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();
718 if (ofproto->sflow) {
719 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
720 netdev_get_name(port->up.netdev));
727 port_destruct(struct ofport *port_)
729 struct ofport_dpif *port = ofport_dpif_cast(port_);
730 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
732 bundle_remove(port_);
733 set_cfm(port_, NULL);
734 if (ofproto->sflow) {
735 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
740 port_modified(struct ofport *port_)
742 struct ofport_dpif *port = ofport_dpif_cast(port_);
744 if (port->bundle && port->bundle->bond) {
745 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
750 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
752 struct ofport_dpif *port = ofport_dpif_cast(port_);
753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
754 ovs_be32 changed = old_config ^ port->up.opp.config;
756 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
757 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
758 ofproto->need_revalidate = true;
763 set_sflow(struct ofproto *ofproto_,
764 const struct ofproto_sflow_options *sflow_options)
766 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
767 struct dpif_sflow *ds = ofproto->sflow;
770 struct ofport_dpif *ofport;
772 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
773 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
774 dpif_sflow_add_port(ds, ofport->odp_port,
775 netdev_get_name(ofport->up.netdev));
778 dpif_sflow_set_options(ds, sflow_options);
780 dpif_sflow_destroy(ds);
781 ofproto->sflow = NULL;
787 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
789 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
796 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
799 if (cfm_configure(ofport->cfm, s)) {
805 cfm_destroy(ofport->cfm);
811 get_cfm_fault(const struct ofport *ofport_)
813 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
815 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
820 /* Expires all MAC learning entries associated with 'port' and forces ofproto
821 * to revalidate every flow. */
823 bundle_flush_macs(struct ofbundle *bundle)
825 struct ofproto_dpif *ofproto = bundle->ofproto;
826 struct mac_learning *ml = ofproto->ml;
827 struct mac_entry *mac, *next_mac;
829 ofproto->need_revalidate = true;
830 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
831 if (mac->port.p == bundle) {
832 mac_learning_expire(ml, mac);
837 static struct ofbundle *
838 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
840 struct ofbundle *bundle;
842 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
844 if (bundle->aux == aux) {
851 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
852 * ones that are found to 'bundles'. */
854 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
855 void **auxes, size_t n_auxes,
856 struct hmapx *bundles)
861 for (i = 0; i < n_auxes; i++) {
862 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
864 hmapx_add(bundles, bundle);
870 bundle_del_port(struct ofport_dpif *port)
872 struct ofbundle *bundle = port->bundle;
874 bundle->ofproto->need_revalidate = true;
876 list_remove(&port->bundle_node);
880 lacp_slave_unregister(bundle->lacp, port);
883 bond_slave_unregister(bundle->bond, port);
886 bundle->floodable = true;
887 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
888 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
889 bundle->floodable = false;
895 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
896 struct lacp_slave_settings *lacp,
897 uint32_t bond_stable_id)
899 struct ofport_dpif *port;
901 port = get_ofp_port(bundle->ofproto, ofp_port);
906 if (port->bundle != bundle) {
907 bundle->ofproto->need_revalidate = true;
909 bundle_del_port(port);
912 port->bundle = bundle;
913 list_push_back(&bundle->ports, &port->bundle_node);
914 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
915 bundle->floodable = false;
919 lacp_slave_register(bundle->lacp, port, lacp);
922 port->bond_stable_id = bond_stable_id;
928 bundle_destroy(struct ofbundle *bundle)
930 struct ofproto_dpif *ofproto;
931 struct ofport_dpif *port, *next_port;
938 ofproto = bundle->ofproto;
939 for (i = 0; i < MAX_MIRRORS; i++) {
940 struct ofmirror *m = ofproto->mirrors[i];
942 if (m->out == bundle) {
944 } else if (hmapx_find_and_delete(&m->srcs, bundle)
945 || hmapx_find_and_delete(&m->dsts, bundle)) {
946 ofproto->need_revalidate = true;
951 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
952 bundle_del_port(port);
955 bundle_flush_macs(bundle);
956 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
958 free(bundle->trunks);
959 lacp_destroy(bundle->lacp);
960 bond_destroy(bundle->bond);
965 bundle_set(struct ofproto *ofproto_, void *aux,
966 const struct ofproto_bundle_settings *s)
968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
969 bool need_flush = false;
970 const unsigned long *trunks;
971 struct ofport_dpif *port;
972 struct ofbundle *bundle;
977 bundle_destroy(bundle_lookup(ofproto, aux));
981 assert(s->n_slaves == 1 || s->bond != NULL);
982 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
984 bundle = bundle_lookup(ofproto, aux);
986 bundle = xmalloc(sizeof *bundle);
988 bundle->ofproto = ofproto;
989 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
990 hash_pointer(aux, 0));
994 list_init(&bundle->ports);
996 bundle->trunks = NULL;
1000 bundle->floodable = true;
1002 bundle->src_mirrors = 0;
1003 bundle->dst_mirrors = 0;
1004 bundle->mirror_out = 0;
1007 if (!bundle->name || strcmp(s->name, bundle->name)) {
1009 bundle->name = xstrdup(s->name);
1014 if (!bundle->lacp) {
1015 bundle->lacp = lacp_create();
1017 lacp_configure(bundle->lacp, s->lacp);
1019 lacp_destroy(bundle->lacp);
1020 bundle->lacp = NULL;
1023 /* Update set of ports. */
1025 for (i = 0; i < s->n_slaves; i++) {
1026 if (!bundle_add_port(bundle, s->slaves[i],
1027 s->lacp ? &s->lacp_slaves[i] : NULL,
1028 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1032 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1033 struct ofport_dpif *next_port;
1035 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1036 for (i = 0; i < s->n_slaves; i++) {
1037 if (s->slaves[i] == port->up.ofp_port) {
1042 bundle_del_port(port);
1046 assert(list_size(&bundle->ports) <= s->n_slaves);
1048 if (list_is_empty(&bundle->ports)) {
1049 bundle_destroy(bundle);
1054 if (s->vlan != bundle->vlan) {
1055 bundle->vlan = s->vlan;
1059 /* Get trunked VLANs. */
1060 trunks = s->vlan == -1 ? NULL : s->trunks;
1061 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1062 free(bundle->trunks);
1063 bundle->trunks = vlan_bitmap_clone(trunks);
1068 if (!list_is_short(&bundle->ports)) {
1069 bundle->ofproto->has_bonded_bundles = true;
1071 if (bond_reconfigure(bundle->bond, s->bond)) {
1072 ofproto->need_revalidate = true;
1075 bundle->bond = bond_create(s->bond);
1076 ofproto->need_revalidate = true;
1079 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1080 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1084 bond_destroy(bundle->bond);
1085 bundle->bond = NULL;
1088 /* If we changed something that would affect MAC learning, un-learn
1089 * everything on this port and force flow revalidation. */
1091 bundle_flush_macs(bundle);
1098 bundle_remove(struct ofport *port_)
1100 struct ofport_dpif *port = ofport_dpif_cast(port_);
1101 struct ofbundle *bundle = port->bundle;
1104 bundle_del_port(port);
1105 if (list_is_empty(&bundle->ports)) {
1106 bundle_destroy(bundle);
1107 } else if (list_is_short(&bundle->ports)) {
1108 bond_destroy(bundle->bond);
1109 bundle->bond = NULL;
1115 send_pdu_cb(void *port_, const struct lacp_pdu *pdu)
1117 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1118 struct ofport_dpif *port = port_;
1119 uint8_t ea[ETH_ADDR_LEN];
1122 error = netdev_get_etheraddr(port->up.netdev, ea);
1124 struct lacp_pdu *packet_pdu;
1125 struct ofpbuf packet;
1127 ofpbuf_init(&packet, 0);
1128 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1129 sizeof *packet_pdu);
1131 error = netdev_send(port->up.netdev, &packet);
1133 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1134 "(%s)", port->bundle->name,
1135 netdev_get_name(port->up.netdev), strerror(error));
1137 ofpbuf_uninit(&packet);
1139 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1140 "%s (%s)", port->bundle->name,
1141 netdev_get_name(port->up.netdev), strerror(error));
1146 bundle_send_learning_packets(struct ofbundle *bundle)
1148 struct ofproto_dpif *ofproto = bundle->ofproto;
1149 int error, n_packets, n_errors;
1150 struct mac_entry *e;
1152 error = n_packets = n_errors = 0;
1153 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1154 if (e->port.p != bundle) {
1155 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1165 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1166 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1167 "packets, last error was: %s",
1168 bundle->name, n_errors, n_packets, strerror(error));
1170 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1171 bundle->name, n_packets);
1176 bundle_run(struct ofbundle *bundle)
1179 lacp_run(bundle->lacp, send_pdu_cb);
1182 struct ofport_dpif *port;
1184 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1185 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1188 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1189 lacp_negotiated(bundle->lacp));
1190 if (bond_should_send_learning_packets(bundle->bond)) {
1191 bundle_send_learning_packets(bundle);
1197 bundle_wait(struct ofbundle *bundle)
1200 lacp_wait(bundle->lacp);
1203 bond_wait(bundle->bond);
1210 mirror_scan(struct ofproto_dpif *ofproto)
1214 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1215 if (!ofproto->mirrors[idx]) {
1222 static struct ofmirror *
1223 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1227 for (i = 0; i < MAX_MIRRORS; i++) {
1228 struct ofmirror *mirror = ofproto->mirrors[i];
1229 if (mirror && mirror->aux == aux) {
1238 mirror_set(struct ofproto *ofproto_, void *aux,
1239 const struct ofproto_mirror_settings *s)
1241 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1242 mirror_mask_t mirror_bit;
1243 struct ofbundle *bundle;
1244 struct ofmirror *mirror;
1245 struct ofbundle *out;
1246 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1247 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1250 mirror = mirror_lookup(ofproto, aux);
1252 mirror_destroy(mirror);
1258 idx = mirror_scan(ofproto);
1260 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1262 ofproto->up.name, MAX_MIRRORS, s->name);
1266 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1267 mirror->ofproto = ofproto;
1269 mirror->out_vlan = -1;
1270 mirror->name = NULL;
1273 if (!mirror->name || strcmp(s->name, mirror->name)) {
1275 mirror->name = xstrdup(s->name);
1278 /* Get the new configuration. */
1279 if (s->out_bundle) {
1280 out = bundle_lookup(ofproto, s->out_bundle);
1282 mirror_destroy(mirror);
1288 out_vlan = s->out_vlan;
1290 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1291 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1293 /* If the configuration has not changed, do nothing. */
1294 if (hmapx_equals(&srcs, &mirror->srcs)
1295 && hmapx_equals(&dsts, &mirror->dsts)
1296 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1297 && mirror->out == out
1298 && mirror->out_vlan == out_vlan)
1300 hmapx_destroy(&srcs);
1301 hmapx_destroy(&dsts);
1305 hmapx_swap(&srcs, &mirror->srcs);
1306 hmapx_destroy(&srcs);
1308 hmapx_swap(&dsts, &mirror->dsts);
1309 hmapx_destroy(&dsts);
1311 free(mirror->vlans);
1312 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1315 mirror->out_vlan = out_vlan;
1317 /* Update bundles. */
1318 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1319 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1320 if (hmapx_contains(&mirror->srcs, bundle)) {
1321 bundle->src_mirrors |= mirror_bit;
1323 bundle->src_mirrors &= ~mirror_bit;
1326 if (hmapx_contains(&mirror->dsts, bundle)) {
1327 bundle->dst_mirrors |= mirror_bit;
1329 bundle->dst_mirrors &= ~mirror_bit;
1332 if (mirror->out == bundle) {
1333 bundle->mirror_out |= mirror_bit;
1335 bundle->mirror_out &= ~mirror_bit;
1339 ofproto->need_revalidate = true;
1340 mac_learning_flush(ofproto->ml);
1346 mirror_destroy(struct ofmirror *mirror)
1348 struct ofproto_dpif *ofproto;
1349 mirror_mask_t mirror_bit;
1350 struct ofbundle *bundle;
1356 ofproto = mirror->ofproto;
1357 ofproto->need_revalidate = true;
1358 mac_learning_flush(ofproto->ml);
1360 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1361 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1362 bundle->src_mirrors &= ~mirror_bit;
1363 bundle->dst_mirrors &= ~mirror_bit;
1364 bundle->mirror_out &= ~mirror_bit;
1367 hmapx_destroy(&mirror->srcs);
1368 hmapx_destroy(&mirror->dsts);
1369 free(mirror->vlans);
1371 ofproto->mirrors[mirror->idx] = NULL;
1377 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1379 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1380 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1381 ofproto->need_revalidate = true;
1382 mac_learning_flush(ofproto->ml);
1388 is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
1390 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1391 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1392 return bundle && bundle->mirror_out != 0;
1397 static struct ofport_dpif *
1398 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1400 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1401 return ofport ? ofport_dpif_cast(ofport) : NULL;
1404 static struct ofport_dpif *
1405 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1407 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1411 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1412 struct dpif_port *dpif_port)
1414 ofproto_port->name = dpif_port->name;
1415 ofproto_port->type = dpif_port->type;
1416 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1420 port_run(struct ofport_dpif *ofport)
1422 bool enable = netdev_get_carrier(ofport->up.netdev);
1425 cfm_run(ofport->cfm);
1427 if (cfm_should_send_ccm(ofport->cfm)) {
1428 struct ofpbuf packet;
1430 ofpbuf_init(&packet, 0);
1431 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1432 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1433 ofport->odp_port, &packet);
1434 ofpbuf_uninit(&packet);
1437 enable = enable && !cfm_get_fault(ofport->cfm);
1440 if (ofport->bundle) {
1441 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1444 if (ofport->may_enable != enable) {
1445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1447 if (ofproto->has_bundle_action) {
1448 ofproto->need_revalidate = true;
1452 ofport->may_enable = enable;
1456 port_wait(struct ofport_dpif *ofport)
1459 cfm_wait(ofport->cfm);
1464 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1465 struct ofproto_port *ofproto_port)
1467 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1468 struct dpif_port dpif_port;
1471 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1473 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1479 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1485 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1487 *ofp_portp = odp_port_to_ofp_port(odp_port);
1493 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1495 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1498 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1500 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1502 /* The caller is going to close ofport->up.netdev. If this is a
1503 * bonded port, then the bond is using that netdev, so remove it
1504 * from the bond. The client will need to reconfigure everything
1505 * after deleting ports, so then the slave will get re-added. */
1506 bundle_remove(&ofport->up);
1512 struct port_dump_state {
1513 struct dpif_port_dump dump;
1518 port_dump_start(const struct ofproto *ofproto_, void **statep)
1520 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1521 struct port_dump_state *state;
1523 *statep = state = xmalloc(sizeof *state);
1524 dpif_port_dump_start(&state->dump, ofproto->dpif);
1525 state->done = false;
1530 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1531 struct ofproto_port *port)
1533 struct port_dump_state *state = state_;
1534 struct dpif_port dpif_port;
1536 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1537 ofproto_port_from_dpif_port(port, &dpif_port);
1540 int error = dpif_port_dump_done(&state->dump);
1542 return error ? error : EOF;
1547 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1549 struct port_dump_state *state = state_;
1552 dpif_port_dump_done(&state->dump);
1559 port_poll(const struct ofproto *ofproto_, char **devnamep)
1561 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1562 return dpif_port_poll(ofproto->dpif, devnamep);
1566 port_poll_wait(const struct ofproto *ofproto_)
1568 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1569 dpif_port_poll_wait(ofproto->dpif);
1573 port_is_lacp_current(const struct ofport *ofport_)
1575 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1576 return (ofport->bundle && ofport->bundle->lacp
1577 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
1581 /* Upcall handling. */
1583 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
1584 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
1585 * their individual configurations.
1587 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
1588 * Otherwise, ownership is transferred to this function. */
1590 send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
1591 const struct flow *flow, bool clone)
1593 struct ofputil_packet_in pin;
1595 pin.packet = upcall->packet;
1596 pin.in_port = flow->in_port;
1597 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
1598 pin.buffer_id = 0; /* not yet known */
1599 pin.send_len = upcall->userdata;
1600 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
1601 clone ? NULL : upcall->packet);
1605 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
1606 const struct ofpbuf *packet)
1608 if (cfm_should_process_flow(flow)) {
1609 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
1610 if (packet && ofport && ofport->cfm) {
1611 cfm_process_heartbeat(ofport->cfm, packet);
1614 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
1615 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
1616 if (packet && port && port->bundle && port->bundle->lacp) {
1617 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
1619 lacp_process_pdu(port->bundle->lacp, port, pdu);
1628 handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1630 struct facet *facet;
1633 /* Obtain in_port and tun_id, at least. */
1634 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1636 /* Set header pointers in 'flow'. */
1637 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
1639 /* Handle 802.1ag and LACP. */
1640 if (process_special(ofproto, &flow, upcall->packet)) {
1641 ofpbuf_delete(upcall->packet);
1642 ofproto->n_matches++;
1646 /* Check with in-band control to see if this packet should be sent
1647 * to the local port regardless of the flow table. */
1648 if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
1649 send_packet(ofproto, ODPP_LOCAL, upcall->packet);
1652 facet = facet_lookup_valid(ofproto, &flow);
1654 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow);
1656 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
1657 struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
1659 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
1660 COVERAGE_INC(ofproto_dpif_no_packet_in);
1661 /* XXX install 'drop' flow entry */
1662 ofpbuf_delete(upcall->packet);
1666 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
1670 send_packet_in(ofproto, upcall, &flow, false);
1674 facet = facet_create(rule, &flow, upcall->packet);
1675 } else if (!facet->may_install) {
1676 /* The facet is not installable, that is, we need to process every
1677 * packet, so process the current packet's actions into 'facet'. */
1678 facet_make_actions(ofproto, facet, upcall->packet);
1681 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
1683 * Extra-special case for fail-open mode.
1685 * We are in fail-open mode and the packet matched the fail-open rule,
1686 * but we are connected to a controller too. We should send the packet
1687 * up to the controller in the hope that it will try to set up a flow
1688 * and thereby allow us to exit fail-open.
1690 * See the top-level comment in fail-open.c for more information.
1692 send_packet_in(ofproto, upcall, &flow, true);
1695 facet_execute(ofproto, facet, upcall->packet);
1696 facet_install(ofproto, facet, false);
1697 ofproto->n_matches++;
1701 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1705 switch (upcall->type) {
1706 case DPIF_UC_ACTION:
1707 COVERAGE_INC(ofproto_dpif_ctlr_action);
1708 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1709 send_packet_in(ofproto, upcall, &flow, false);
1712 case DPIF_UC_SAMPLE:
1713 if (ofproto->sflow) {
1714 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1715 dpif_sflow_received(ofproto->sflow, upcall, &flow);
1717 ofpbuf_delete(upcall->packet);
1721 handle_miss_upcall(ofproto, upcall);
1724 case DPIF_N_UC_TYPES:
1726 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
1731 /* Flow expiration. */
1733 static int facet_max_idle(const struct ofproto_dpif *);
1734 static void update_stats(struct ofproto_dpif *);
1735 static void rule_expire(struct rule_dpif *);
1736 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
1738 /* This function is called periodically by run(). Its job is to collect
1739 * updates for the flows that have been installed into the datapath, most
1740 * importantly when they last were used, and then use that information to
1741 * expire flows that have not been used recently.
1743 * Returns the number of milliseconds after which it should be called again. */
1745 expire(struct ofproto_dpif *ofproto)
1747 struct rule_dpif *rule, *next_rule;
1748 struct cls_cursor cursor;
1751 /* Update stats for each flow in the datapath. */
1752 update_stats(ofproto);
1754 /* Expire facets that have been idle too long. */
1755 dp_max_idle = facet_max_idle(ofproto);
1756 expire_facets(ofproto, dp_max_idle);
1758 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
1759 cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
1760 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1764 /* All outstanding data in existing flows has been accounted, so it's a
1765 * good time to do bond rebalancing. */
1766 if (ofproto->has_bonded_bundles) {
1767 struct ofbundle *bundle;
1769 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1771 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
1776 return MIN(dp_max_idle, 1000);
1779 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
1781 * This function also pushes statistics updates to rules which each facet
1782 * resubmits into. Generally these statistics will be accurate. However, if a
1783 * facet changes the rule it resubmits into at some time in between
1784 * update_stats() runs, it is possible that statistics accrued to the
1785 * old rule will be incorrectly attributed to the new rule. This could be
1786 * avoided by calling update_stats() whenever rules are created or
1787 * deleted. However, the performance impact of making so many calls to the
1788 * datapath do not justify the benefit of having perfectly accurate statistics.
1791 update_stats(struct ofproto_dpif *p)
1793 const struct dpif_flow_stats *stats;
1794 struct dpif_flow_dump dump;
1795 const struct nlattr *key;
1798 dpif_flow_dump_start(&dump, p->dpif);
1799 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
1800 struct facet *facet;
1803 if (odp_flow_key_to_flow(key, key_len, &flow)) {
1807 odp_flow_key_format(key, key_len, &s);
1808 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
1814 facet = facet_find(p, &flow);
1816 if (facet && facet->installed) {
1818 if (stats->n_packets >= facet->dp_packet_count) {
1819 uint64_t extra = stats->n_packets - facet->dp_packet_count;
1820 facet->packet_count += extra;
1822 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
1825 if (stats->n_bytes >= facet->dp_byte_count) {
1826 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
1828 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
1831 facet->dp_packet_count = stats->n_packets;
1832 facet->dp_byte_count = stats->n_bytes;
1834 facet_update_time(p, facet, stats->used);
1835 facet_account(p, facet, stats->n_bytes);
1836 facet_push_stats(facet);
1838 /* There's a flow in the datapath that we know nothing about.
1840 COVERAGE_INC(facet_unexpected);
1841 dpif_flow_del(p->dpif, key, key_len, NULL);
1844 dpif_flow_dump_done(&dump);
1847 /* Calculates and returns the number of milliseconds of idle time after which
1848 * facets should expire from the datapath and we should fold their statistics
1849 * into their parent rules in userspace. */
1851 facet_max_idle(const struct ofproto_dpif *ofproto)
1854 * Idle time histogram.
1856 * Most of the time a switch has a relatively small number of facets. When
1857 * this is the case we might as well keep statistics for all of them in
1858 * userspace and to cache them in the kernel datapath for performance as
1861 * As the number of facets increases, the memory required to maintain
1862 * statistics about them in userspace and in the kernel becomes
1863 * significant. However, with a large number of facets it is likely that
1864 * only a few of them are "heavy hitters" that consume a large amount of
1865 * bandwidth. At this point, only heavy hitters are worth caching in the
1866 * kernel and maintaining in userspaces; other facets we can discard.
1868 * The technique used to compute the idle time is to build a histogram with
1869 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
1870 * that is installed in the kernel gets dropped in the appropriate bucket.
1871 * After the histogram has been built, we compute the cutoff so that only
1872 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
1873 * cached. At least the most-recently-used bucket of facets is kept, so
1874 * actually an arbitrary number of facets can be kept in any given
1875 * expiration run (though the next run will delete most of those unless
1876 * they receive additional data).
1878 * This requires a second pass through the facets, in addition to the pass
1879 * made by update_stats(), because the former function never looks
1880 * at uninstallable facets.
1882 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
1883 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
1884 int buckets[N_BUCKETS] = { 0 };
1885 int total, subtotal, bucket;
1886 struct facet *facet;
1890 total = hmap_count(&ofproto->facets);
1891 if (total <= 1000) {
1892 return N_BUCKETS * BUCKET_WIDTH;
1895 /* Build histogram. */
1897 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1898 long long int idle = now - facet->used;
1899 int bucket = (idle <= 0 ? 0
1900 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
1901 : (unsigned int) idle / BUCKET_WIDTH);
1905 /* Find the first bucket whose flows should be expired. */
1906 subtotal = bucket = 0;
1908 subtotal += buckets[bucket++];
1909 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
1911 if (VLOG_IS_DBG_ENABLED()) {
1915 ds_put_cstr(&s, "keep");
1916 for (i = 0; i < N_BUCKETS; i++) {
1918 ds_put_cstr(&s, ", drop");
1921 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
1924 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
1928 return bucket * BUCKET_WIDTH;
1932 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
1934 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
1935 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
1936 struct ofexpired expired;
1938 if (facet->installed) {
1939 struct dpif_flow_stats stats;
1941 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
1943 facet_update_stats(ofproto, facet, &stats);
1946 expired.flow = facet->flow;
1947 expired.packet_count = facet->packet_count;
1948 expired.byte_count = facet->byte_count;
1949 expired.used = facet->used;
1950 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1955 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
1957 long long int cutoff = time_msec() - dp_max_idle;
1958 struct facet *facet, *next_facet;
1960 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1961 facet_active_timeout(ofproto, facet);
1962 if (facet->used < cutoff) {
1963 facet_remove(ofproto, facet);
1968 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
1969 * then delete it entirely. */
1971 rule_expire(struct rule_dpif *rule)
1973 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
1974 struct facet *facet, *next_facet;
1978 /* Has 'rule' expired? */
1980 if (rule->up.hard_timeout
1981 && now > rule->up.created + rule->up.hard_timeout * 1000) {
1982 reason = OFPRR_HARD_TIMEOUT;
1983 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
1984 && now > rule->used + rule->up.idle_timeout * 1000) {
1985 reason = OFPRR_IDLE_TIMEOUT;
1990 COVERAGE_INC(ofproto_dpif_expired);
1992 /* Update stats. (This is a no-op if the rule expired due to an idle
1993 * timeout, because that only happens when the rule has no facets left.) */
1994 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1995 facet_remove(ofproto, facet);
1998 /* Get rid of the rule. */
1999 ofproto_rule_expire(&rule->up, reason);
2004 /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
2005 * example 'packet' within that flow.
2007 * The caller must already have determined that no facet with an identical
2008 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2009 * the ofproto's classifier table. */
2010 static struct facet *
2011 facet_create(struct rule_dpif *rule, const struct flow *flow,
2012 const struct ofpbuf *packet)
2014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2015 struct facet *facet;
2017 facet = xzalloc(sizeof *facet);
2018 facet->used = time_msec();
2019 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2020 list_push_back(&rule->facets, &facet->list_node);
2022 facet->flow = *flow;
2023 netflow_flow_init(&facet->nf_flow);
2024 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2026 facet_make_actions(ofproto, facet, packet);
2032 facet_free(struct facet *facet)
2034 free(facet->actions);
2038 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2039 * 'packet', which arrived on 'in_port'.
2041 * Takes ownership of 'packet'. */
2043 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2044 const struct nlattr *odp_actions, size_t actions_len,
2045 struct ofpbuf *packet)
2047 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2048 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2049 /* As an optimization, avoid a round-trip from userspace to kernel to
2050 * userspace. This also avoids possibly filling up kernel packet
2051 * buffers along the way. */
2052 struct dpif_upcall upcall;
2054 upcall.type = DPIF_UC_ACTION;
2055 upcall.packet = packet;
2058 upcall.userdata = nl_attr_get_u64(odp_actions);
2059 upcall.sample_pool = 0;
2060 upcall.actions = NULL;
2061 upcall.actions_len = 0;
2063 send_packet_in(ofproto, &upcall, flow, false);
2067 struct odputil_keybuf keybuf;
2071 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2072 odp_flow_key_from_flow(&key, flow);
2074 error = dpif_execute(ofproto->dpif, key.data, key.size,
2075 odp_actions, actions_len, packet);
2077 ofpbuf_delete(packet);
2082 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2083 * statistics appropriately. 'packet' must have at least sizeof(struct
2084 * ofp_packet_in) bytes of headroom.
2086 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2087 * applying flow_extract() to 'packet' would yield the same flow as
2090 * 'facet' must have accurately composed ODP actions; that is, it must not be
2091 * in need of revalidation.
2093 * Takes ownership of 'packet'. */
2095 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2096 struct ofpbuf *packet)
2098 struct dpif_flow_stats stats;
2100 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2102 flow_extract_stats(&facet->flow, packet, &stats);
2103 stats.used = time_msec();
2104 if (execute_odp_actions(ofproto, &facet->flow,
2105 facet->actions, facet->actions_len, packet)) {
2106 facet_update_stats(ofproto, facet, &stats);
2110 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2112 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2113 * rule's statistics, via facet_uninstall().
2115 * - Removes 'facet' from its rule and from ofproto->facets.
2118 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2120 facet_uninstall(ofproto, facet);
2121 facet_flush_stats(ofproto, facet);
2122 hmap_remove(&ofproto->facets, &facet->hmap_node);
2123 list_remove(&facet->list_node);
2127 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2129 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2130 const struct ofpbuf *packet)
2132 const struct rule_dpif *rule = facet->rule;
2133 struct ofpbuf *odp_actions;
2134 struct action_xlate_ctx ctx;
2136 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2137 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2138 facet->tags = ctx.tags;
2139 facet->may_install = ctx.may_set_up_flow;
2140 facet->nf_flow.output_iface = ctx.nf_output_iface;
2142 if (facet->actions_len != odp_actions->size
2143 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2144 free(facet->actions);
2145 facet->actions_len = odp_actions->size;
2146 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2149 ofpbuf_delete(odp_actions);
2152 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2153 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2154 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2155 * since 'facet' was last updated.
2157 * Returns 0 if successful, otherwise a positive errno value.*/
2159 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2160 const struct nlattr *actions, size_t actions_len,
2161 struct dpif_flow_stats *stats)
2163 struct odputil_keybuf keybuf;
2164 enum dpif_flow_put_flags flags;
2168 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2170 flags |= DPIF_FP_ZERO_STATS;
2173 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2174 odp_flow_key_from_flow(&key, &facet->flow);
2176 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2177 actions, actions_len, stats);
2180 facet_reset_dp_stats(facet, stats);
2186 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2187 * 'zero_stats' is true, clears any existing statistics from the datapath for
2190 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2192 struct dpif_flow_stats stats;
2194 if (facet->may_install
2195 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2196 zero_stats ? &stats : NULL)) {
2197 facet->installed = true;
2202 vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci)
2204 return vlan_tci != htons(0) ? vlan_tci_to_vid(vlan_tci) : OFP_VLAN_NONE;
2208 facet_account(struct ofproto_dpif *ofproto,
2209 struct facet *facet, uint64_t extra_bytes)
2211 uint64_t total_bytes, n_bytes;
2212 struct ofbundle *in_bundle;
2213 const struct nlattr *a;
2219 total_bytes = facet->byte_count + extra_bytes;
2220 if (total_bytes <= facet->accounted_bytes) {
2223 n_bytes = total_bytes - facet->accounted_bytes;
2224 facet->accounted_bytes = total_bytes;
2226 /* Test that 'tags' is nonzero to ensure that only flows that include an
2227 * OFPP_NORMAL action are used for learning and bond slave rebalancing.
2228 * This works because OFPP_NORMAL always sets a nonzero tag value.
2230 * Feed information from the active flows back into the learning table to
2231 * ensure that table is always in sync with what is actually flowing
2232 * through the datapath. */
2234 || !is_admissible(ofproto, &facet->flow, false, &dummy,
2235 &vlan, &in_bundle)) {
2239 update_learning_table(ofproto, &facet->flow, vlan, in_bundle);
2241 if (!ofproto->has_bonded_bundles) {
2245 /* This loop feeds byte counters to bond_account() for rebalancing to use
2246 * as a basis. We also need to track the actual VLAN on which the packet
2247 * is going to be sent to ensure that it matches the one passed to
2248 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2250 vlan_tci = facet->flow.vlan_tci;
2251 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2252 struct ofport_dpif *port;
2254 switch (nl_attr_type(a)) {
2255 case ODP_ACTION_ATTR_OUTPUT:
2256 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2257 if (port && port->bundle && port->bundle->bond) {
2258 bond_account(port->bundle->bond, &facet->flow,
2259 vlan_tci_to_openflow_vlan(vlan_tci), n_bytes);
2263 case ODP_ACTION_ATTR_STRIP_VLAN:
2264 vlan_tci = htons(0);
2267 case ODP_ACTION_ATTR_SET_DL_TCI:
2268 vlan_tci = nl_attr_get_be16(a);
2274 /* If 'rule' is installed in the datapath, uninstalls it. */
2276 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2278 if (facet->installed) {
2279 struct odputil_keybuf keybuf;
2280 struct dpif_flow_stats stats;
2284 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2285 odp_flow_key_from_flow(&key, &facet->flow);
2287 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2288 facet_reset_dp_stats(facet, &stats);
2290 facet_update_stats(p, facet, &stats);
2292 facet->installed = false;
2294 assert(facet->dp_packet_count == 0);
2295 assert(facet->dp_byte_count == 0);
2299 /* Returns true if the only action for 'facet' is to send to the controller.
2300 * (We don't report NetFlow expiration messages for such facets because they
2301 * are just part of the control logic for the network, not real traffic). */
2303 facet_is_controller_flow(struct facet *facet)
2306 && facet->rule->up.n_actions == 1
2307 && action_outputs_to_port(&facet->rule->up.actions[0],
2308 htons(OFPP_CONTROLLER)));
2311 /* Resets 'facet''s datapath statistics counters. This should be called when
2312 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2313 * it should contain the statistics returned by dpif when 'facet' was reset in
2314 * the datapath. 'stats' will be modified to only included statistics new
2315 * since 'facet' was last updated. */
2317 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2319 if (stats && facet->dp_packet_count <= stats->n_packets
2320 && facet->dp_byte_count <= stats->n_bytes) {
2321 stats->n_packets -= facet->dp_packet_count;
2322 stats->n_bytes -= facet->dp_byte_count;
2325 facet->dp_packet_count = 0;
2326 facet->dp_byte_count = 0;
2329 /* Folds all of 'facet''s statistics into its rule. Also updates the
2330 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2331 * 'facet''s statistics in the datapath should have been zeroed and folded into
2332 * its packet and byte counts before this function is called. */
2334 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2336 assert(!facet->dp_byte_count);
2337 assert(!facet->dp_packet_count);
2339 facet_push_stats(facet);
2340 facet_account(ofproto, facet, 0);
2342 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2343 struct ofexpired expired;
2344 expired.flow = facet->flow;
2345 expired.packet_count = facet->packet_count;
2346 expired.byte_count = facet->byte_count;
2347 expired.used = facet->used;
2348 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2351 facet->rule->packet_count += facet->packet_count;
2352 facet->rule->byte_count += facet->byte_count;
2354 /* Reset counters to prevent double counting if 'facet' ever gets
2356 facet->packet_count = 0;
2357 facet->byte_count = 0;
2358 facet->rs_packet_count = 0;
2359 facet->rs_byte_count = 0;
2360 facet->accounted_bytes = 0;
2362 netflow_flow_clear(&facet->nf_flow);
2365 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2366 * Returns it if found, otherwise a null pointer.
2368 * The returned facet might need revalidation; use facet_lookup_valid()
2369 * instead if that is important. */
2370 static struct facet *
2371 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
2373 struct facet *facet;
2375 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2377 if (flow_equal(flow, &facet->flow)) {
2385 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2386 * Returns it if found, otherwise a null pointer.
2388 * The returned facet is guaranteed to be valid. */
2389 static struct facet *
2390 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
2392 struct facet *facet = facet_find(ofproto, flow);
2394 /* The facet we found might not be valid, since we could be in need of
2395 * revalidation. If it is not valid, don't return it. */
2397 && ofproto->need_revalidate
2398 && !facet_revalidate(ofproto, facet)) {
2399 COVERAGE_INC(facet_invalidated);
2406 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2408 * - If the rule found is different from 'facet''s current rule, moves
2409 * 'facet' to the new rule and recompiles its actions.
2411 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2412 * where it is and recompiles its actions anyway.
2414 * - If there is none, destroys 'facet'.
2416 * Returns true if 'facet' still exists, false if it has been destroyed. */
2418 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
2420 struct action_xlate_ctx ctx;
2421 struct ofpbuf *odp_actions;
2422 struct rule_dpif *new_rule;
2423 bool actions_changed;
2425 COVERAGE_INC(facet_revalidate);
2427 /* Determine the new rule. */
2428 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
2430 /* No new rule, so delete the facet. */
2431 facet_remove(ofproto, facet);
2435 /* Calculate new ODP actions.
2437 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2438 * emit a NetFlow expiration and, if so, we need to have the old state
2439 * around to properly compose it. */
2440 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2441 odp_actions = xlate_actions(&ctx,
2442 new_rule->up.actions, new_rule->up.n_actions);
2443 actions_changed = (facet->actions_len != odp_actions->size
2444 || memcmp(facet->actions, odp_actions->data,
2445 facet->actions_len));
2447 /* If the ODP actions changed or the installability changed, then we need
2448 * to talk to the datapath. */
2449 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2450 if (ctx.may_set_up_flow) {
2451 struct dpif_flow_stats stats;
2453 facet_put__(ofproto, facet,
2454 odp_actions->data, odp_actions->size, &stats);
2455 facet_update_stats(ofproto, facet, &stats);
2457 facet_uninstall(ofproto, facet);
2460 /* The datapath flow is gone or has zeroed stats, so push stats out of
2461 * 'facet' into 'rule'. */
2462 facet_flush_stats(ofproto, facet);
2465 /* Update 'facet' now that we've taken care of all the old state. */
2466 facet->tags = ctx.tags;
2467 facet->nf_flow.output_iface = ctx.nf_output_iface;
2468 facet->may_install = ctx.may_set_up_flow;
2469 if (actions_changed) {
2470 free(facet->actions);
2471 facet->actions_len = odp_actions->size;
2472 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2474 if (facet->rule != new_rule) {
2475 COVERAGE_INC(facet_changed_rule);
2476 list_remove(&facet->list_node);
2477 list_push_back(&new_rule->facets, &facet->list_node);
2478 facet->rule = new_rule;
2479 facet->used = new_rule->up.created;
2480 facet->rs_used = facet->used;
2483 ofpbuf_delete(odp_actions);
2488 /* Updates 'facet''s used time. Caller is responsible for calling
2489 * facet_push_stats() to update the flows which 'facet' resubmits into. */
2491 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
2494 if (used > facet->used) {
2496 if (used > facet->rule->used) {
2497 facet->rule->used = used;
2499 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
2503 /* Folds the statistics from 'stats' into the counters in 'facet'.
2505 * Because of the meaning of a facet's counters, it only makes sense to do this
2506 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
2507 * packet that was sent by hand or if it represents statistics that have been
2508 * cleared out of the datapath. */
2510 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
2511 const struct dpif_flow_stats *stats)
2513 if (stats->n_packets || stats->used > facet->used) {
2514 facet_update_time(ofproto, facet, stats->used);
2515 facet->packet_count += stats->n_packets;
2516 facet->byte_count += stats->n_bytes;
2517 facet_push_stats(facet);
2518 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
2523 facet_push_stats(struct facet *facet)
2525 uint64_t rs_packets, rs_bytes;
2527 assert(facet->packet_count >= facet->rs_packet_count);
2528 assert(facet->byte_count >= facet->rs_byte_count);
2529 assert(facet->used >= facet->rs_used);
2531 rs_packets = facet->packet_count - facet->rs_packet_count;
2532 rs_bytes = facet->byte_count - facet->rs_byte_count;
2534 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
2535 facet->rs_packet_count = facet->packet_count;
2536 facet->rs_byte_count = facet->byte_count;
2537 facet->rs_used = facet->used;
2539 flow_push_stats(facet->rule, &facet->flow,
2540 rs_packets, rs_bytes, facet->used);
2544 struct ofproto_push {
2545 struct action_xlate_ctx ctx;
2552 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
2554 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
2557 rule->packet_count += push->packets;
2558 rule->byte_count += push->bytes;
2559 rule->used = MAX(push->used, rule->used);
2563 /* Pushes flow statistics to the rules which 'flow' resubmits into given
2564 * 'rule''s actions. */
2566 flow_push_stats(const struct rule_dpif *rule,
2567 struct flow *flow, uint64_t packets, uint64_t bytes,
2570 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2571 struct ofproto_push push;
2573 push.packets = packets;
2577 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
2578 push.ctx.resubmit_hook = push_resubmit;
2579 ofpbuf_delete(xlate_actions(&push.ctx,
2580 rule->up.actions, rule->up.n_actions));
2585 static struct rule_dpif *
2586 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
2588 return rule_dpif_cast(rule_from_cls_rule(
2589 classifier_lookup(&ofproto->up.tables[0],
2594 complete_operation(struct rule_dpif *rule)
2596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2598 ofproto->need_revalidate = true;
2600 struct dpif_completion *c = xmalloc(sizeof *c);
2601 c->op = rule->up.pending;
2602 list_push_back(&ofproto->completions, &c->list_node);
2604 ofoperation_complete(rule->up.pending, 0);
2608 static struct rule *
2611 struct rule_dpif *rule = xmalloc(sizeof *rule);
2616 rule_dealloc(struct rule *rule_)
2618 struct rule_dpif *rule = rule_dpif_cast(rule_);
2623 rule_construct(struct rule *rule_)
2625 struct rule_dpif *rule = rule_dpif_cast(rule_);
2626 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2627 struct rule_dpif *victim;
2630 error = validate_actions(rule->up.actions, rule->up.n_actions,
2631 &rule->up.cr.flow, ofproto->max_ports);
2636 rule->used = rule->up.created;
2637 rule->packet_count = 0;
2638 rule->byte_count = 0;
2640 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
2641 if (victim && !list_is_empty(&victim->facets)) {
2642 struct facet *facet;
2644 rule->facets = victim->facets;
2645 list_moved(&rule->facets);
2646 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2650 /* Must avoid list_moved() in this case. */
2651 list_init(&rule->facets);
2654 complete_operation(rule);
2659 rule_destruct(struct rule *rule_)
2661 struct rule_dpif *rule = rule_dpif_cast(rule_);
2662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2663 struct facet *facet, *next_facet;
2665 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2666 facet_revalidate(ofproto, facet);
2669 complete_operation(rule);
2673 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
2675 struct rule_dpif *rule = rule_dpif_cast(rule_);
2676 struct facet *facet;
2678 /* Start from historical data for 'rule' itself that are no longer tracked
2679 * in facets. This counts, for example, facets that have expired. */
2680 *packets = rule->packet_count;
2681 *bytes = rule->byte_count;
2683 /* Add any statistics that are tracked by facets. This includes
2684 * statistical data recently updated by ofproto_update_stats() as well as
2685 * stats for packets that were executed "by hand" via dpif_execute(). */
2686 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2687 *packets += facet->packet_count;
2688 *bytes += facet->byte_count;
2693 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
2695 struct rule_dpif *rule = rule_dpif_cast(rule_);
2696 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2697 struct action_xlate_ctx ctx;
2698 struct ofpbuf *odp_actions;
2699 struct facet *facet;
2702 /* First look for a related facet. If we find one, account it to that. */
2703 facet = facet_lookup_valid(ofproto, flow);
2704 if (facet && facet->rule == rule) {
2705 facet_execute(ofproto, facet, packet);
2709 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2710 * create a new facet for it and use that. */
2711 if (rule_dpif_lookup(ofproto, flow) == rule) {
2712 facet = facet_create(rule, flow, packet);
2713 facet_execute(ofproto, facet, packet);
2714 facet_install(ofproto, facet, true);
2718 /* We can't account anything to a facet. If we were to try, then that
2719 * facet would have a non-matching rule, busting our invariants. */
2720 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
2721 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2722 size = packet->size;
2723 if (execute_odp_actions(ofproto, flow, odp_actions->data,
2724 odp_actions->size, packet)) {
2725 rule->used = time_msec();
2726 rule->packet_count++;
2727 rule->byte_count += size;
2728 flow_push_stats(rule, flow, 1, size, rule->used);
2730 ofpbuf_delete(odp_actions);
2736 rule_modify_actions(struct rule *rule_)
2738 struct rule_dpif *rule = rule_dpif_cast(rule_);
2739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2742 error = validate_actions(rule->up.actions, rule->up.n_actions,
2743 &rule->up.cr.flow, ofproto->max_ports);
2745 ofoperation_complete(rule->up.pending, error);
2749 complete_operation(rule);
2752 /* Sends 'packet' out of port 'odp_port' within 'p'.
2753 * Returns 0 if successful, otherwise a positive errno value. */
2755 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
2756 const struct ofpbuf *packet)
2758 struct ofpbuf key, odp_actions;
2759 struct odputil_keybuf keybuf;
2763 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
2764 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2765 odp_flow_key_from_flow(&key, &flow);
2767 ofpbuf_init(&odp_actions, 32);
2768 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2769 error = dpif_execute(ofproto->dpif,
2771 odp_actions.data, odp_actions.size,
2773 ofpbuf_uninit(&odp_actions);
2776 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
2777 ofproto->up.name, odp_port, strerror(error));
2782 /* OpenFlow to ODP action translation. */
2784 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2785 struct action_xlate_ctx *ctx);
2786 static bool xlate_normal(struct action_xlate_ctx *);
2789 commit_odp_actions(struct action_xlate_ctx *ctx)
2791 const struct flow *flow = &ctx->flow;
2792 struct flow *base = &ctx->base_flow;
2793 struct ofpbuf *odp_actions = ctx->odp_actions;
2795 if (base->tun_id != flow->tun_id) {
2796 nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id);
2797 base->tun_id = flow->tun_id;
2800 if (base->nw_src != flow->nw_src) {
2801 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src);
2802 base->nw_src = flow->nw_src;
2805 if (base->nw_dst != flow->nw_dst) {
2806 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst);
2807 base->nw_dst = flow->nw_dst;
2810 if (base->vlan_tci != flow->vlan_tci) {
2811 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
2812 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2814 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2815 flow->vlan_tci & ~htons(VLAN_CFI));
2817 base->vlan_tci = flow->vlan_tci;
2820 if (base->tp_src != flow->tp_src) {
2821 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src);
2822 base->tp_src = flow->tp_src;
2825 if (base->tp_dst != flow->tp_dst) {
2826 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst);
2827 base->tp_dst = flow->tp_dst;
2830 if (!eth_addr_equals(base->dl_src, flow->dl_src)) {
2831 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2832 flow->dl_src, ETH_ADDR_LEN);
2833 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
2836 if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) {
2837 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2838 flow->dl_dst, ETH_ADDR_LEN);
2839 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
2842 if (ctx->base_priority != ctx->priority) {
2843 if (ctx->priority) {
2844 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY,
2847 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2849 ctx->base_priority = ctx->priority;
2854 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
2856 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
2857 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
2860 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
2861 /* Forwarding disabled on port. */
2866 * We don't have an ofport record for this port, but it doesn't hurt to
2867 * allow forwarding to it anyhow. Maybe such a port will appear later
2868 * and we're pre-populating the flow table.
2872 commit_odp_actions(ctx);
2873 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2874 ctx->nf_output_iface = ofp_port;
2878 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2880 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2881 struct rule_dpif *rule;
2882 uint16_t old_in_port;
2884 /* Look up a flow with 'in_port' as the input port. Then restore the
2885 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2886 * have surprising behavior). */
2887 old_in_port = ctx->flow.in_port;
2888 ctx->flow.in_port = in_port;
2889 rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow);
2890 ctx->flow.in_port = old_in_port;
2892 if (ctx->resubmit_hook) {
2893 ctx->resubmit_hook(ctx, rule);
2898 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
2902 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2904 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2905 MAX_RESUBMIT_RECURSION);
2910 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
2912 struct ofport_dpif *ofport;
2914 commit_odp_actions(ctx);
2915 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
2916 uint16_t ofp_port = ofport->up.ofp_port;
2917 if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) {
2918 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT,
2923 ctx->nf_output_iface = NF_OUT_FLOOD;
2927 xlate_output_action__(struct action_xlate_ctx *ctx,
2928 uint16_t port, uint16_t max_len)
2930 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2932 ctx->nf_output_iface = NF_OUT_DROP;
2936 add_output_action(ctx, ctx->flow.in_port);
2939 xlate_table_action(ctx, ctx->flow.in_port);
2945 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
2948 flood_packets(ctx, htonl(0));
2950 case OFPP_CONTROLLER:
2951 commit_odp_actions(ctx);
2952 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2955 add_output_action(ctx, OFPP_LOCAL);
2960 if (port != ctx->flow.in_port) {
2961 add_output_action(ctx, port);
2966 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2967 ctx->nf_output_iface = NF_OUT_FLOOD;
2968 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2969 ctx->nf_output_iface = prev_nf_output_iface;
2970 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2971 ctx->nf_output_iface != NF_OUT_FLOOD) {
2972 ctx->nf_output_iface = NF_OUT_MULTI;
2977 xlate_output_action(struct action_xlate_ctx *ctx,
2978 const struct ofp_action_output *oao)
2980 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2984 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2985 const struct ofp_action_enqueue *oae)
2987 uint16_t ofp_port, odp_port;
2988 uint32_t ctx_priority, priority;
2991 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2994 /* Fall back to ordinary output action. */
2995 xlate_output_action__(ctx, ntohs(oae->port), 0);
2999 /* Figure out ODP output port. */
3000 ofp_port = ntohs(oae->port);
3001 if (ofp_port == OFPP_IN_PORT) {
3002 ofp_port = ctx->flow.in_port;
3004 odp_port = ofp_port_to_odp_port(ofp_port);
3006 /* Add ODP actions. */
3007 ctx_priority = ctx->priority;
3008 ctx->priority = priority;
3009 add_output_action(ctx, odp_port);
3010 ctx->priority = ctx_priority;
3012 /* Update NetFlow output port. */
3013 if (ctx->nf_output_iface == NF_OUT_DROP) {
3014 ctx->nf_output_iface = odp_port;
3015 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3016 ctx->nf_output_iface = NF_OUT_MULTI;
3021 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3022 const struct nx_action_set_queue *nasq)
3027 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3030 /* Couldn't translate queue to a priority, so ignore. A warning
3031 * has already been logged. */
3035 ctx->priority = priority;
3038 struct xlate_reg_state {
3044 xlate_autopath(struct action_xlate_ctx *ctx,
3045 const struct nx_action_autopath *naa)
3047 uint16_t ofp_port = ntohl(naa->id);
3048 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
3050 if (!port || !port->bundle) {
3051 ofp_port = OFPP_NONE;
3052 } else if (port->bundle->bond) {
3053 /* Autopath does not support VLAN hashing. */
3054 struct ofport_dpif *slave = bond_choose_output_slave(
3055 port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags);
3057 ofp_port = slave->up.ofp_port;
3060 autopath_execute(naa, &ctx->flow, ofp_port);
3064 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
3066 struct ofproto_dpif *ofproto = ofproto_;
3067 struct ofport_dpif *port;
3077 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3080 port = get_ofp_port(ofproto, ofp_port);
3081 return port ? port->may_enable : false;
3086 do_xlate_actions(const union ofp_action *in, size_t n_in,
3087 struct action_xlate_ctx *ctx)
3089 const struct ofport_dpif *port;
3090 const union ofp_action *ia;
3093 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
3095 && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3096 port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3097 ? htonl(OFPPC_NO_RECV_STP)
3098 : htonl(OFPPC_NO_RECV))) {
3099 /* Drop this flow. */
3103 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
3104 const struct ofp_action_dl_addr *oada;
3105 const struct nx_action_resubmit *nar;
3106 const struct nx_action_set_tunnel *nast;
3107 const struct nx_action_set_queue *nasq;
3108 const struct nx_action_multipath *nam;
3109 const struct nx_action_autopath *naa;
3110 const struct nx_action_bundle *nab;
3111 enum ofputil_action_code code;
3114 code = ofputil_decode_action_unsafe(ia);
3116 case OFPUTIL_OFPAT_OUTPUT:
3117 xlate_output_action(ctx, &ia->output);
3120 case OFPUTIL_OFPAT_SET_VLAN_VID:
3121 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3122 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3125 case OFPUTIL_OFPAT_SET_VLAN_PCP:
3126 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3127 ctx->flow.vlan_tci |= htons(
3128 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3131 case OFPUTIL_OFPAT_STRIP_VLAN:
3132 ctx->flow.vlan_tci = htons(0);
3135 case OFPUTIL_OFPAT_SET_DL_SRC:
3136 oada = ((struct ofp_action_dl_addr *) ia);
3137 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3140 case OFPUTIL_OFPAT_SET_DL_DST:
3141 oada = ((struct ofp_action_dl_addr *) ia);
3142 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3145 case OFPUTIL_OFPAT_SET_NW_SRC:
3146 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3149 case OFPUTIL_OFPAT_SET_NW_DST:
3150 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3153 case OFPUTIL_OFPAT_SET_NW_TOS:
3154 ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK;
3157 case OFPUTIL_OFPAT_SET_TP_SRC:
3158 ctx->flow.tp_src = ia->tp_port.tp_port;
3161 case OFPUTIL_OFPAT_SET_TP_DST:
3162 ctx->flow.tp_dst = ia->tp_port.tp_port;
3165 case OFPUTIL_OFPAT_ENQUEUE:
3166 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3169 case OFPUTIL_NXAST_RESUBMIT:
3170 nar = (const struct nx_action_resubmit *) ia;
3171 xlate_table_action(ctx, ntohs(nar->in_port));
3174 case OFPUTIL_NXAST_SET_TUNNEL:
3175 nast = (const struct nx_action_set_tunnel *) ia;
3176 tun_id = htonll(ntohl(nast->tun_id));
3177 ctx->flow.tun_id = tun_id;
3180 case OFPUTIL_NXAST_SET_QUEUE:
3181 nasq = (const struct nx_action_set_queue *) ia;
3182 xlate_set_queue_action(ctx, nasq);
3185 case OFPUTIL_NXAST_POP_QUEUE:
3189 case OFPUTIL_NXAST_REG_MOVE:
3190 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
3194 case OFPUTIL_NXAST_REG_LOAD:
3195 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
3199 case OFPUTIL_NXAST_NOTE:
3200 /* Nothing to do. */
3203 case OFPUTIL_NXAST_SET_TUNNEL64:
3204 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
3205 ctx->flow.tun_id = tun_id;
3208 case OFPUTIL_NXAST_MULTIPATH:
3209 nam = (const struct nx_action_multipath *) ia;
3210 multipath_execute(nam, &ctx->flow);
3213 case OFPUTIL_NXAST_AUTOPATH:
3214 naa = (const struct nx_action_autopath *) ia;
3215 xlate_autopath(ctx, naa);
3218 case OFPUTIL_NXAST_BUNDLE:
3219 ctx->ofproto->has_bundle_action = true;
3220 nab = (const struct nx_action_bundle *) ia;
3221 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
3230 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3231 struct ofproto_dpif *ofproto, const struct flow *flow,
3232 const struct ofpbuf *packet)
3234 ctx->ofproto = ofproto;
3236 ctx->packet = packet;
3237 ctx->resubmit_hook = NULL;
3240 static struct ofpbuf *
3241 xlate_actions(struct action_xlate_ctx *ctx,
3242 const union ofp_action *in, size_t n_in)
3244 COVERAGE_INC(ofproto_dpif_xlate);
3246 ctx->odp_actions = ofpbuf_new(512);
3248 ctx->may_set_up_flow = true;
3249 ctx->nf_output_iface = NF_OUT_DROP;
3252 ctx->base_priority = 0;
3253 ctx->base_flow = ctx->flow;
3255 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
3256 ctx->may_set_up_flow = false;
3258 do_xlate_actions(in, n_in, ctx);
3261 /* Check with in-band control to see if we're allowed to set up this
3263 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
3264 ctx->odp_actions->data,
3265 ctx->odp_actions->size)) {
3266 ctx->may_set_up_flow = false;
3269 return ctx->odp_actions;
3272 /* OFPP_NORMAL implementation. */
3275 struct ofport_dpif *port;
3280 struct dst builtin[32];
3282 size_t n, allocated;
3285 static void dst_set_init(struct dst_set *);
3286 static void dst_set_add(struct dst_set *, const struct dst *);
3287 static void dst_set_free(struct dst_set *);
3289 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
3292 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
3293 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
3295 dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE
3296 : in_bundle->vlan >= 0 ? in_bundle->vlan
3297 : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE
3298 : vlan_tci_to_vid(ctx->flow.vlan_tci));
3300 dst->port = (!out_bundle->bond
3301 ? ofbundle_get_a_port(out_bundle)
3302 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
3303 dst->vlan, &ctx->tags));
3305 return dst->port != NULL;
3309 mirror_mask_ffs(mirror_mask_t mask)
3311 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
3316 dst_set_init(struct dst_set *set)
3318 set->dsts = set->builtin;
3320 set->allocated = ARRAY_SIZE(set->builtin);
3324 dst_set_add(struct dst_set *set, const struct dst *dst)
3326 if (set->n >= set->allocated) {
3327 size_t new_allocated;
3328 struct dst *new_dsts;
3330 new_allocated = set->allocated * 2;
3331 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
3332 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
3336 set->dsts = new_dsts;
3337 set->allocated = new_allocated;
3339 set->dsts[set->n++] = *dst;
3343 dst_set_free(struct dst_set *set)
3345 if (set->dsts != set->builtin) {
3351 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
3354 for (i = 0; i < set->n; i++) {
3355 if (set->dsts[i].vlan == test->vlan
3356 && set->dsts[i].port == test->port) {
3364 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
3366 return bundle->vlan < 0 && vlan_bitmap_contains(bundle->trunks, vlan);
3370 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
3372 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
3375 /* Returns an arbitrary interface within 'bundle'. */
3376 static struct ofport_dpif *
3377 ofbundle_get_a_port(const struct ofbundle *bundle)
3379 return CONTAINER_OF(list_front(&bundle->ports),
3380 struct ofport_dpif, bundle_node);
3384 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
3385 const struct ofbundle *in_bundle,
3386 const struct ofbundle *out_bundle, struct dst_set *set)
3390 if (out_bundle == OFBUNDLE_FLOOD) {
3391 struct ofbundle *bundle;
3393 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
3394 if (bundle != in_bundle
3395 && ofbundle_includes_vlan(bundle, vlan)
3396 && bundle->floodable
3397 && !bundle->mirror_out
3398 && set_dst(ctx, &dst, in_bundle, bundle)) {
3399 dst_set_add(set, &dst);
3402 ctx->nf_output_iface = NF_OUT_FLOOD;
3403 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
3404 dst_set_add(set, &dst);
3405 ctx->nf_output_iface = dst.port->odp_port;
3410 vlan_is_mirrored(const struct ofmirror *m, int vlan)
3412 return vlan_bitmap_contains(m->vlans, vlan);
3415 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
3416 * to a VLAN. In general most packets may be mirrored but we want to drop
3417 * protocols that may confuse switches. */
3419 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
3421 /* If you change this function's behavior, please update corresponding
3422 * documentation in vswitch.xml at the same time. */
3423 if (dst[0] != 0x01) {
3424 /* All the currently banned MACs happen to start with 01 currently, so
3425 * this is a quick way to eliminate most of the good ones. */
3427 if (eth_addr_is_reserved(dst)) {
3428 /* Drop STP, IEEE pause frames, and other reserved protocols
3429 * (01-80-c2-00-00-0x). */
3433 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
3435 if ((dst[3] & 0xfe) == 0xcc &&
3436 (dst[4] & 0xfe) == 0xcc &&
3437 (dst[5] & 0xfe) == 0xcc) {
3438 /* Drop the following protocols plus others following the same
3441 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
3442 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
3443 STP Uplink Fast (01-00-0c-cd-cd-cd) */
3447 if (!(dst[3] | dst[4] | dst[5])) {
3448 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
3457 compose_mirror_dsts(struct action_xlate_ctx *ctx,
3458 uint16_t vlan, const struct ofbundle *in_bundle,
3459 struct dst_set *set)
3461 struct ofproto_dpif *ofproto = ctx->ofproto;
3462 mirror_mask_t mirrors;
3466 mirrors = in_bundle->src_mirrors;
3467 for (i = 0; i < set->n; i++) {
3468 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
3475 flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3476 if (flow_vlan == 0) {
3477 flow_vlan = OFP_VLAN_NONE;
3481 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
3482 if (vlan_is_mirrored(m, vlan)) {
3486 if (set_dst(ctx, &dst, in_bundle, m->out)
3487 && !dst_is_duplicate(set, &dst)) {
3488 dst_set_add(set, &dst);
3490 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
3491 struct ofbundle *bundle;
3493 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3494 if (ofbundle_includes_vlan(bundle, m->out_vlan)
3495 && set_dst(ctx, &dst, in_bundle, bundle))
3497 if (bundle->vlan < 0) {
3498 dst.vlan = m->out_vlan;
3500 if (dst_is_duplicate(set, &dst)) {
3504 /* Use the vlan tag on the original flow instead of
3505 * the one passed in the vlan parameter. This ensures
3506 * that we compare the vlan from before any implicit
3507 * tagging tags place. This is necessary because
3508 * dst->vlan is the final vlan, after removing implicit
3510 if (bundle == in_bundle && dst.vlan == flow_vlan) {
3511 /* Don't send out input port on same VLAN. */
3514 dst_set_add(set, &dst);
3519 mirrors &= mirrors - 1;
3524 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
3525 const struct ofbundle *in_bundle,
3526 const struct ofbundle *out_bundle)
3528 uint16_t initial_vlan, cur_vlan;
3529 const struct dst *dst;
3533 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
3534 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
3536 /* Output all the packets we can without having to change the VLAN. */
3537 initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3538 if (initial_vlan == 0) {
3539 initial_vlan = OFP_VLAN_NONE;
3541 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3542 if (dst->vlan != initial_vlan) {
3545 nl_msg_put_u32(ctx->odp_actions,
3546 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3549 /* Then output the rest. */
3550 cur_vlan = initial_vlan;
3551 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3552 if (dst->vlan == initial_vlan) {
3555 if (dst->vlan != cur_vlan) {
3556 if (dst->vlan == OFP_VLAN_NONE) {
3557 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
3560 tci = htons(dst->vlan & VLAN_VID_MASK);
3561 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
3562 nl_msg_put_be16(ctx->odp_actions,
3563 ODP_ACTION_ATTR_SET_DL_TCI, tci);
3565 cur_vlan = dst->vlan;
3567 nl_msg_put_u32(ctx->odp_actions,
3568 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3574 /* Returns the effective vlan of a packet, taking into account both the
3575 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
3576 * the packet is untagged and -1 indicates it has an invalid header and
3577 * should be dropped. */
3579 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
3580 struct ofbundle *in_bundle, bool have_packet)
3582 int vlan = vlan_tci_to_vid(flow->vlan_tci);
3583 if (in_bundle->vlan >= 0) {
3586 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3587 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3588 "packet received on port %s configured with "
3589 "implicit VLAN %"PRIu16,
3590 ofproto->up.name, vlan,
3591 in_bundle->name, in_bundle->vlan);
3595 vlan = in_bundle->vlan;
3597 if (!ofbundle_includes_vlan(in_bundle, vlan)) {
3599 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3600 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3601 "packet received on port %s not configured for "
3603 ofproto->up.name, vlan, in_bundle->name, vlan);
3612 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
3613 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
3614 * indicate this; newer upstream kernels use gratuitous ARP requests. */
3616 is_gratuitous_arp(const struct flow *flow)
3618 return (flow->dl_type == htons(ETH_TYPE_ARP)
3619 && eth_addr_is_broadcast(flow->dl_dst)
3620 && (flow->nw_proto == ARP_OP_REPLY
3621 || (flow->nw_proto == ARP_OP_REQUEST
3622 && flow->nw_src == flow->nw_dst)));
3626 update_learning_table(struct ofproto_dpif *ofproto,
3627 const struct flow *flow, int vlan,
3628 struct ofbundle *in_bundle)
3630 struct mac_entry *mac;
3632 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
3636 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
3637 if (is_gratuitous_arp(flow)) {
3638 /* We don't want to learn from gratuitous ARP packets that are
3639 * reflected back over bond slaves so we lock the learning table. */
3640 if (!in_bundle->bond) {
3641 mac_entry_set_grat_arp_lock(mac);
3642 } else if (mac_entry_is_grat_arp_locked(mac)) {
3647 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
3648 /* The log messages here could actually be useful in debugging,
3649 * so keep the rate limit relatively high. */
3650 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3651 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
3652 "on port %s in VLAN %d",
3653 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
3654 in_bundle->name, vlan);
3656 mac->port.p = in_bundle;
3657 tag_set_add(&ofproto->revalidate_set,
3658 mac_learning_changed(ofproto->ml, mac));
3662 /* Determines whether packets in 'flow' within 'br' should be forwarded or
3663 * dropped. Returns true if they may be forwarded, false if they should be
3666 * If 'have_packet' is true, it indicates that the caller is processing a
3667 * received packet. If 'have_packet' is false, then the caller is just
3668 * revalidating an existing flow because configuration has changed. Either
3669 * way, 'have_packet' only affects logging (there is no point in logging errors
3670 * during revalidation).
3672 * Sets '*in_portp' to the input port. This will be a null pointer if
3673 * flow->in_port does not designate a known input port (in which case
3674 * is_admissible() returns false).
3676 * When returning true, sets '*vlanp' to the effective VLAN of the input
3677 * packet, as returned by flow_get_vlan().
3679 * May also add tags to '*tags', although the current implementation only does
3680 * so in one special case.
3683 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
3685 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
3687 struct ofport_dpif *in_port;
3688 struct ofbundle *in_bundle;
3691 /* Find the port and bundle for the received packet. */
3692 in_port = get_ofp_port(ofproto, flow->in_port);
3693 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
3694 if (!in_port || !in_bundle) {
3695 /* No interface? Something fishy... */
3697 /* Odd. A few possible reasons here:
3699 * - We deleted a port but there are still a few packets queued up
3702 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
3703 * we don't know about.
3705 * - Packet arrived on the local port but the local port is not
3708 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3710 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3712 ofproto->up.name, flow->in_port);
3716 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
3721 /* Drop frames for reserved multicast addresses. */
3722 if (eth_addr_is_reserved(flow->dl_dst)) {
3726 /* Drop frames on bundles reserved for mirroring. */
3727 if (in_bundle->mirror_out) {
3729 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3730 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3731 "%s, which is reserved exclusively for mirroring",
3732 ofproto->up.name, in_bundle->name);
3737 if (in_bundle->bond) {
3738 struct mac_entry *mac;
3740 switch (bond_check_admissibility(in_bundle->bond, in_port,
3741 flow->dl_dst, tags)) {
3748 case BV_DROP_IF_MOVED:
3749 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
3750 if (mac && mac->port.p != in_bundle &&
3751 (!is_gratuitous_arp(flow)
3752 || mac_entry_is_grat_arp_locked(mac))) {
3762 /* If the composed actions may be applied to any packet in the given 'flow',
3763 * returns true. Otherwise, the actions should only be applied to 'packet', or
3764 * not at all, if 'packet' was NULL. */
3766 xlate_normal(struct action_xlate_ctx *ctx)
3768 struct ofbundle *in_bundle;
3769 struct ofbundle *out_bundle;
3770 struct mac_entry *mac;
3773 /* Check whether we should drop packets in this flow. */
3774 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
3775 &ctx->tags, &vlan, &in_bundle)) {
3780 /* Learn source MAC (but don't try to learn from revalidation). */
3782 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
3785 /* Determine output bundle. */
3786 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
3789 out_bundle = mac->port.p;
3790 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
3791 /* If we are revalidating but don't have a learning entry then eject
3792 * the flow. Installing a flow that floods packets opens up a window
3793 * of time where we could learn from a packet reflected on a bond and
3794 * blackhole packets before the learning table is updated to reflect
3795 * the correct port. */
3798 out_bundle = OFBUNDLE_FLOOD;
3801 /* Don't send packets out their input bundles. */
3802 if (in_bundle == out_bundle) {
3808 compose_actions(ctx, vlan, in_bundle, out_bundle);
3815 get_drop_frags(struct ofproto *ofproto_)
3817 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3820 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
3825 set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
3827 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3829 dpif_set_drop_frags(ofproto->dpif, drop_frags);
3833 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3834 const struct flow *flow,
3835 const union ofp_action *ofp_actions, size_t n_ofp_actions)
3837 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3840 error = validate_actions(ofp_actions, n_ofp_actions, flow,
3841 ofproto->max_ports);
3843 struct odputil_keybuf keybuf;
3844 struct action_xlate_ctx ctx;
3845 struct ofpbuf *odp_actions;
3848 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3849 odp_flow_key_from_flow(&key, flow);
3851 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3852 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3853 dpif_execute(ofproto->dpif, key.data, key.size,
3854 odp_actions->data, odp_actions->size, packet);
3855 ofpbuf_delete(odp_actions);
3861 get_netflow_ids(const struct ofproto *ofproto_,
3862 uint8_t *engine_type, uint8_t *engine_id)
3864 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3866 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3869 static struct ofproto_dpif *
3870 ofproto_dpif_lookup(const char *name)
3872 struct ofproto *ofproto = ofproto_lookup(name);
3873 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
3874 ? ofproto_dpif_cast(ofproto)
3879 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
3880 const char *args, void *aux OVS_UNUSED)
3882 struct ds ds = DS_EMPTY_INITIALIZER;
3883 const struct ofproto_dpif *ofproto;
3884 const struct mac_entry *e;
3886 ofproto = ofproto_dpif_lookup(args);
3888 unixctl_command_reply(conn, 501, "no such bridge");
3892 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3893 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3894 struct ofbundle *bundle = e->port.p;
3895 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
3896 ofbundle_get_a_port(bundle)->odp_port,
3897 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
3899 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3903 struct ofproto_trace {
3904 struct action_xlate_ctx ctx;
3910 trace_format_rule(struct ds *result, int level, const struct rule *rule)
3912 ds_put_char_multiple(result, '\t', level);
3914 ds_put_cstr(result, "No match\n");
3918 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
3919 ntohll(rule->flow_cookie));
3920 cls_rule_format(&rule->cr, result);
3921 ds_put_char(result, '\n');
3923 ds_put_char_multiple(result, '\t', level);
3924 ds_put_cstr(result, "OpenFlow ");
3925 ofp_print_actions(result, rule->actions, rule->n_actions);
3926 ds_put_char(result, '\n');
3930 trace_format_flow(struct ds *result, int level, const char *title,
3931 struct ofproto_trace *trace)
3933 ds_put_char_multiple(result, '\t', level);
3934 ds_put_format(result, "%s: ", title);
3935 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
3936 ds_put_cstr(result, "unchanged");
3938 flow_format(result, &trace->ctx.flow);
3939 trace->flow = trace->ctx.flow;
3941 ds_put_char(result, '\n');
3945 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3947 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
3948 struct ds *result = trace->result;
3950 ds_put_char(result, '\n');
3951 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
3952 trace_format_rule(result, ctx->recurse + 1, &rule->up);
3956 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
3957 void *aux OVS_UNUSED)
3959 char *dpname, *in_port_s, *tun_id_s, *packet_s;
3960 char *args = xstrdup(args_);
3961 char *save_ptr = NULL;
3962 struct ofproto_dpif *ofproto;
3963 struct ofpbuf packet;
3964 struct rule_dpif *rule;
3971 ofpbuf_init(&packet, strlen(args) / 2);
3974 dpname = strtok_r(args, " ", &save_ptr);
3975 tun_id_s = strtok_r(NULL, " ", &save_ptr);
3976 in_port_s = strtok_r(NULL, " ", &save_ptr);
3977 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
3978 if (!dpname || !in_port_s || !packet_s) {
3979 unixctl_command_reply(conn, 501, "Bad command syntax");
3983 ofproto = ofproto_dpif_lookup(dpname);
3985 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
3990 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
3991 in_port = ofp_port_to_odp_port(atoi(in_port_s));
3993 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
3994 packet_s += strspn(packet_s, " ");
3995 if (*packet_s != '\0') {
3996 unixctl_command_reply(conn, 501, "Trailing garbage in command");
3999 if (packet.size < ETH_HEADER_LEN) {
4000 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4004 ds_put_cstr(&result, "Packet: ");
4005 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4006 ds_put_cstr(&result, s);
4009 flow_extract(&packet, tun_id, in_port, &flow);
4010 ds_put_cstr(&result, "Flow: ");
4011 flow_format(&result, &flow);
4012 ds_put_char(&result, '\n');
4014 rule = rule_dpif_lookup(ofproto, &flow);
4015 trace_format_rule(&result, 0, &rule->up);
4017 struct ofproto_trace trace;
4018 struct ofpbuf *odp_actions;
4020 trace.result = &result;
4022 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4023 trace.ctx.resubmit_hook = trace_resubmit;
4024 odp_actions = xlate_actions(&trace.ctx,
4025 rule->up.actions, rule->up.n_actions);
4027 ds_put_char(&result, '\n');
4028 trace_format_flow(&result, 0, "Final flow", &trace);
4029 ds_put_cstr(&result, "Datapath actions: ");
4030 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4031 ofpbuf_delete(odp_actions);
4034 unixctl_command_reply(conn, 200, ds_cstr(&result));
4037 ds_destroy(&result);
4038 ofpbuf_uninit(&packet);
4043 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
4044 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4047 unixctl_command_reply(conn, 200, NULL);
4051 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
4052 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4055 unixctl_command_reply(conn, 200, NULL);
4059 ofproto_dpif_unixctl_init(void)
4061 static bool registered;
4067 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4068 unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
4070 unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL);
4071 unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL);
4074 const struct ofproto_class ofproto_dpif_class = {
4101 port_is_lacp_current,
4102 NULL, /* rule_choose_table */
4109 rule_modify_actions,
4122 is_mirror_output_bundle,