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"
35 #include "mac-learning.h"
36 #include "multipath.h"
43 #include "ofp-print.h"
44 #include "ofproto-dpif-sflow.h"
45 #include "poll-loop.h"
47 #include "unaligned.h"
49 #include "vlan-bitmap.h"
52 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
54 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
55 COVERAGE_DEFINE(ofproto_dpif_expired);
56 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
57 COVERAGE_DEFINE(ofproto_dpif_xlate);
58 COVERAGE_DEFINE(facet_changed_rule);
59 COVERAGE_DEFINE(facet_invalidated);
60 COVERAGE_DEFINE(facet_revalidate);
61 COVERAGE_DEFINE(facet_unexpected);
63 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
64 * flow translation. */
65 #define MAX_RESUBMIT_RECURSION 16
73 long long int used; /* Time last used; time created if not used. */
77 * - Do include packets and bytes from facets that have been deleted or
78 * whose own statistics have been folded into the rule.
80 * - Do include packets and bytes sent "by hand" that were accounted to
81 * the rule without any facet being involved (this is a rare corner
82 * case in rule_execute()).
84 * - Do not include packet or bytes that can be obtained from any facet's
85 * packet_count or byte_count member or that can be obtained from the
86 * datapath by, e.g., dpif_flow_get() for any facet.
88 uint64_t packet_count; /* Number of packets received. */
89 uint64_t byte_count; /* Number of bytes received. */
91 struct list facets; /* List of "struct facet"s. */
94 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
96 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
99 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *ofproto,
100 const struct flow *flow);
102 #define MAX_MIRRORS 32
103 typedef uint32_t mirror_mask_t;
104 #define MIRROR_MASK_C(X) UINT32_C(X)
105 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
107 struct ofproto_dpif *ofproto; /* Owning ofproto. */
108 size_t idx; /* In ofproto's "mirrors" array. */
109 void *aux; /* Key supplied by ofproto's client. */
110 char *name; /* Identifier for log messages. */
112 /* Selection criteria. */
113 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
114 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
115 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
117 /* Output (mutually exclusive). */
118 struct ofbundle *out; /* Output port or NULL. */
119 int out_vlan; /* Output VLAN or -1. */
122 static void mirror_destroy(struct ofmirror *);
124 /* A group of one or more OpenFlow ports. */
125 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
127 struct ofproto_dpif *ofproto; /* Owning ofproto. */
128 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
129 void *aux; /* Key supplied by ofproto's client. */
130 char *name; /* Identifier for log messages. */
133 struct list ports; /* Contains "struct ofport"s. */
134 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
135 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
136 * NULL if all VLANs are trunked. */
137 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
138 struct bond *bond; /* Nonnull iff more than one port. */
141 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
143 /* Port mirroring info. */
144 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
145 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
146 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
149 static void bundle_remove(struct ofport *);
150 static void bundle_destroy(struct ofbundle *);
151 static void bundle_del_port(struct ofport_dpif *);
152 static void bundle_run(struct ofbundle *);
153 static void bundle_wait(struct ofbundle *);
155 struct action_xlate_ctx {
156 /* action_xlate_ctx_init() initializes these members. */
159 struct ofproto_dpif *ofproto;
161 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
162 * this flow when actions change header fields. */
165 /* The packet corresponding to 'flow', or a null pointer if we are
166 * revalidating without a packet to refer to. */
167 const struct ofpbuf *packet;
169 /* If nonnull, called just before executing a resubmit action.
171 * This is normally null so the client has to set it manually after
172 * calling action_xlate_ctx_init(). */
173 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
175 /* xlate_actions() initializes and uses these members. The client might want
176 * to look at them after it returns. */
178 struct ofpbuf *odp_actions; /* Datapath actions. */
179 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
180 bool may_set_up_flow; /* True ordinarily; false if the actions must
181 * be reassessed for every packet. */
182 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
184 /* xlate_actions() initializes and uses these members, but the client has no
185 * reason to look at them. */
187 int recurse; /* Recursion level, via xlate_table_action. */
188 uint32_t priority; /* Current flow priority. 0 if none. */
189 struct flow base_flow; /* Flow at the last commit. */
190 uint32_t base_priority; /* Priority at the last commit. */
193 static void action_xlate_ctx_init(struct action_xlate_ctx *,
194 struct ofproto_dpif *, const struct flow *,
195 const struct ofpbuf *);
196 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
197 const union ofp_action *in, size_t n_in);
199 /* An exact-match instantiation of an OpenFlow flow. */
201 long long int used; /* Time last used; time created if not used. */
205 * - Do include packets and bytes sent "by hand", e.g. with
208 * - Do include packets and bytes that were obtained from the datapath
209 * when a flow was deleted (e.g. dpif_flow_del()) or when its
210 * statistics were reset (e.g. dpif_flow_put() with
211 * DPIF_FP_ZERO_STATS).
213 * - Do not include any packets or bytes that can currently be obtained
214 * from the datapath by, e.g., dpif_flow_get().
216 uint64_t packet_count; /* Number of packets received. */
217 uint64_t byte_count; /* Number of bytes received. */
219 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
220 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
222 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
223 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
224 long long int rs_used; /* Used time pushed to resubmit children. */
226 /* Number of bytes passed to account_cb. This may include bytes that can
227 * currently obtained from the datapath (thus, it can be greater than
229 uint64_t accounted_bytes;
231 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
232 struct list list_node; /* In owning rule's 'facets' list. */
233 struct rule_dpif *rule; /* Owning rule. */
234 struct flow flow; /* Exact-match flow. */
235 bool installed; /* Installed in datapath? */
236 bool may_install; /* True ordinarily; false if actions must
237 * be reassessed for every packet. */
238 size_t actions_len; /* Number of bytes in actions[]. */
239 struct nlattr *actions; /* Datapath actions. */
240 tag_type tags; /* Tags. */
241 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
244 static struct facet *facet_create(struct rule_dpif *, const struct flow *,
245 const struct ofpbuf *packet);
246 static void facet_remove(struct ofproto_dpif *, struct facet *);
247 static void facet_free(struct facet *);
249 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
250 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
251 const struct flow *);
252 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
254 static void facet_execute(struct ofproto_dpif *, struct facet *,
255 struct ofpbuf *packet);
257 static int facet_put__(struct ofproto_dpif *, struct facet *,
258 const struct nlattr *actions, size_t actions_len,
259 struct dpif_flow_stats *);
260 static void facet_install(struct ofproto_dpif *, struct facet *,
262 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
263 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
265 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
266 const struct ofpbuf *packet);
267 static void facet_update_time(struct ofproto_dpif *, struct facet *,
269 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
270 const struct dpif_flow_stats *);
271 static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
272 static void facet_push_stats(struct facet *);
273 static void facet_account(struct ofproto_dpif *, struct facet *,
274 uint64_t extra_bytes);
276 static bool facet_is_controller_flow(struct facet *);
278 static void flow_push_stats(const struct rule_dpif *,
279 struct flow *, uint64_t packets, uint64_t bytes,
286 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
287 struct list bundle_node; /* In struct ofbundle's "ports" list. */
288 struct cfm *cfm; /* Connectivity Fault Management, if any. */
289 tag_type tag; /* Tag associated with this port. */
290 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
291 bool may_enable; /* May be enabled in bonds. */
294 static struct ofport_dpif *
295 ofport_dpif_cast(const struct ofport *ofport)
297 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
298 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
301 static void port_run(struct ofport_dpif *);
302 static void port_wait(struct ofport_dpif *);
303 static int set_cfm(struct ofport *, const struct cfm_settings *);
305 struct dpif_completion {
306 struct list list_node;
307 struct ofoperation *op;
310 struct ofproto_dpif {
319 struct netflow *netflow;
320 struct dpif_sflow *sflow;
321 struct hmap bundles; /* Contains "struct ofbundle"s. */
322 struct mac_learning *ml;
323 struct ofmirror *mirrors[MAX_MIRRORS];
324 bool has_bonded_bundles;
327 struct timer next_expiration;
331 bool need_revalidate;
332 struct tag_set revalidate_set;
334 /* Support for debugging async flow mods. */
335 struct list completions;
337 bool has_bundle_action; /* True when the first bundle action appears. */
340 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
341 * for debugging the asynchronous flow_mod implementation.) */
344 static void ofproto_dpif_unixctl_init(void);
346 static struct ofproto_dpif *
347 ofproto_dpif_cast(const struct ofproto *ofproto)
349 assert(ofproto->ofproto_class == &ofproto_dpif_class);
350 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
353 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
355 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
358 /* Packet processing. */
359 static void update_learning_table(struct ofproto_dpif *,
360 const struct flow *, int vlan,
362 static bool is_admissible(struct ofproto_dpif *, const struct flow *,
363 bool have_packet, tag_type *, int *vlanp,
364 struct ofbundle **in_bundlep);
365 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
367 /* Flow expiration. */
368 static int expire(struct ofproto_dpif *);
371 static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
372 const struct ofpbuf *packet);
374 /* Global variables. */
375 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
377 /* Factory functions. */
380 enumerate_types(struct sset *types)
382 dp_enumerate_types(types);
386 enumerate_names(const char *type, struct sset *names)
388 return dp_enumerate_names(type, names);
392 del(const char *type, const char *name)
397 error = dpif_open(name, type, &dpif);
399 error = dpif_delete(dpif);
405 /* Basic life-cycle. */
407 static struct ofproto *
410 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
415 dealloc(struct ofproto *ofproto_)
417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
422 construct(struct ofproto *ofproto_)
424 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
425 const char *name = ofproto->up.name;
429 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
431 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
435 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
436 ofproto->n_matches = 0;
438 error = dpif_recv_set_mask(ofproto->dpif,
439 ((1u << DPIF_UC_MISS) |
440 (1u << DPIF_UC_ACTION) |
441 (1u << DPIF_UC_SAMPLE)));
443 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
444 dpif_close(ofproto->dpif);
447 dpif_flow_flush(ofproto->dpif);
448 dpif_recv_purge(ofproto->dpif);
450 ofproto->netflow = NULL;
451 ofproto->sflow = NULL;
452 hmap_init(&ofproto->bundles);
453 ofproto->ml = mac_learning_create();
454 for (i = 0; i < MAX_MIRRORS; i++) {
455 ofproto->mirrors[i] = NULL;
457 ofproto->has_bonded_bundles = false;
459 timer_set_duration(&ofproto->next_expiration, 1000);
461 hmap_init(&ofproto->facets);
462 ofproto->need_revalidate = false;
463 tag_set_init(&ofproto->revalidate_set);
465 list_init(&ofproto->completions);
467 ofproto->up.tables = xmalloc(sizeof *ofproto->up.tables);
468 classifier_init(&ofproto->up.tables[0]);
469 ofproto->up.n_tables = 1;
471 ofproto_dpif_unixctl_init();
473 ofproto->has_bundle_action = false;
479 complete_operations(struct ofproto_dpif *ofproto)
481 struct dpif_completion *c, *next;
483 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
484 ofoperation_complete(c->op, 0);
485 list_remove(&c->list_node);
491 destruct(struct ofproto *ofproto_)
493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
494 struct rule_dpif *rule, *next_rule;
495 struct cls_cursor cursor;
498 complete_operations(ofproto);
500 cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
501 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
502 ofproto_rule_destroy(&rule->up);
505 for (i = 0; i < MAX_MIRRORS; i++) {
506 mirror_destroy(ofproto->mirrors[i]);
509 netflow_destroy(ofproto->netflow);
510 dpif_sflow_destroy(ofproto->sflow);
511 hmap_destroy(&ofproto->bundles);
512 mac_learning_destroy(ofproto->ml);
514 hmap_destroy(&ofproto->facets);
516 dpif_close(ofproto->dpif);
520 run(struct ofproto *ofproto_)
522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
523 struct ofport_dpif *ofport;
524 struct ofbundle *bundle;
528 complete_operations(ofproto);
530 dpif_run(ofproto->dpif);
532 for (i = 0; i < 50; i++) {
533 struct dpif_upcall packet;
536 error = dpif_recv(ofproto->dpif, &packet);
538 if (error == ENODEV) {
539 /* Datapath destroyed. */
545 handle_upcall(ofproto, &packet);
548 if (timer_expired(&ofproto->next_expiration)) {
549 int delay = expire(ofproto);
550 timer_set_duration(&ofproto->next_expiration, delay);
553 if (ofproto->netflow) {
554 netflow_run(ofproto->netflow);
556 if (ofproto->sflow) {
557 dpif_sflow_run(ofproto->sflow);
560 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
563 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
567 /* Now revalidate if there's anything to do. */
568 if (ofproto->need_revalidate
569 || !tag_set_is_empty(&ofproto->revalidate_set)) {
570 struct tag_set revalidate_set = ofproto->revalidate_set;
571 bool revalidate_all = ofproto->need_revalidate;
572 struct facet *facet, *next;
574 /* Clear the revalidation flags. */
575 tag_set_init(&ofproto->revalidate_set);
576 ofproto->need_revalidate = false;
578 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
580 || tag_set_intersects(&revalidate_set, facet->tags)) {
581 facet_revalidate(ofproto, facet);
590 wait(struct ofproto *ofproto_)
592 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
593 struct ofport_dpif *ofport;
594 struct ofbundle *bundle;
596 if (!clogged && !list_is_empty(&ofproto->completions)) {
597 poll_immediate_wake();
600 dpif_wait(ofproto->dpif);
601 dpif_recv_wait(ofproto->dpif);
602 if (ofproto->sflow) {
603 dpif_sflow_wait(ofproto->sflow);
605 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
606 poll_immediate_wake();
608 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
611 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
614 if (ofproto->need_revalidate) {
615 /* Shouldn't happen, but if it does just go around again. */
616 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
617 poll_immediate_wake();
619 timer_wait(&ofproto->next_expiration);
624 flush(struct ofproto *ofproto_)
626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
627 struct facet *facet, *next_facet;
629 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
630 /* Mark the facet as not installed so that facet_remove() doesn't
631 * bother trying to uninstall it. There is no point in uninstalling it
632 * individually since we are about to blow away all the facets with
633 * dpif_flow_flush(). */
634 facet->installed = false;
635 facet->dp_packet_count = 0;
636 facet->dp_byte_count = 0;
637 facet_remove(ofproto, facet);
639 dpif_flow_flush(ofproto->dpif);
643 get_features(struct ofproto *ofproto_ OVS_UNUSED,
644 bool *arp_match_ip, uint32_t *actions)
646 *arp_match_ip = true;
647 *actions = ((1u << OFPAT_OUTPUT) |
648 (1u << OFPAT_SET_VLAN_VID) |
649 (1u << OFPAT_SET_VLAN_PCP) |
650 (1u << OFPAT_STRIP_VLAN) |
651 (1u << OFPAT_SET_DL_SRC) |
652 (1u << OFPAT_SET_DL_DST) |
653 (1u << OFPAT_SET_NW_SRC) |
654 (1u << OFPAT_SET_NW_DST) |
655 (1u << OFPAT_SET_NW_TOS) |
656 (1u << OFPAT_SET_TP_SRC) |
657 (1u << OFPAT_SET_TP_DST) |
658 (1u << OFPAT_ENQUEUE));
662 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
667 strcpy(ots->name, "classifier");
669 dpif_get_dp_stats(ofproto->dpif, &s);
670 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
671 put_32aligned_be64(&ots->matched_count,
672 htonll(s.n_hit + ofproto->n_matches));
676 set_netflow(struct ofproto *ofproto_,
677 const struct netflow_options *netflow_options)
679 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
681 if (netflow_options) {
682 if (!ofproto->netflow) {
683 ofproto->netflow = netflow_create();
685 return netflow_set_options(ofproto->netflow, netflow_options);
687 netflow_destroy(ofproto->netflow);
688 ofproto->netflow = NULL;
693 static struct ofport *
696 struct ofport_dpif *port = xmalloc(sizeof *port);
701 port_dealloc(struct ofport *port_)
703 struct ofport_dpif *port = ofport_dpif_cast(port_);
708 port_construct(struct ofport *port_)
710 struct ofport_dpif *port = ofport_dpif_cast(port_);
711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
713 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
716 port->tag = tag_create_random();
717 port->may_enable = true;
719 if (ofproto->sflow) {
720 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
721 netdev_get_name(port->up.netdev));
728 port_destruct(struct ofport *port_)
730 struct ofport_dpif *port = ofport_dpif_cast(port_);
731 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
733 bundle_remove(port_);
734 set_cfm(port_, NULL);
735 if (ofproto->sflow) {
736 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
741 port_modified(struct ofport *port_)
743 struct ofport_dpif *port = ofport_dpif_cast(port_);
745 if (port->bundle && port->bundle->bond) {
746 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
751 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
753 struct ofport_dpif *port = ofport_dpif_cast(port_);
754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
755 ovs_be32 changed = old_config ^ port->up.opp.config;
757 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
758 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
759 ofproto->need_revalidate = true;
764 set_sflow(struct ofproto *ofproto_,
765 const struct ofproto_sflow_options *sflow_options)
767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
768 struct dpif_sflow *ds = ofproto->sflow;
771 struct ofport_dpif *ofport;
773 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
774 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
775 dpif_sflow_add_port(ds, ofport->odp_port,
776 netdev_get_name(ofport->up.netdev));
779 dpif_sflow_set_options(ds, sflow_options);
781 dpif_sflow_destroy(ds);
782 ofproto->sflow = NULL;
788 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
790 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
797 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
800 if (cfm_configure(ofport->cfm, s)) {
806 cfm_destroy(ofport->cfm);
812 get_cfm_fault(const struct ofport *ofport_)
814 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
816 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
821 /* Expires all MAC learning entries associated with 'port' and forces ofproto
822 * to revalidate every flow. */
824 bundle_flush_macs(struct ofbundle *bundle)
826 struct ofproto_dpif *ofproto = bundle->ofproto;
827 struct mac_learning *ml = ofproto->ml;
828 struct mac_entry *mac, *next_mac;
830 ofproto->need_revalidate = true;
831 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
832 if (mac->port.p == bundle) {
833 mac_learning_expire(ml, mac);
838 static struct ofbundle *
839 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
841 struct ofbundle *bundle;
843 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
845 if (bundle->aux == aux) {
852 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
853 * ones that are found to 'bundles'. */
855 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
856 void **auxes, size_t n_auxes,
857 struct hmapx *bundles)
862 for (i = 0; i < n_auxes; i++) {
863 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
865 hmapx_add(bundles, bundle);
871 bundle_del_port(struct ofport_dpif *port)
873 struct ofbundle *bundle = port->bundle;
875 bundle->ofproto->need_revalidate = true;
877 list_remove(&port->bundle_node);
881 lacp_slave_unregister(bundle->lacp, port);
884 bond_slave_unregister(bundle->bond, port);
887 bundle->floodable = true;
888 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
889 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
890 bundle->floodable = false;
896 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
897 struct lacp_slave_settings *lacp,
898 uint32_t bond_stable_id)
900 struct ofport_dpif *port;
902 port = get_ofp_port(bundle->ofproto, ofp_port);
907 if (port->bundle != bundle) {
908 bundle->ofproto->need_revalidate = true;
910 bundle_del_port(port);
913 port->bundle = bundle;
914 list_push_back(&bundle->ports, &port->bundle_node);
915 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
916 bundle->floodable = false;
920 lacp_slave_register(bundle->lacp, port, lacp);
923 port->bond_stable_id = bond_stable_id;
929 bundle_destroy(struct ofbundle *bundle)
931 struct ofproto_dpif *ofproto;
932 struct ofport_dpif *port, *next_port;
939 ofproto = bundle->ofproto;
940 for (i = 0; i < MAX_MIRRORS; i++) {
941 struct ofmirror *m = ofproto->mirrors[i];
943 if (m->out == bundle) {
945 } else if (hmapx_find_and_delete(&m->srcs, bundle)
946 || hmapx_find_and_delete(&m->dsts, bundle)) {
947 ofproto->need_revalidate = true;
952 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
953 bundle_del_port(port);
956 bundle_flush_macs(bundle);
957 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
959 free(bundle->trunks);
960 lacp_destroy(bundle->lacp);
961 bond_destroy(bundle->bond);
966 bundle_set(struct ofproto *ofproto_, void *aux,
967 const struct ofproto_bundle_settings *s)
969 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
970 bool need_flush = false;
971 const unsigned long *trunks;
972 struct ofport_dpif *port;
973 struct ofbundle *bundle;
978 bundle_destroy(bundle_lookup(ofproto, aux));
982 assert(s->n_slaves == 1 || s->bond != NULL);
983 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
985 bundle = bundle_lookup(ofproto, aux);
987 bundle = xmalloc(sizeof *bundle);
989 bundle->ofproto = ofproto;
990 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
991 hash_pointer(aux, 0));
995 list_init(&bundle->ports);
997 bundle->trunks = NULL;
1001 bundle->floodable = true;
1003 bundle->src_mirrors = 0;
1004 bundle->dst_mirrors = 0;
1005 bundle->mirror_out = 0;
1008 if (!bundle->name || strcmp(s->name, bundle->name)) {
1010 bundle->name = xstrdup(s->name);
1015 if (!bundle->lacp) {
1016 bundle->lacp = lacp_create();
1018 lacp_configure(bundle->lacp, s->lacp);
1020 lacp_destroy(bundle->lacp);
1021 bundle->lacp = NULL;
1024 /* Update set of ports. */
1026 for (i = 0; i < s->n_slaves; i++) {
1027 if (!bundle_add_port(bundle, s->slaves[i],
1028 s->lacp ? &s->lacp_slaves[i] : NULL,
1029 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1033 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1034 struct ofport_dpif *next_port;
1036 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1037 for (i = 0; i < s->n_slaves; i++) {
1038 if (s->slaves[i] == port->up.ofp_port) {
1043 bundle_del_port(port);
1047 assert(list_size(&bundle->ports) <= s->n_slaves);
1049 if (list_is_empty(&bundle->ports)) {
1050 bundle_destroy(bundle);
1055 if (s->vlan != bundle->vlan) {
1056 bundle->vlan = s->vlan;
1060 /* Get trunked VLANs. */
1061 trunks = s->vlan == -1 ? NULL : s->trunks;
1062 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1063 free(bundle->trunks);
1064 bundle->trunks = vlan_bitmap_clone(trunks);
1069 if (!list_is_short(&bundle->ports)) {
1070 bundle->ofproto->has_bonded_bundles = true;
1072 if (bond_reconfigure(bundle->bond, s->bond)) {
1073 ofproto->need_revalidate = true;
1076 bundle->bond = bond_create(s->bond);
1077 ofproto->need_revalidate = true;
1080 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1081 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1085 bond_destroy(bundle->bond);
1086 bundle->bond = NULL;
1089 /* If we changed something that would affect MAC learning, un-learn
1090 * everything on this port and force flow revalidation. */
1092 bundle_flush_macs(bundle);
1099 bundle_remove(struct ofport *port_)
1101 struct ofport_dpif *port = ofport_dpif_cast(port_);
1102 struct ofbundle *bundle = port->bundle;
1105 bundle_del_port(port);
1106 if (list_is_empty(&bundle->ports)) {
1107 bundle_destroy(bundle);
1108 } else if (list_is_short(&bundle->ports)) {
1109 bond_destroy(bundle->bond);
1110 bundle->bond = NULL;
1116 send_pdu_cb(void *port_, const struct lacp_pdu *pdu)
1118 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1119 struct ofport_dpif *port = port_;
1120 uint8_t ea[ETH_ADDR_LEN];
1123 error = netdev_get_etheraddr(port->up.netdev, ea);
1125 struct lacp_pdu *packet_pdu;
1126 struct ofpbuf packet;
1128 ofpbuf_init(&packet, 0);
1129 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1130 sizeof *packet_pdu);
1132 error = netdev_send(port->up.netdev, &packet);
1134 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1135 "(%s)", port->bundle->name,
1136 netdev_get_name(port->up.netdev), strerror(error));
1138 ofpbuf_uninit(&packet);
1140 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1141 "%s (%s)", port->bundle->name,
1142 netdev_get_name(port->up.netdev), strerror(error));
1147 bundle_send_learning_packets(struct ofbundle *bundle)
1149 struct ofproto_dpif *ofproto = bundle->ofproto;
1150 int error, n_packets, n_errors;
1151 struct mac_entry *e;
1153 error = n_packets = n_errors = 0;
1154 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1155 if (e->port.p != bundle) {
1156 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1166 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1167 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1168 "packets, last error was: %s",
1169 bundle->name, n_errors, n_packets, strerror(error));
1171 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1172 bundle->name, n_packets);
1177 bundle_run(struct ofbundle *bundle)
1180 lacp_run(bundle->lacp, send_pdu_cb);
1183 struct ofport_dpif *port;
1185 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1186 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1189 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1190 lacp_negotiated(bundle->lacp));
1191 if (bond_should_send_learning_packets(bundle->bond)) {
1192 bundle_send_learning_packets(bundle);
1198 bundle_wait(struct ofbundle *bundle)
1201 lacp_wait(bundle->lacp);
1204 bond_wait(bundle->bond);
1211 mirror_scan(struct ofproto_dpif *ofproto)
1215 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1216 if (!ofproto->mirrors[idx]) {
1223 static struct ofmirror *
1224 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1228 for (i = 0; i < MAX_MIRRORS; i++) {
1229 struct ofmirror *mirror = ofproto->mirrors[i];
1230 if (mirror && mirror->aux == aux) {
1239 mirror_set(struct ofproto *ofproto_, void *aux,
1240 const struct ofproto_mirror_settings *s)
1242 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1243 mirror_mask_t mirror_bit;
1244 struct ofbundle *bundle;
1245 struct ofmirror *mirror;
1246 struct ofbundle *out;
1247 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1248 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1251 mirror = mirror_lookup(ofproto, aux);
1253 mirror_destroy(mirror);
1259 idx = mirror_scan(ofproto);
1261 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1263 ofproto->up.name, MAX_MIRRORS, s->name);
1267 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1268 mirror->ofproto = ofproto;
1271 mirror->out_vlan = -1;
1272 mirror->name = NULL;
1275 if (!mirror->name || strcmp(s->name, mirror->name)) {
1277 mirror->name = xstrdup(s->name);
1280 /* Get the new configuration. */
1281 if (s->out_bundle) {
1282 out = bundle_lookup(ofproto, s->out_bundle);
1284 mirror_destroy(mirror);
1290 out_vlan = s->out_vlan;
1292 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1293 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1295 /* If the configuration has not changed, do nothing. */
1296 if (hmapx_equals(&srcs, &mirror->srcs)
1297 && hmapx_equals(&dsts, &mirror->dsts)
1298 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1299 && mirror->out == out
1300 && mirror->out_vlan == out_vlan)
1302 hmapx_destroy(&srcs);
1303 hmapx_destroy(&dsts);
1307 hmapx_swap(&srcs, &mirror->srcs);
1308 hmapx_destroy(&srcs);
1310 hmapx_swap(&dsts, &mirror->dsts);
1311 hmapx_destroy(&dsts);
1313 free(mirror->vlans);
1314 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1317 mirror->out_vlan = out_vlan;
1319 /* Update bundles. */
1320 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1321 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1322 if (hmapx_contains(&mirror->srcs, bundle)) {
1323 bundle->src_mirrors |= mirror_bit;
1325 bundle->src_mirrors &= ~mirror_bit;
1328 if (hmapx_contains(&mirror->dsts, bundle)) {
1329 bundle->dst_mirrors |= mirror_bit;
1331 bundle->dst_mirrors &= ~mirror_bit;
1334 if (mirror->out == bundle) {
1335 bundle->mirror_out |= mirror_bit;
1337 bundle->mirror_out &= ~mirror_bit;
1341 ofproto->need_revalidate = true;
1342 mac_learning_flush(ofproto->ml);
1348 mirror_destroy(struct ofmirror *mirror)
1350 struct ofproto_dpif *ofproto;
1351 mirror_mask_t mirror_bit;
1352 struct ofbundle *bundle;
1358 ofproto = mirror->ofproto;
1359 ofproto->need_revalidate = true;
1360 mac_learning_flush(ofproto->ml);
1362 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1363 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1364 bundle->src_mirrors &= ~mirror_bit;
1365 bundle->dst_mirrors &= ~mirror_bit;
1366 bundle->mirror_out &= ~mirror_bit;
1369 hmapx_destroy(&mirror->srcs);
1370 hmapx_destroy(&mirror->dsts);
1371 free(mirror->vlans);
1373 ofproto->mirrors[mirror->idx] = NULL;
1379 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1382 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1383 ofproto->need_revalidate = true;
1384 mac_learning_flush(ofproto->ml);
1390 is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
1392 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1393 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1394 return bundle && bundle->mirror_out != 0;
1399 static struct ofport_dpif *
1400 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1402 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1403 return ofport ? ofport_dpif_cast(ofport) : NULL;
1406 static struct ofport_dpif *
1407 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1409 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1413 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1414 struct dpif_port *dpif_port)
1416 ofproto_port->name = dpif_port->name;
1417 ofproto_port->type = dpif_port->type;
1418 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1422 port_run(struct ofport_dpif *ofport)
1424 bool enable = netdev_get_carrier(ofport->up.netdev);
1427 cfm_run(ofport->cfm);
1429 if (cfm_should_send_ccm(ofport->cfm)) {
1430 struct ofpbuf packet;
1432 ofpbuf_init(&packet, 0);
1433 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1434 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1435 ofport->odp_port, &packet);
1436 ofpbuf_uninit(&packet);
1439 enable = enable && !cfm_get_fault(ofport->cfm);
1442 if (ofport->bundle) {
1443 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1446 if (ofport->may_enable != enable) {
1447 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1449 if (ofproto->has_bundle_action) {
1450 ofproto->need_revalidate = true;
1454 ofport->may_enable = enable;
1458 port_wait(struct ofport_dpif *ofport)
1461 cfm_wait(ofport->cfm);
1466 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1467 struct ofproto_port *ofproto_port)
1469 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1470 struct dpif_port dpif_port;
1473 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1475 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1481 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1483 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1487 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1489 *ofp_portp = odp_port_to_ofp_port(odp_port);
1495 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1497 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1500 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1502 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1504 /* The caller is going to close ofport->up.netdev. If this is a
1505 * bonded port, then the bond is using that netdev, so remove it
1506 * from the bond. The client will need to reconfigure everything
1507 * after deleting ports, so then the slave will get re-added. */
1508 bundle_remove(&ofport->up);
1514 struct port_dump_state {
1515 struct dpif_port_dump dump;
1520 port_dump_start(const struct ofproto *ofproto_, void **statep)
1522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1523 struct port_dump_state *state;
1525 *statep = state = xmalloc(sizeof *state);
1526 dpif_port_dump_start(&state->dump, ofproto->dpif);
1527 state->done = false;
1532 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1533 struct ofproto_port *port)
1535 struct port_dump_state *state = state_;
1536 struct dpif_port dpif_port;
1538 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1539 ofproto_port_from_dpif_port(port, &dpif_port);
1542 int error = dpif_port_dump_done(&state->dump);
1544 return error ? error : EOF;
1549 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1551 struct port_dump_state *state = state_;
1554 dpif_port_dump_done(&state->dump);
1561 port_poll(const struct ofproto *ofproto_, char **devnamep)
1563 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1564 return dpif_port_poll(ofproto->dpif, devnamep);
1568 port_poll_wait(const struct ofproto *ofproto_)
1570 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1571 dpif_port_poll_wait(ofproto->dpif);
1575 port_is_lacp_current(const struct ofport *ofport_)
1577 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1578 return (ofport->bundle && ofport->bundle->lacp
1579 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
1583 /* Upcall handling. */
1585 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
1586 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
1587 * their individual configurations.
1589 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
1590 * Otherwise, ownership is transferred to this function. */
1592 send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
1593 const struct flow *flow, bool clone)
1595 struct ofputil_packet_in pin;
1597 pin.packet = upcall->packet;
1598 pin.in_port = flow->in_port;
1599 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
1600 pin.buffer_id = 0; /* not yet known */
1601 pin.send_len = upcall->userdata;
1602 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
1603 clone ? NULL : upcall->packet);
1607 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
1608 const struct ofpbuf *packet)
1610 if (cfm_should_process_flow(flow)) {
1611 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
1612 if (packet && ofport && ofport->cfm) {
1613 cfm_process_heartbeat(ofport->cfm, packet);
1616 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
1617 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
1618 if (packet && port && port->bundle && port->bundle->lacp) {
1619 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
1621 lacp_process_pdu(port->bundle->lacp, port, pdu);
1630 handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1632 struct facet *facet;
1635 /* Obtain in_port and tun_id, at least. */
1636 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1638 /* Set header pointers in 'flow'. */
1639 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
1641 /* Handle 802.1ag and LACP. */
1642 if (process_special(ofproto, &flow, upcall->packet)) {
1643 ofpbuf_delete(upcall->packet);
1644 ofproto->n_matches++;
1648 /* Check with in-band control to see if this packet should be sent
1649 * to the local port regardless of the flow table. */
1650 if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
1651 send_packet(ofproto, ODPP_LOCAL, upcall->packet);
1654 facet = facet_lookup_valid(ofproto, &flow);
1656 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow);
1658 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
1659 struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
1661 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
1662 COVERAGE_INC(ofproto_dpif_no_packet_in);
1663 /* XXX install 'drop' flow entry */
1664 ofpbuf_delete(upcall->packet);
1668 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
1672 send_packet_in(ofproto, upcall, &flow, false);
1676 facet = facet_create(rule, &flow, upcall->packet);
1677 } else if (!facet->may_install) {
1678 /* The facet is not installable, that is, we need to process every
1679 * packet, so process the current packet's actions into 'facet'. */
1680 facet_make_actions(ofproto, facet, upcall->packet);
1683 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
1685 * Extra-special case for fail-open mode.
1687 * We are in fail-open mode and the packet matched the fail-open rule,
1688 * but we are connected to a controller too. We should send the packet
1689 * up to the controller in the hope that it will try to set up a flow
1690 * and thereby allow us to exit fail-open.
1692 * See the top-level comment in fail-open.c for more information.
1694 send_packet_in(ofproto, upcall, &flow, true);
1697 facet_execute(ofproto, facet, upcall->packet);
1698 facet_install(ofproto, facet, false);
1699 ofproto->n_matches++;
1703 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1707 switch (upcall->type) {
1708 case DPIF_UC_ACTION:
1709 COVERAGE_INC(ofproto_dpif_ctlr_action);
1710 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1711 send_packet_in(ofproto, upcall, &flow, false);
1714 case DPIF_UC_SAMPLE:
1715 if (ofproto->sflow) {
1716 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1717 dpif_sflow_received(ofproto->sflow, upcall, &flow);
1719 ofpbuf_delete(upcall->packet);
1723 handle_miss_upcall(ofproto, upcall);
1726 case DPIF_N_UC_TYPES:
1728 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
1733 /* Flow expiration. */
1735 static int facet_max_idle(const struct ofproto_dpif *);
1736 static void update_stats(struct ofproto_dpif *);
1737 static void rule_expire(struct rule_dpif *);
1738 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
1740 /* This function is called periodically by run(). Its job is to collect
1741 * updates for the flows that have been installed into the datapath, most
1742 * importantly when they last were used, and then use that information to
1743 * expire flows that have not been used recently.
1745 * Returns the number of milliseconds after which it should be called again. */
1747 expire(struct ofproto_dpif *ofproto)
1749 struct rule_dpif *rule, *next_rule;
1750 struct cls_cursor cursor;
1753 /* Update stats for each flow in the datapath. */
1754 update_stats(ofproto);
1756 /* Expire facets that have been idle too long. */
1757 dp_max_idle = facet_max_idle(ofproto);
1758 expire_facets(ofproto, dp_max_idle);
1760 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
1761 cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
1762 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1766 /* All outstanding data in existing flows has been accounted, so it's a
1767 * good time to do bond rebalancing. */
1768 if (ofproto->has_bonded_bundles) {
1769 struct ofbundle *bundle;
1771 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1773 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
1778 return MIN(dp_max_idle, 1000);
1781 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
1783 * This function also pushes statistics updates to rules which each facet
1784 * resubmits into. Generally these statistics will be accurate. However, if a
1785 * facet changes the rule it resubmits into at some time in between
1786 * update_stats() runs, it is possible that statistics accrued to the
1787 * old rule will be incorrectly attributed to the new rule. This could be
1788 * avoided by calling update_stats() whenever rules are created or
1789 * deleted. However, the performance impact of making so many calls to the
1790 * datapath do not justify the benefit of having perfectly accurate statistics.
1793 update_stats(struct ofproto_dpif *p)
1795 const struct dpif_flow_stats *stats;
1796 struct dpif_flow_dump dump;
1797 const struct nlattr *key;
1800 dpif_flow_dump_start(&dump, p->dpif);
1801 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
1802 struct facet *facet;
1805 if (odp_flow_key_to_flow(key, key_len, &flow)) {
1809 odp_flow_key_format(key, key_len, &s);
1810 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
1816 facet = facet_find(p, &flow);
1818 if (facet && facet->installed) {
1820 if (stats->n_packets >= facet->dp_packet_count) {
1821 uint64_t extra = stats->n_packets - facet->dp_packet_count;
1822 facet->packet_count += extra;
1824 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
1827 if (stats->n_bytes >= facet->dp_byte_count) {
1828 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
1830 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
1833 facet->dp_packet_count = stats->n_packets;
1834 facet->dp_byte_count = stats->n_bytes;
1836 facet_update_time(p, facet, stats->used);
1837 facet_account(p, facet, stats->n_bytes);
1838 facet_push_stats(facet);
1840 /* There's a flow in the datapath that we know nothing about.
1842 COVERAGE_INC(facet_unexpected);
1843 dpif_flow_del(p->dpif, key, key_len, NULL);
1846 dpif_flow_dump_done(&dump);
1849 /* Calculates and returns the number of milliseconds of idle time after which
1850 * facets should expire from the datapath and we should fold their statistics
1851 * into their parent rules in userspace. */
1853 facet_max_idle(const struct ofproto_dpif *ofproto)
1856 * Idle time histogram.
1858 * Most of the time a switch has a relatively small number of facets. When
1859 * this is the case we might as well keep statistics for all of them in
1860 * userspace and to cache them in the kernel datapath for performance as
1863 * As the number of facets increases, the memory required to maintain
1864 * statistics about them in userspace and in the kernel becomes
1865 * significant. However, with a large number of facets it is likely that
1866 * only a few of them are "heavy hitters" that consume a large amount of
1867 * bandwidth. At this point, only heavy hitters are worth caching in the
1868 * kernel and maintaining in userspaces; other facets we can discard.
1870 * The technique used to compute the idle time is to build a histogram with
1871 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
1872 * that is installed in the kernel gets dropped in the appropriate bucket.
1873 * After the histogram has been built, we compute the cutoff so that only
1874 * the most-recently-used 1% of facets (but at least
1875 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
1876 * the most-recently-used bucket of facets is kept, so actually an
1877 * arbitrary number of facets can be kept in any given expiration run
1878 * (though the next run will delete most of those unless they receive
1881 * This requires a second pass through the facets, in addition to the pass
1882 * made by update_stats(), because the former function never looks
1883 * at uninstallable facets.
1885 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
1886 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
1887 int buckets[N_BUCKETS] = { 0 };
1888 int total, subtotal, bucket;
1889 struct facet *facet;
1893 total = hmap_count(&ofproto->facets);
1894 if (total <= ofproto->up.flow_eviction_threshold) {
1895 return N_BUCKETS * BUCKET_WIDTH;
1898 /* Build histogram. */
1900 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1901 long long int idle = now - facet->used;
1902 int bucket = (idle <= 0 ? 0
1903 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
1904 : (unsigned int) idle / BUCKET_WIDTH);
1908 /* Find the first bucket whose flows should be expired. */
1909 subtotal = bucket = 0;
1911 subtotal += buckets[bucket++];
1912 } while (bucket < N_BUCKETS &&
1913 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
1915 if (VLOG_IS_DBG_ENABLED()) {
1919 ds_put_cstr(&s, "keep");
1920 for (i = 0; i < N_BUCKETS; i++) {
1922 ds_put_cstr(&s, ", drop");
1925 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
1928 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
1932 return bucket * BUCKET_WIDTH;
1936 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
1938 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
1939 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
1940 struct ofexpired expired;
1942 if (facet->installed) {
1943 struct dpif_flow_stats stats;
1945 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
1947 facet_update_stats(ofproto, facet, &stats);
1950 expired.flow = facet->flow;
1951 expired.packet_count = facet->packet_count;
1952 expired.byte_count = facet->byte_count;
1953 expired.used = facet->used;
1954 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1959 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
1961 long long int cutoff = time_msec() - dp_max_idle;
1962 struct facet *facet, *next_facet;
1964 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1965 facet_active_timeout(ofproto, facet);
1966 if (facet->used < cutoff) {
1967 facet_remove(ofproto, facet);
1972 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
1973 * then delete it entirely. */
1975 rule_expire(struct rule_dpif *rule)
1977 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
1978 struct facet *facet, *next_facet;
1982 /* Has 'rule' expired? */
1984 if (rule->up.hard_timeout
1985 && now > rule->up.created + rule->up.hard_timeout * 1000) {
1986 reason = OFPRR_HARD_TIMEOUT;
1987 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
1988 && now > rule->used + rule->up.idle_timeout * 1000) {
1989 reason = OFPRR_IDLE_TIMEOUT;
1994 COVERAGE_INC(ofproto_dpif_expired);
1996 /* Update stats. (This is a no-op if the rule expired due to an idle
1997 * timeout, because that only happens when the rule has no facets left.) */
1998 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1999 facet_remove(ofproto, facet);
2002 /* Get rid of the rule. */
2003 ofproto_rule_expire(&rule->up, reason);
2008 /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
2009 * example 'packet' within that flow.
2011 * The caller must already have determined that no facet with an identical
2012 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2013 * the ofproto's classifier table. */
2014 static struct facet *
2015 facet_create(struct rule_dpif *rule, const struct flow *flow,
2016 const struct ofpbuf *packet)
2018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2019 struct facet *facet;
2021 facet = xzalloc(sizeof *facet);
2022 facet->used = time_msec();
2023 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2024 list_push_back(&rule->facets, &facet->list_node);
2026 facet->flow = *flow;
2027 netflow_flow_init(&facet->nf_flow);
2028 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2030 facet_make_actions(ofproto, facet, packet);
2036 facet_free(struct facet *facet)
2038 free(facet->actions);
2042 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2043 * 'packet', which arrived on 'in_port'.
2045 * Takes ownership of 'packet'. */
2047 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2048 const struct nlattr *odp_actions, size_t actions_len,
2049 struct ofpbuf *packet)
2051 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2052 && odp_actions->nla_type == ODP_ACTION_ATTR_USERSPACE) {
2053 /* As an optimization, avoid a round-trip from userspace to kernel to
2054 * userspace. This also avoids possibly filling up kernel packet
2055 * buffers along the way. */
2056 struct dpif_upcall upcall;
2058 upcall.type = DPIF_UC_ACTION;
2059 upcall.packet = packet;
2062 upcall.userdata = nl_attr_get_u64(odp_actions);
2063 upcall.sample_pool = 0;
2064 upcall.actions = NULL;
2065 upcall.actions_len = 0;
2067 send_packet_in(ofproto, &upcall, flow, false);
2071 struct odputil_keybuf keybuf;
2075 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2076 odp_flow_key_from_flow(&key, flow);
2078 error = dpif_execute(ofproto->dpif, key.data, key.size,
2079 odp_actions, actions_len, packet);
2081 ofpbuf_delete(packet);
2086 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2087 * statistics appropriately. 'packet' must have at least sizeof(struct
2088 * ofp_packet_in) bytes of headroom.
2090 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2091 * applying flow_extract() to 'packet' would yield the same flow as
2094 * 'facet' must have accurately composed ODP actions; that is, it must not be
2095 * in need of revalidation.
2097 * Takes ownership of 'packet'. */
2099 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2100 struct ofpbuf *packet)
2102 struct dpif_flow_stats stats;
2104 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2106 flow_extract_stats(&facet->flow, packet, &stats);
2107 stats.used = time_msec();
2108 if (execute_odp_actions(ofproto, &facet->flow,
2109 facet->actions, facet->actions_len, packet)) {
2110 facet_update_stats(ofproto, facet, &stats);
2114 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2116 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2117 * rule's statistics, via facet_uninstall().
2119 * - Removes 'facet' from its rule and from ofproto->facets.
2122 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2124 facet_uninstall(ofproto, facet);
2125 facet_flush_stats(ofproto, facet);
2126 hmap_remove(&ofproto->facets, &facet->hmap_node);
2127 list_remove(&facet->list_node);
2131 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2133 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2134 const struct ofpbuf *packet)
2136 const struct rule_dpif *rule = facet->rule;
2137 struct ofpbuf *odp_actions;
2138 struct action_xlate_ctx ctx;
2140 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2141 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2142 facet->tags = ctx.tags;
2143 facet->may_install = ctx.may_set_up_flow;
2144 facet->nf_flow.output_iface = ctx.nf_output_iface;
2146 if (facet->actions_len != odp_actions->size
2147 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2148 free(facet->actions);
2149 facet->actions_len = odp_actions->size;
2150 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2153 ofpbuf_delete(odp_actions);
2156 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2157 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2158 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2159 * since 'facet' was last updated.
2161 * Returns 0 if successful, otherwise a positive errno value.*/
2163 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2164 const struct nlattr *actions, size_t actions_len,
2165 struct dpif_flow_stats *stats)
2167 struct odputil_keybuf keybuf;
2168 enum dpif_flow_put_flags flags;
2172 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2174 flags |= DPIF_FP_ZERO_STATS;
2177 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2178 odp_flow_key_from_flow(&key, &facet->flow);
2180 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2181 actions, actions_len, stats);
2184 facet_reset_dp_stats(facet, stats);
2190 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2191 * 'zero_stats' is true, clears any existing statistics from the datapath for
2194 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2196 struct dpif_flow_stats stats;
2198 if (facet->may_install
2199 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2200 zero_stats ? &stats : NULL)) {
2201 facet->installed = true;
2206 vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci)
2208 return vlan_tci != htons(0) ? vlan_tci_to_vid(vlan_tci) : OFP_VLAN_NONE;
2212 facet_account(struct ofproto_dpif *ofproto,
2213 struct facet *facet, uint64_t extra_bytes)
2215 uint64_t total_bytes, n_bytes;
2216 struct ofbundle *in_bundle;
2217 const struct nlattr *a;
2223 total_bytes = facet->byte_count + extra_bytes;
2224 if (total_bytes <= facet->accounted_bytes) {
2227 n_bytes = total_bytes - facet->accounted_bytes;
2228 facet->accounted_bytes = total_bytes;
2230 /* Test that 'tags' is nonzero to ensure that only flows that include an
2231 * OFPP_NORMAL action are used for learning and bond slave rebalancing.
2232 * This works because OFPP_NORMAL always sets a nonzero tag value.
2234 * Feed information from the active flows back into the learning table to
2235 * ensure that table is always in sync with what is actually flowing
2236 * through the datapath. */
2238 || !is_admissible(ofproto, &facet->flow, false, &dummy,
2239 &vlan, &in_bundle)) {
2243 update_learning_table(ofproto, &facet->flow, vlan, in_bundle);
2245 if (!ofproto->has_bonded_bundles) {
2249 /* This loop feeds byte counters to bond_account() for rebalancing to use
2250 * as a basis. We also need to track the actual VLAN on which the packet
2251 * is going to be sent to ensure that it matches the one passed to
2252 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2254 vlan_tci = facet->flow.vlan_tci;
2255 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2256 struct ofport_dpif *port;
2258 switch (nl_attr_type(a)) {
2259 case ODP_ACTION_ATTR_OUTPUT:
2260 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2261 if (port && port->bundle && port->bundle->bond) {
2262 bond_account(port->bundle->bond, &facet->flow,
2263 vlan_tci_to_openflow_vlan(vlan_tci), n_bytes);
2267 case ODP_ACTION_ATTR_STRIP_VLAN:
2268 vlan_tci = htons(0);
2271 case ODP_ACTION_ATTR_SET_DL_TCI:
2272 vlan_tci = nl_attr_get_be16(a);
2278 /* If 'rule' is installed in the datapath, uninstalls it. */
2280 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2282 if (facet->installed) {
2283 struct odputil_keybuf keybuf;
2284 struct dpif_flow_stats stats;
2288 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2289 odp_flow_key_from_flow(&key, &facet->flow);
2291 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2292 facet_reset_dp_stats(facet, &stats);
2294 facet_update_stats(p, facet, &stats);
2296 facet->installed = false;
2298 assert(facet->dp_packet_count == 0);
2299 assert(facet->dp_byte_count == 0);
2303 /* Returns true if the only action for 'facet' is to send to the controller.
2304 * (We don't report NetFlow expiration messages for such facets because they
2305 * are just part of the control logic for the network, not real traffic). */
2307 facet_is_controller_flow(struct facet *facet)
2310 && facet->rule->up.n_actions == 1
2311 && action_outputs_to_port(&facet->rule->up.actions[0],
2312 htons(OFPP_CONTROLLER)));
2315 /* Resets 'facet''s datapath statistics counters. This should be called when
2316 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2317 * it should contain the statistics returned by dpif when 'facet' was reset in
2318 * the datapath. 'stats' will be modified to only included statistics new
2319 * since 'facet' was last updated. */
2321 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2323 if (stats && facet->dp_packet_count <= stats->n_packets
2324 && facet->dp_byte_count <= stats->n_bytes) {
2325 stats->n_packets -= facet->dp_packet_count;
2326 stats->n_bytes -= facet->dp_byte_count;
2329 facet->dp_packet_count = 0;
2330 facet->dp_byte_count = 0;
2333 /* Folds all of 'facet''s statistics into its rule. Also updates the
2334 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2335 * 'facet''s statistics in the datapath should have been zeroed and folded into
2336 * its packet and byte counts before this function is called. */
2338 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2340 assert(!facet->dp_byte_count);
2341 assert(!facet->dp_packet_count);
2343 facet_push_stats(facet);
2344 facet_account(ofproto, facet, 0);
2346 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2347 struct ofexpired expired;
2348 expired.flow = facet->flow;
2349 expired.packet_count = facet->packet_count;
2350 expired.byte_count = facet->byte_count;
2351 expired.used = facet->used;
2352 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2355 facet->rule->packet_count += facet->packet_count;
2356 facet->rule->byte_count += facet->byte_count;
2358 /* Reset counters to prevent double counting if 'facet' ever gets
2360 facet->packet_count = 0;
2361 facet->byte_count = 0;
2362 facet->rs_packet_count = 0;
2363 facet->rs_byte_count = 0;
2364 facet->accounted_bytes = 0;
2366 netflow_flow_clear(&facet->nf_flow);
2369 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2370 * Returns it if found, otherwise a null pointer.
2372 * The returned facet might need revalidation; use facet_lookup_valid()
2373 * instead if that is important. */
2374 static struct facet *
2375 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
2377 struct facet *facet;
2379 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2381 if (flow_equal(flow, &facet->flow)) {
2389 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2390 * Returns it if found, otherwise a null pointer.
2392 * The returned facet is guaranteed to be valid. */
2393 static struct facet *
2394 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
2396 struct facet *facet = facet_find(ofproto, flow);
2398 /* The facet we found might not be valid, since we could be in need of
2399 * revalidation. If it is not valid, don't return it. */
2401 && ofproto->need_revalidate
2402 && !facet_revalidate(ofproto, facet)) {
2403 COVERAGE_INC(facet_invalidated);
2410 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2412 * - If the rule found is different from 'facet''s current rule, moves
2413 * 'facet' to the new rule and recompiles its actions.
2415 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2416 * where it is and recompiles its actions anyway.
2418 * - If there is none, destroys 'facet'.
2420 * Returns true if 'facet' still exists, false if it has been destroyed. */
2422 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
2424 struct action_xlate_ctx ctx;
2425 struct ofpbuf *odp_actions;
2426 struct rule_dpif *new_rule;
2427 bool actions_changed;
2429 COVERAGE_INC(facet_revalidate);
2431 /* Determine the new rule. */
2432 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
2434 /* No new rule, so delete the facet. */
2435 facet_remove(ofproto, facet);
2439 /* Calculate new ODP actions.
2441 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2442 * emit a NetFlow expiration and, if so, we need to have the old state
2443 * around to properly compose it. */
2444 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2445 odp_actions = xlate_actions(&ctx,
2446 new_rule->up.actions, new_rule->up.n_actions);
2447 actions_changed = (facet->actions_len != odp_actions->size
2448 || memcmp(facet->actions, odp_actions->data,
2449 facet->actions_len));
2451 /* If the ODP actions changed or the installability changed, then we need
2452 * to talk to the datapath. */
2453 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2454 if (ctx.may_set_up_flow) {
2455 struct dpif_flow_stats stats;
2457 facet_put__(ofproto, facet,
2458 odp_actions->data, odp_actions->size, &stats);
2459 facet_update_stats(ofproto, facet, &stats);
2461 facet_uninstall(ofproto, facet);
2464 /* The datapath flow is gone or has zeroed stats, so push stats out of
2465 * 'facet' into 'rule'. */
2466 facet_flush_stats(ofproto, facet);
2469 /* Update 'facet' now that we've taken care of all the old state. */
2470 facet->tags = ctx.tags;
2471 facet->nf_flow.output_iface = ctx.nf_output_iface;
2472 facet->may_install = ctx.may_set_up_flow;
2473 if (actions_changed) {
2474 free(facet->actions);
2475 facet->actions_len = odp_actions->size;
2476 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2478 if (facet->rule != new_rule) {
2479 COVERAGE_INC(facet_changed_rule);
2480 list_remove(&facet->list_node);
2481 list_push_back(&new_rule->facets, &facet->list_node);
2482 facet->rule = new_rule;
2483 facet->used = new_rule->up.created;
2484 facet->rs_used = facet->used;
2487 ofpbuf_delete(odp_actions);
2492 /* Updates 'facet''s used time. Caller is responsible for calling
2493 * facet_push_stats() to update the flows which 'facet' resubmits into. */
2495 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
2498 if (used > facet->used) {
2500 if (used > facet->rule->used) {
2501 facet->rule->used = used;
2503 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
2507 /* Folds the statistics from 'stats' into the counters in 'facet'.
2509 * Because of the meaning of a facet's counters, it only makes sense to do this
2510 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
2511 * packet that was sent by hand or if it represents statistics that have been
2512 * cleared out of the datapath. */
2514 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
2515 const struct dpif_flow_stats *stats)
2517 if (stats->n_packets || stats->used > facet->used) {
2518 facet_update_time(ofproto, facet, stats->used);
2519 facet->packet_count += stats->n_packets;
2520 facet->byte_count += stats->n_bytes;
2521 facet_push_stats(facet);
2522 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
2527 facet_push_stats(struct facet *facet)
2529 uint64_t rs_packets, rs_bytes;
2531 assert(facet->packet_count >= facet->rs_packet_count);
2532 assert(facet->byte_count >= facet->rs_byte_count);
2533 assert(facet->used >= facet->rs_used);
2535 rs_packets = facet->packet_count - facet->rs_packet_count;
2536 rs_bytes = facet->byte_count - facet->rs_byte_count;
2538 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
2539 facet->rs_packet_count = facet->packet_count;
2540 facet->rs_byte_count = facet->byte_count;
2541 facet->rs_used = facet->used;
2543 flow_push_stats(facet->rule, &facet->flow,
2544 rs_packets, rs_bytes, facet->used);
2548 struct ofproto_push {
2549 struct action_xlate_ctx ctx;
2556 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
2558 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
2561 rule->packet_count += push->packets;
2562 rule->byte_count += push->bytes;
2563 rule->used = MAX(push->used, rule->used);
2567 /* Pushes flow statistics to the rules which 'flow' resubmits into given
2568 * 'rule''s actions. */
2570 flow_push_stats(const struct rule_dpif *rule,
2571 struct flow *flow, uint64_t packets, uint64_t bytes,
2574 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2575 struct ofproto_push push;
2577 push.packets = packets;
2581 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
2582 push.ctx.resubmit_hook = push_resubmit;
2583 ofpbuf_delete(xlate_actions(&push.ctx,
2584 rule->up.actions, rule->up.n_actions));
2589 static struct rule_dpif *
2590 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
2592 return rule_dpif_cast(rule_from_cls_rule(
2593 classifier_lookup(&ofproto->up.tables[0],
2598 complete_operation(struct rule_dpif *rule)
2600 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2602 ofproto->need_revalidate = true;
2604 struct dpif_completion *c = xmalloc(sizeof *c);
2605 c->op = rule->up.pending;
2606 list_push_back(&ofproto->completions, &c->list_node);
2608 ofoperation_complete(rule->up.pending, 0);
2612 static struct rule *
2615 struct rule_dpif *rule = xmalloc(sizeof *rule);
2620 rule_dealloc(struct rule *rule_)
2622 struct rule_dpif *rule = rule_dpif_cast(rule_);
2627 rule_construct(struct rule *rule_)
2629 struct rule_dpif *rule = rule_dpif_cast(rule_);
2630 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2631 struct rule_dpif *victim;
2634 error = validate_actions(rule->up.actions, rule->up.n_actions,
2635 &rule->up.cr.flow, ofproto->max_ports);
2640 rule->used = rule->up.created;
2641 rule->packet_count = 0;
2642 rule->byte_count = 0;
2644 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
2645 if (victim && !list_is_empty(&victim->facets)) {
2646 struct facet *facet;
2648 rule->facets = victim->facets;
2649 list_moved(&rule->facets);
2650 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2654 /* Must avoid list_moved() in this case. */
2655 list_init(&rule->facets);
2658 complete_operation(rule);
2663 rule_destruct(struct rule *rule_)
2665 struct rule_dpif *rule = rule_dpif_cast(rule_);
2666 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2667 struct facet *facet, *next_facet;
2669 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2670 facet_revalidate(ofproto, facet);
2673 complete_operation(rule);
2677 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
2679 struct rule_dpif *rule = rule_dpif_cast(rule_);
2680 struct facet *facet;
2682 /* Start from historical data for 'rule' itself that are no longer tracked
2683 * in facets. This counts, for example, facets that have expired. */
2684 *packets = rule->packet_count;
2685 *bytes = rule->byte_count;
2687 /* Add any statistics that are tracked by facets. This includes
2688 * statistical data recently updated by ofproto_update_stats() as well as
2689 * stats for packets that were executed "by hand" via dpif_execute(). */
2690 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2691 *packets += facet->packet_count;
2692 *bytes += facet->byte_count;
2697 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
2699 struct rule_dpif *rule = rule_dpif_cast(rule_);
2700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2701 struct action_xlate_ctx ctx;
2702 struct ofpbuf *odp_actions;
2703 struct facet *facet;
2706 /* First look for a related facet. If we find one, account it to that. */
2707 facet = facet_lookup_valid(ofproto, flow);
2708 if (facet && facet->rule == rule) {
2709 facet_execute(ofproto, facet, packet);
2713 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2714 * create a new facet for it and use that. */
2715 if (rule_dpif_lookup(ofproto, flow) == rule) {
2716 facet = facet_create(rule, flow, packet);
2717 facet_execute(ofproto, facet, packet);
2718 facet_install(ofproto, facet, true);
2722 /* We can't account anything to a facet. If we were to try, then that
2723 * facet would have a non-matching rule, busting our invariants. */
2724 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
2725 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2726 size = packet->size;
2727 if (execute_odp_actions(ofproto, flow, odp_actions->data,
2728 odp_actions->size, packet)) {
2729 rule->used = time_msec();
2730 rule->packet_count++;
2731 rule->byte_count += size;
2732 flow_push_stats(rule, flow, 1, size, rule->used);
2734 ofpbuf_delete(odp_actions);
2740 rule_modify_actions(struct rule *rule_)
2742 struct rule_dpif *rule = rule_dpif_cast(rule_);
2743 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2746 error = validate_actions(rule->up.actions, rule->up.n_actions,
2747 &rule->up.cr.flow, ofproto->max_ports);
2749 ofoperation_complete(rule->up.pending, error);
2753 complete_operation(rule);
2756 /* Sends 'packet' out of port 'odp_port' within 'p'.
2757 * Returns 0 if successful, otherwise a positive errno value. */
2759 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
2760 const struct ofpbuf *packet)
2762 struct ofpbuf key, odp_actions;
2763 struct odputil_keybuf keybuf;
2767 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
2768 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2769 odp_flow_key_from_flow(&key, &flow);
2771 ofpbuf_init(&odp_actions, 32);
2772 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2773 error = dpif_execute(ofproto->dpif,
2775 odp_actions.data, odp_actions.size,
2777 ofpbuf_uninit(&odp_actions);
2780 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
2781 ofproto->up.name, odp_port, strerror(error));
2786 /* OpenFlow to ODP action translation. */
2788 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2789 struct action_xlate_ctx *ctx);
2790 static void xlate_normal(struct action_xlate_ctx *);
2793 commit_odp_actions(struct action_xlate_ctx *ctx)
2795 const struct flow *flow = &ctx->flow;
2796 struct flow *base = &ctx->base_flow;
2797 struct ofpbuf *odp_actions = ctx->odp_actions;
2799 if (base->tun_id != flow->tun_id) {
2800 nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id);
2801 base->tun_id = flow->tun_id;
2804 if (base->nw_src != flow->nw_src) {
2805 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src);
2806 base->nw_src = flow->nw_src;
2809 if (base->nw_dst != flow->nw_dst) {
2810 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst);
2811 base->nw_dst = flow->nw_dst;
2814 if (base->nw_tos != flow->nw_tos) {
2815 nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos);
2816 base->nw_tos = flow->nw_tos;
2819 if (base->vlan_tci != flow->vlan_tci) {
2820 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
2821 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2823 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2824 flow->vlan_tci & ~htons(VLAN_CFI));
2826 base->vlan_tci = flow->vlan_tci;
2829 if (base->tp_src != flow->tp_src) {
2830 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src);
2831 base->tp_src = flow->tp_src;
2834 if (base->tp_dst != flow->tp_dst) {
2835 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst);
2836 base->tp_dst = flow->tp_dst;
2839 if (!eth_addr_equals(base->dl_src, flow->dl_src)) {
2840 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2841 flow->dl_src, ETH_ADDR_LEN);
2842 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
2845 if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) {
2846 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2847 flow->dl_dst, ETH_ADDR_LEN);
2848 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
2851 if (ctx->base_priority != ctx->priority) {
2852 if (ctx->priority) {
2853 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY,
2856 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2858 ctx->base_priority = ctx->priority;
2863 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
2865 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
2866 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
2869 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
2870 /* Forwarding disabled on port. */
2875 * We don't have an ofport record for this port, but it doesn't hurt to
2876 * allow forwarding to it anyhow. Maybe such a port will appear later
2877 * and we're pre-populating the flow table.
2881 commit_odp_actions(ctx);
2882 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2883 ctx->nf_output_iface = ofp_port;
2887 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2889 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2890 struct rule_dpif *rule;
2891 uint16_t old_in_port;
2893 /* Look up a flow with 'in_port' as the input port. Then restore the
2894 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2895 * have surprising behavior). */
2896 old_in_port = ctx->flow.in_port;
2897 ctx->flow.in_port = in_port;
2898 rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow);
2899 ctx->flow.in_port = old_in_port;
2901 if (ctx->resubmit_hook) {
2902 ctx->resubmit_hook(ctx, rule);
2907 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
2911 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2913 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2914 MAX_RESUBMIT_RECURSION);
2919 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
2921 struct ofport_dpif *ofport;
2923 commit_odp_actions(ctx);
2924 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
2925 uint16_t ofp_port = ofport->up.ofp_port;
2926 if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) {
2927 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT,
2932 ctx->nf_output_iface = NF_OUT_FLOOD;
2936 xlate_output_action__(struct action_xlate_ctx *ctx,
2937 uint16_t port, uint16_t max_len)
2939 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2941 ctx->nf_output_iface = NF_OUT_DROP;
2945 add_output_action(ctx, ctx->flow.in_port);
2948 xlate_table_action(ctx, ctx->flow.in_port);
2954 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
2957 flood_packets(ctx, htonl(0));
2959 case OFPP_CONTROLLER:
2960 commit_odp_actions(ctx);
2961 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_USERSPACE, max_len);
2964 add_output_action(ctx, OFPP_LOCAL);
2969 if (port != ctx->flow.in_port) {
2970 add_output_action(ctx, port);
2975 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2976 ctx->nf_output_iface = NF_OUT_FLOOD;
2977 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2978 ctx->nf_output_iface = prev_nf_output_iface;
2979 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2980 ctx->nf_output_iface != NF_OUT_FLOOD) {
2981 ctx->nf_output_iface = NF_OUT_MULTI;
2986 xlate_output_action(struct action_xlate_ctx *ctx,
2987 const struct ofp_action_output *oao)
2989 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2993 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2994 const struct ofp_action_enqueue *oae)
2996 uint16_t ofp_port, odp_port;
2997 uint32_t ctx_priority, priority;
3000 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3003 /* Fall back to ordinary output action. */
3004 xlate_output_action__(ctx, ntohs(oae->port), 0);
3008 /* Figure out ODP output port. */
3009 ofp_port = ntohs(oae->port);
3010 if (ofp_port == OFPP_IN_PORT) {
3011 ofp_port = ctx->flow.in_port;
3013 odp_port = ofp_port_to_odp_port(ofp_port);
3015 /* Add ODP actions. */
3016 ctx_priority = ctx->priority;
3017 ctx->priority = priority;
3018 add_output_action(ctx, odp_port);
3019 ctx->priority = ctx_priority;
3021 /* Update NetFlow output port. */
3022 if (ctx->nf_output_iface == NF_OUT_DROP) {
3023 ctx->nf_output_iface = odp_port;
3024 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3025 ctx->nf_output_iface = NF_OUT_MULTI;
3030 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3031 const struct nx_action_set_queue *nasq)
3036 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3039 /* Couldn't translate queue to a priority, so ignore. A warning
3040 * has already been logged. */
3044 ctx->priority = priority;
3047 struct xlate_reg_state {
3053 xlate_autopath(struct action_xlate_ctx *ctx,
3054 const struct nx_action_autopath *naa)
3056 uint16_t ofp_port = ntohl(naa->id);
3057 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
3059 if (!port || !port->bundle) {
3060 ofp_port = OFPP_NONE;
3061 } else if (port->bundle->bond) {
3062 /* Autopath does not support VLAN hashing. */
3063 struct ofport_dpif *slave = bond_choose_output_slave(
3064 port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags);
3066 ofp_port = slave->up.ofp_port;
3069 autopath_execute(naa, &ctx->flow, ofp_port);
3073 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
3075 struct ofproto_dpif *ofproto = ofproto_;
3076 struct ofport_dpif *port;
3086 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3089 port = get_ofp_port(ofproto, ofp_port);
3090 return port ? port->may_enable : false;
3095 do_xlate_actions(const union ofp_action *in, size_t n_in,
3096 struct action_xlate_ctx *ctx)
3098 const struct ofport_dpif *port;
3099 const union ofp_action *ia;
3102 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
3104 && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3105 port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3106 ? htonl(OFPPC_NO_RECV_STP)
3107 : htonl(OFPPC_NO_RECV))) {
3108 /* Drop this flow. */
3112 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
3113 const struct ofp_action_dl_addr *oada;
3114 const struct nx_action_resubmit *nar;
3115 const struct nx_action_set_tunnel *nast;
3116 const struct nx_action_set_queue *nasq;
3117 const struct nx_action_multipath *nam;
3118 const struct nx_action_autopath *naa;
3119 const struct nx_action_bundle *nab;
3120 enum ofputil_action_code code;
3123 code = ofputil_decode_action_unsafe(ia);
3125 case OFPUTIL_OFPAT_OUTPUT:
3126 xlate_output_action(ctx, &ia->output);
3129 case OFPUTIL_OFPAT_SET_VLAN_VID:
3130 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3131 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3134 case OFPUTIL_OFPAT_SET_VLAN_PCP:
3135 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3136 ctx->flow.vlan_tci |= htons(
3137 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3140 case OFPUTIL_OFPAT_STRIP_VLAN:
3141 ctx->flow.vlan_tci = htons(0);
3144 case OFPUTIL_OFPAT_SET_DL_SRC:
3145 oada = ((struct ofp_action_dl_addr *) ia);
3146 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3149 case OFPUTIL_OFPAT_SET_DL_DST:
3150 oada = ((struct ofp_action_dl_addr *) ia);
3151 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3154 case OFPUTIL_OFPAT_SET_NW_SRC:
3155 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3158 case OFPUTIL_OFPAT_SET_NW_DST:
3159 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3162 case OFPUTIL_OFPAT_SET_NW_TOS:
3163 ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK;
3166 case OFPUTIL_OFPAT_SET_TP_SRC:
3167 ctx->flow.tp_src = ia->tp_port.tp_port;
3170 case OFPUTIL_OFPAT_SET_TP_DST:
3171 ctx->flow.tp_dst = ia->tp_port.tp_port;
3174 case OFPUTIL_OFPAT_ENQUEUE:
3175 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3178 case OFPUTIL_NXAST_RESUBMIT:
3179 nar = (const struct nx_action_resubmit *) ia;
3180 xlate_table_action(ctx, ntohs(nar->in_port));
3183 case OFPUTIL_NXAST_SET_TUNNEL:
3184 nast = (const struct nx_action_set_tunnel *) ia;
3185 tun_id = htonll(ntohl(nast->tun_id));
3186 ctx->flow.tun_id = tun_id;
3189 case OFPUTIL_NXAST_SET_QUEUE:
3190 nasq = (const struct nx_action_set_queue *) ia;
3191 xlate_set_queue_action(ctx, nasq);
3194 case OFPUTIL_NXAST_POP_QUEUE:
3198 case OFPUTIL_NXAST_REG_MOVE:
3199 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
3203 case OFPUTIL_NXAST_REG_LOAD:
3204 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
3208 case OFPUTIL_NXAST_NOTE:
3209 /* Nothing to do. */
3212 case OFPUTIL_NXAST_SET_TUNNEL64:
3213 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
3214 ctx->flow.tun_id = tun_id;
3217 case OFPUTIL_NXAST_MULTIPATH:
3218 nam = (const struct nx_action_multipath *) ia;
3219 multipath_execute(nam, &ctx->flow);
3222 case OFPUTIL_NXAST_AUTOPATH:
3223 naa = (const struct nx_action_autopath *) ia;
3224 xlate_autopath(ctx, naa);
3227 case OFPUTIL_NXAST_BUNDLE:
3228 ctx->ofproto->has_bundle_action = true;
3229 nab = (const struct nx_action_bundle *) ia;
3230 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
3235 case OFPUTIL_NXAST_BUNDLE_LOAD:
3236 ctx->ofproto->has_bundle_action = true;
3237 nab = (const struct nx_action_bundle *) ia;
3238 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
3246 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3247 struct ofproto_dpif *ofproto, const struct flow *flow,
3248 const struct ofpbuf *packet)
3250 ctx->ofproto = ofproto;
3252 ctx->packet = packet;
3253 ctx->resubmit_hook = NULL;
3256 static struct ofpbuf *
3257 xlate_actions(struct action_xlate_ctx *ctx,
3258 const union ofp_action *in, size_t n_in)
3260 COVERAGE_INC(ofproto_dpif_xlate);
3262 ctx->odp_actions = ofpbuf_new(512);
3264 ctx->may_set_up_flow = true;
3265 ctx->nf_output_iface = NF_OUT_DROP;
3268 ctx->base_priority = 0;
3269 ctx->base_flow = ctx->flow;
3271 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
3272 ctx->may_set_up_flow = false;
3274 do_xlate_actions(in, n_in, ctx);
3277 /* Check with in-band control to see if we're allowed to set up this
3279 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
3280 ctx->odp_actions->data,
3281 ctx->odp_actions->size)) {
3282 ctx->may_set_up_flow = false;
3285 return ctx->odp_actions;
3288 /* OFPP_NORMAL implementation. */
3291 struct ofport_dpif *port;
3296 struct dst builtin[32];
3298 size_t n, allocated;
3301 static void dst_set_init(struct dst_set *);
3302 static void dst_set_add(struct dst_set *, const struct dst *);
3303 static void dst_set_free(struct dst_set *);
3305 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
3308 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
3309 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
3311 dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE
3312 : in_bundle->vlan >= 0 ? in_bundle->vlan
3313 : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE
3314 : vlan_tci_to_vid(ctx->flow.vlan_tci));
3316 dst->port = (!out_bundle->bond
3317 ? ofbundle_get_a_port(out_bundle)
3318 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
3319 dst->vlan, &ctx->tags));
3321 return dst->port != NULL;
3325 mirror_mask_ffs(mirror_mask_t mask)
3327 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
3332 dst_set_init(struct dst_set *set)
3334 set->dsts = set->builtin;
3336 set->allocated = ARRAY_SIZE(set->builtin);
3340 dst_set_add(struct dst_set *set, const struct dst *dst)
3342 if (set->n >= set->allocated) {
3343 size_t new_allocated;
3344 struct dst *new_dsts;
3346 new_allocated = set->allocated * 2;
3347 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
3348 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
3352 set->dsts = new_dsts;
3353 set->allocated = new_allocated;
3355 set->dsts[set->n++] = *dst;
3359 dst_set_free(struct dst_set *set)
3361 if (set->dsts != set->builtin) {
3367 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
3370 for (i = 0; i < set->n; i++) {
3371 if (set->dsts[i].vlan == test->vlan
3372 && set->dsts[i].port == test->port) {
3380 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
3382 return (bundle->vlan < 0
3383 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
3387 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
3389 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
3392 /* Returns an arbitrary interface within 'bundle'. */
3393 static struct ofport_dpif *
3394 ofbundle_get_a_port(const struct ofbundle *bundle)
3396 return CONTAINER_OF(list_front(&bundle->ports),
3397 struct ofport_dpif, bundle_node);
3401 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
3402 const struct ofbundle *in_bundle,
3403 const struct ofbundle *out_bundle, struct dst_set *set)
3407 if (out_bundle == OFBUNDLE_FLOOD) {
3408 struct ofbundle *bundle;
3410 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
3411 if (bundle != in_bundle
3412 && ofbundle_includes_vlan(bundle, vlan)
3413 && bundle->floodable
3414 && !bundle->mirror_out
3415 && set_dst(ctx, &dst, in_bundle, bundle)) {
3416 dst_set_add(set, &dst);
3419 ctx->nf_output_iface = NF_OUT_FLOOD;
3420 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
3421 dst_set_add(set, &dst);
3422 ctx->nf_output_iface = dst.port->odp_port;
3427 vlan_is_mirrored(const struct ofmirror *m, int vlan)
3429 return !m->vlans || bitmap_is_set(m->vlans, vlan);
3432 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
3433 * to a VLAN. In general most packets may be mirrored but we want to drop
3434 * protocols that may confuse switches. */
3436 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
3438 /* If you change this function's behavior, please update corresponding
3439 * documentation in vswitch.xml at the same time. */
3440 if (dst[0] != 0x01) {
3441 /* All the currently banned MACs happen to start with 01 currently, so
3442 * this is a quick way to eliminate most of the good ones. */
3444 if (eth_addr_is_reserved(dst)) {
3445 /* Drop STP, IEEE pause frames, and other reserved protocols
3446 * (01-80-c2-00-00-0x). */
3450 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
3452 if ((dst[3] & 0xfe) == 0xcc &&
3453 (dst[4] & 0xfe) == 0xcc &&
3454 (dst[5] & 0xfe) == 0xcc) {
3455 /* Drop the following protocols plus others following the same
3458 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
3459 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
3460 STP Uplink Fast (01-00-0c-cd-cd-cd) */
3464 if (!(dst[3] | dst[4] | dst[5])) {
3465 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
3474 compose_mirror_dsts(struct action_xlate_ctx *ctx,
3475 uint16_t vlan, const struct ofbundle *in_bundle,
3476 struct dst_set *set)
3478 struct ofproto_dpif *ofproto = ctx->ofproto;
3479 mirror_mask_t mirrors;
3483 mirrors = in_bundle->src_mirrors;
3484 for (i = 0; i < set->n; i++) {
3485 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
3492 flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3493 if (flow_vlan == 0) {
3494 flow_vlan = OFP_VLAN_NONE;
3498 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
3499 if (vlan_is_mirrored(m, vlan)) {
3503 if (set_dst(ctx, &dst, in_bundle, m->out)
3504 && !dst_is_duplicate(set, &dst)) {
3505 dst_set_add(set, &dst);
3507 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
3508 struct ofbundle *bundle;
3510 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3511 if (ofbundle_includes_vlan(bundle, m->out_vlan)
3512 && set_dst(ctx, &dst, in_bundle, bundle))
3514 if (bundle->vlan < 0) {
3515 dst.vlan = m->out_vlan;
3517 if (dst_is_duplicate(set, &dst)) {
3521 /* Use the vlan tag on the original flow instead of
3522 * the one passed in the vlan parameter. This ensures
3523 * that we compare the vlan from before any implicit
3524 * tagging tags place. This is necessary because
3525 * dst->vlan is the final vlan, after removing implicit
3527 if (bundle == in_bundle && dst.vlan == flow_vlan) {
3528 /* Don't send out input port on same VLAN. */
3531 dst_set_add(set, &dst);
3536 mirrors &= mirrors - 1;
3541 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
3542 const struct ofbundle *in_bundle,
3543 const struct ofbundle *out_bundle)
3545 uint16_t initial_vlan, cur_vlan;
3546 const struct dst *dst;
3550 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
3551 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
3553 /* Output all the packets we can without having to change the VLAN. */
3554 initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3555 if (initial_vlan == 0) {
3556 initial_vlan = OFP_VLAN_NONE;
3558 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3559 if (dst->vlan != initial_vlan) {
3562 nl_msg_put_u32(ctx->odp_actions,
3563 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3566 /* Then output the rest. */
3567 cur_vlan = initial_vlan;
3568 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3569 if (dst->vlan == initial_vlan) {
3572 if (dst->vlan != cur_vlan) {
3573 if (dst->vlan == OFP_VLAN_NONE) {
3574 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
3577 tci = htons(dst->vlan & VLAN_VID_MASK);
3578 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
3579 nl_msg_put_be16(ctx->odp_actions,
3580 ODP_ACTION_ATTR_SET_DL_TCI, tci);
3582 cur_vlan = dst->vlan;
3584 nl_msg_put_u32(ctx->odp_actions,
3585 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3591 /* Returns the effective vlan of a packet, taking into account both the
3592 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
3593 * the packet is untagged and -1 indicates it has an invalid header and
3594 * should be dropped. */
3596 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
3597 struct ofbundle *in_bundle, bool have_packet)
3599 int vlan = vlan_tci_to_vid(flow->vlan_tci);
3600 if (in_bundle->vlan >= 0) {
3603 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3604 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3605 "packet received on port %s configured with "
3606 "implicit VLAN %"PRIu16,
3607 ofproto->up.name, vlan,
3608 in_bundle->name, in_bundle->vlan);
3612 vlan = in_bundle->vlan;
3614 if (!ofbundle_includes_vlan(in_bundle, vlan)) {
3616 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3617 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3618 "packet received on port %s not configured for "
3620 ofproto->up.name, vlan, in_bundle->name, vlan);
3629 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
3630 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
3631 * indicate this; newer upstream kernels use gratuitous ARP requests. */
3633 is_gratuitous_arp(const struct flow *flow)
3635 return (flow->dl_type == htons(ETH_TYPE_ARP)
3636 && eth_addr_is_broadcast(flow->dl_dst)
3637 && (flow->nw_proto == ARP_OP_REPLY
3638 || (flow->nw_proto == ARP_OP_REQUEST
3639 && flow->nw_src == flow->nw_dst)));
3643 update_learning_table(struct ofproto_dpif *ofproto,
3644 const struct flow *flow, int vlan,
3645 struct ofbundle *in_bundle)
3647 struct mac_entry *mac;
3649 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
3653 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
3654 if (is_gratuitous_arp(flow)) {
3655 /* We don't want to learn from gratuitous ARP packets that are
3656 * reflected back over bond slaves so we lock the learning table. */
3657 if (!in_bundle->bond) {
3658 mac_entry_set_grat_arp_lock(mac);
3659 } else if (mac_entry_is_grat_arp_locked(mac)) {
3664 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
3665 /* The log messages here could actually be useful in debugging,
3666 * so keep the rate limit relatively high. */
3667 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3668 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
3669 "on port %s in VLAN %d",
3670 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
3671 in_bundle->name, vlan);
3673 mac->port.p = in_bundle;
3674 tag_set_add(&ofproto->revalidate_set,
3675 mac_learning_changed(ofproto->ml, mac));
3679 /* Determines whether packets in 'flow' within 'br' should be forwarded or
3680 * dropped. Returns true if they may be forwarded, false if they should be
3683 * If 'have_packet' is true, it indicates that the caller is processing a
3684 * received packet. If 'have_packet' is false, then the caller is just
3685 * revalidating an existing flow because configuration has changed. Either
3686 * way, 'have_packet' only affects logging (there is no point in logging errors
3687 * during revalidation).
3689 * Sets '*in_portp' to the input port. This will be a null pointer if
3690 * flow->in_port does not designate a known input port (in which case
3691 * is_admissible() returns false).
3693 * When returning true, sets '*vlanp' to the effective VLAN of the input
3694 * packet, as returned by flow_get_vlan().
3696 * May also add tags to '*tags', although the current implementation only does
3697 * so in one special case.
3700 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
3702 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
3704 struct ofport_dpif *in_port;
3705 struct ofbundle *in_bundle;
3708 /* Find the port and bundle for the received packet. */
3709 in_port = get_ofp_port(ofproto, flow->in_port);
3710 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
3711 if (!in_port || !in_bundle) {
3712 /* No interface? Something fishy... */
3714 /* Odd. A few possible reasons here:
3716 * - We deleted a port but there are still a few packets queued up
3719 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
3720 * we don't know about.
3722 * - Packet arrived on the local port but the local port is not
3725 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3727 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3729 ofproto->up.name, flow->in_port);
3733 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
3738 /* Drop frames for reserved multicast addresses. */
3739 if (eth_addr_is_reserved(flow->dl_dst)) {
3743 /* Drop frames on bundles reserved for mirroring. */
3744 if (in_bundle->mirror_out) {
3746 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3747 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3748 "%s, which is reserved exclusively for mirroring",
3749 ofproto->up.name, in_bundle->name);
3754 if (in_bundle->bond) {
3755 struct mac_entry *mac;
3757 switch (bond_check_admissibility(in_bundle->bond, in_port,
3758 flow->dl_dst, tags)) {
3765 case BV_DROP_IF_MOVED:
3766 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
3767 if (mac && mac->port.p != in_bundle &&
3768 (!is_gratuitous_arp(flow)
3769 || mac_entry_is_grat_arp_locked(mac))) {
3780 xlate_normal(struct action_xlate_ctx *ctx)
3782 struct ofbundle *in_bundle;
3783 struct ofbundle *out_bundle;
3784 struct mac_entry *mac;
3787 /* Check whether we should drop packets in this flow. */
3788 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
3789 &ctx->tags, &vlan, &in_bundle)) {
3794 /* Learn source MAC (but don't try to learn from revalidation). */
3796 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
3799 /* Determine output bundle. */
3800 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
3803 out_bundle = mac->port.p;
3804 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
3805 /* If we are revalidating but don't have a learning entry then eject
3806 * the flow. Installing a flow that floods packets opens up a window
3807 * of time where we could learn from a packet reflected on a bond and
3808 * blackhole packets before the learning table is updated to reflect
3809 * the correct port. */
3810 ctx->may_set_up_flow = false;
3813 out_bundle = OFBUNDLE_FLOOD;
3816 /* Don't send packets out their input bundles. */
3817 if (in_bundle == out_bundle) {
3823 compose_actions(ctx, vlan, in_bundle, out_bundle);
3828 get_drop_frags(struct ofproto *ofproto_)
3830 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3833 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
3838 set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
3840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3842 dpif_set_drop_frags(ofproto->dpif, drop_frags);
3846 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3847 const struct flow *flow,
3848 const union ofp_action *ofp_actions, size_t n_ofp_actions)
3850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3853 error = validate_actions(ofp_actions, n_ofp_actions, flow,
3854 ofproto->max_ports);
3856 struct odputil_keybuf keybuf;
3857 struct action_xlate_ctx ctx;
3858 struct ofpbuf *odp_actions;
3861 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3862 odp_flow_key_from_flow(&key, flow);
3864 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3865 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3866 dpif_execute(ofproto->dpif, key.data, key.size,
3867 odp_actions->data, odp_actions->size, packet);
3868 ofpbuf_delete(odp_actions);
3874 get_netflow_ids(const struct ofproto *ofproto_,
3875 uint8_t *engine_type, uint8_t *engine_id)
3877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3879 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3882 static struct ofproto_dpif *
3883 ofproto_dpif_lookup(const char *name)
3885 struct ofproto *ofproto = ofproto_lookup(name);
3886 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
3887 ? ofproto_dpif_cast(ofproto)
3892 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
3893 const char *args, void *aux OVS_UNUSED)
3895 struct ds ds = DS_EMPTY_INITIALIZER;
3896 const struct ofproto_dpif *ofproto;
3897 const struct mac_entry *e;
3899 ofproto = ofproto_dpif_lookup(args);
3901 unixctl_command_reply(conn, 501, "no such bridge");
3905 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3906 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3907 struct ofbundle *bundle = e->port.p;
3908 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
3909 ofbundle_get_a_port(bundle)->odp_port,
3910 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
3912 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3916 struct ofproto_trace {
3917 struct action_xlate_ctx ctx;
3923 trace_format_rule(struct ds *result, int level, const struct rule *rule)
3925 ds_put_char_multiple(result, '\t', level);
3927 ds_put_cstr(result, "No match\n");
3931 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
3932 ntohll(rule->flow_cookie));
3933 cls_rule_format(&rule->cr, result);
3934 ds_put_char(result, '\n');
3936 ds_put_char_multiple(result, '\t', level);
3937 ds_put_cstr(result, "OpenFlow ");
3938 ofp_print_actions(result, rule->actions, rule->n_actions);
3939 ds_put_char(result, '\n');
3943 trace_format_flow(struct ds *result, int level, const char *title,
3944 struct ofproto_trace *trace)
3946 ds_put_char_multiple(result, '\t', level);
3947 ds_put_format(result, "%s: ", title);
3948 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
3949 ds_put_cstr(result, "unchanged");
3951 flow_format(result, &trace->ctx.flow);
3952 trace->flow = trace->ctx.flow;
3954 ds_put_char(result, '\n');
3958 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3960 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
3961 struct ds *result = trace->result;
3963 ds_put_char(result, '\n');
3964 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
3965 trace_format_rule(result, ctx->recurse + 1, &rule->up);
3969 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
3970 void *aux OVS_UNUSED)
3972 char *dpname, *in_port_s, *tun_id_s, *packet_s;
3973 char *args = xstrdup(args_);
3974 char *save_ptr = NULL;
3975 struct ofproto_dpif *ofproto;
3976 struct ofpbuf packet;
3977 struct rule_dpif *rule;
3984 ofpbuf_init(&packet, strlen(args) / 2);
3987 dpname = strtok_r(args, " ", &save_ptr);
3988 tun_id_s = strtok_r(NULL, " ", &save_ptr);
3989 in_port_s = strtok_r(NULL, " ", &save_ptr);
3990 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
3991 if (!dpname || !in_port_s || !packet_s) {
3992 unixctl_command_reply(conn, 501, "Bad command syntax");
3996 ofproto = ofproto_dpif_lookup(dpname);
3998 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4003 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4004 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4006 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4007 packet_s += strspn(packet_s, " ");
4008 if (*packet_s != '\0') {
4009 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4012 if (packet.size < ETH_HEADER_LEN) {
4013 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4017 ds_put_cstr(&result, "Packet: ");
4018 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4019 ds_put_cstr(&result, s);
4022 flow_extract(&packet, tun_id, in_port, &flow);
4023 ds_put_cstr(&result, "Flow: ");
4024 flow_format(&result, &flow);
4025 ds_put_char(&result, '\n');
4027 rule = rule_dpif_lookup(ofproto, &flow);
4028 trace_format_rule(&result, 0, &rule->up);
4030 struct ofproto_trace trace;
4031 struct ofpbuf *odp_actions;
4033 trace.result = &result;
4035 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4036 trace.ctx.resubmit_hook = trace_resubmit;
4037 odp_actions = xlate_actions(&trace.ctx,
4038 rule->up.actions, rule->up.n_actions);
4040 ds_put_char(&result, '\n');
4041 trace_format_flow(&result, 0, "Final flow", &trace);
4042 ds_put_cstr(&result, "Datapath actions: ");
4043 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4044 ofpbuf_delete(odp_actions);
4047 unixctl_command_reply(conn, 200, ds_cstr(&result));
4050 ds_destroy(&result);
4051 ofpbuf_uninit(&packet);
4056 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
4057 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4060 unixctl_command_reply(conn, 200, NULL);
4064 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
4065 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4068 unixctl_command_reply(conn, 200, NULL);
4072 ofproto_dpif_unixctl_init(void)
4074 static bool registered;
4080 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4081 unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
4083 unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL);
4084 unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL);
4087 const struct ofproto_class ofproto_dpif_class = {
4114 port_is_lacp_current,
4115 NULL, /* rule_choose_table */
4122 rule_modify_actions,
4135 is_mirror_output_bundle,