2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 16
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any facet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 #define MAX_MIRRORS 32
110 typedef uint32_t mirror_mask_t;
111 #define MIRROR_MASK_C(X) UINT32_C(X)
112 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
114 struct ofproto_dpif *ofproto; /* Owning ofproto. */
115 size_t idx; /* In ofproto's "mirrors" array. */
116 void *aux; /* Key supplied by ofproto's client. */
117 char *name; /* Identifier for log messages. */
119 /* Selection criteria. */
120 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
121 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
122 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
124 /* Output (mutually exclusive). */
125 struct ofbundle *out; /* Output port or NULL. */
126 int out_vlan; /* Output VLAN or -1. */
129 static void mirror_destroy(struct ofmirror *);
131 /* A group of one or more OpenFlow ports. */
132 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
140 struct list ports; /* Contains "struct ofport"s. */
141 enum port_vlan_mode vlan_mode; /* VLAN mode */
142 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
143 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
144 * NULL if all VLANs are trunked. */
145 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
146 struct bond *bond; /* Nonnull iff more than one port. */
149 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
151 /* Port mirroring info. */
152 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
153 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
154 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
157 static void bundle_remove(struct ofport *);
158 static void bundle_destroy(struct ofbundle *);
159 static void bundle_del_port(struct ofport_dpif *);
160 static void bundle_run(struct ofbundle *);
161 static void bundle_wait(struct ofbundle *);
163 struct action_xlate_ctx {
164 /* action_xlate_ctx_init() initializes these members. */
167 struct ofproto_dpif *ofproto;
169 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
170 * this flow when actions change header fields. */
173 /* The packet corresponding to 'flow', or a null pointer if we are
174 * revalidating without a packet to refer to. */
175 const struct ofpbuf *packet;
177 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
178 * want to execute them if we are actually processing a packet, or if we
179 * are accounting for packets that the datapath has processed, but not if
180 * we are just revalidating. */
183 /* If nonnull, called just before executing a resubmit action.
185 * This is normally null so the client has to set it manually after
186 * calling action_xlate_ctx_init(). */
187 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
189 /* xlate_actions() initializes and uses these members. The client might want
190 * to look at them after it returns. */
192 struct ofpbuf *odp_actions; /* Datapath actions. */
193 tag_type tags; /* Tags associated with actions. */
194 bool may_set_up_flow; /* True ordinarily; false if the actions must
195 * be reassessed for every packet. */
196 bool has_learn; /* Actions include NXAST_LEARN? */
197 bool has_normal; /* Actions output to OFPP_NORMAL? */
198 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
200 /* xlate_actions() initializes and uses these members, but the client has no
201 * reason to look at them. */
203 int recurse; /* Recursion level, via xlate_table_action. */
204 uint32_t priority; /* Current flow priority. 0 if none. */
205 struct flow base_flow; /* Flow at the last commit. */
206 uint32_t base_priority; /* Priority at the last commit. */
207 uint8_t table_id; /* OpenFlow table ID where flow was found. */
210 static void action_xlate_ctx_init(struct action_xlate_ctx *,
211 struct ofproto_dpif *, const struct flow *,
212 const struct ofpbuf *);
213 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
214 const union ofp_action *in, size_t n_in);
216 /* An exact-match instantiation of an OpenFlow flow. */
218 long long int used; /* Time last used; time created if not used. */
222 * - Do include packets and bytes sent "by hand", e.g. with
225 * - Do include packets and bytes that were obtained from the datapath
226 * when its statistics were reset (e.g. dpif_flow_put() with
227 * DPIF_FP_ZERO_STATS).
229 uint64_t packet_count; /* Number of packets received. */
230 uint64_t byte_count; /* Number of bytes received. */
232 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
233 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
235 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
236 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
237 long long int rs_used; /* Used time pushed to resubmit children. */
239 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
241 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
242 struct list list_node; /* In owning rule's 'facets' list. */
243 struct rule_dpif *rule; /* Owning rule. */
244 struct flow flow; /* Exact-match flow. */
245 bool installed; /* Installed in datapath? */
246 bool may_install; /* True ordinarily; false if actions must
247 * be reassessed for every packet. */
248 bool has_learn; /* Actions include NXAST_LEARN? */
249 bool has_normal; /* Actions output to OFPP_NORMAL? */
250 size_t actions_len; /* Number of bytes in actions[]. */
251 struct nlattr *actions; /* Datapath actions. */
252 tag_type tags; /* Tags. */
253 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
256 static struct facet *facet_create(struct rule_dpif *, const struct flow *,
257 const struct ofpbuf *packet);
258 static void facet_remove(struct ofproto_dpif *, struct facet *);
259 static void facet_free(struct facet *);
261 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
262 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
263 const struct flow *);
264 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
266 static void facet_execute(struct ofproto_dpif *, struct facet *,
267 struct ofpbuf *packet);
269 static int facet_put__(struct ofproto_dpif *, struct facet *,
270 const struct nlattr *actions, size_t actions_len,
271 struct dpif_flow_stats *);
272 static void facet_install(struct ofproto_dpif *, struct facet *,
274 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
275 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
277 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
278 const struct ofpbuf *packet);
279 static void facet_update_time(struct ofproto_dpif *, struct facet *,
281 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
282 const struct dpif_flow_stats *);
283 static void facet_reset_counters(struct facet *);
284 static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
285 static void facet_push_stats(struct facet *);
286 static void facet_account(struct ofproto_dpif *, struct facet *);
288 static bool facet_is_controller_flow(struct facet *);
290 static void flow_push_stats(const struct rule_dpif *,
291 struct flow *, uint64_t packets, uint64_t bytes,
294 static uint32_t rule_calculate_tag(const struct flow *,
295 const struct flow_wildcards *,
297 static void rule_invalidate(const struct rule_dpif *);
303 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
304 struct list bundle_node; /* In struct ofbundle's "ports" list. */
305 struct cfm *cfm; /* Connectivity Fault Management, if any. */
306 tag_type tag; /* Tag associated with this port. */
307 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
308 bool may_enable; /* May be enabled in bonds. */
311 static struct ofport_dpif *
312 ofport_dpif_cast(const struct ofport *ofport)
314 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
315 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
318 static void port_run(struct ofport_dpif *);
319 static void port_wait(struct ofport_dpif *);
320 static int set_cfm(struct ofport *, const struct cfm_settings *);
322 struct dpif_completion {
323 struct list list_node;
324 struct ofoperation *op;
327 /* Extra information about a classifier table.
328 * Currently used just for optimized flow revalidation. */
330 /* If either of these is nonnull, then this table has a form that allows
331 * flows to be tagged to avoid revalidating most flows for the most common
332 * kinds of flow table changes. */
333 struct cls_table *catchall_table; /* Table that wildcards all fields. */
334 struct cls_table *other_table; /* Table with any other wildcard set. */
335 uint32_t basis; /* Keeps each table's tags separate. */
338 struct ofproto_dpif {
347 struct netflow *netflow;
348 struct dpif_sflow *sflow;
349 struct hmap bundles; /* Contains "struct ofbundle"s. */
350 struct mac_learning *ml;
351 struct ofmirror *mirrors[MAX_MIRRORS];
352 bool has_bonded_bundles;
355 struct timer next_expiration;
361 struct table_dpif tables[N_TABLES];
362 bool need_revalidate;
363 struct tag_set revalidate_set;
365 /* Support for debugging async flow mods. */
366 struct list completions;
368 bool has_bundle_action; /* True when the first bundle action appears. */
371 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
372 * for debugging the asynchronous flow_mod implementation.) */
375 static void ofproto_dpif_unixctl_init(void);
377 static struct ofproto_dpif *
378 ofproto_dpif_cast(const struct ofproto *ofproto)
380 assert(ofproto->ofproto_class == &ofproto_dpif_class);
381 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
384 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
386 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
389 /* Packet processing. */
390 static void update_learning_table(struct ofproto_dpif *,
391 const struct flow *, int vlan,
393 static bool is_admissible(struct ofproto_dpif *, const struct flow *,
394 bool have_packet, tag_type *, int *vlanp,
395 struct ofbundle **in_bundlep);
396 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
398 /* Flow expiration. */
399 static int expire(struct ofproto_dpif *);
402 static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
403 const struct ofpbuf *packet);
405 /* Global variables. */
406 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
408 /* Factory functions. */
411 enumerate_types(struct sset *types)
413 dp_enumerate_types(types);
417 enumerate_names(const char *type, struct sset *names)
419 return dp_enumerate_names(type, names);
423 del(const char *type, const char *name)
428 error = dpif_open(name, type, &dpif);
430 error = dpif_delete(dpif);
436 /* Basic life-cycle. */
438 static struct ofproto *
441 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
446 dealloc(struct ofproto *ofproto_)
448 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
453 construct(struct ofproto *ofproto_, int *n_tablesp)
455 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
456 const char *name = ofproto->up.name;
460 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
462 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
466 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
467 ofproto->n_matches = 0;
469 dpif_flow_flush(ofproto->dpif);
470 dpif_recv_purge(ofproto->dpif);
472 error = dpif_recv_set_mask(ofproto->dpif,
473 ((1u << DPIF_UC_MISS) |
474 (1u << DPIF_UC_ACTION) |
475 (1u << DPIF_UC_SAMPLE)));
477 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
478 dpif_close(ofproto->dpif);
482 ofproto->netflow = NULL;
483 ofproto->sflow = NULL;
484 hmap_init(&ofproto->bundles);
485 ofproto->ml = mac_learning_create();
486 for (i = 0; i < MAX_MIRRORS; i++) {
487 ofproto->mirrors[i] = NULL;
489 ofproto->has_bonded_bundles = false;
491 timer_set_duration(&ofproto->next_expiration, 1000);
493 hmap_init(&ofproto->facets);
495 for (i = 0; i < N_TABLES; i++) {
496 struct table_dpif *table = &ofproto->tables[i];
498 table->catchall_table = NULL;
499 table->other_table = NULL;
500 table->basis = random_uint32();
502 ofproto->need_revalidate = false;
503 tag_set_init(&ofproto->revalidate_set);
505 list_init(&ofproto->completions);
507 ofproto_dpif_unixctl_init();
509 ofproto->has_bundle_action = false;
511 *n_tablesp = N_TABLES;
516 complete_operations(struct ofproto_dpif *ofproto)
518 struct dpif_completion *c, *next;
520 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
521 ofoperation_complete(c->op, 0);
522 list_remove(&c->list_node);
528 destruct(struct ofproto *ofproto_)
530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
531 struct rule_dpif *rule, *next_rule;
532 struct classifier *table;
535 complete_operations(ofproto);
537 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
538 struct cls_cursor cursor;
540 cls_cursor_init(&cursor, table, NULL);
541 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
542 ofproto_rule_destroy(&rule->up);
546 for (i = 0; i < MAX_MIRRORS; i++) {
547 mirror_destroy(ofproto->mirrors[i]);
550 netflow_destroy(ofproto->netflow);
551 dpif_sflow_destroy(ofproto->sflow);
552 hmap_destroy(&ofproto->bundles);
553 mac_learning_destroy(ofproto->ml);
555 hmap_destroy(&ofproto->facets);
557 dpif_close(ofproto->dpif);
561 run(struct ofproto *ofproto_)
563 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
564 struct ofport_dpif *ofport;
565 struct ofbundle *bundle;
569 complete_operations(ofproto);
571 dpif_run(ofproto->dpif);
573 for (i = 0; i < 50; i++) {
574 struct dpif_upcall packet;
577 error = dpif_recv(ofproto->dpif, &packet);
579 if (error == ENODEV) {
580 /* Datapath destroyed. */
586 handle_upcall(ofproto, &packet);
589 if (timer_expired(&ofproto->next_expiration)) {
590 int delay = expire(ofproto);
591 timer_set_duration(&ofproto->next_expiration, delay);
594 if (ofproto->netflow) {
595 netflow_run(ofproto->netflow);
597 if (ofproto->sflow) {
598 dpif_sflow_run(ofproto->sflow);
601 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
604 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
608 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
610 /* Now revalidate if there's anything to do. */
611 if (ofproto->need_revalidate
612 || !tag_set_is_empty(&ofproto->revalidate_set)) {
613 struct tag_set revalidate_set = ofproto->revalidate_set;
614 bool revalidate_all = ofproto->need_revalidate;
615 struct facet *facet, *next;
617 /* Clear the revalidation flags. */
618 tag_set_init(&ofproto->revalidate_set);
619 ofproto->need_revalidate = false;
621 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
623 || tag_set_intersects(&revalidate_set, facet->tags)) {
624 facet_revalidate(ofproto, facet);
633 wait(struct ofproto *ofproto_)
635 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
636 struct ofport_dpif *ofport;
637 struct ofbundle *bundle;
639 if (!clogged && !list_is_empty(&ofproto->completions)) {
640 poll_immediate_wake();
643 dpif_wait(ofproto->dpif);
644 dpif_recv_wait(ofproto->dpif);
645 if (ofproto->sflow) {
646 dpif_sflow_wait(ofproto->sflow);
648 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
649 poll_immediate_wake();
651 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
654 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
657 mac_learning_wait(ofproto->ml);
658 if (ofproto->need_revalidate) {
659 /* Shouldn't happen, but if it does just go around again. */
660 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
661 poll_immediate_wake();
663 timer_wait(&ofproto->next_expiration);
668 flush(struct ofproto *ofproto_)
670 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
671 struct facet *facet, *next_facet;
673 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
674 /* Mark the facet as not installed so that facet_remove() doesn't
675 * bother trying to uninstall it. There is no point in uninstalling it
676 * individually since we are about to blow away all the facets with
677 * dpif_flow_flush(). */
678 facet->installed = false;
679 facet->dp_packet_count = 0;
680 facet->dp_byte_count = 0;
681 facet_remove(ofproto, facet);
683 dpif_flow_flush(ofproto->dpif);
687 get_features(struct ofproto *ofproto_ OVS_UNUSED,
688 bool *arp_match_ip, uint32_t *actions)
690 *arp_match_ip = true;
691 *actions = ((1u << OFPAT_OUTPUT) |
692 (1u << OFPAT_SET_VLAN_VID) |
693 (1u << OFPAT_SET_VLAN_PCP) |
694 (1u << OFPAT_STRIP_VLAN) |
695 (1u << OFPAT_SET_DL_SRC) |
696 (1u << OFPAT_SET_DL_DST) |
697 (1u << OFPAT_SET_NW_SRC) |
698 (1u << OFPAT_SET_NW_DST) |
699 (1u << OFPAT_SET_NW_TOS) |
700 (1u << OFPAT_SET_TP_SRC) |
701 (1u << OFPAT_SET_TP_DST) |
702 (1u << OFPAT_ENQUEUE));
706 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
709 struct ovs_dp_stats s;
711 strcpy(ots->name, "classifier");
713 dpif_get_dp_stats(ofproto->dpif, &s);
714 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
715 put_32aligned_be64(&ots->matched_count,
716 htonll(s.n_hit + ofproto->n_matches));
720 set_netflow(struct ofproto *ofproto_,
721 const struct netflow_options *netflow_options)
723 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
725 if (netflow_options) {
726 if (!ofproto->netflow) {
727 ofproto->netflow = netflow_create();
729 return netflow_set_options(ofproto->netflow, netflow_options);
731 netflow_destroy(ofproto->netflow);
732 ofproto->netflow = NULL;
737 static struct ofport *
740 struct ofport_dpif *port = xmalloc(sizeof *port);
745 port_dealloc(struct ofport *port_)
747 struct ofport_dpif *port = ofport_dpif_cast(port_);
752 port_construct(struct ofport *port_)
754 struct ofport_dpif *port = ofport_dpif_cast(port_);
755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
757 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
760 port->tag = tag_create_random();
761 port->may_enable = true;
763 if (ofproto->sflow) {
764 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
765 netdev_get_name(port->up.netdev));
772 port_destruct(struct ofport *port_)
774 struct ofport_dpif *port = ofport_dpif_cast(port_);
775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
777 bundle_remove(port_);
778 set_cfm(port_, NULL);
779 if (ofproto->sflow) {
780 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
785 port_modified(struct ofport *port_)
787 struct ofport_dpif *port = ofport_dpif_cast(port_);
789 if (port->bundle && port->bundle->bond) {
790 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
795 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
797 struct ofport_dpif *port = ofport_dpif_cast(port_);
798 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
799 ovs_be32 changed = old_config ^ port->up.opp.config;
801 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
802 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
803 ofproto->need_revalidate = true;
808 set_sflow(struct ofproto *ofproto_,
809 const struct ofproto_sflow_options *sflow_options)
811 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
812 struct dpif_sflow *ds = ofproto->sflow;
815 struct ofport_dpif *ofport;
817 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
818 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
819 dpif_sflow_add_port(ds, ofport->odp_port,
820 netdev_get_name(ofport->up.netdev));
823 dpif_sflow_set_options(ds, sflow_options);
825 dpif_sflow_destroy(ds);
826 ofproto->sflow = NULL;
832 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
834 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
841 struct ofproto_dpif *ofproto;
843 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
844 ofproto->need_revalidate = true;
845 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
848 if (cfm_configure(ofport->cfm, s)) {
854 cfm_destroy(ofport->cfm);
860 get_cfm_fault(const struct ofport *ofport_)
862 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
864 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
868 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
871 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
874 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
883 /* Expires all MAC learning entries associated with 'port' and forces ofproto
884 * to revalidate every flow. */
886 bundle_flush_macs(struct ofbundle *bundle)
888 struct ofproto_dpif *ofproto = bundle->ofproto;
889 struct mac_learning *ml = ofproto->ml;
890 struct mac_entry *mac, *next_mac;
892 ofproto->need_revalidate = true;
893 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
894 if (mac->port.p == bundle) {
895 mac_learning_expire(ml, mac);
900 static struct ofbundle *
901 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
903 struct ofbundle *bundle;
905 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
907 if (bundle->aux == aux) {
914 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
915 * ones that are found to 'bundles'. */
917 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
918 void **auxes, size_t n_auxes,
919 struct hmapx *bundles)
924 for (i = 0; i < n_auxes; i++) {
925 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
927 hmapx_add(bundles, bundle);
933 bundle_del_port(struct ofport_dpif *port)
935 struct ofbundle *bundle = port->bundle;
937 bundle->ofproto->need_revalidate = true;
939 list_remove(&port->bundle_node);
943 lacp_slave_unregister(bundle->lacp, port);
946 bond_slave_unregister(bundle->bond, port);
949 bundle->floodable = true;
950 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
951 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
952 bundle->floodable = false;
958 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
959 struct lacp_slave_settings *lacp,
960 uint32_t bond_stable_id)
962 struct ofport_dpif *port;
964 port = get_ofp_port(bundle->ofproto, ofp_port);
969 if (port->bundle != bundle) {
970 bundle->ofproto->need_revalidate = true;
972 bundle_del_port(port);
975 port->bundle = bundle;
976 list_push_back(&bundle->ports, &port->bundle_node);
977 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
978 bundle->floodable = false;
982 lacp_slave_register(bundle->lacp, port, lacp);
985 port->bond_stable_id = bond_stable_id;
991 bundle_destroy(struct ofbundle *bundle)
993 struct ofproto_dpif *ofproto;
994 struct ofport_dpif *port, *next_port;
1001 ofproto = bundle->ofproto;
1002 for (i = 0; i < MAX_MIRRORS; i++) {
1003 struct ofmirror *m = ofproto->mirrors[i];
1005 if (m->out == bundle) {
1007 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1008 || hmapx_find_and_delete(&m->dsts, bundle)) {
1009 ofproto->need_revalidate = true;
1014 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1015 bundle_del_port(port);
1018 bundle_flush_macs(bundle);
1019 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1021 free(bundle->trunks);
1022 lacp_destroy(bundle->lacp);
1023 bond_destroy(bundle->bond);
1028 bundle_set(struct ofproto *ofproto_, void *aux,
1029 const struct ofproto_bundle_settings *s)
1031 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1032 bool need_flush = false;
1033 struct ofport_dpif *port;
1034 struct ofbundle *bundle;
1035 unsigned long *trunks;
1041 bundle_destroy(bundle_lookup(ofproto, aux));
1045 assert(s->n_slaves == 1 || s->bond != NULL);
1046 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1048 bundle = bundle_lookup(ofproto, aux);
1050 bundle = xmalloc(sizeof *bundle);
1052 bundle->ofproto = ofproto;
1053 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1054 hash_pointer(aux, 0));
1056 bundle->name = NULL;
1058 list_init(&bundle->ports);
1059 bundle->vlan_mode = PORT_VLAN_TRUNK;
1061 bundle->trunks = NULL;
1062 bundle->lacp = NULL;
1063 bundle->bond = NULL;
1065 bundle->floodable = true;
1067 bundle->src_mirrors = 0;
1068 bundle->dst_mirrors = 0;
1069 bundle->mirror_out = 0;
1072 if (!bundle->name || strcmp(s->name, bundle->name)) {
1074 bundle->name = xstrdup(s->name);
1079 if (!bundle->lacp) {
1080 ofproto->need_revalidate = true;
1081 bundle->lacp = lacp_create();
1083 lacp_configure(bundle->lacp, s->lacp);
1085 lacp_destroy(bundle->lacp);
1086 bundle->lacp = NULL;
1089 /* Update set of ports. */
1091 for (i = 0; i < s->n_slaves; i++) {
1092 if (!bundle_add_port(bundle, s->slaves[i],
1093 s->lacp ? &s->lacp_slaves[i] : NULL,
1094 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1098 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1099 struct ofport_dpif *next_port;
1101 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1102 for (i = 0; i < s->n_slaves; i++) {
1103 if (s->slaves[i] == port->up.ofp_port) {
1108 bundle_del_port(port);
1112 assert(list_size(&bundle->ports) <= s->n_slaves);
1114 if (list_is_empty(&bundle->ports)) {
1115 bundle_destroy(bundle);
1119 /* Set VLAN tagging mode */
1120 if (s->vlan_mode != bundle->vlan_mode) {
1121 bundle->vlan_mode = s->vlan_mode;
1126 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1127 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1129 if (vlan != bundle->vlan) {
1130 bundle->vlan = vlan;
1134 /* Get trunked VLANs. */
1135 switch (s->vlan_mode) {
1136 case PORT_VLAN_ACCESS:
1140 case PORT_VLAN_TRUNK:
1141 trunks = (unsigned long *) s->trunks;
1144 case PORT_VLAN_NATIVE_UNTAGGED:
1145 case PORT_VLAN_NATIVE_TAGGED:
1146 if (vlan != 0 && (!s->trunks
1147 || !bitmap_is_set(s->trunks, vlan)
1148 || bitmap_is_set(s->trunks, 0))) {
1149 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1151 trunks = bitmap_clone(s->trunks, 4096);
1153 trunks = bitmap_allocate1(4096);
1155 bitmap_set1(trunks, vlan);
1156 bitmap_set0(trunks, 0);
1158 trunks = (unsigned long *) s->trunks;
1165 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1166 free(bundle->trunks);
1167 if (trunks == s->trunks) {
1168 bundle->trunks = vlan_bitmap_clone(trunks);
1170 bundle->trunks = trunks;
1175 if (trunks != s->trunks) {
1180 if (!list_is_short(&bundle->ports)) {
1181 bundle->ofproto->has_bonded_bundles = true;
1183 if (bond_reconfigure(bundle->bond, s->bond)) {
1184 ofproto->need_revalidate = true;
1187 bundle->bond = bond_create(s->bond);
1188 ofproto->need_revalidate = true;
1191 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1192 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1196 bond_destroy(bundle->bond);
1197 bundle->bond = NULL;
1200 /* If we changed something that would affect MAC learning, un-learn
1201 * everything on this port and force flow revalidation. */
1203 bundle_flush_macs(bundle);
1210 bundle_remove(struct ofport *port_)
1212 struct ofport_dpif *port = ofport_dpif_cast(port_);
1213 struct ofbundle *bundle = port->bundle;
1216 bundle_del_port(port);
1217 if (list_is_empty(&bundle->ports)) {
1218 bundle_destroy(bundle);
1219 } else if (list_is_short(&bundle->ports)) {
1220 bond_destroy(bundle->bond);
1221 bundle->bond = NULL;
1227 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1230 struct ofport_dpif *port = port_;
1231 uint8_t ea[ETH_ADDR_LEN];
1234 error = netdev_get_etheraddr(port->up.netdev, ea);
1236 struct ofpbuf packet;
1239 ofpbuf_init(&packet, 0);
1240 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1242 memcpy(packet_pdu, pdu, pdu_size);
1244 error = netdev_send(port->up.netdev, &packet);
1246 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1247 "(%s)", port->bundle->name,
1248 netdev_get_name(port->up.netdev), strerror(error));
1250 ofpbuf_uninit(&packet);
1252 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1253 "%s (%s)", port->bundle->name,
1254 netdev_get_name(port->up.netdev), strerror(error));
1259 bundle_send_learning_packets(struct ofbundle *bundle)
1261 struct ofproto_dpif *ofproto = bundle->ofproto;
1262 int error, n_packets, n_errors;
1263 struct mac_entry *e;
1265 error = n_packets = n_errors = 0;
1266 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1267 if (e->port.p != bundle) {
1268 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1278 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1279 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1280 "packets, last error was: %s",
1281 bundle->name, n_errors, n_packets, strerror(error));
1283 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1284 bundle->name, n_packets);
1289 bundle_run(struct ofbundle *bundle)
1292 lacp_run(bundle->lacp, send_pdu_cb);
1295 struct ofport_dpif *port;
1297 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1298 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1301 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1302 lacp_negotiated(bundle->lacp));
1303 if (bond_should_send_learning_packets(bundle->bond)) {
1304 bundle_send_learning_packets(bundle);
1310 bundle_wait(struct ofbundle *bundle)
1313 lacp_wait(bundle->lacp);
1316 bond_wait(bundle->bond);
1323 mirror_scan(struct ofproto_dpif *ofproto)
1327 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1328 if (!ofproto->mirrors[idx]) {
1335 static struct ofmirror *
1336 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1340 for (i = 0; i < MAX_MIRRORS; i++) {
1341 struct ofmirror *mirror = ofproto->mirrors[i];
1342 if (mirror && mirror->aux == aux) {
1351 mirror_set(struct ofproto *ofproto_, void *aux,
1352 const struct ofproto_mirror_settings *s)
1354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1355 mirror_mask_t mirror_bit;
1356 struct ofbundle *bundle;
1357 struct ofmirror *mirror;
1358 struct ofbundle *out;
1359 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1360 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1363 mirror = mirror_lookup(ofproto, aux);
1365 mirror_destroy(mirror);
1371 idx = mirror_scan(ofproto);
1373 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1375 ofproto->up.name, MAX_MIRRORS, s->name);
1379 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1380 mirror->ofproto = ofproto;
1383 mirror->out_vlan = -1;
1384 mirror->name = NULL;
1387 if (!mirror->name || strcmp(s->name, mirror->name)) {
1389 mirror->name = xstrdup(s->name);
1392 /* Get the new configuration. */
1393 if (s->out_bundle) {
1394 out = bundle_lookup(ofproto, s->out_bundle);
1396 mirror_destroy(mirror);
1402 out_vlan = s->out_vlan;
1404 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1405 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1407 /* If the configuration has not changed, do nothing. */
1408 if (hmapx_equals(&srcs, &mirror->srcs)
1409 && hmapx_equals(&dsts, &mirror->dsts)
1410 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1411 && mirror->out == out
1412 && mirror->out_vlan == out_vlan)
1414 hmapx_destroy(&srcs);
1415 hmapx_destroy(&dsts);
1419 hmapx_swap(&srcs, &mirror->srcs);
1420 hmapx_destroy(&srcs);
1422 hmapx_swap(&dsts, &mirror->dsts);
1423 hmapx_destroy(&dsts);
1425 free(mirror->vlans);
1426 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1429 mirror->out_vlan = out_vlan;
1431 /* Update bundles. */
1432 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1433 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1434 if (hmapx_contains(&mirror->srcs, bundle)) {
1435 bundle->src_mirrors |= mirror_bit;
1437 bundle->src_mirrors &= ~mirror_bit;
1440 if (hmapx_contains(&mirror->dsts, bundle)) {
1441 bundle->dst_mirrors |= mirror_bit;
1443 bundle->dst_mirrors &= ~mirror_bit;
1446 if (mirror->out == bundle) {
1447 bundle->mirror_out |= mirror_bit;
1449 bundle->mirror_out &= ~mirror_bit;
1453 ofproto->need_revalidate = true;
1454 mac_learning_flush(ofproto->ml);
1460 mirror_destroy(struct ofmirror *mirror)
1462 struct ofproto_dpif *ofproto;
1463 mirror_mask_t mirror_bit;
1464 struct ofbundle *bundle;
1470 ofproto = mirror->ofproto;
1471 ofproto->need_revalidate = true;
1472 mac_learning_flush(ofproto->ml);
1474 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1475 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1476 bundle->src_mirrors &= ~mirror_bit;
1477 bundle->dst_mirrors &= ~mirror_bit;
1478 bundle->mirror_out &= ~mirror_bit;
1481 hmapx_destroy(&mirror->srcs);
1482 hmapx_destroy(&mirror->dsts);
1483 free(mirror->vlans);
1485 ofproto->mirrors[mirror->idx] = NULL;
1491 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1494 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1495 ofproto->need_revalidate = true;
1496 mac_learning_flush(ofproto->ml);
1502 is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
1504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1505 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1506 return bundle && bundle->mirror_out != 0;
1510 forward_bpdu_changed(struct ofproto *ofproto_)
1512 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1513 /* Revalidate cached flows whenever forward_bpdu option changes. */
1514 ofproto->need_revalidate = true;
1519 static struct ofport_dpif *
1520 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1522 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1523 return ofport ? ofport_dpif_cast(ofport) : NULL;
1526 static struct ofport_dpif *
1527 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1529 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1533 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1534 struct dpif_port *dpif_port)
1536 ofproto_port->name = dpif_port->name;
1537 ofproto_port->type = dpif_port->type;
1538 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1542 port_run(struct ofport_dpif *ofport)
1544 bool enable = netdev_get_carrier(ofport->up.netdev);
1547 cfm_run(ofport->cfm);
1549 if (cfm_should_send_ccm(ofport->cfm)) {
1550 struct ofpbuf packet;
1552 ofpbuf_init(&packet, 0);
1553 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1554 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1555 ofport->odp_port, &packet);
1556 ofpbuf_uninit(&packet);
1559 enable = enable && !cfm_get_fault(ofport->cfm);
1562 if (ofport->bundle) {
1563 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1566 if (ofport->may_enable != enable) {
1567 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1569 if (ofproto->has_bundle_action) {
1570 ofproto->need_revalidate = true;
1574 ofport->may_enable = enable;
1578 port_wait(struct ofport_dpif *ofport)
1581 cfm_wait(ofport->cfm);
1586 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1587 struct ofproto_port *ofproto_port)
1589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1590 struct dpif_port dpif_port;
1593 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1595 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1601 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1603 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1607 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1609 *ofp_portp = odp_port_to_ofp_port(odp_port);
1615 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1617 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1620 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1622 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1624 /* The caller is going to close ofport->up.netdev. If this is a
1625 * bonded port, then the bond is using that netdev, so remove it
1626 * from the bond. The client will need to reconfigure everything
1627 * after deleting ports, so then the slave will get re-added. */
1628 bundle_remove(&ofport->up);
1634 struct port_dump_state {
1635 struct dpif_port_dump dump;
1640 port_dump_start(const struct ofproto *ofproto_, void **statep)
1642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1643 struct port_dump_state *state;
1645 *statep = state = xmalloc(sizeof *state);
1646 dpif_port_dump_start(&state->dump, ofproto->dpif);
1647 state->done = false;
1652 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1653 struct ofproto_port *port)
1655 struct port_dump_state *state = state_;
1656 struct dpif_port dpif_port;
1658 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1659 ofproto_port_from_dpif_port(port, &dpif_port);
1662 int error = dpif_port_dump_done(&state->dump);
1664 return error ? error : EOF;
1669 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1671 struct port_dump_state *state = state_;
1674 dpif_port_dump_done(&state->dump);
1681 port_poll(const struct ofproto *ofproto_, char **devnamep)
1683 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1684 return dpif_port_poll(ofproto->dpif, devnamep);
1688 port_poll_wait(const struct ofproto *ofproto_)
1690 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1691 dpif_port_poll_wait(ofproto->dpif);
1695 port_is_lacp_current(const struct ofport *ofport_)
1697 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1698 return (ofport->bundle && ofport->bundle->lacp
1699 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
1703 /* Upcall handling. */
1705 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
1706 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
1707 * their individual configurations.
1709 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
1710 * Otherwise, ownership is transferred to this function. */
1712 send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
1713 const struct flow *flow, bool clone)
1715 struct ofputil_packet_in pin;
1717 pin.packet = upcall->packet;
1718 pin.in_port = flow->in_port;
1719 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
1720 pin.buffer_id = 0; /* not yet known */
1721 pin.send_len = upcall->userdata;
1722 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
1723 clone ? NULL : upcall->packet);
1727 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
1728 const struct ofpbuf *packet)
1730 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
1736 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
1738 cfm_process_heartbeat(ofport->cfm, packet);
1741 } else if (ofport->bundle && ofport->bundle->lacp
1742 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1744 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
1752 handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1754 struct facet *facet;
1757 /* Obtain in_port and tun_id, at least. */
1758 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1760 /* Set header pointers in 'flow'. */
1761 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
1763 /* Handle 802.1ag and LACP. */
1764 if (process_special(ofproto, &flow, upcall->packet)) {
1765 ofpbuf_delete(upcall->packet);
1766 ofproto->n_matches++;
1770 /* Check with in-band control to see if this packet should be sent
1771 * to the local port regardless of the flow table. */
1772 if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
1773 send_packet(ofproto, OVSP_LOCAL, upcall->packet);
1776 facet = facet_lookup_valid(ofproto, &flow);
1778 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow, 0);
1780 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
1781 struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
1783 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
1784 COVERAGE_INC(ofproto_dpif_no_packet_in);
1785 /* XXX install 'drop' flow entry */
1786 ofpbuf_delete(upcall->packet);
1790 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
1794 send_packet_in(ofproto, upcall, &flow, false);
1798 facet = facet_create(rule, &flow, upcall->packet);
1799 } else if (!facet->may_install) {
1800 /* The facet is not installable, that is, we need to process every
1801 * packet, so process the current packet's actions into 'facet'. */
1802 facet_make_actions(ofproto, facet, upcall->packet);
1805 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
1807 * Extra-special case for fail-open mode.
1809 * We are in fail-open mode and the packet matched the fail-open rule,
1810 * but we are connected to a controller too. We should send the packet
1811 * up to the controller in the hope that it will try to set up a flow
1812 * and thereby allow us to exit fail-open.
1814 * See the top-level comment in fail-open.c for more information.
1816 send_packet_in(ofproto, upcall, &flow, true);
1819 facet_execute(ofproto, facet, upcall->packet);
1820 facet_install(ofproto, facet, false);
1821 ofproto->n_matches++;
1825 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1829 switch (upcall->type) {
1830 case DPIF_UC_ACTION:
1831 COVERAGE_INC(ofproto_dpif_ctlr_action);
1832 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1833 send_packet_in(ofproto, upcall, &flow, false);
1836 case DPIF_UC_SAMPLE:
1837 if (ofproto->sflow) {
1838 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1839 dpif_sflow_received(ofproto->sflow, upcall, &flow);
1841 ofpbuf_delete(upcall->packet);
1845 handle_miss_upcall(ofproto, upcall);
1848 case DPIF_N_UC_TYPES:
1850 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
1855 /* Flow expiration. */
1857 static int facet_max_idle(const struct ofproto_dpif *);
1858 static void update_stats(struct ofproto_dpif *);
1859 static void rule_expire(struct rule_dpif *);
1860 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
1862 /* This function is called periodically by run(). Its job is to collect
1863 * updates for the flows that have been installed into the datapath, most
1864 * importantly when they last were used, and then use that information to
1865 * expire flows that have not been used recently.
1867 * Returns the number of milliseconds after which it should be called again. */
1869 expire(struct ofproto_dpif *ofproto)
1871 struct rule_dpif *rule, *next_rule;
1872 struct classifier *table;
1875 /* Update stats for each flow in the datapath. */
1876 update_stats(ofproto);
1878 /* Expire facets that have been idle too long. */
1879 dp_max_idle = facet_max_idle(ofproto);
1880 expire_facets(ofproto, dp_max_idle);
1882 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
1883 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1884 struct cls_cursor cursor;
1886 cls_cursor_init(&cursor, table, NULL);
1887 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1892 /* All outstanding data in existing flows has been accounted, so it's a
1893 * good time to do bond rebalancing. */
1894 if (ofproto->has_bonded_bundles) {
1895 struct ofbundle *bundle;
1897 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1899 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
1904 return MIN(dp_max_idle, 1000);
1907 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
1909 * This function also pushes statistics updates to rules which each facet
1910 * resubmits into. Generally these statistics will be accurate. However, if a
1911 * facet changes the rule it resubmits into at some time in between
1912 * update_stats() runs, it is possible that statistics accrued to the
1913 * old rule will be incorrectly attributed to the new rule. This could be
1914 * avoided by calling update_stats() whenever rules are created or
1915 * deleted. However, the performance impact of making so many calls to the
1916 * datapath do not justify the benefit of having perfectly accurate statistics.
1919 update_stats(struct ofproto_dpif *p)
1921 const struct dpif_flow_stats *stats;
1922 struct dpif_flow_dump dump;
1923 const struct nlattr *key;
1926 dpif_flow_dump_start(&dump, p->dpif);
1927 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
1928 struct facet *facet;
1931 if (odp_flow_key_to_flow(key, key_len, &flow)) {
1935 odp_flow_key_format(key, key_len, &s);
1936 VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
1942 facet = facet_find(p, &flow);
1944 if (facet && facet->installed) {
1946 if (stats->n_packets >= facet->dp_packet_count) {
1947 uint64_t extra = stats->n_packets - facet->dp_packet_count;
1948 facet->packet_count += extra;
1950 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
1953 if (stats->n_bytes >= facet->dp_byte_count) {
1954 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
1956 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
1959 facet->dp_packet_count = stats->n_packets;
1960 facet->dp_byte_count = stats->n_bytes;
1962 facet_update_time(p, facet, stats->used);
1963 facet_account(p, facet);
1964 facet_push_stats(facet);
1966 /* There's a flow in the datapath that we know nothing about.
1968 COVERAGE_INC(facet_unexpected);
1969 dpif_flow_del(p->dpif, key, key_len, NULL);
1972 dpif_flow_dump_done(&dump);
1975 /* Calculates and returns the number of milliseconds of idle time after which
1976 * facets should expire from the datapath and we should fold their statistics
1977 * into their parent rules in userspace. */
1979 facet_max_idle(const struct ofproto_dpif *ofproto)
1982 * Idle time histogram.
1984 * Most of the time a switch has a relatively small number of facets. When
1985 * this is the case we might as well keep statistics for all of them in
1986 * userspace and to cache them in the kernel datapath for performance as
1989 * As the number of facets increases, the memory required to maintain
1990 * statistics about them in userspace and in the kernel becomes
1991 * significant. However, with a large number of facets it is likely that
1992 * only a few of them are "heavy hitters" that consume a large amount of
1993 * bandwidth. At this point, only heavy hitters are worth caching in the
1994 * kernel and maintaining in userspaces; other facets we can discard.
1996 * The technique used to compute the idle time is to build a histogram with
1997 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
1998 * that is installed in the kernel gets dropped in the appropriate bucket.
1999 * After the histogram has been built, we compute the cutoff so that only
2000 * the most-recently-used 1% of facets (but at least
2001 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2002 * the most-recently-used bucket of facets is kept, so actually an
2003 * arbitrary number of facets can be kept in any given expiration run
2004 * (though the next run will delete most of those unless they receive
2007 * This requires a second pass through the facets, in addition to the pass
2008 * made by update_stats(), because the former function never looks
2009 * at uninstallable facets.
2011 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2012 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2013 int buckets[N_BUCKETS] = { 0 };
2014 int total, subtotal, bucket;
2015 struct facet *facet;
2019 total = hmap_count(&ofproto->facets);
2020 if (total <= ofproto->up.flow_eviction_threshold) {
2021 return N_BUCKETS * BUCKET_WIDTH;
2024 /* Build histogram. */
2026 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
2027 long long int idle = now - facet->used;
2028 int bucket = (idle <= 0 ? 0
2029 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2030 : (unsigned int) idle / BUCKET_WIDTH);
2034 /* Find the first bucket whose flows should be expired. */
2035 subtotal = bucket = 0;
2037 subtotal += buckets[bucket++];
2038 } while (bucket < N_BUCKETS &&
2039 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2041 if (VLOG_IS_DBG_ENABLED()) {
2045 ds_put_cstr(&s, "keep");
2046 for (i = 0; i < N_BUCKETS; i++) {
2048 ds_put_cstr(&s, ", drop");
2051 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2054 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2058 return bucket * BUCKET_WIDTH;
2062 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
2064 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
2065 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
2066 struct ofexpired expired;
2068 if (facet->installed) {
2069 struct dpif_flow_stats stats;
2071 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
2073 facet_update_stats(ofproto, facet, &stats);
2076 expired.flow = facet->flow;
2077 expired.packet_count = facet->packet_count;
2078 expired.byte_count = facet->byte_count;
2079 expired.used = facet->used;
2080 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2085 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
2087 long long int cutoff = time_msec() - dp_max_idle;
2088 struct facet *facet, *next_facet;
2090 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
2091 facet_active_timeout(ofproto, facet);
2092 if (facet->used < cutoff) {
2093 facet_remove(ofproto, facet);
2098 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2099 * then delete it entirely. */
2101 rule_expire(struct rule_dpif *rule)
2103 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2104 struct facet *facet, *next_facet;
2108 /* Has 'rule' expired? */
2110 if (rule->up.hard_timeout
2111 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
2112 reason = OFPRR_HARD_TIMEOUT;
2113 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2114 && now > rule->used + rule->up.idle_timeout * 1000) {
2115 reason = OFPRR_IDLE_TIMEOUT;
2120 COVERAGE_INC(ofproto_dpif_expired);
2122 /* Update stats. (This is a no-op if the rule expired due to an idle
2123 * timeout, because that only happens when the rule has no facets left.) */
2124 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2125 facet_remove(ofproto, facet);
2128 /* Get rid of the rule. */
2129 ofproto_rule_expire(&rule->up, reason);
2134 /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
2135 * example 'packet' within that flow.
2137 * The caller must already have determined that no facet with an identical
2138 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2139 * the ofproto's classifier table. */
2140 static struct facet *
2141 facet_create(struct rule_dpif *rule, const struct flow *flow,
2142 const struct ofpbuf *packet)
2144 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2145 struct facet *facet;
2147 facet = xzalloc(sizeof *facet);
2148 facet->used = time_msec();
2149 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2150 list_push_back(&rule->facets, &facet->list_node);
2152 facet->flow = *flow;
2153 netflow_flow_init(&facet->nf_flow);
2154 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2156 facet_make_actions(ofproto, facet, packet);
2162 facet_free(struct facet *facet)
2164 free(facet->actions);
2168 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2169 * 'packet', which arrived on 'in_port'.
2171 * Takes ownership of 'packet'. */
2173 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2174 const struct nlattr *odp_actions, size_t actions_len,
2175 struct ofpbuf *packet)
2177 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2178 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE) {
2179 /* As an optimization, avoid a round-trip from userspace to kernel to
2180 * userspace. This also avoids possibly filling up kernel packet
2181 * buffers along the way. */
2182 struct dpif_upcall upcall;
2184 upcall.type = DPIF_UC_ACTION;
2185 upcall.packet = packet;
2188 upcall.userdata = nl_attr_get_u64(odp_actions);
2189 upcall.sample_pool = 0;
2190 upcall.actions = NULL;
2191 upcall.actions_len = 0;
2193 send_packet_in(ofproto, &upcall, flow, false);
2197 struct odputil_keybuf keybuf;
2201 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2202 odp_flow_key_from_flow(&key, flow);
2204 error = dpif_execute(ofproto->dpif, key.data, key.size,
2205 odp_actions, actions_len, packet);
2207 ofpbuf_delete(packet);
2212 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2213 * statistics appropriately. 'packet' must have at least sizeof(struct
2214 * ofp_packet_in) bytes of headroom.
2216 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2217 * applying flow_extract() to 'packet' would yield the same flow as
2220 * 'facet' must have accurately composed datapath actions; that is, it must
2221 * not be in need of revalidation.
2223 * Takes ownership of 'packet'. */
2225 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2226 struct ofpbuf *packet)
2228 struct dpif_flow_stats stats;
2230 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2232 flow_extract_stats(&facet->flow, packet, &stats);
2233 stats.used = time_msec();
2234 if (execute_odp_actions(ofproto, &facet->flow,
2235 facet->actions, facet->actions_len, packet)) {
2236 facet_update_stats(ofproto, facet, &stats);
2240 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2242 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2243 * rule's statistics, via facet_uninstall().
2245 * - Removes 'facet' from its rule and from ofproto->facets.
2248 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2250 facet_uninstall(ofproto, facet);
2251 facet_flush_stats(ofproto, facet);
2252 hmap_remove(&ofproto->facets, &facet->hmap_node);
2253 list_remove(&facet->list_node);
2257 /* Composes the datapath actions for 'facet' based on its rule's actions. */
2259 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2260 const struct ofpbuf *packet)
2262 const struct rule_dpif *rule = facet->rule;
2263 struct ofpbuf *odp_actions;
2264 struct action_xlate_ctx ctx;
2266 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2267 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2268 facet->tags = ctx.tags;
2269 facet->may_install = ctx.may_set_up_flow;
2270 facet->has_learn = ctx.has_learn;
2271 facet->has_normal = ctx.has_normal;
2272 facet->nf_flow.output_iface = ctx.nf_output_iface;
2274 if (facet->actions_len != odp_actions->size
2275 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2276 free(facet->actions);
2277 facet->actions_len = odp_actions->size;
2278 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2281 ofpbuf_delete(odp_actions);
2284 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2285 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2286 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2287 * since 'facet' was last updated.
2289 * Returns 0 if successful, otherwise a positive errno value.*/
2291 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2292 const struct nlattr *actions, size_t actions_len,
2293 struct dpif_flow_stats *stats)
2295 struct odputil_keybuf keybuf;
2296 enum dpif_flow_put_flags flags;
2300 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2302 flags |= DPIF_FP_ZERO_STATS;
2305 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2306 odp_flow_key_from_flow(&key, &facet->flow);
2308 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2309 actions, actions_len, stats);
2312 facet_reset_dp_stats(facet, stats);
2318 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2319 * 'zero_stats' is true, clears any existing statistics from the datapath for
2322 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2324 struct dpif_flow_stats stats;
2326 if (facet->may_install
2327 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2328 zero_stats ? &stats : NULL)) {
2329 facet->installed = true;
2334 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
2337 const struct nlattr *a;
2341 if (facet->byte_count <= facet->accounted_bytes) {
2344 n_bytes = facet->byte_count - facet->accounted_bytes;
2345 facet->accounted_bytes = facet->byte_count;
2347 /* Feed information from the active flows back into the learning table to
2348 * ensure that table is always in sync with what is actually flowing
2349 * through the datapath. */
2350 if (facet->has_learn || facet->has_normal) {
2351 struct action_xlate_ctx ctx;
2353 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2354 ctx.may_learn = true;
2355 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
2356 facet->rule->up.n_actions));
2359 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
2363 /* This loop feeds byte counters to bond_account() for rebalancing to use
2364 * as a basis. We also need to track the actual VLAN on which the packet
2365 * is going to be sent to ensure that it matches the one passed to
2366 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2368 vlan_tci = facet->flow.vlan_tci;
2369 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2370 struct ofport_dpif *port;
2372 switch (nl_attr_type(a)) {
2373 case OVS_ACTION_ATTR_OUTPUT:
2374 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2375 if (port && port->bundle && port->bundle->bond) {
2376 bond_account(port->bundle->bond, &facet->flow,
2377 vlan_tci_to_vid(vlan_tci), n_bytes);
2381 case OVS_ACTION_ATTR_POP_VLAN:
2382 vlan_tci = htons(0);
2385 case OVS_ACTION_ATTR_PUSH_VLAN:
2386 vlan_tci = nl_attr_get_be16(a);
2392 /* If 'rule' is installed in the datapath, uninstalls it. */
2394 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2396 if (facet->installed) {
2397 struct odputil_keybuf keybuf;
2398 struct dpif_flow_stats stats;
2402 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2403 odp_flow_key_from_flow(&key, &facet->flow);
2405 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2406 facet_reset_dp_stats(facet, &stats);
2408 facet_update_stats(p, facet, &stats);
2410 facet->installed = false;
2412 assert(facet->dp_packet_count == 0);
2413 assert(facet->dp_byte_count == 0);
2417 /* Returns true if the only action for 'facet' is to send to the controller.
2418 * (We don't report NetFlow expiration messages for such facets because they
2419 * are just part of the control logic for the network, not real traffic). */
2421 facet_is_controller_flow(struct facet *facet)
2424 && facet->rule->up.n_actions == 1
2425 && action_outputs_to_port(&facet->rule->up.actions[0],
2426 htons(OFPP_CONTROLLER)));
2429 /* Resets 'facet''s datapath statistics counters. This should be called when
2430 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2431 * it should contain the statistics returned by dpif when 'facet' was reset in
2432 * the datapath. 'stats' will be modified to only included statistics new
2433 * since 'facet' was last updated. */
2435 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2437 if (stats && facet->dp_packet_count <= stats->n_packets
2438 && facet->dp_byte_count <= stats->n_bytes) {
2439 stats->n_packets -= facet->dp_packet_count;
2440 stats->n_bytes -= facet->dp_byte_count;
2443 facet->dp_packet_count = 0;
2444 facet->dp_byte_count = 0;
2447 /* Folds all of 'facet''s statistics into its rule. Also updates the
2448 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2449 * 'facet''s statistics in the datapath should have been zeroed and folded into
2450 * its packet and byte counts before this function is called. */
2452 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2454 assert(!facet->dp_byte_count);
2455 assert(!facet->dp_packet_count);
2457 facet_push_stats(facet);
2458 facet_account(ofproto, facet);
2460 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2461 struct ofexpired expired;
2462 expired.flow = facet->flow;
2463 expired.packet_count = facet->packet_count;
2464 expired.byte_count = facet->byte_count;
2465 expired.used = facet->used;
2466 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2469 facet->rule->packet_count += facet->packet_count;
2470 facet->rule->byte_count += facet->byte_count;
2472 /* Reset counters to prevent double counting if 'facet' ever gets
2474 facet_reset_counters(facet);
2476 netflow_flow_clear(&facet->nf_flow);
2479 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2480 * Returns it if found, otherwise a null pointer.
2482 * The returned facet might need revalidation; use facet_lookup_valid()
2483 * instead if that is important. */
2484 static struct facet *
2485 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
2487 struct facet *facet;
2489 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2491 if (flow_equal(flow, &facet->flow)) {
2499 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2500 * Returns it if found, otherwise a null pointer.
2502 * The returned facet is guaranteed to be valid. */
2503 static struct facet *
2504 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
2506 struct facet *facet = facet_find(ofproto, flow);
2508 /* The facet we found might not be valid, since we could be in need of
2509 * revalidation. If it is not valid, don't return it. */
2511 && ofproto->need_revalidate
2512 && !facet_revalidate(ofproto, facet)) {
2513 COVERAGE_INC(facet_invalidated);
2520 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2522 * - If the rule found is different from 'facet''s current rule, moves
2523 * 'facet' to the new rule and recompiles its actions.
2525 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2526 * where it is and recompiles its actions anyway.
2528 * - If there is none, destroys 'facet'.
2530 * Returns true if 'facet' still exists, false if it has been destroyed. */
2532 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
2534 struct action_xlate_ctx ctx;
2535 struct ofpbuf *odp_actions;
2536 struct rule_dpif *new_rule;
2537 bool actions_changed;
2539 COVERAGE_INC(facet_revalidate);
2541 /* Determine the new rule. */
2542 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
2544 /* No new rule, so delete the facet. */
2545 facet_remove(ofproto, facet);
2549 /* Calculate new datapath actions.
2551 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2552 * emit a NetFlow expiration and, if so, we need to have the old state
2553 * around to properly compose it. */
2554 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2555 odp_actions = xlate_actions(&ctx,
2556 new_rule->up.actions, new_rule->up.n_actions);
2557 actions_changed = (facet->actions_len != odp_actions->size
2558 || memcmp(facet->actions, odp_actions->data,
2559 facet->actions_len));
2561 /* If the datapath actions changed or the installability changed,
2562 * then we need to talk to the datapath. */
2563 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2564 if (ctx.may_set_up_flow) {
2565 struct dpif_flow_stats stats;
2567 facet_put__(ofproto, facet,
2568 odp_actions->data, odp_actions->size, &stats);
2569 facet_update_stats(ofproto, facet, &stats);
2571 facet_uninstall(ofproto, facet);
2574 /* The datapath flow is gone or has zeroed stats, so push stats out of
2575 * 'facet' into 'rule'. */
2576 facet_flush_stats(ofproto, facet);
2579 /* Update 'facet' now that we've taken care of all the old state. */
2580 facet->tags = ctx.tags;
2581 facet->nf_flow.output_iface = ctx.nf_output_iface;
2582 facet->may_install = ctx.may_set_up_flow;
2583 facet->has_learn = ctx.has_learn;
2584 facet->has_normal = ctx.has_normal;
2585 if (actions_changed) {
2586 free(facet->actions);
2587 facet->actions_len = odp_actions->size;
2588 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2590 if (facet->rule != new_rule) {
2591 COVERAGE_INC(facet_changed_rule);
2592 list_remove(&facet->list_node);
2593 list_push_back(&new_rule->facets, &facet->list_node);
2594 facet->rule = new_rule;
2595 facet->used = new_rule->up.created;
2596 facet->rs_used = facet->used;
2599 ofpbuf_delete(odp_actions);
2604 /* Updates 'facet''s used time. Caller is responsible for calling
2605 * facet_push_stats() to update the flows which 'facet' resubmits into. */
2607 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
2610 if (used > facet->used) {
2612 if (used > facet->rule->used) {
2613 facet->rule->used = used;
2615 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
2619 /* Folds the statistics from 'stats' into the counters in 'facet'.
2621 * Because of the meaning of a facet's counters, it only makes sense to do this
2622 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
2623 * packet that was sent by hand or if it represents statistics that have been
2624 * cleared out of the datapath. */
2626 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
2627 const struct dpif_flow_stats *stats)
2629 if (stats->n_packets || stats->used > facet->used) {
2630 facet_update_time(ofproto, facet, stats->used);
2631 facet->packet_count += stats->n_packets;
2632 facet->byte_count += stats->n_bytes;
2633 facet_push_stats(facet);
2634 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
2639 facet_reset_counters(struct facet *facet)
2641 facet->packet_count = 0;
2642 facet->byte_count = 0;
2643 facet->rs_packet_count = 0;
2644 facet->rs_byte_count = 0;
2645 facet->accounted_bytes = 0;
2649 facet_push_stats(struct facet *facet)
2651 uint64_t rs_packets, rs_bytes;
2653 assert(facet->packet_count >= facet->rs_packet_count);
2654 assert(facet->byte_count >= facet->rs_byte_count);
2655 assert(facet->used >= facet->rs_used);
2657 rs_packets = facet->packet_count - facet->rs_packet_count;
2658 rs_bytes = facet->byte_count - facet->rs_byte_count;
2660 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
2661 facet->rs_packet_count = facet->packet_count;
2662 facet->rs_byte_count = facet->byte_count;
2663 facet->rs_used = facet->used;
2665 flow_push_stats(facet->rule, &facet->flow,
2666 rs_packets, rs_bytes, facet->used);
2670 struct ofproto_push {
2671 struct action_xlate_ctx ctx;
2678 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
2680 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
2683 rule->packet_count += push->packets;
2684 rule->byte_count += push->bytes;
2685 rule->used = MAX(push->used, rule->used);
2689 /* Pushes flow statistics to the rules which 'flow' resubmits into given
2690 * 'rule''s actions. */
2692 flow_push_stats(const struct rule_dpif *rule,
2693 struct flow *flow, uint64_t packets, uint64_t bytes,
2696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2697 struct ofproto_push push;
2699 push.packets = packets;
2703 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
2704 push.ctx.resubmit_hook = push_resubmit;
2705 ofpbuf_delete(xlate_actions(&push.ctx,
2706 rule->up.actions, rule->up.n_actions));
2711 static struct rule_dpif *
2712 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
2715 if (table_id >= N_TABLES) {
2719 return rule_dpif_cast(rule_from_cls_rule(
2720 classifier_lookup(&ofproto->up.tables[table_id],
2725 complete_operation(struct rule_dpif *rule)
2727 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2729 rule_invalidate(rule);
2731 struct dpif_completion *c = xmalloc(sizeof *c);
2732 c->op = rule->up.pending;
2733 list_push_back(&ofproto->completions, &c->list_node);
2735 ofoperation_complete(rule->up.pending, 0);
2739 static struct rule *
2742 struct rule_dpif *rule = xmalloc(sizeof *rule);
2747 rule_dealloc(struct rule *rule_)
2749 struct rule_dpif *rule = rule_dpif_cast(rule_);
2754 rule_construct(struct rule *rule_)
2756 struct rule_dpif *rule = rule_dpif_cast(rule_);
2757 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2758 struct rule_dpif *victim;
2762 error = validate_actions(rule->up.actions, rule->up.n_actions,
2763 &rule->up.cr.flow, ofproto->max_ports);
2768 rule->used = rule->up.created;
2769 rule->packet_count = 0;
2770 rule->byte_count = 0;
2772 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
2773 if (victim && !list_is_empty(&victim->facets)) {
2774 struct facet *facet;
2776 rule->facets = victim->facets;
2777 list_moved(&rule->facets);
2778 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2779 /* XXX: We're only clearing our local counters here. It's possible
2780 * that quite a few packets are unaccounted for in the datapath
2781 * statistics. These will be accounted to the new rule instead of
2782 * cleared as required. This could be fixed by clearing out the
2783 * datapath statistics for this facet, but currently it doesn't
2785 facet_reset_counters(facet);
2789 /* Must avoid list_moved() in this case. */
2790 list_init(&rule->facets);
2793 table_id = rule->up.table_id;
2794 rule->tag = (victim ? victim->tag
2796 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
2797 ofproto->tables[table_id].basis));
2799 complete_operation(rule);
2804 rule_destruct(struct rule *rule_)
2806 struct rule_dpif *rule = rule_dpif_cast(rule_);
2807 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2808 struct facet *facet, *next_facet;
2810 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2811 facet_revalidate(ofproto, facet);
2814 complete_operation(rule);
2818 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
2820 struct rule_dpif *rule = rule_dpif_cast(rule_);
2821 struct facet *facet;
2823 /* Start from historical data for 'rule' itself that are no longer tracked
2824 * in facets. This counts, for example, facets that have expired. */
2825 *packets = rule->packet_count;
2826 *bytes = rule->byte_count;
2828 /* Add any statistics that are tracked by facets. This includes
2829 * statistical data recently updated by ofproto_update_stats() as well as
2830 * stats for packets that were executed "by hand" via dpif_execute(). */
2831 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2832 *packets += facet->packet_count;
2833 *bytes += facet->byte_count;
2838 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
2840 struct rule_dpif *rule = rule_dpif_cast(rule_);
2841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2842 struct action_xlate_ctx ctx;
2843 struct ofpbuf *odp_actions;
2844 struct facet *facet;
2847 /* First look for a related facet. If we find one, account it to that. */
2848 facet = facet_lookup_valid(ofproto, flow);
2849 if (facet && facet->rule == rule) {
2850 facet_execute(ofproto, facet, packet);
2854 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2855 * create a new facet for it and use that. */
2856 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
2857 facet = facet_create(rule, flow, packet);
2858 facet_execute(ofproto, facet, packet);
2859 facet_install(ofproto, facet, true);
2863 /* We can't account anything to a facet. If we were to try, then that
2864 * facet would have a non-matching rule, busting our invariants. */
2865 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
2866 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2867 size = packet->size;
2868 if (execute_odp_actions(ofproto, flow, odp_actions->data,
2869 odp_actions->size, packet)) {
2870 rule->used = time_msec();
2871 rule->packet_count++;
2872 rule->byte_count += size;
2873 flow_push_stats(rule, flow, 1, size, rule->used);
2875 ofpbuf_delete(odp_actions);
2881 rule_modify_actions(struct rule *rule_)
2883 struct rule_dpif *rule = rule_dpif_cast(rule_);
2884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2887 error = validate_actions(rule->up.actions, rule->up.n_actions,
2888 &rule->up.cr.flow, ofproto->max_ports);
2890 ofoperation_complete(rule->up.pending, error);
2894 complete_operation(rule);
2897 /* Sends 'packet' out of port 'odp_port' within 'p'.
2898 * Returns 0 if successful, otherwise a positive errno value. */
2900 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
2901 const struct ofpbuf *packet)
2903 struct ofpbuf key, odp_actions;
2904 struct odputil_keybuf keybuf;
2908 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
2909 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2910 odp_flow_key_from_flow(&key, &flow);
2912 ofpbuf_init(&odp_actions, 32);
2913 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
2914 error = dpif_execute(ofproto->dpif,
2916 odp_actions.data, odp_actions.size,
2918 ofpbuf_uninit(&odp_actions);
2921 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
2922 ofproto->up.name, odp_port, strerror(error));
2927 /* OpenFlow to datapath action translation. */
2929 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2930 struct action_xlate_ctx *ctx);
2931 static void xlate_normal(struct action_xlate_ctx *);
2934 commit_vlan_tci(struct action_xlate_ctx *ctx, ovs_be16 vlan_tci)
2936 struct flow *base = &ctx->base_flow;
2937 struct ofpbuf *odp_actions = ctx->odp_actions;
2939 if (base->vlan_tci != vlan_tci) {
2940 if (!(vlan_tci & htons(VLAN_CFI))) {
2941 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
2943 if (base->vlan_tci != htons(0)) {
2944 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN);
2946 nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
2947 vlan_tci & ~htons(VLAN_CFI));
2949 base->vlan_tci = vlan_tci;
2954 commit_odp_actions(struct action_xlate_ctx *ctx)
2956 const struct flow *flow = &ctx->flow;
2957 struct flow *base = &ctx->base_flow;
2958 struct ofpbuf *odp_actions = ctx->odp_actions;
2960 if (base->tun_id != flow->tun_id) {
2961 nl_msg_put_be64(odp_actions, OVS_ACTION_ATTR_SET_TUNNEL, flow->tun_id);
2962 base->tun_id = flow->tun_id;
2965 if (base->nw_src != flow->nw_src) {
2966 nl_msg_put_be32(odp_actions, OVS_ACTION_ATTR_SET_NW_SRC, flow->nw_src);
2967 base->nw_src = flow->nw_src;
2970 if (base->nw_dst != flow->nw_dst) {
2971 nl_msg_put_be32(odp_actions, OVS_ACTION_ATTR_SET_NW_DST, flow->nw_dst);
2972 base->nw_dst = flow->nw_dst;
2975 if (base->nw_tos != flow->nw_tos) {
2976 nl_msg_put_u8(odp_actions, OVS_ACTION_ATTR_SET_NW_TOS, flow->nw_tos);
2977 base->nw_tos = flow->nw_tos;
2980 commit_vlan_tci(ctx, flow->vlan_tci);
2982 if (base->tp_src != flow->tp_src) {
2983 nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_SET_TP_SRC, flow->tp_src);
2984 base->tp_src = flow->tp_src;
2987 if (base->tp_dst != flow->tp_dst) {
2988 nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_SET_TP_DST, flow->tp_dst);
2989 base->tp_dst = flow->tp_dst;
2992 if (!eth_addr_equals(base->dl_src, flow->dl_src)) {
2993 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_SET_DL_SRC,
2994 flow->dl_src, ETH_ADDR_LEN);
2995 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
2998 if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) {
2999 nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_SET_DL_DST,
3000 flow->dl_dst, ETH_ADDR_LEN);
3001 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
3004 if (ctx->base_priority != ctx->priority) {
3005 if (ctx->priority) {
3006 nl_msg_put_u32(odp_actions, OVS_ACTION_ATTR_SET_PRIORITY,
3009 nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_PRIORITY);
3011 ctx->base_priority = ctx->priority;
3016 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3018 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3019 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
3022 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
3023 /* Forwarding disabled on port. */
3028 * We don't have an ofport record for this port, but it doesn't hurt to
3029 * allow forwarding to it anyhow. Maybe such a port will appear later
3030 * and we're pre-populating the flow table.
3034 commit_odp_actions(ctx);
3035 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3036 ctx->nf_output_iface = ofp_port;
3040 xlate_table_action(struct action_xlate_ctx *ctx,
3041 uint16_t in_port, uint8_t table_id)
3043 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
3044 struct ofproto_dpif *ofproto = ctx->ofproto;
3045 struct rule_dpif *rule;
3046 uint16_t old_in_port;
3047 uint8_t old_table_id;
3049 old_table_id = ctx->table_id;
3050 ctx->table_id = table_id;
3052 /* Look up a flow with 'in_port' as the input port. */
3053 old_in_port = ctx->flow.in_port;
3054 ctx->flow.in_port = in_port;
3055 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
3058 if (table_id > 0 && table_id < N_TABLES) {
3059 struct table_dpif *table = &ofproto->tables[table_id];
3060 if (table->other_table) {
3063 : rule_calculate_tag(&ctx->flow,
3064 &table->other_table->wc,
3069 /* Restore the original input port. Otherwise OFPP_NORMAL and
3070 * OFPP_IN_PORT will have surprising behavior. */
3071 ctx->flow.in_port = old_in_port;
3073 if (ctx->resubmit_hook) {
3074 ctx->resubmit_hook(ctx, rule);
3079 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
3083 ctx->table_id = old_table_id;
3085 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3087 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
3088 MAX_RESUBMIT_RECURSION);
3093 xlate_resubmit_table(struct action_xlate_ctx *ctx,
3094 const struct nx_action_resubmit *nar)
3099 in_port = (nar->in_port == htons(OFPP_IN_PORT)
3101 : ntohs(nar->in_port));
3102 table_id = nar->table == 255 ? ctx->table_id : nar->table;
3104 xlate_table_action(ctx, in_port, table_id);
3108 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
3110 struct ofport_dpif *ofport;
3112 commit_odp_actions(ctx);
3113 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
3114 uint16_t ofp_port = ofport->up.ofp_port;
3115 if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) {
3116 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT,
3121 ctx->nf_output_iface = NF_OUT_FLOOD;
3125 xlate_output_action__(struct action_xlate_ctx *ctx,
3126 uint16_t port, uint16_t max_len)
3128 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
3130 ctx->nf_output_iface = NF_OUT_DROP;
3134 add_output_action(ctx, ctx->flow.in_port);
3137 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
3143 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
3146 flood_packets(ctx, htonl(0));
3148 case OFPP_CONTROLLER:
3149 commit_odp_actions(ctx);
3150 nl_msg_put_u64(ctx->odp_actions, OVS_ACTION_ATTR_USERSPACE, max_len);
3153 add_output_action(ctx, OFPP_LOCAL);
3158 if (port != ctx->flow.in_port) {
3159 add_output_action(ctx, port);
3164 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3165 ctx->nf_output_iface = NF_OUT_FLOOD;
3166 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3167 ctx->nf_output_iface = prev_nf_output_iface;
3168 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3169 ctx->nf_output_iface != NF_OUT_FLOOD) {
3170 ctx->nf_output_iface = NF_OUT_MULTI;
3175 xlate_output_reg_action(struct action_xlate_ctx *ctx,
3176 const struct nx_action_output_reg *naor)
3180 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
3182 if (ofp_port <= UINT16_MAX) {
3183 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
3188 xlate_output_action(struct action_xlate_ctx *ctx,
3189 const struct ofp_action_output *oao)
3191 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3195 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3196 const struct ofp_action_enqueue *oae)
3198 uint16_t ofp_port, odp_port;
3199 uint32_t ctx_priority, priority;
3202 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3205 /* Fall back to ordinary output action. */
3206 xlate_output_action__(ctx, ntohs(oae->port), 0);
3210 /* Figure out datapath output port. */
3211 ofp_port = ntohs(oae->port);
3212 if (ofp_port == OFPP_IN_PORT) {
3213 ofp_port = ctx->flow.in_port;
3215 odp_port = ofp_port_to_odp_port(ofp_port);
3217 /* Add datapath actions. */
3218 ctx_priority = ctx->priority;
3219 ctx->priority = priority;
3220 add_output_action(ctx, odp_port);
3221 ctx->priority = ctx_priority;
3223 /* Update NetFlow output port. */
3224 if (ctx->nf_output_iface == NF_OUT_DROP) {
3225 ctx->nf_output_iface = odp_port;
3226 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3227 ctx->nf_output_iface = NF_OUT_MULTI;
3232 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3233 const struct nx_action_set_queue *nasq)
3238 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3241 /* Couldn't translate queue to a priority, so ignore. A warning
3242 * has already been logged. */
3246 ctx->priority = priority;
3249 struct xlate_reg_state {
3255 xlate_autopath(struct action_xlate_ctx *ctx,
3256 const struct nx_action_autopath *naa)
3258 uint16_t ofp_port = ntohl(naa->id);
3259 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
3261 if (!port || !port->bundle) {
3262 ofp_port = OFPP_NONE;
3263 } else if (port->bundle->bond) {
3264 /* Autopath does not support VLAN hashing. */
3265 struct ofport_dpif *slave = bond_choose_output_slave(
3266 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
3268 ofp_port = slave->up.ofp_port;
3271 autopath_execute(naa, &ctx->flow, ofp_port);
3275 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
3277 struct ofproto_dpif *ofproto = ofproto_;
3278 struct ofport_dpif *port;
3288 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3291 port = get_ofp_port(ofproto, ofp_port);
3292 return port ? port->may_enable : false;
3297 xlate_learn_action(struct action_xlate_ctx *ctx,
3298 const struct nx_action_learn *learn)
3300 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
3301 struct ofputil_flow_mod fm;
3304 learn_execute(learn, &ctx->flow, &fm);
3306 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
3307 if (error && !VLOG_DROP_WARN(&rl)) {
3308 char *msg = ofputil_error_to_string(error);
3309 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
3317 do_xlate_actions(const union ofp_action *in, size_t n_in,
3318 struct action_xlate_ctx *ctx)
3320 const struct ofport_dpif *port;
3321 const union ofp_action *ia;
3324 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
3326 && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3327 port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3328 ? htonl(OFPPC_NO_RECV_STP)
3329 : htonl(OFPPC_NO_RECV))) {
3330 /* Drop this flow. */
3334 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
3335 const struct ofp_action_dl_addr *oada;
3336 const struct nx_action_resubmit *nar;
3337 const struct nx_action_set_tunnel *nast;
3338 const struct nx_action_set_queue *nasq;
3339 const struct nx_action_multipath *nam;
3340 const struct nx_action_autopath *naa;
3341 const struct nx_action_bundle *nab;
3342 const struct nx_action_output_reg *naor;
3343 enum ofputil_action_code code;
3346 code = ofputil_decode_action_unsafe(ia);
3348 case OFPUTIL_OFPAT_OUTPUT:
3349 xlate_output_action(ctx, &ia->output);
3352 case OFPUTIL_OFPAT_SET_VLAN_VID:
3353 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3354 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3357 case OFPUTIL_OFPAT_SET_VLAN_PCP:
3358 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3359 ctx->flow.vlan_tci |= htons(
3360 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3363 case OFPUTIL_OFPAT_STRIP_VLAN:
3364 ctx->flow.vlan_tci = htons(0);
3367 case OFPUTIL_OFPAT_SET_DL_SRC:
3368 oada = ((struct ofp_action_dl_addr *) ia);
3369 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3372 case OFPUTIL_OFPAT_SET_DL_DST:
3373 oada = ((struct ofp_action_dl_addr *) ia);
3374 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3377 case OFPUTIL_OFPAT_SET_NW_SRC:
3378 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3381 case OFPUTIL_OFPAT_SET_NW_DST:
3382 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3385 case OFPUTIL_OFPAT_SET_NW_TOS:
3386 ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK;
3389 case OFPUTIL_OFPAT_SET_TP_SRC:
3390 ctx->flow.tp_src = ia->tp_port.tp_port;
3393 case OFPUTIL_OFPAT_SET_TP_DST:
3394 ctx->flow.tp_dst = ia->tp_port.tp_port;
3397 case OFPUTIL_OFPAT_ENQUEUE:
3398 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3401 case OFPUTIL_NXAST_RESUBMIT:
3402 nar = (const struct nx_action_resubmit *) ia;
3403 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
3406 case OFPUTIL_NXAST_RESUBMIT_TABLE:
3407 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
3410 case OFPUTIL_NXAST_SET_TUNNEL:
3411 nast = (const struct nx_action_set_tunnel *) ia;
3412 tun_id = htonll(ntohl(nast->tun_id));
3413 ctx->flow.tun_id = tun_id;
3416 case OFPUTIL_NXAST_SET_QUEUE:
3417 nasq = (const struct nx_action_set_queue *) ia;
3418 xlate_set_queue_action(ctx, nasq);
3421 case OFPUTIL_NXAST_POP_QUEUE:
3425 case OFPUTIL_NXAST_REG_MOVE:
3426 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
3430 case OFPUTIL_NXAST_REG_LOAD:
3431 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
3435 case OFPUTIL_NXAST_NOTE:
3436 /* Nothing to do. */
3439 case OFPUTIL_NXAST_SET_TUNNEL64:
3440 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
3441 ctx->flow.tun_id = tun_id;
3444 case OFPUTIL_NXAST_MULTIPATH:
3445 nam = (const struct nx_action_multipath *) ia;
3446 multipath_execute(nam, &ctx->flow);
3449 case OFPUTIL_NXAST_AUTOPATH:
3450 naa = (const struct nx_action_autopath *) ia;
3451 xlate_autopath(ctx, naa);
3454 case OFPUTIL_NXAST_BUNDLE:
3455 ctx->ofproto->has_bundle_action = true;
3456 nab = (const struct nx_action_bundle *) ia;
3457 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
3462 case OFPUTIL_NXAST_BUNDLE_LOAD:
3463 ctx->ofproto->has_bundle_action = true;
3464 nab = (const struct nx_action_bundle *) ia;
3465 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
3469 case OFPUTIL_NXAST_OUTPUT_REG:
3470 naor = (const struct nx_action_output_reg *) ia;
3471 xlate_output_reg_action(ctx, naor);
3474 case OFPUTIL_NXAST_LEARN:
3475 ctx->has_learn = true;
3476 if (ctx->may_learn) {
3477 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
3485 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3486 struct ofproto_dpif *ofproto, const struct flow *flow,
3487 const struct ofpbuf *packet)
3489 ctx->ofproto = ofproto;
3491 ctx->packet = packet;
3492 ctx->may_learn = packet != NULL;
3493 ctx->resubmit_hook = NULL;
3496 static struct ofpbuf *
3497 xlate_actions(struct action_xlate_ctx *ctx,
3498 const union ofp_action *in, size_t n_in)
3500 COVERAGE_INC(ofproto_dpif_xlate);
3502 ctx->odp_actions = ofpbuf_new(512);
3504 ctx->may_set_up_flow = true;
3505 ctx->has_learn = false;
3506 ctx->has_normal = false;
3507 ctx->nf_output_iface = NF_OUT_DROP;
3510 ctx->base_priority = 0;
3511 ctx->base_flow = ctx->flow;
3512 ctx->base_flow.tun_id = 0;
3515 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
3516 ctx->may_set_up_flow = false;
3518 do_xlate_actions(in, n_in, ctx);
3521 /* Check with in-band control to see if we're allowed to set up this
3523 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
3524 ctx->odp_actions->data,
3525 ctx->odp_actions->size)) {
3526 ctx->may_set_up_flow = false;
3529 return ctx->odp_actions;
3532 /* OFPP_NORMAL implementation. */
3535 struct ofport_dpif *port;
3540 struct dst builtin[32];
3542 size_t n, allocated;
3545 static void dst_set_init(struct dst_set *);
3546 static void dst_set_add(struct dst_set *, const struct dst *);
3547 static void dst_set_free(struct dst_set *);
3549 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
3551 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
3552 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
3553 * the bundle on which the packet was received, returns the VLAN to which the
3556 * Both 'vid' and the return value are in the range 0...4095. */
3558 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
3560 switch (in_bundle->vlan_mode) {
3561 case PORT_VLAN_ACCESS:
3562 return in_bundle->vlan;
3565 case PORT_VLAN_TRUNK:
3568 case PORT_VLAN_NATIVE_UNTAGGED:
3569 case PORT_VLAN_NATIVE_TAGGED:
3570 return vid ? vid : in_bundle->vlan;
3577 /* Given 'vlan', the VLAN that a packet belongs to, and
3578 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
3579 * that should be included in the 802.1Q header. (If the return value is 0,
3580 * then the 802.1Q header should only be included in the packet if there is a
3583 * Both 'vlan' and the return value are in the range 0...4095. */
3585 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
3587 switch (out_bundle->vlan_mode) {
3588 case PORT_VLAN_ACCESS:
3591 case PORT_VLAN_TRUNK:
3592 case PORT_VLAN_NATIVE_TAGGED:
3595 case PORT_VLAN_NATIVE_UNTAGGED:
3596 return vlan == out_bundle->vlan ? 0 : vlan;
3604 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
3605 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
3609 vlan = input_vid_to_vlan(in_bundle, vlan_tci_to_vid(ctx->flow.vlan_tci));
3610 dst->vid = output_vlan_to_vid(out_bundle, vlan);
3612 dst->port = (!out_bundle->bond
3613 ? ofbundle_get_a_port(out_bundle)
3614 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
3615 dst->vid, &ctx->tags));
3616 return dst->port != NULL;
3620 mirror_mask_ffs(mirror_mask_t mask)
3622 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
3627 dst_set_init(struct dst_set *set)
3629 set->dsts = set->builtin;
3631 set->allocated = ARRAY_SIZE(set->builtin);
3635 dst_set_add(struct dst_set *set, const struct dst *dst)
3637 if (set->n >= set->allocated) {
3638 size_t new_allocated;
3639 struct dst *new_dsts;
3641 new_allocated = set->allocated * 2;
3642 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
3643 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
3647 set->dsts = new_dsts;
3648 set->allocated = new_allocated;
3650 set->dsts[set->n++] = *dst;
3654 dst_set_free(struct dst_set *set)
3656 if (set->dsts != set->builtin) {
3662 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
3665 for (i = 0; i < set->n; i++) {
3666 if (set->dsts[i].vid == test->vid
3667 && set->dsts[i].port == test->port) {
3675 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
3677 return (bundle->vlan_mode != PORT_VLAN_ACCESS
3678 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
3682 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
3684 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
3687 /* Returns an arbitrary interface within 'bundle'. */
3688 static struct ofport_dpif *
3689 ofbundle_get_a_port(const struct ofbundle *bundle)
3691 return CONTAINER_OF(list_front(&bundle->ports),
3692 struct ofport_dpif, bundle_node);
3696 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
3697 const struct ofbundle *in_bundle,
3698 const struct ofbundle *out_bundle, struct dst_set *set)
3702 if (out_bundle == OFBUNDLE_FLOOD) {
3703 struct ofbundle *bundle;
3705 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
3706 if (bundle != in_bundle
3707 && ofbundle_includes_vlan(bundle, vlan)
3708 && bundle->floodable
3709 && !bundle->mirror_out
3710 && set_dst(ctx, &dst, in_bundle, bundle)) {
3711 dst_set_add(set, &dst);
3714 ctx->nf_output_iface = NF_OUT_FLOOD;
3715 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
3716 dst_set_add(set, &dst);
3717 ctx->nf_output_iface = dst.port->odp_port;
3722 vlan_is_mirrored(const struct ofmirror *m, int vlan)
3724 return !m->vlans || bitmap_is_set(m->vlans, vlan);
3727 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
3728 * to a VLAN. In general most packets may be mirrored but we want to drop
3729 * protocols that may confuse switches. */
3731 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
3733 /* If you change this function's behavior, please update corresponding
3734 * documentation in vswitch.xml at the same time. */
3735 if (dst[0] != 0x01) {
3736 /* All the currently banned MACs happen to start with 01 currently, so
3737 * this is a quick way to eliminate most of the good ones. */
3739 if (eth_addr_is_reserved(dst)) {
3740 /* Drop STP, IEEE pause frames, and other reserved protocols
3741 * (01-80-c2-00-00-0x). */
3745 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
3747 if ((dst[3] & 0xfe) == 0xcc &&
3748 (dst[4] & 0xfe) == 0xcc &&
3749 (dst[5] & 0xfe) == 0xcc) {
3750 /* Drop the following protocols plus others following the same
3753 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
3754 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
3755 STP Uplink Fast (01-00-0c-cd-cd-cd) */
3759 if (!(dst[3] | dst[4] | dst[5])) {
3760 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
3769 compose_mirror_dsts(struct action_xlate_ctx *ctx,
3770 uint16_t vlan, const struct ofbundle *in_bundle,
3771 struct dst_set *set)
3773 struct ofproto_dpif *ofproto = ctx->ofproto;
3774 mirror_mask_t mirrors;
3778 mirrors = in_bundle->src_mirrors;
3779 for (i = 0; i < set->n; i++) {
3780 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
3787 flow_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
3789 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
3790 if (vlan_is_mirrored(m, vlan)) {
3794 if (set_dst(ctx, &dst, in_bundle, m->out)
3795 && !dst_is_duplicate(set, &dst)) {
3796 dst_set_add(set, &dst);
3798 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
3799 struct ofbundle *bundle;
3801 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3802 if (ofbundle_includes_vlan(bundle, m->out_vlan)
3803 && set_dst(ctx, &dst, in_bundle, bundle))
3805 /* set_dst() got dst->vid from the input packet's VLAN,
3806 * not from m->out_vlan, so recompute it. */
3807 dst.vid = output_vlan_to_vid(bundle, m->out_vlan);
3809 if (dst_is_duplicate(set, &dst)) {
3813 if (bundle == in_bundle && dst.vid == flow_vid) {
3814 /* Don't send out input port on same VLAN. */
3817 dst_set_add(set, &dst);
3822 mirrors &= mirrors - 1;
3827 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
3828 const struct ofbundle *in_bundle,
3829 const struct ofbundle *out_bundle)
3831 uint16_t initial_vid, cur_vid;
3832 const struct dst *dst;
3836 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
3837 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
3843 /* Output all the packets we can without having to change the VLAN. */
3844 commit_odp_actions(ctx);
3845 initial_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
3846 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3847 if (dst->vid != initial_vid) {
3850 nl_msg_put_u32(ctx->odp_actions,
3851 OVS_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3854 /* Then output the rest. */
3855 cur_vid = initial_vid;
3856 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3857 if (dst->vid == initial_vid) {
3860 if (dst->vid != cur_vid) {
3863 tci = htons(dst->vid);
3864 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
3866 tci |= htons(VLAN_CFI);
3868 commit_vlan_tci(ctx, tci);
3872 nl_msg_put_u32(ctx->odp_actions,
3873 OVS_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3879 /* Returns the effective vlan of a packet, taking into account both the
3880 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
3881 * the packet is untagged and -1 indicates it has an invalid header and
3882 * should be dropped. */
3884 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
3885 struct ofbundle *in_bundle, bool have_packet)
3887 int vlan = vlan_tci_to_vid(flow->vlan_tci);
3889 if (in_bundle->vlan_mode == PORT_VLAN_ACCESS) {
3890 /* Drop tagged packet on access port */
3892 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3893 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3894 "packet received on port %s configured with "
3895 "implicit VLAN %"PRIu16,
3896 ofproto->up.name, vlan,
3897 in_bundle->name, in_bundle->vlan);
3900 } else if (ofbundle_includes_vlan(in_bundle, vlan)) {
3903 /* Drop packets from a VLAN not member of the trunk */
3905 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3906 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3907 "packet received on port %s not configured for "
3909 ofproto->up.name, vlan, in_bundle->name, vlan);
3914 if (in_bundle->vlan_mode != PORT_VLAN_TRUNK) {
3915 return in_bundle->vlan;
3917 return ofbundle_includes_vlan(in_bundle, 0) ? 0 : -1;
3922 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
3923 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
3924 * indicate this; newer upstream kernels use gratuitous ARP requests. */
3926 is_gratuitous_arp(const struct flow *flow)
3928 return (flow->dl_type == htons(ETH_TYPE_ARP)
3929 && eth_addr_is_broadcast(flow->dl_dst)
3930 && (flow->nw_proto == ARP_OP_REPLY
3931 || (flow->nw_proto == ARP_OP_REQUEST
3932 && flow->nw_src == flow->nw_dst)));
3936 update_learning_table(struct ofproto_dpif *ofproto,
3937 const struct flow *flow, int vlan,
3938 struct ofbundle *in_bundle)
3940 struct mac_entry *mac;
3942 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
3946 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
3947 if (is_gratuitous_arp(flow)) {
3948 /* We don't want to learn from gratuitous ARP packets that are
3949 * reflected back over bond slaves so we lock the learning table. */
3950 if (!in_bundle->bond) {
3951 mac_entry_set_grat_arp_lock(mac);
3952 } else if (mac_entry_is_grat_arp_locked(mac)) {
3957 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
3958 /* The log messages here could actually be useful in debugging,
3959 * so keep the rate limit relatively high. */
3960 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3961 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
3962 "on port %s in VLAN %d",
3963 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
3964 in_bundle->name, vlan);
3966 mac->port.p = in_bundle;
3967 tag_set_add(&ofproto->revalidate_set,
3968 mac_learning_changed(ofproto->ml, mac));
3972 /* Determines whether packets in 'flow' within 'br' should be forwarded or
3973 * dropped. Returns true if they may be forwarded, false if they should be
3976 * If 'have_packet' is true, it indicates that the caller is processing a
3977 * received packet. If 'have_packet' is false, then the caller is just
3978 * revalidating an existing flow because configuration has changed. Either
3979 * way, 'have_packet' only affects logging (there is no point in logging errors
3980 * during revalidation).
3982 * Sets '*in_portp' to the input port. This will be a null pointer if
3983 * flow->in_port does not designate a known input port (in which case
3984 * is_admissible() returns false).
3986 * When returning true, sets '*vlanp' to the effective VLAN of the input
3987 * packet, as returned by flow_get_vlan().
3989 * May also add tags to '*tags', although the current implementation only does
3990 * so in one special case.
3993 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
3995 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
3997 struct ofport_dpif *in_port;
3998 struct ofbundle *in_bundle;
4001 /* Find the port and bundle for the received packet. */
4002 in_port = get_ofp_port(ofproto, flow->in_port);
4003 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
4004 if (!in_port || !in_bundle) {
4005 /* No interface? Something fishy... */
4007 /* Odd. A few possible reasons here:
4009 * - We deleted a port but there are still a few packets queued up
4012 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
4013 * we don't know about.
4015 * - Packet arrived on the local port but the local port is not
4018 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4020 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
4022 ofproto->up.name, flow->in_port);
4027 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
4032 /* Drop frames for reserved multicast addresses
4033 * only if forward_bpdu option is absent. */
4034 if (eth_addr_is_reserved(flow->dl_dst) &&
4035 !ofproto->up.forward_bpdu) {
4039 /* Drop frames on bundles reserved for mirroring. */
4040 if (in_bundle->mirror_out) {
4042 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4043 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4044 "%s, which is reserved exclusively for mirroring",
4045 ofproto->up.name, in_bundle->name);
4050 if (in_bundle->bond) {
4051 struct mac_entry *mac;
4053 switch (bond_check_admissibility(in_bundle->bond, in_port,
4054 flow->dl_dst, tags)) {
4061 case BV_DROP_IF_MOVED:
4062 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
4063 if (mac && mac->port.p != in_bundle &&
4064 (!is_gratuitous_arp(flow)
4065 || mac_entry_is_grat_arp_locked(mac))) {
4076 xlate_normal(struct action_xlate_ctx *ctx)
4078 struct ofbundle *in_bundle;
4079 struct ofbundle *out_bundle;
4080 struct mac_entry *mac;
4083 ctx->has_normal = true;
4085 /* Check whether we should drop packets in this flow. */
4086 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
4087 &ctx->tags, &vlan, &in_bundle)) {
4092 /* Learn source MAC. */
4093 if (ctx->may_learn) {
4094 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
4097 /* Determine output bundle. */
4098 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
4101 out_bundle = mac->port.p;
4102 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
4103 /* If we are revalidating but don't have a learning entry then eject
4104 * the flow. Installing a flow that floods packets opens up a window
4105 * of time where we could learn from a packet reflected on a bond and
4106 * blackhole packets before the learning table is updated to reflect
4107 * the correct port. */
4108 ctx->may_set_up_flow = false;
4111 out_bundle = OFBUNDLE_FLOOD;
4114 /* Don't send packets out their input bundles. */
4115 if (in_bundle == out_bundle) {
4121 compose_actions(ctx, vlan, in_bundle, out_bundle);
4125 /* Optimized flow revalidation.
4127 * It's a difficult problem, in general, to tell which facets need to have
4128 * their actions recalculated whenever the OpenFlow flow table changes. We
4129 * don't try to solve that general problem: for most kinds of OpenFlow flow
4130 * table changes, we recalculate the actions for every facet. This is
4131 * relatively expensive, but it's good enough if the OpenFlow flow table
4132 * doesn't change very often.
4134 * However, we can expect one particular kind of OpenFlow flow table change to
4135 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
4136 * of CPU on revalidating every facet whenever MAC learning modifies the flow
4137 * table, we add a special case that applies to flow tables in which every rule
4138 * has the same form (that is, the same wildcards), except that the table is
4139 * also allowed to have a single "catch-all" flow that matches all packets. We
4140 * optimize this case by tagging all of the facets that resubmit into the table
4141 * and invalidating the same tag whenever a flow changes in that table. The
4142 * end result is that we revalidate just the facets that need it (and sometimes
4143 * a few more, but not all of the facets or even all of the facets that
4144 * resubmit to the table modified by MAC learning). */
4146 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
4147 * into an OpenFlow table with the given 'basis'. */
4149 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
4152 if (flow_wildcards_is_catchall(wc)) {
4155 struct flow tag_flow = *flow;
4156 flow_zero_wildcards(&tag_flow, wc);
4157 return tag_create_deterministic(flow_hash(&tag_flow, secret));
4161 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
4162 * taggability of that table.
4164 * This function must be called after *each* change to a flow table. If you
4165 * skip calling it on some changes then the pointer comparisons at the end can
4166 * be invalid if you get unlucky. For example, if a flow removal causes a
4167 * cls_table to be destroyed and then a flow insertion causes a cls_table with
4168 * different wildcards to be created with the same address, then this function
4169 * will incorrectly skip revalidation. */
4171 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
4173 struct table_dpif *table = &ofproto->tables[table_id];
4174 const struct classifier *cls = &ofproto->up.tables[table_id];
4175 struct cls_table *catchall, *other;
4176 struct cls_table *t;
4178 catchall = other = NULL;
4180 switch (hmap_count(&cls->tables)) {
4182 /* We could tag this OpenFlow table but it would make the logic a
4183 * little harder and it's a corner case that doesn't seem worth it
4189 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
4190 if (cls_table_is_catchall(t)) {
4192 } else if (!other) {
4195 /* Indicate that we can't tag this by setting both tables to
4196 * NULL. (We know that 'catchall' is already NULL.) */
4203 /* Can't tag this table. */
4207 if (table->catchall_table != catchall || table->other_table != other) {
4208 table->catchall_table = catchall;
4209 table->other_table = other;
4210 ofproto->need_revalidate = true;
4214 /* Given 'rule' that has changed in some way (either it is a rule being
4215 * inserted, a rule being deleted, or a rule whose actions are being
4216 * modified), marks facets for revalidation to ensure that packets will be
4217 * forwarded correctly according to the new state of the flow table.
4219 * This function must be called after *each* change to a flow table. See
4220 * the comment on table_update_taggable() for more information. */
4222 rule_invalidate(const struct rule_dpif *rule)
4224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4226 table_update_taggable(ofproto, rule->up.table_id);
4228 if (!ofproto->need_revalidate) {
4229 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
4231 if (table->other_table && rule->tag) {
4232 tag_set_add(&ofproto->revalidate_set, rule->tag);
4234 ofproto->need_revalidate = true;
4240 get_drop_frags(struct ofproto *ofproto_)
4242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4245 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
4250 set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
4252 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4254 dpif_set_drop_frags(ofproto->dpif, drop_frags);
4258 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4259 const struct flow *flow,
4260 const union ofp_action *ofp_actions, size_t n_ofp_actions)
4262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4265 error = validate_actions(ofp_actions, n_ofp_actions, flow,
4266 ofproto->max_ports);
4268 struct odputil_keybuf keybuf;
4269 struct action_xlate_ctx ctx;
4270 struct ofpbuf *odp_actions;
4273 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4274 odp_flow_key_from_flow(&key, flow);
4276 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
4277 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
4278 dpif_execute(ofproto->dpif, key.data, key.size,
4279 odp_actions->data, odp_actions->size, packet);
4280 ofpbuf_delete(odp_actions);
4286 get_netflow_ids(const struct ofproto *ofproto_,
4287 uint8_t *engine_type, uint8_t *engine_id)
4289 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4291 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
4294 static struct ofproto_dpif *
4295 ofproto_dpif_lookup(const char *name)
4297 struct ofproto *ofproto = ofproto_lookup(name);
4298 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
4299 ? ofproto_dpif_cast(ofproto)
4304 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
4305 const char *args, void *aux OVS_UNUSED)
4307 struct ds ds = DS_EMPTY_INITIALIZER;
4308 const struct ofproto_dpif *ofproto;
4309 const struct mac_entry *e;
4311 ofproto = ofproto_dpif_lookup(args);
4313 unixctl_command_reply(conn, 501, "no such bridge");
4317 ds_put_cstr(&ds, " port VLAN MAC Age\n");
4318 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4319 struct ofbundle *bundle = e->port.p;
4320 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
4321 ofbundle_get_a_port(bundle)->odp_port,
4322 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
4324 unixctl_command_reply(conn, 200, ds_cstr(&ds));
4328 struct ofproto_trace {
4329 struct action_xlate_ctx ctx;
4335 trace_format_rule(struct ds *result, uint8_t table_id, int level,
4336 const struct rule_dpif *rule)
4338 ds_put_char_multiple(result, '\t', level);
4340 ds_put_cstr(result, "No match\n");
4344 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4345 table_id, ntohll(rule->up.flow_cookie));
4346 cls_rule_format(&rule->up.cr, result);
4347 ds_put_char(result, '\n');
4349 ds_put_char_multiple(result, '\t', level);
4350 ds_put_cstr(result, "OpenFlow ");
4351 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
4352 ds_put_char(result, '\n');
4356 trace_format_flow(struct ds *result, int level, const char *title,
4357 struct ofproto_trace *trace)
4359 ds_put_char_multiple(result, '\t', level);
4360 ds_put_format(result, "%s: ", title);
4361 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4362 ds_put_cstr(result, "unchanged");
4364 flow_format(result, &trace->ctx.flow);
4365 trace->flow = trace->ctx.flow;
4367 ds_put_char(result, '\n');
4371 trace_format_regs(struct ds *result, int level, const char *title,
4372 struct ofproto_trace *trace)
4376 ds_put_char_multiple(result, '\t', level);
4377 ds_put_format(result, "%s:", title);
4378 for (i = 0; i < FLOW_N_REGS; i++) {
4379 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
4381 ds_put_char(result, '\n');
4385 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
4387 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4388 struct ds *result = trace->result;
4390 ds_put_char(result, '\n');
4391 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4392 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
4393 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
4397 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4398 void *aux OVS_UNUSED)
4400 char *dpname, *arg1, *arg2, *arg3;
4401 char *args = xstrdup(args_);
4402 char *save_ptr = NULL;
4403 struct ofproto_dpif *ofproto;
4404 struct ofpbuf odp_key;
4405 struct ofpbuf *packet;
4406 struct rule_dpif *rule;
4412 ofpbuf_init(&odp_key, 0);
4415 dpname = strtok_r(args, " ", &save_ptr);
4416 arg1 = strtok_r(NULL, " ", &save_ptr);
4417 arg2 = strtok_r(NULL, " ", &save_ptr);
4418 arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4419 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
4420 /* ofproto/trace dpname flow [-generate] */
4423 /* Convert string to datapath key. */
4424 ofpbuf_init(&odp_key, 0);
4425 error = odp_flow_key_from_string(arg1, &odp_key);
4427 unixctl_command_reply(conn, 501, "Bad flow syntax");
4431 /* Convert odp_key to flow. */
4432 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
4434 unixctl_command_reply(conn, 501, "Invalid flow");
4438 /* Generate a packet, if requested. */
4440 packet = ofpbuf_new(0);
4441 flow_compose(packet, &flow);
4443 } else if (dpname && arg1 && arg2 && arg3) {
4444 /* ofproto/trace dpname tun_id in_port packet */
4448 tun_id = htonll(strtoull(arg1, NULL, 0));
4449 in_port = ofp_port_to_odp_port(atoi(arg2));
4451 packet = ofpbuf_new(strlen(args) / 2);
4452 arg3 = ofpbuf_put_hex(packet, arg3, NULL);
4453 arg3 += strspn(arg3, " ");
4454 if (*arg3 != '\0') {
4455 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4458 if (packet->size < ETH_HEADER_LEN) {
4459 unixctl_command_reply(conn, 501,
4460 "Packet data too short for Ethernet");
4464 ds_put_cstr(&result, "Packet: ");
4465 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
4466 ds_put_cstr(&result, s);
4469 flow_extract(packet, tun_id, in_port, &flow);
4471 unixctl_command_reply(conn, 501, "Bad command syntax");
4475 ofproto = ofproto_dpif_lookup(dpname);
4477 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4482 ds_put_cstr(&result, "Flow: ");
4483 flow_format(&result, &flow);
4484 ds_put_char(&result, '\n');
4486 rule = rule_dpif_lookup(ofproto, &flow, 0);
4487 trace_format_rule(&result, 0, 0, rule);
4489 struct ofproto_trace trace;
4490 struct ofpbuf *odp_actions;
4492 trace.result = &result;
4494 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
4495 trace.ctx.resubmit_hook = trace_resubmit;
4496 odp_actions = xlate_actions(&trace.ctx,
4497 rule->up.actions, rule->up.n_actions);
4499 ds_put_char(&result, '\n');
4500 trace_format_flow(&result, 0, "Final flow", &trace);
4501 ds_put_cstr(&result, "Datapath actions: ");
4502 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4503 ofpbuf_delete(odp_actions);
4505 if (!trace.ctx.may_set_up_flow) {
4507 ds_put_cstr(&result, "\nThis flow is not cachable.");
4509 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
4510 "for complete actions, please supply a packet.");
4515 unixctl_command_reply(conn, 200, ds_cstr(&result));
4518 ds_destroy(&result);
4519 ofpbuf_delete(packet);
4520 ofpbuf_uninit(&odp_key);
4525 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
4526 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4529 unixctl_command_reply(conn, 200, NULL);
4533 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
4534 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4537 unixctl_command_reply(conn, 200, NULL);
4541 ofproto_dpif_unixctl_init(void)
4543 static bool registered;
4549 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4550 unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
4552 unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL);
4553 unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL);
4556 const struct ofproto_class ofproto_dpif_class = {
4583 port_is_lacp_current,
4584 NULL, /* rule_choose_table */
4591 rule_modify_actions,
4600 get_cfm_remote_mpids,
4605 is_mirror_output_bundle,
4606 forward_bpdu_changed,