow_add(lflows, od, S_SWITCH_IN_PORT_SEC_L2, 100, "vlan.present",
1888 /* Broadcast/multicast source address is invalid. */
1889 ovn_lflow_add(lflows, od, S_SWITCH_IN_PORT_SEC_L2, 100, "eth.src[40]",
1892 /* Port security flows have priority 50 (see below) and will continue
1893 * to the next table if packet source is acceptable. */
1896 /* Logical switch ingress table 0: Ingress port security - L2
1898 * Ingress table 1: Ingress port security - IP (priority 90 and 80)
1899 * Ingress table 2: Ingress port security - ND (priority 90 and 80)
1901 struct ovn_port *op;
1902 HMAP_FOR_EACH (op, key_node, ports) {
1907 if (!lsp_is_enabled(op->nbsp)) {
1908 /* Drop packets from disabled logical ports (since logical flow
1909 * tables are default-drop). */
1914 ds_put_format(&match, "inport == %s", op->json_key);
1915 build_port_security_l2("eth.src", op->ps_addrs, op->n_ps_addrs,
1917 ovn_lflow_add(lflows, op->od, S_SWITCH_IN_PORT_SEC_L2, 50,
1918 ds_cstr(&match), "next;");
1920 if (op->nbsp->n_port_security) {
1921 build_port_security_ip(P_IN, op, lflows);
1922 build_port_security_nd(op, lflows);
1926 /* Ingress table 1 and 2: Port security - IP and ND, by default goto next.
1928 HMAP_FOR_EACH (od, key_node, datapaths) {
1933 ovn_lflow_add(lflows, od, S_SWITCH_IN_PORT_SEC_ND, 0, "1", "next;");
1934 ovn_lflow_add(lflows, od, S_SWITCH_IN_PORT_SEC_IP, 0, "1", "next;");
1937 /* Ingress table 9: ARP responder, skip requests coming from localnet ports.
1938 * (priority 100). */
1939 HMAP_FOR_EACH (op, key_node, ports) {
1944 if (!strcmp(op->nbsp->type, "localnet")) {
1946 ds_put_format(&match, "inport == %s", op->json_key);
1947 ovn_lflow_add(lflows, op->od, S_SWITCH_IN_ARP_ND_RSP, 100,
1948 ds_cstr(&match), "next;");
1952 /* Ingress table 9: ARP/ND responder, reply for known IPs.
1954 HMAP_FOR_EACH (op, key_node, ports) {
1960 * Add ARP/ND reply flows if either the
1962 * - port type is router
1964 if (!lsp_is_up(op->nbsp) && strcmp(op->nbsp->type, "router")) {
1968 for (size_t i = 0; i < op->n_lsp_addrs; i++) {
1969 for (size_t j = 0; j < op->lsp_addrs[i].n_ipv4_addrs; j++) {
1971 ds_put_format(&match, "arp.tpa == %s && arp.op == 1",
1972 op->lsp_addrs[i].ipv4_addrs[j].addr_s);
1974 ds_put_format(&actions,
1975 "eth.dst = eth.src; "
1977 "arp.op = 2; /* ARP reply */ "
1978 "arp.tha = arp.sha; "
1980 "arp.tpa = arp.spa; "
1982 "outport = inport; "
1983 "inport = \"\"; /* Allow sending out inport. */ "
1985 op->lsp_addrs[i].ea_s, op->lsp_addrs[i].ea_s,
1986 op->lsp_addrs[i].ipv4_addrs[j].addr_s);
1987 ovn_lflow_add(lflows, op->od, S_SWITCH_IN_ARP_ND_RSP, 50,
1988 ds_cstr(&match), ds_cstr(&actions));
1991 if (op->lsp_addrs[i].n_ipv6_addrs > 0) {
1993 ds_put_cstr(&match, "icmp6 && icmp6.type == 135 && ");
1994 if (op->lsp_addrs[i].n_ipv6_addrs == 1) {
1995 ds_put_format(&match, "nd.target == %s",
1996 op->lsp_addrs[i].ipv6_addrs[0].addr_s);
1998 ds_put_format(&match, "nd.target == {");
1999 for (size_t j = 0; j < op->lsp_addrs[i].n_ipv6_addrs; j++) {
2001 op->lsp_addrs[i].ipv6_addrs[j].addr_s);
2003 ds_chomp(&match, ' ');
2004 ds_chomp(&match, ',');
2005 ds_put_cstr(&match, "}");
2008 ds_put_format(&actions,
2009 "na { eth.src = %s; "
2011 "outport = inport; "
2012 "inport = \"\"; /* Allow sending out inport. */ "
2014 op->lsp_addrs[i].ea_s,
2015 op->lsp_addrs[i].ea_s);
2017 ovn_lflow_add(lflows, op->od, S_SWITCH_IN_ARP_ND_RSP, 50,
2018 ds_cstr(&match), ds_cstr(&actions));
2024 /* Ingress table 9: ARP/ND responder, by default goto next.
2026 HMAP_FOR_EACH (od, key_node, datapaths) {
2031 ovn_lflow_add(lflows, od, S_SWITCH_IN_ARP_ND_RSP, 0, "1", "next;");
2034 /* Ingress table 10: Destination lookup, broadcast and multicast handling
2035 * (priority 100). */
2036 HMAP_FOR_EACH (op, key_node, ports) {
2041 if (lsp_is_enabled(op->nbsp)) {
2042 ovn_multicast_add(mcgroups, &mc_flood, op);
2045 HMAP_FOR_EACH (od, key_node, datapaths) {
2050 ovn_lflow_add(lflows, od, S_SWITCH_IN_L2_LKUP, 100, "eth.mcast",
2051 "outport = \""MC_FLOOD"\"; output;");
2054 /* Ingress table 10: Destination lookup, unicast handling (priority 50), */
2055 HMAP_FOR_EACH (op, key_node, ports) {
2060 for (size_t i = 0; i < op->nbsp->n_addresses; i++) {
2061 struct eth_addr mac;
2063 if (eth_addr_from_string(op->nbsp->addresses[i], &mac)) {
2065 ds_put_format(&match, "eth.dst == "ETH_ADDR_FMT,
2066 ETH_ADDR_ARGS(mac));
2069 ds_put_format(&actions, "outport = %s; output;", op->json_key);
2070 ovn_lflow_add(lflows, op->od, S_SWITCH_IN_L2_LKUP, 50,
2071 ds_cstr(&match), ds_cstr(&actions));
2072 } else if (!strcmp(op->nbsp->addresses[i], "unknown")) {
2073 if (lsp_is_enabled(op->nbsp)) {
2074 ovn_multicast_add(mcgroups, &mc_unknown, op);
2075 op->od->has_unknown = true;
2078 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2081 "%s: invalid syntax '%s' in addresses column",
2082 op->nbsp->name, op->nbsp->addresses[i]);
2087 /* Ingress table 10: Destination lookup for unknown MACs (priority 0). */
2088 HMAP_FOR_EACH (od, key_node, datapaths) {
2093 if (od->has_unknown) {
2094 ovn_lflow_add(lflows, od, S_SWITCH_IN_L2_LKUP, 0, "1",
2095 "outport = \""MC_UNKNOWN"\"; output;");
2099 /* Egress tables 6: Egress port security - IP (priority 0)
2100 * Egress table 7: Egress port security L2 - multicast/broadcast
2101 * (priority 100). */
2102 HMAP_FOR_EACH (od, key_node, datapaths) {
2107 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PORT_SEC_IP, 0, "1", "next;");
2108 ovn_lflow_add(lflows, od, S_SWITCH_OUT_PORT_SEC_L2, 100, "eth.mcast",
2112 /* Egress table 6: Egress port security - IP (priorities 90 and 80)
2113 * if port security enabled.
2115 * Egress table 7: Egress port security - L2 (priorities 50 and 150).
2117 * Priority 50 rules implement port security for enabled logical port.
2119 * Priority 150 rules drop packets to disabled logical ports, so that they
2120 * don't even receive multicast or broadcast packets. */
2121 HMAP_FOR_EACH (op, key_node, ports) {
2127 ds_put_format(&match, "outport == %s", op->json_key);
2128 if (lsp_is_enabled(op->nbsp)) {
2129 build_port_security_l2("eth.dst", op->ps_addrs, op->n_ps_addrs,
2131 ovn_lflow_add(lflows, op->od, S_SWITCH_OUT_PORT_SEC_L2, 50,
2132 ds_cstr(&match), "output;");
2134 ovn_lflow_add(lflows, op->od, S_SWITCH_OUT_PORT_SEC_L2, 150,
2135 ds_cstr(&match), "drop;");
2138 if (op->nbsp->n_port_security) {
2139 build_port_security_ip(P_OUT, op, lflows);
2144 ds_destroy(&actions);
2148 lrport_is_enabled(const struct nbrec_logical_router_port *lrport)
2150 return !lrport->enabled || *lrport->enabled;
2153 /* Returns a string of the IP address of the router port 'op' that
2154 * overlaps with 'ip_s". If one is not found, returns NULL.
2156 * The caller must not free the returned string. */
2158 find_lrp_member_ip(const struct ovn_port *op, const char *ip_s)
2162 if (!ip_parse(ip_s, &ip)) {
2163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
2164 VLOG_WARN_RL(&rl, "bad ip address %s", ip_s);
2168 for (int i = 0; i < op->lrp_networks.n_ipv4_addrs; i++) {
2169 const struct ipv4_netaddr *na = &op->lrp_networks.ipv4_addrs[i];
2171 if (!((na->network ^ ip) & na->mask)) {
2172 /* There should be only 1 interface that matches the
2173 * next hop. Otherwise, it's a configuration error,
2174 * because subnets of router's interfaces should NOT
2184 add_route(struct hmap *lflows, const struct ovn_port *op,
2185 const char *lrp_addr_s, const char *network_s, int plen,
2186 const char *gateway)
2188 char *match = xasprintf("ip4.dst == %s/%d", network_s, plen);
2190 struct ds actions = DS_EMPTY_INITIALIZER;
2191 ds_put_cstr(&actions, "ip.ttl--; reg0 = ");
2193 ds_put_cstr(&actions, gateway);
2195 ds_put_cstr(&actions, "ip4.dst");
2197 ds_put_format(&actions, "; "
2201 "inport = \"\"; /* Allow sending out inport. */ "
2204 op->lrp_networks.ea_s,
2207 /* The priority here is calculated to implement longest-prefix-match
2209 ovn_lflow_add(lflows, op->od, S_ROUTER_IN_IP_ROUTING, plen, match,
2211 ds_destroy(&actions);
2216 build_static_route_flow(struct hmap *lflows, struct ovn_datapath *od,
2218 const struct nbrec_logical_router_static_route *route)
2220 ovs_be32 prefix, nexthop, mask;
2221 const char *lrp_addr_s;
2223 /* Verify that next hop is an IP address with 32 bits mask. */
2224 char *error = ip_parse_masked(route->nexthop, &nexthop, &mask);
2225 if (error || mask != OVS_BE32_MAX) {
2226 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
2227 VLOG_WARN_RL(&rl, "bad next hop ip address %s", route->nexthop);
2232 /* Verify that ip prefix is a valid CIDR address. */
2233 error = ip_parse_masked(route->ip_prefix, &prefix, &mask);
2234 if (error || !ip_is_cidr(mask)) {
2235 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
2236 VLOG_WARN_RL(&rl, "bad 'ip_prefix' in static routes %s",
2242 /* Find the outgoing port. */
2243 struct ovn_port *out_port = NULL;
2244 if (route->output_port) {
2245 out_port = ovn_port_find(ports, route->output_port);
2247 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
2248 VLOG_WARN_RL(&rl, "Bad out port %s for static route %s",
2249 route->output_port, route->ip_prefix);
2252 lrp_addr_s = find_lrp_member_ip(out_port, route->nexthop);
2254 /* output_port is not specified, find the
2255 * router port matching the next hop. */
2257 for (i = 0; i < od->nbr->n_ports; i++) {
2258 struct nbrec_logical_router_port *lrp = od->nbr->ports[i];
2259 out_port = ovn_port_find(ports, lrp->name);
2261 /* This should not happen. */
2265 lrp_addr_s = find_lrp_member_ip(out_port, route->nexthop);
2273 /* There is no matched out port. */
2274 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
2275 VLOG_WARN_RL(&rl, "No path for static route %s; next hop %s",
2276 route->ip_prefix, route->nexthop);
2280 char *prefix_s = xasprintf(IP_FMT, IP_ARGS(prefix & mask));
2281 add_route(lflows, out_port, lrp_addr_s, prefix_s,
2282 ip_count_cidr_bits(mask), route->nexthop);
2287 op_put_networks(struct ds *ds, const struct ovn_port *op, bool add_bcast)
2289 if (!add_bcast && op->lrp_networks.n_ipv4_addrs == 1) {
2290 ds_put_format(ds, "%s", op->lrp_networks.ipv4_addrs[0].addr_s);
2294 ds_put_cstr(ds, "{");
2295 for (int i = 0; i < op->lrp_networks.n_ipv4_addrs; i++) {
2296 ds_put_format(ds, "%s, ", op->lrp_networks.ipv4_addrs[i].addr_s);
2298 ds_put_format(ds, "%s, ", op->lrp_networks.ipv4_addrs[i].bcast_s);
2303 ds_put_cstr(ds, "}");
2307 build_lrouter_flows(struct hmap *datapaths, struct hmap *ports,
2308 struct hmap *lflows)
2310 /* This flow table structure is documented in ovn-northd(8), so please
2311 * update ovn-northd.8.xml if you change anything. */
2313 struct ds match = DS_EMPTY_INITIALIZER;
2314 struct ds actions = DS_EMPTY_INITIALIZER;
2316 /* Logical router ingress table 0: Admission control framework. */
2317 struct ovn_datapath *od;
2318 HMAP_FOR_EACH (od, key_node, datapaths) {
2323 /* Logical VLANs not supported.
2324 * Broadcast/multicast source address is invalid. */
2325 ovn_lflow_add(lflows, od, S_ROUTER_IN_ADMISSION, 100,
2326 "vlan.present || eth.src[40]", "drop;");
2329 /* Logical router ingress table 0: match (priority 50). */
2330 struct ovn_port *op;
2331 HMAP_FOR_EACH (op, key_node, ports) {
2336 if (!lrport_is_enabled(op->nbrp)) {
2337 /* Drop packets from disabled logical ports (since logical flow
2338 * tables are default-drop). */
2343 ds_put_format(&match, "(eth.mcast || eth.dst == %s) && inport == %s",
2344 op->lrp_networks.ea_s, op->json_key);
2345 ovn_lflow_add(lflows, op->od, S_ROUTER_IN_ADMISSION, 50,
2346 ds_cstr(&match), "next;");
2349 /* Logical router ingress table 1: IP Input. */
2350 HMAP_FOR_EACH (od, key_node, datapaths) {
2355 /* L3 admission control: drop multicast and broadcast source, localhost
2356 * source or destination, and zero network source or destination
2357 * (priority 100). */
2358 ovn_lflow_add(lflows, od, S_ROUTER_IN_IP_INPUT, 100,
2360 "ip4.src == 255.255.255.255 || "
2361 "ip4.src == 127.0.0.0/8 || "
2362 "ip4.dst == 127.0.0.0/8 || "
2363 "ip4.src == 0.0.0.0/8 || "
2364 "ip4.dst == 0.0.0.0/8",
2367 /* ARP reply handling. Use ARP replies to populate the logical
2368 * router's ARP table. */
2369 ovn_lflow_add(lflows, od, S_ROUTER_IN_IP_INPUT, 90, "arp.op == 2",
2370 "put_arp(inport, arp.spa, arp.sha);");
2372 /* Drop Ethernet local broadcast. By definition this traffic should
2373 * not be forwarded.*/
2374 ovn_lflow_add(lflows, od, S_ROUTER_IN_IP_INPUT, 50,
2375 "eth.bcast", "drop;");
2379 * XXX Need to send ICMP time exceeded if !ip.later_frag. */
2381 ds_put_cstr(&match, "ip4 && ip.ttl == {0, 1}");
2382 ovn_lflow_add(lflows, od, S_ROUTER_IN_IP_INPUT, 30,
2383 ds_cstr(&match), "drop;");
2385 /* Pass other traffic not already handled to the next table for
2387 ovn_lflow_add(lflows, od, S_ROUTER_IN_IP_INPUT, 0, "1", "next;");
2390 HMAP_FOR_EACH (op, key_node, ports) {
2395 /* L3 admission control: drop packets that originate from an IP address
2396 * owned by the router or a broadcast address known to the router
2397 * (priority 100). */
2399 ds_put_cstr(&match, "ip4.src == ");
2400 op_put_networks(&match, op, true);
2401 ovn_lflow_add(lflows, op->od, S_ROUTER_IN_IP_INPUT, 100,
2402 ds_cstr(&match), "drop;");
2404 /* ICMP echo reply. These flows reply to ICMP echo requests
2405 * received for the router's IP address. Since packets only
2406 * get here as part of the logical router datapath, the inport
2407 * (i.e. the incoming locally attached net) does not matter.
2408 * The ip.ttl also does not matter (RFC1812 section 4.2.2.9) */
2410 ds_put_cstr(&match, "ip4.dst == ");
2411 op_put_networks(&match, op, false);
2412 ds_put_cstr(&match, " && icmp4.type == 8 && icmp4.code == 0");
2415 ds_put_format(&actions,
2416 "ip4.dst <-> ip4.src; "
2419 "inport = \"\"; /* Allow sending out inport. */ "
2421 ovn_lflow_add(lflows, op->od, S_ROUTER_IN_IP_INPUT, 90,
2422 ds_cstr(&match), ds_cstr(&actions));
2424 /* ARP reply. These flows reply to ARP requests for the router's own
2426 for (int i = 0; i < op->lrp_networks.n_ipv4_addrs; i++) {
2428 ds_put_format(&match,
2429 "inport == %s && arp.tpa == %s && arp.op == 1",
2430 op->json_key, op->lrp_networks.ipv4_addrs[i].addr_s);
2433 ds_put_format(&actions,
2434 "eth.dst = eth.src; "
2436 "arp.op = 2; /* ARP reply */ "
2437 "arp.tha = arp.sha; "
2439 "arp.tpa = arp.spa; "
2442 "inport = \"\"; /* Allow sending out inport. */ "
2444 op->lrp_networks.ea_s,
2445 op->lrp_networks.ea_s,
2446 op->lrp_networks.ipv4_addrs[i].addr_s,
2448 ovn_lflow_add(lflows, op->od, S_ROUTER_IN_IP_INPUT, 90,
2449 ds_cstr(&match), ds_cstr(&actions));
2452 /* ARP handling for external IP addresses.
2454 * DNAT IP addresses are external IP addresses that need ARP
2456 for (int i = 0; i < op->od->nbr->n_nat; i++) {
2457 const struct nbrec_nat *nat;
2459 nat = op->od->nbr->nat[i];
2461 if(!strcmp(nat->type, "snat")) {
2466 if (!ip_parse(nat->external_ip, &ip) || !ip) {
2467 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
2468 VLOG_WARN_RL(&rl, "bad ip address %s in dnat configuration "
2469 "for router %s", nat->external_ip, op->key);
2474 ds_put_format(&match,
2475 "inport == %s && arp.tpa == "IP_FMT" && arp.op == 1",
2476 op->json_key, IP_ARGS(ip));
2479 ds_put_format(&actions,
2480 "eth.dst = eth.src; "
2482 "arp.op = 2; /* ARP reply */ "
2483 "arp.tha = arp.sha; "
2485 "arp.tpa = arp.spa; "
2486 "arp.spa = "IP_FMT"; "
2488 "inport = \"\"; /* Allow sending out inport. */ "
2490 op->lrp_networks.ea_s,
2491 op->lrp_networks.ea_s,
2494 ovn_lflow_add(lflows, op->od, S_ROUTER_IN_IP_INPUT, 90,
2495 ds_cstr(&match), ds_cstr(&actions));
2498 /* Drop IP traffic to this router, unless the router ip is used as
2500 ovs_be32 *nat_ips = xmalloc(sizeof *nat_ips * op->od->nbr->n_nat);
2501 size_t n_nat_ips = 0;
2502 for (int i = 0; i < op->od->nbr->n_nat; i++) {
2503 const struct nbrec_nat *nat;
2506 nat = op->od->nbr->nat[i];
2507 if (strcmp(nat->type, "snat")) {
2511 if (!ip_parse(nat->external_ip, &ip) || !ip) {
2512 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
2513 VLOG_WARN_RL(&rl, "bad ip address %s in snat configuration "
2514 "for router %s", nat->external_ip, op->key);
2518 nat_ips[n_nat_ips++] = ip;
2522 ds_put_cstr(&match, "ip4.dst == {");
2523 bool has_drop_ips = false;
2524 for (int i = 0; i < op->lrp_networks.n_ipv4_addrs; i++) {
2525 for (int j = 0; j < n_nat_ips; j++) {
2526 if (op->lrp_networks.ipv4_addrs[i].addr == nat_ips[j]) {
2530 ds_put_format(&match, "%s, ",
2531 op->lrp_networks.ipv4_addrs[i].addr_s);
2532 has_drop_ips = true;
2534 ds_chomp(&match, ' ');
2535 ds_chomp(&match, ',');
2536 ds_put_cstr(&match, "}");
2539 /* Drop IP traffic to this router. */
2540 ovn_lflow_add(lflows, op->od, S_ROUTER_IN_IP_INPUT, 60,
2541 ds_cstr(&match), "drop;");
2547 /* NAT in Gateway routers. */
2548 HMAP_FOR_EACH (od, key_node, datapaths) {
2553 /* Packets are allowed by default. */
2554 ovn_lflow_add(lflows, od, S_ROUTER_IN_UNSNAT, 0, "1", "next;");
2555 ovn_lflow_add(lflows, od, S_ROUTER_OUT_SNAT, 0, "1", "next;");
2556 ovn_lflow_add(lflows, od, S_ROUTER_IN_DNAT, 0, "1", "next;");
2558 /* NAT rules are only valid on Gateway routers. */
2559 if (!smap_get(&od->nbr->options, "chassis")) {
2563 for (int i = 0; i < od->nbr->n_nat; i++) {
2564 const struct nbrec_nat *nat;
2566 nat = od->nbr->nat[i];
2570 char *error = ip_parse_masked(nat->external_ip, &ip, &mask);
2571 if (error || mask != OVS_BE32_MAX) {
2572 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
2573 VLOG_WARN_RL(&rl, "bad external ip %s for nat",
2579 /* Check the validity of nat->logical_ip. 'logical_ip' can
2580 * be a subnet when the type is "snat". */
2581 error = ip_parse_masked(nat->logical_ip, &ip, &mask);
2582 if (!strcmp(nat->type, "snat")) {
2584 static struct vlog_rate_limit rl =
2585 VLOG_RATE_LIMIT_INIT(5, 1);
2586 VLOG_WARN_RL(&rl, "bad ip network or ip %s for snat "
2587 "in router "UUID_FMT"",
2588 nat->logical_ip, UUID_ARGS(&od->key));
2593 if (error || mask != OVS_BE32_MAX) {
2594 static struct vlog_rate_limit rl =
2595 VLOG_RATE_LIMIT_INIT(5, 1);
2596 VLOG_WARN_RL(&rl, "bad ip %s for dnat in router "
2597 ""UUID_FMT"", nat->logical_ip, UUID_ARGS(&od->key));
2603 /* Ingress UNSNAT table: It is for already established connections'
2604 * reverse traffic. i.e., SNAT has already been done in egress
2605 * pipeline and now the packet has entered the ingress pipeline as
2606 * part of a reply. We undo the SNAT here.
2608 * Undoing SNAT has to happen before DNAT processing. This is
2609 * because when the packet was DNATed in ingress pipeline, it did
2610 * not know about the possibility of eventual additional SNAT in
2611 * egress pipeline. */
2612 if (!strcmp(nat->type, "snat")
2613 || !strcmp(nat->type, "dnat_and_snat")) {
2615 ds_put_format(&match, "ip && ip4.dst == %s", nat->external_ip);
2616 ovn_lflow_add(lflows, od, S_ROUTER_IN_UNSNAT, 100,
2617 ds_cstr(&match), "ct_snat; next;");
2620 /* Ingress DNAT table: Packets enter the pipeline with destination
2621 * IP address that needs to be DNATted from a external IP address
2622 * to a logical IP address. */
2623 if (!strcmp(nat->type, "dnat")
2624 || !strcmp(nat->type, "dnat_and_snat")) {
2625 /* Packet when it goes from the initiator to destination.
2626 * We need to zero the inport because the router can
2627 * send the packet back through the same interface. */
2629 ds_put_format(&match, "ip && ip4.dst == %s", nat->external_ip);
2631 ds_put_format(&actions,"inport = \"\"; ct_dnat(%s);",
2633 ovn_lflow_add(lflows, od, S_ROUTER_IN_DNAT, 100,
2634 ds_cstr(&match), ds_cstr(&actions));
2637 /* Egress SNAT table: Packets enter the egress pipeline with
2638 * source ip address that needs to be SNATted to a external ip
2640 if (!strcmp(nat->type, "snat")
2641 || !strcmp(nat->type, "dnat_and_snat")) {
2643 ds_put_format(&match, "ip && ip4.src == %s", nat->logical_ip);
2645 ds_put_format(&actions, "ct_snat(%s);", nat->external_ip);
2647 /* The priority here is calculated such that the
2648 * nat->logical_ip with the longest mask gets a higher
2650 ovn_lflow_add(lflows, od, S_ROUTER_OUT_SNAT,
2651 count_1bits(ntohl(mask)) + 1,
2652 ds_cstr(&match), ds_cstr(&actions));
2656 /* Re-circulate every packet through the DNAT zone.
2657 * This helps with two things.
2659 * 1. Any packet that needs to be unDNATed in the reverse
2660 * direction gets unDNATed. Ideally this could be done in
2661 * the egress pipeline. But since the gateway router
2662 * does not have any feature that depends on the source
2663 * ip address being external IP address for IP routing,
2664 * we can do it here, saving a future re-circulation.
2666 * 2. Any packet that was sent through SNAT zone in the
2667 * previous table automatically gets re-circulated to get
2668 * back the new destination IP address that is needed for
2669 * routing in the openflow pipeline. */
2670 ovn_lflow_add(lflows, od, S_ROUTER_IN_DNAT, 50,
2671 "ip", "inport = \"\"; ct_dnat;");
2674 /* Logical router ingress table 4: IP Routing.
2676 * A packet that arrives at this table is an IP packet that should be
2677 * routed to the address in ip4.dst. This table sets outport to the correct
2678 * output port, eth.src to the output port's MAC address, and reg0 to the
2679 * next-hop IP address (leaving ip4.dst, the packet’s final destination,
2680 * unchanged), and advances to the next table for ARP resolution. */
2681 HMAP_FOR_EACH (op, key_node, ports) {
2686 for (int i = 0; i < op->lrp_networks.n_ipv4_addrs; i++) {
2687 add_route(lflows, op, op->lrp_networks.ipv4_addrs[i].addr_s,
2688 op->lrp_networks.ipv4_addrs[i].network_s,
2689 op->lrp_networks.ipv4_addrs[i].plen, NULL);
2693 HMAP_FOR_EACH (od, key_node, datapaths) {
2698 /* Convert the static routes to flows. */
2699 for (int i = 0; i < od->nbr->n_static_routes; i++) {
2700 const struct nbrec_logical_router_static_route *route;
2702 route = od->nbr->static_routes[i];
2703 build_static_route_flow(lflows, od, ports, route);
2706 /* XXX destination unreachable */
2708 /* Local router ingress table 5: ARP Resolution.
2710 * Any packet that reaches this table is an IP packet whose next-hop IP
2711 * address is in reg0. (ip4.dst is the final destination.) This table
2712 * resolves the IP address in reg0 into an output port in outport and an
2713 * Ethernet address in eth.dst. */
2714 HMAP_FOR_EACH (op, key_node, ports) {
2716 /* This is a logical router port. If next-hop IP address in 'reg0'
2717 * matches ip address of this router port, then the packet is
2718 * intended to eventually be sent to this logical port. Set the
2719 * destination mac address using this port's mac address.
2721 * The packet is still in peer's logical pipeline. So the match
2722 * should be on peer's outport. */
2723 if (op->nbrp->peer) {
2724 struct ovn_port *peer = ovn_port_find(ports, op->nbrp->peer);
2730 ds_put_format(&match, "outport == %s && reg0 == ",
2732 op_put_networks(&match, op, false);
2735 ds_put_format(&actions, "eth.dst = %s; next;",
2736 op->lrp_networks.ea_s);
2737 ovn_lflow_add(lflows, peer->od, S_ROUTER_IN_ARP_RESOLVE,
2738 100, ds_cstr(&match), ds_cstr(&actions));
2740 } else if (op->od->n_router_ports && strcmp(op->nbsp->type, "router")) {
2741 /* This is a logical switch port that backs a VM or a container.
2742 * Extract its addresses. For each of the address, go through all
2743 * the router ports attached to the switch (to which this port
2744 * connects) and if the address in question is reachable from the
2745 * router port, add an ARP entry in that router's pipeline. */
2747 for (size_t i = 0; i < op->n_lsp_addrs; i++) {
2748 const char *ea_s = op->lsp_addrs[i].ea_s;
2749 for (size_t j = 0; j < op->lsp_addrs[i].n_ipv4_addrs; j++) {
2750 const char *ip_s = op->lsp_addrs[i].ipv4_addrs[j].addr_s;
2751 for (size_t k = 0; k < op->od->n_router_ports; k++) {
2752 /* Get the Logical_Router_Port that the
2753 * Logical_Switch_Port is connected to, as
2755 const char *peer_name = smap_get(
2756 &op->od->router_ports[k]->nbsp->options,
2762 struct ovn_port *peer = ovn_port_find(ports, peer_name);
2763 if (!peer || !peer->nbrp) {
2767 if (!find_lrp_member_ip(peer, ip_s)) {
2772 ds_put_format(&match, "outport == %s && reg0 == %s",
2773 peer->json_key, ip_s);
2776 ds_put_format(&actions, "eth.dst = %s; next;", ea_s);
2777 ovn_lflow_add(lflows, peer->od,
2778 S_ROUTER_IN_ARP_RESOLVE, 100,
2779 ds_cstr(&match), ds_cstr(&actions));
2783 } else if (!strcmp(op->nbsp->type, "router")) {
2784 /* This is a logical switch port that connects to a router. */
2786 /* The peer of this switch port is the router port for which
2787 * we need to add logical flows such that it can resolve
2788 * ARP entries for all the other router ports connected to
2789 * the switch in question. */
2791 const char *peer_name = smap_get(&op->nbsp->options,
2797 struct ovn_port *peer = ovn_port_find(ports, peer_name);
2798 if (!peer || !peer->nbrp) {
2802 for (size_t i = 0; i < op->od->n_router_ports; i++) {
2803 const char *router_port_name = smap_get(
2804 &op->od->router_ports[i]->nbsp->options,
2806 struct ovn_port *router_port = ovn_port_find(ports,
2808 if (!router_port || !router_port->nbrp) {
2812 /* Skip the router port under consideration. */
2813 if (router_port == peer) {
2818 ds_put_format(&match, "outport == %s && reg0 == ",
2820 op_put_networks(&match, router_port, false);
2823 ds_put_format(&actions, "eth.dst = %s; next;",
2824 router_port->lrp_networks.ea_s);
2825 ovn_lflow_add(lflows, peer->od, S_ROUTER_IN_ARP_RESOLVE,
2826 100, ds_cstr(&match), ds_cstr(&actions));
2831 HMAP_FOR_EACH (od, key_node, datapaths) {
2836 ovn_lflow_add(lflows, od, S_ROUTER_IN_ARP_RESOLVE, 0, "1",
2837 "get_arp(outport, reg0); next;");
2840 /* Local router ingress table 6: ARP request.
2842 * In the common case where the Ethernet destination has been resolved,
2843 * this table outputs the packet (priority 0). Otherwise, it composes
2844 * and sends an ARP request (priority 100). */
2845 HMAP_FOR_EACH (od, key_node, datapaths) {
2850 ovn_lflow_add(lflows, od, S_ROUTER_IN_ARP_REQUEST, 100,
2851 "eth.dst == 00:00:00:00:00:00",
2853 "eth.dst = ff:ff:ff:ff:ff:ff; "
2855 "arp.op = 1; " /* ARP request */
2858 ovn_lflow_add(lflows, od, S_ROUTER_IN_ARP_REQUEST, 0, "1", "output;");
2861 /* Logical router egress table 1: Delivery (priority 100).
2863 * Priority 100 rules deliver packets to enabled logical ports. */
2864 HMAP_FOR_EACH (op, key_node, ports) {
2869 if (!lrport_is_enabled(op->nbrp)) {
2870 /* Drop packets to disabled logical ports (since logical flow
2871 * tables are default-drop). */
2876 ds_put_format(&match, "outport == %s", op->json_key);
2877 ovn_lflow_add(lflows, op->od, S_ROUTER_OUT_DELIVERY, 100,
2878 ds_cstr(&match), "output;");
2882 ds_destroy(&actions);
2885 /* Updates the Logical_Flow and Multicast_Group tables in the OVN_SB database,
2886 * constructing their contents based on the OVN_NB database. */
2888 build_lflows(struct northd_context *ctx, struct hmap *datapaths,
2891 struct hmap lflows = HMAP_INITIALIZER(&lflows);
2892 struct hmap mcgroups = HMAP_INITIALIZER(&mcgroups);
2894 build_lswitch_flows(datapaths, ports, &lflows, &mcgroups);
2895 build_lrouter_flows(datapaths, ports, &lflows);
2897 /* Push changes to the Logical_Flow table to database. */
2898 const struct sbrec_logical_flow *sbflow, *next_sbflow;
2899 SBREC_LOGICAL_FLOW_FOR_EACH_SAFE (sbflow, next_sbflow, ctx->ovnsb_idl) {
2900 struct ovn_datapath *od
2901 = ovn_datapath_from_sbrec(datapaths, sbflow->logical_datapath);
2903 sbrec_logical_flow_delete(sbflow);
2907 enum ovn_datapath_type dp_type = od->nbs ? DP_SWITCH : DP_ROUTER;
2908 enum ovn_pipeline pipeline
2909 = !strcmp(sbflow->pipeline, "ingress") ? P_IN : P_OUT;
2910 struct ovn_lflow *lflow = ovn_lflow_find(
2911 &lflows, od, ovn_stage_build(dp_type, pipeline, sbflow->table_id),
2912 sbflow->priority, sbflow->match, sbflow->actions);
2914 ovn_lflow_destroy(&lflows, lflow);
2916 sbrec_logical_flow_delete(sbflow);
2919 struct ovn_lflow *lflow, *next_lflow;
2920 HMAP_FOR_EACH_SAFE (lflow, next_lflow, hmap_node, &lflows) {
2921 enum ovn_pipeline pipeline = ovn_stage_get_pipeline(lflow->stage);
2922 uint8_t table = ovn_stage_get_table(lflow->stage);
2924 sbflow = sbrec_logical_flow_insert(ctx->ovnsb_txn);
2925 sbrec_logical_flow_set_logical_datapath(sbflow, lflow->od->sb);
2926 sbrec_logical_flow_set_pipeline(
2927 sbflow, pipeline == P_IN ? "ingress" : "egress");
2928 sbrec_logical_flow_set_table_id(sbflow, table);
2929 sbrec_logical_flow_set_priority(sbflow, lflow->priority);
2930 sbrec_logical_flow_set_match(sbflow, lflow->match);
2931 sbrec_logical_flow_set_actions(sbflow, lflow->actions);
2933 const struct smap ids = SMAP_CONST1(&ids, "stage-name",
2934 ovn_stage_to_str(lflow->stage));
2935 sbrec_logical_flow_set_external_ids(sbflow, &ids);
2937 ovn_lflow_destroy(&lflows, lflow);
2939 hmap_destroy(&lflows);
2941 /* Push changes to the Multicast_Group table to database. */
2942 const struct sbrec_multicast_group *sbmc, *next_sbmc;
2943 SBREC_MULTICAST_GROUP_FOR_EACH_SAFE (sbmc, next_sbmc, ctx->ovnsb_idl) {
2944 struct ovn_datapath *od = ovn_datapath_from_sbrec(datapaths,
2947 sbrec_multicast_group_delete(sbmc);
2951 struct multicast_group group = { .name = sbmc->name,
2952 .key = sbmc->tunnel_key };
2953 struct ovn_multicast *mc = ovn_multicast_find(&mcgroups, od, &group);
2955 ovn_multicast_update_sbrec(mc, sbmc);
2956 ovn_multicast_destroy(&mcgroups, mc);
2958 sbrec_multicast_group_delete(sbmc);
2961 struct ovn_multicast *mc, *next_mc;
2962 HMAP_FOR_EACH_SAFE (mc, next_mc, hmap_node, &mcgroups) {
2963 sbmc = sbrec_multicast_group_insert(ctx->ovnsb_txn);
2964 sbrec_multicast_group_set_datapath(sbmc, mc->datapath->sb);
2965 sbrec_multicast_group_set_name(sbmc, mc->group->name);
2966 sbrec_multicast_group_set_tunnel_key(sbmc, mc->group->key);
2967 ovn_multicast_update_sbrec(mc, sbmc);
2968 ovn_multicast_destroy(&mcgroups, mc);
2970 hmap_destroy(&mcgroups);
2973 /* OVN_Northbound and OVN_Southbound have an identical Address_Set table.
2974 * We always update OVN_Southbound to match the current data in
2975 * OVN_Northbound, so that the address sets used in Logical_Flows in
2976 * OVN_Southbound is checked against the proper set.*/
2978 sync_address_sets(struct northd_context *ctx)
2980 struct shash sb_address_sets = SHASH_INITIALIZER(&sb_address_sets);
2982 const struct sbrec_address_set *sb_address_set;
2983 SBREC_ADDRESS_SET_FOR_EACH (sb_address_set, ctx->ovnsb_idl) {
2984 shash_add(&sb_address_sets, sb_address_set->name, sb_address_set);
2987 const struct nbrec_address_set *nb_address_set;
2988 NBREC_ADDRESS_SET_FOR_EACH (nb_address_set, ctx->ovnnb_idl) {
2989 sb_address_set = shash_find_and_delete(&sb_address_sets,
2990 nb_address_set->name);
2991 if (!sb_address_set) {
2992 sb_address_set = sbrec_address_set_insert(ctx->ovnsb_txn);
2993 sbrec_address_set_set_name(sb_address_set, nb_address_set->name);
2996 sbrec_address_set_set_addresses(sb_address_set,
2997 /* "char **" is not compatible with "const char **" */
2998 (const char **) nb_address_set->addresses,
2999 nb_address_set->n_addresses);
3002 struct shash_node *node, *next;
3003 SHASH_FOR_EACH_SAFE (node, next, &sb_address_sets) {
3004 sbrec_address_set_delete(node->data);
3005 shash_delete(&sb_address_sets, node);
3007 shash_destroy(&sb_address_sets);
3011 ovnnb_db_run(struct northd_context *ctx)
3013 if (!ctx->ovnsb_txn) {
3016 struct hmap datapaths, ports;
3017 build_datapaths(ctx, &datapaths);
3018 build_ports(ctx, &datapaths, &ports);
3019 build_lflows(ctx, &datapaths, &ports);
3021 sync_address_sets(ctx);
3023 struct ovn_datapath *dp, *next_dp;
3024 HMAP_FOR_EACH_SAFE (dp, next_dp, key_node, &datapaths) {
3025 ovn_datapath_destroy(&datapaths, dp);
3027 hmap_destroy(&datapaths);
3029 struct ovn_port *port, *next_port;
3030 HMAP_FOR_EACH_SAFE (port, next_port, key_node, &ports) {
3031 ovn_port_destroy(&ports, port);
3033 hmap_destroy(&ports);
3037 * The only change we get notified about is if the 'chassis' column of the
3038 * 'Port_Binding' table changes. When this column is not empty, it means we
3039 * need to set the corresponding logical port as 'up' in the northbound DB.
3042 ovnsb_db_run(struct northd_context *ctx)
3044 if (!ctx->ovnnb_txn) {
3047 struct hmap lports_hmap;
3048 const struct sbrec_port_binding *sb;
3049 const struct nbrec_logical_switch_port *nbsp;
3051 struct lport_hash_node {
3052 struct hmap_node node;
3053 const struct nbrec_logical_switch_port *nbsp;
3056 hmap_init(&lports_hmap);
3058 NBREC_LOGICAL_SWITCH_PORT_FOR_EACH(nbsp, ctx->ovnnb_idl) {
3059 hash_node = xzalloc(sizeof *hash_node);
3060 hash_node->nbsp = nbsp;
3061 hmap_insert(&lports_hmap, &hash_node->node, hash_string(nbsp->name, 0));
3064 SBREC_PORT_BINDING_FOR_EACH(sb, ctx->ovnsb_idl) {
3066 HMAP_FOR_EACH_WITH_HASH(hash_node, node,
3067 hash_string(sb->logical_port, 0),
3069 if (!strcmp(sb->logical_port, hash_node->nbsp->name)) {
3070 nbsp = hash_node->nbsp;
3076 /* The logical port doesn't exist for this port binding. This can
3077 * happen under normal circumstances when ovn-northd hasn't gotten
3078 * around to pruning the Port_Binding yet. */
3082 if (sb->chassis && (!nbsp->up || !*nbsp->up)) {
3084 nbrec_logical_switch_port_set_up(nbsp, &up, 1);
3085 } else if (!sb->chassis && (!nbsp->up || *nbsp->up)) {
3087 nbrec_logical_switch_port_set_up(nbsp, &up, 1);
3091 HMAP_FOR_EACH_POP(hash_node, node, &lports_hmap) {
3094 hmap_destroy(&lports_hmap);
3098 static char *default_nb_db_;
3103 if (!default_nb_db_) {
3104 default_nb_db_ = xasprintf("unix:%s/ovnnb_db.sock", ovs_rundir());
3106 return default_nb_db_;
3109 static char *default_sb_db_;
3114 if (!default_sb_db_) {
3115 default_sb_db_ = xasprintf("unix:%s/ovnsb_db.sock", ovs_rundir());
3117 return default_sb_db_;
3121 parse_options(int argc OVS_UNUSED, char *argv[] OVS_UNUSED)
3124 DAEMON_OPTION_ENUMS,
3127 static const struct option long_options[] = {
3128 {"ovnsb-db", required_argument, NULL, 'd'},
3129 {"ovnnb-db", required_argument, NULL, 'D'},
3130 {"help", no_argument, NULL, 'h'},
3131 {"options", no_argument, NULL, 'o'},
3132 {"version", no_argument, NULL, 'V'},
3133 DAEMON_LONG_OPTIONS,
3135 STREAM_SSL_LONG_OPTIONS,
3138 char *short_options = ovs_cmdl_long_options_to_short_options(long_options);
3143 c = getopt_long(argc, argv, short_options, long_options, NULL);
3149 DAEMON_OPTION_HANDLERS;
3150 VLOG_OPTION_HANDLERS;
3151 STREAM_SSL_OPTION_HANDLERS;
3166 ovs_cmdl_print_options(long_options);
3170 ovs_print_version(0, 0);
3179 ovnsb_db = default_sb_db();
3183 ovnnb_db = default_nb_db();
3186 free(short_options);
3190 add_column_noalert(struct ovsdb_idl *idl,
3191 const struct ovsdb_idl_column *column)
3193 ovsdb_idl_add_column(idl, column);
3194 ovsdb_idl_omit_alert(idl, column);
3198 main(int argc, char *argv[])
3200 int res = EXIT_SUCCESS;
3201 struct unixctl_server *unixctl;
3205 fatal_ignore_sigpipe();
3206 set_program_name(argv[0]);
3207 service_start(&argc, &argv);
3208 parse_options(argc, argv);
3210 daemonize_start(false);
3212 retval = unixctl_server_create(NULL, &unixctl);
3216 unixctl_command_register("exit", "", 0, 0, ovn_northd_exit, &exiting);
3218 daemonize_complete();
3223 /* We want to detect all changes to the ovn-nb db. */
3224 struct ovsdb_idl_loop ovnnb_idl_loop = OVSDB_IDL_LOOP_INITIALIZER(
3225 ovsdb_idl_create(ovnnb_db, &nbrec_idl_class, true, true));
3227 struct ovsdb_idl_loop ovnsb_idl_loop = OVSDB_IDL_LOOP_INITIALIZER(
3228 ovsdb_idl_create(ovnsb_db, &sbrec_idl_class, false, true));
3230 ovsdb_idl_add_table(ovnsb_idl_loop.idl, &sbrec_table_logical_flow);
3231 add_column_noalert(ovnsb_idl_loop.idl,
3232 &sbrec_logical_flow_col_logical_datapath);
3233 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_logical_flow_col_pipeline);
3234 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_logical_flow_col_table_id);
3235 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_logical_flow_col_priority);
3236 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_logical_flow_col_match);
3237 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_logical_flow_col_actions);
3239 ovsdb_idl_add_table(ovnsb_idl_loop.idl, &sbrec_table_multicast_group);
3240 add_column_noalert(ovnsb_idl_loop.idl,
3241 &sbrec_multicast_group_col_datapath);
3242 add_column_noalert(ovnsb_idl_loop.idl,
3243 &sbrec_multicast_group_col_tunnel_key);
3244 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_multicast_group_col_name);
3245 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_multicast_group_col_ports);
3247 ovsdb_idl_add_table(ovnsb_idl_loop.idl, &sbrec_table_datapath_binding);
3248 add_column_noalert(ovnsb_idl_loop.idl,
3249 &sbrec_datapath_binding_col_tunnel_key);
3250 add_column_noalert(ovnsb_idl_loop.idl,
3251 &sbrec_datapath_binding_col_external_ids);
3253 ovsdb_idl_add_table(ovnsb_idl_loop.idl, &sbrec_table_port_binding);
3254 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_port_binding_col_datapath);
3255 add_column_noalert(ovnsb_idl_loop.idl,
3256 &sbrec_port_binding_col_logical_port);
3257 add_column_noalert(ovnsb_idl_loop.idl,
3258 &sbrec_port_binding_col_tunnel_key);
3259 add_column_noalert(ovnsb_idl_loop.idl,
3260 &sbrec_port_binding_col_parent_port);
3261 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_port_binding_col_tag);
3262 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_port_binding_col_type);
3263 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_port_binding_col_options);
3264 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_port_binding_col_mac);
3265 ovsdb_idl_add_column(ovnsb_idl_loop.idl, &sbrec_port_binding_col_chassis);
3267 ovsdb_idl_add_table(ovnsb_idl_loop.idl, &sbrec_table_address_set);
3268 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_address_set_col_name);
3269 add_column_noalert(ovnsb_idl_loop.idl, &sbrec_address_set_col_addresses);
3274 struct northd_context ctx = {
3275 .ovnnb_idl = ovnnb_idl_loop.idl,
3276 .ovnnb_txn = ovsdb_idl_loop_run(&ovnnb_idl_loop),
3277 .ovnsb_idl = ovnsb_idl_loop.idl,
3278 .ovnsb_txn = ovsdb_idl_loop_run(&ovnsb_idl_loop),
3284 unixctl_server_run(unixctl);
3285 unixctl_server_wait(unixctl);
3287 poll_immediate_wake();
3289 ovsdb_idl_loop_commit_and_wait(&ovnnb_idl_loop);
3290 ovsdb_idl_loop_commit_and_wait(&ovnsb_idl_loop);
3293 if (should_service_stop()) {
3298 unixctl_server_destroy(unixctl);
3299 ovsdb_idl_loop_destroy(&ovnnb_idl_loop);
3300 ovsdb_idl_loop_destroy(&ovnsb_idl_loop);
3303 free(default_nb_db_);
3304 free(default_sb_db_);
3309 ovn_northd_exit(struct unixctl_conn *conn, int argc OVS_UNUSED,
3310 const char *argv[] OVS_UNUSED, void *exiting_)
3312 bool *exiting = exiting_;
3315 unixctl_command_reply(conn, NULL);