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_update(struct ofbundle *);
151 static void bundle_destroy(struct ofbundle *);
152 static void bundle_del_port(struct ofport_dpif *);
153 static void bundle_run(struct ofbundle *);
154 static void bundle_wait(struct ofbundle *);
156 struct action_xlate_ctx {
157 /* action_xlate_ctx_init() initializes these members. */
160 struct ofproto_dpif *ofproto;
162 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
163 * this flow when actions change header fields. */
166 /* The packet corresponding to 'flow', or a null pointer if we are
167 * revalidating without a packet to refer to. */
168 const struct ofpbuf *packet;
170 /* If nonnull, called just before executing a resubmit action.
172 * This is normally null so the client has to set it manually after
173 * calling action_xlate_ctx_init(). */
174 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
176 /* xlate_actions() initializes and uses these members. The client might want
177 * to look at them after it returns. */
179 struct ofpbuf *odp_actions; /* Datapath actions. */
180 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
181 bool may_set_up_flow; /* True ordinarily; false if the actions must
182 * be reassessed for every packet. */
183 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
185 /* xlate_actions() initializes and uses these members, but the client has no
186 * reason to look at them. */
188 int recurse; /* Recursion level, via xlate_table_action. */
189 uint32_t priority; /* Current flow priority. 0 if none. */
190 struct flow base_flow; /* Flow at the last commit. */
191 uint32_t base_priority; /* Priority at the last commit. */
194 static void action_xlate_ctx_init(struct action_xlate_ctx *,
195 struct ofproto_dpif *, const struct flow *,
196 const struct ofpbuf *);
197 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
198 const union ofp_action *in, size_t n_in);
200 /* An exact-match instantiation of an OpenFlow flow. */
202 long long int used; /* Time last used; time created if not used. */
206 * - Do include packets and bytes sent "by hand", e.g. with
209 * - Do include packets and bytes that were obtained from the datapath
210 * when a flow was deleted (e.g. dpif_flow_del()) or when its
211 * statistics were reset (e.g. dpif_flow_put() with
212 * DPIF_FP_ZERO_STATS).
214 * - Do not include any packets or bytes that can currently be obtained
215 * from the datapath by, e.g., dpif_flow_get().
217 uint64_t packet_count; /* Number of packets received. */
218 uint64_t byte_count; /* Number of bytes received. */
220 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
221 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
223 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
224 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
225 long long int rs_used; /* Used time pushed to resubmit children. */
227 /* Number of bytes passed to account_cb. This may include bytes that can
228 * currently obtained from the datapath (thus, it can be greater than
230 uint64_t accounted_bytes;
232 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
233 struct list list_node; /* In owning rule's 'facets' list. */
234 struct rule_dpif *rule; /* Owning rule. */
235 struct flow flow; /* Exact-match flow. */
236 bool installed; /* Installed in datapath? */
237 bool may_install; /* True ordinarily; false if actions must
238 * be reassessed for every packet. */
239 size_t actions_len; /* Number of bytes in actions[]. */
240 struct nlattr *actions; /* Datapath actions. */
241 tag_type tags; /* Tags. */
242 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
245 static struct facet *facet_create(struct rule_dpif *, const struct flow *,
246 const struct ofpbuf *packet);
247 static void facet_remove(struct ofproto_dpif *, struct facet *);
248 static void facet_free(struct facet *);
250 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
251 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
252 const struct flow *);
253 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
255 static void facet_execute(struct ofproto_dpif *, struct facet *,
256 struct ofpbuf *packet);
258 static int facet_put__(struct ofproto_dpif *, struct facet *,
259 const struct nlattr *actions, size_t actions_len,
260 struct dpif_flow_stats *);
261 static void facet_install(struct ofproto_dpif *, struct facet *,
263 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
264 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
266 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
267 const struct ofpbuf *packet);
268 static void facet_update_time(struct ofproto_dpif *, struct facet *,
270 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
271 const struct dpif_flow_stats *);
272 static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
273 static void facet_push_stats(struct facet *);
274 static void facet_account(struct ofproto_dpif *, struct facet *,
275 uint64_t extra_bytes);
277 static bool facet_is_controller_flow(struct facet *);
279 static void flow_push_stats(const struct rule_dpif *,
280 struct flow *, uint64_t packets, uint64_t bytes,
287 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
288 struct list bundle_node; /* In struct ofbundle's "ports" list. */
289 struct cfm *cfm; /* Connectivity Fault Management, if any. */
290 tag_type tag; /* Tag associated with this port. */
291 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
292 bool may_enable; /* May be enabled in bonds. */
295 static struct ofport_dpif *
296 ofport_dpif_cast(const struct ofport *ofport)
298 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
299 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
302 static void port_run(struct ofport_dpif *);
303 static void port_wait(struct ofport_dpif *);
304 static int set_cfm(struct ofport *, const struct cfm_settings *);
306 struct dpif_completion {
307 struct list list_node;
308 struct ofoperation *op;
311 struct ofproto_dpif {
320 struct netflow *netflow;
321 struct dpif_sflow *sflow;
322 struct hmap bundles; /* Contains "struct ofbundle"s. */
323 struct mac_learning *ml;
324 struct ofmirror *mirrors[MAX_MIRRORS];
325 bool has_bonded_bundles;
328 struct timer next_expiration;
332 bool need_revalidate;
333 struct tag_set revalidate_set;
335 /* Support for debugging async flow mods. */
336 struct list completions;
338 bool has_bundle_action; /* True when the first bundle action appears. */
341 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
342 * for debugging the asynchronous flow_mod implementation.) */
345 static void ofproto_dpif_unixctl_init(void);
347 static struct ofproto_dpif *
348 ofproto_dpif_cast(const struct ofproto *ofproto)
350 assert(ofproto->ofproto_class == &ofproto_dpif_class);
351 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
354 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
356 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
359 /* Packet processing. */
360 static void update_learning_table(struct ofproto_dpif *,
361 const struct flow *, int vlan,
363 static bool is_admissible(struct ofproto_dpif *, const struct flow *,
364 bool have_packet, tag_type *, int *vlanp,
365 struct ofbundle **in_bundlep);
366 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
368 /* Flow expiration. */
369 static int expire(struct ofproto_dpif *);
372 static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
373 const struct ofpbuf *packet);
375 /* Global variables. */
376 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
378 /* Factory functions. */
381 enumerate_types(struct sset *types)
383 dp_enumerate_types(types);
387 enumerate_names(const char *type, struct sset *names)
389 return dp_enumerate_names(type, names);
393 del(const char *type, const char *name)
398 error = dpif_open(name, type, &dpif);
400 error = dpif_delete(dpif);
406 /* Basic life-cycle. */
408 static struct ofproto *
411 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
416 dealloc(struct ofproto *ofproto_)
418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
423 construct(struct ofproto *ofproto_)
425 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
426 const char *name = ofproto->up.name;
430 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
432 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
436 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
437 ofproto->n_matches = 0;
439 error = dpif_recv_set_mask(ofproto->dpif,
440 ((1u << DPIF_UC_MISS) |
441 (1u << DPIF_UC_ACTION) |
442 (1u << DPIF_UC_SAMPLE)));
444 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
445 dpif_close(ofproto->dpif);
448 dpif_flow_flush(ofproto->dpif);
449 dpif_recv_purge(ofproto->dpif);
451 ofproto->netflow = NULL;
452 ofproto->sflow = NULL;
453 hmap_init(&ofproto->bundles);
454 ofproto->ml = mac_learning_create();
455 for (i = 0; i < MAX_MIRRORS; i++) {
456 ofproto->mirrors[i] = NULL;
458 ofproto->has_bonded_bundles = false;
460 timer_set_duration(&ofproto->next_expiration, 1000);
462 hmap_init(&ofproto->facets);
463 ofproto->need_revalidate = false;
464 tag_set_init(&ofproto->revalidate_set);
466 list_init(&ofproto->completions);
468 ofproto->up.tables = xmalloc(sizeof *ofproto->up.tables);
469 classifier_init(&ofproto->up.tables[0]);
470 ofproto->up.n_tables = 1;
472 ofproto_dpif_unixctl_init();
474 ofproto->has_bundle_action = false;
480 complete_operations(struct ofproto_dpif *ofproto)
482 struct dpif_completion *c, *next;
484 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
485 ofoperation_complete(c->op, 0);
486 list_remove(&c->list_node);
492 destruct(struct ofproto *ofproto_)
494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
495 struct rule_dpif *rule, *next_rule;
496 struct cls_cursor cursor;
499 complete_operations(ofproto);
501 cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
502 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
503 ofproto_rule_destroy(&rule->up);
506 for (i = 0; i < MAX_MIRRORS; i++) {
507 mirror_destroy(ofproto->mirrors[i]);
510 netflow_destroy(ofproto->netflow);
511 dpif_sflow_destroy(ofproto->sflow);
512 hmap_destroy(&ofproto->bundles);
513 mac_learning_destroy(ofproto->ml);
515 hmap_destroy(&ofproto->facets);
517 dpif_close(ofproto->dpif);
521 run(struct ofproto *ofproto_)
523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
524 struct ofport_dpif *ofport;
525 struct ofbundle *bundle;
529 complete_operations(ofproto);
531 dpif_run(ofproto->dpif);
533 for (i = 0; i < 50; i++) {
534 struct dpif_upcall packet;
537 error = dpif_recv(ofproto->dpif, &packet);
539 if (error == ENODEV) {
540 /* Datapath destroyed. */
546 handle_upcall(ofproto, &packet);
549 if (timer_expired(&ofproto->next_expiration)) {
550 int delay = expire(ofproto);
551 timer_set_duration(&ofproto->next_expiration, delay);
554 if (ofproto->netflow) {
555 netflow_run(ofproto->netflow);
557 if (ofproto->sflow) {
558 dpif_sflow_run(ofproto->sflow);
561 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
564 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
568 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
570 /* Now revalidate if there's anything to do. */
571 if (ofproto->need_revalidate
572 || !tag_set_is_empty(&ofproto->revalidate_set)) {
573 struct tag_set revalidate_set = ofproto->revalidate_set;
574 bool revalidate_all = ofproto->need_revalidate;
575 struct facet *facet, *next;
577 /* Clear the revalidation flags. */
578 tag_set_init(&ofproto->revalidate_set);
579 ofproto->need_revalidate = false;
581 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
583 || tag_set_intersects(&revalidate_set, facet->tags)) {
584 facet_revalidate(ofproto, facet);
593 wait(struct ofproto *ofproto_)
595 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
596 struct ofport_dpif *ofport;
597 struct ofbundle *bundle;
599 if (!clogged && !list_is_empty(&ofproto->completions)) {
600 poll_immediate_wake();
603 dpif_wait(ofproto->dpif);
604 dpif_recv_wait(ofproto->dpif);
605 if (ofproto->sflow) {
606 dpif_sflow_wait(ofproto->sflow);
608 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
609 poll_immediate_wake();
611 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
614 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
617 mac_learning_wait(ofproto->ml);
618 if (ofproto->need_revalidate) {
619 /* Shouldn't happen, but if it does just go around again. */
620 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
621 poll_immediate_wake();
623 timer_wait(&ofproto->next_expiration);
628 flush(struct ofproto *ofproto_)
630 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
631 struct facet *facet, *next_facet;
633 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
634 /* Mark the facet as not installed so that facet_remove() doesn't
635 * bother trying to uninstall it. There is no point in uninstalling it
636 * individually since we are about to blow away all the facets with
637 * dpif_flow_flush(). */
638 facet->installed = false;
639 facet->dp_packet_count = 0;
640 facet->dp_byte_count = 0;
641 facet_remove(ofproto, facet);
643 dpif_flow_flush(ofproto->dpif);
647 get_features(struct ofproto *ofproto_ OVS_UNUSED,
648 bool *arp_match_ip, uint32_t *actions)
650 *arp_match_ip = true;
651 *actions = ((1u << OFPAT_OUTPUT) |
652 (1u << OFPAT_SET_VLAN_VID) |
653 (1u << OFPAT_SET_VLAN_PCP) |
654 (1u << OFPAT_STRIP_VLAN) |
655 (1u << OFPAT_SET_DL_SRC) |
656 (1u << OFPAT_SET_DL_DST) |
657 (1u << OFPAT_SET_NW_SRC) |
658 (1u << OFPAT_SET_NW_DST) |
659 (1u << OFPAT_SET_NW_TOS) |
660 (1u << OFPAT_SET_TP_SRC) |
661 (1u << OFPAT_SET_TP_DST) |
662 (1u << OFPAT_ENQUEUE));
666 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
668 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
671 strcpy(ots->name, "classifier");
673 dpif_get_dp_stats(ofproto->dpif, &s);
674 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
675 put_32aligned_be64(&ots->matched_count,
676 htonll(s.n_hit + ofproto->n_matches));
680 set_netflow(struct ofproto *ofproto_,
681 const struct netflow_options *netflow_options)
683 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
685 if (netflow_options) {
686 if (!ofproto->netflow) {
687 ofproto->netflow = netflow_create();
689 return netflow_set_options(ofproto->netflow, netflow_options);
691 netflow_destroy(ofproto->netflow);
692 ofproto->netflow = NULL;
697 static struct ofport *
700 struct ofport_dpif *port = xmalloc(sizeof *port);
705 port_dealloc(struct ofport *port_)
707 struct ofport_dpif *port = ofport_dpif_cast(port_);
712 port_construct(struct ofport *port_)
714 struct ofport_dpif *port = ofport_dpif_cast(port_);
715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
717 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
720 port->tag = tag_create_random();
721 port->may_enable = true;
723 if (ofproto->sflow) {
724 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
725 netdev_get_name(port->up.netdev));
732 port_destruct(struct ofport *port_)
734 struct ofport_dpif *port = ofport_dpif_cast(port_);
735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
737 bundle_remove(port_);
738 set_cfm(port_, NULL);
739 if (ofproto->sflow) {
740 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
745 port_modified(struct ofport *port_)
747 struct ofport_dpif *port = ofport_dpif_cast(port_);
749 if (port->bundle && port->bundle->bond) {
750 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
755 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
757 struct ofport_dpif *port = ofport_dpif_cast(port_);
758 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
759 ovs_be32 changed = old_config ^ port->up.opp.config;
761 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
762 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
763 ofproto->need_revalidate = true;
765 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
766 bundle_update(port->bundle);
772 set_sflow(struct ofproto *ofproto_,
773 const struct ofproto_sflow_options *sflow_options)
775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
776 struct dpif_sflow *ds = ofproto->sflow;
779 struct ofport_dpif *ofport;
781 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
782 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
783 dpif_sflow_add_port(ds, ofport->odp_port,
784 netdev_get_name(ofport->up.netdev));
787 dpif_sflow_set_options(ds, sflow_options);
789 dpif_sflow_destroy(ds);
790 ofproto->sflow = NULL;
796 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
798 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
805 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
808 if (cfm_configure(ofport->cfm, s)) {
814 cfm_destroy(ofport->cfm);
820 get_cfm_fault(const struct ofport *ofport_)
822 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
824 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
829 /* Expires all MAC learning entries associated with 'port' and forces ofproto
830 * to revalidate every flow. */
832 bundle_flush_macs(struct ofbundle *bundle)
834 struct ofproto_dpif *ofproto = bundle->ofproto;
835 struct mac_learning *ml = ofproto->ml;
836 struct mac_entry *mac, *next_mac;
838 ofproto->need_revalidate = true;
839 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
840 if (mac->port.p == bundle) {
841 mac_learning_expire(ml, mac);
846 static struct ofbundle *
847 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
849 struct ofbundle *bundle;
851 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
853 if (bundle->aux == aux) {
860 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
861 * ones that are found to 'bundles'. */
863 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
864 void **auxes, size_t n_auxes,
865 struct hmapx *bundles)
870 for (i = 0; i < n_auxes; i++) {
871 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
873 hmapx_add(bundles, bundle);
879 bundle_update(struct ofbundle *bundle)
881 struct ofport_dpif *port;
883 bundle->floodable = true;
884 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
885 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
886 bundle->floodable = false;
893 bundle_del_port(struct ofport_dpif *port)
895 struct ofbundle *bundle = port->bundle;
897 bundle->ofproto->need_revalidate = true;
899 list_remove(&port->bundle_node);
903 lacp_slave_unregister(bundle->lacp, port);
906 bond_slave_unregister(bundle->bond, port);
909 bundle_update(bundle);
913 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
914 struct lacp_slave_settings *lacp,
915 uint32_t bond_stable_id)
917 struct ofport_dpif *port;
919 port = get_ofp_port(bundle->ofproto, ofp_port);
924 if (port->bundle != bundle) {
925 bundle->ofproto->need_revalidate = true;
927 bundle_del_port(port);
930 port->bundle = bundle;
931 list_push_back(&bundle->ports, &port->bundle_node);
932 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
933 bundle->floodable = false;
937 lacp_slave_register(bundle->lacp, port, lacp);
940 port->bond_stable_id = bond_stable_id;
946 bundle_destroy(struct ofbundle *bundle)
948 struct ofproto_dpif *ofproto;
949 struct ofport_dpif *port, *next_port;
956 ofproto = bundle->ofproto;
957 for (i = 0; i < MAX_MIRRORS; i++) {
958 struct ofmirror *m = ofproto->mirrors[i];
960 if (m->out == bundle) {
962 } else if (hmapx_find_and_delete(&m->srcs, bundle)
963 || hmapx_find_and_delete(&m->dsts, bundle)) {
964 ofproto->need_revalidate = true;
969 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
970 bundle_del_port(port);
973 bundle_flush_macs(bundle);
974 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
976 free(bundle->trunks);
977 lacp_destroy(bundle->lacp);
978 bond_destroy(bundle->bond);
983 bundle_set(struct ofproto *ofproto_, void *aux,
984 const struct ofproto_bundle_settings *s)
986 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
987 bool need_flush = false;
988 const unsigned long *trunks;
989 struct ofport_dpif *port;
990 struct ofbundle *bundle;
995 bundle_destroy(bundle_lookup(ofproto, aux));
999 assert(s->n_slaves == 1 || s->bond != NULL);
1000 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1002 bundle = bundle_lookup(ofproto, aux);
1004 bundle = xmalloc(sizeof *bundle);
1006 bundle->ofproto = ofproto;
1007 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1008 hash_pointer(aux, 0));
1010 bundle->name = NULL;
1012 list_init(&bundle->ports);
1014 bundle->trunks = NULL;
1015 bundle->lacp = NULL;
1016 bundle->bond = NULL;
1018 bundle->floodable = true;
1020 bundle->src_mirrors = 0;
1021 bundle->dst_mirrors = 0;
1022 bundle->mirror_out = 0;
1025 if (!bundle->name || strcmp(s->name, bundle->name)) {
1027 bundle->name = xstrdup(s->name);
1032 if (!bundle->lacp) {
1033 bundle->lacp = lacp_create();
1035 lacp_configure(bundle->lacp, s->lacp);
1037 lacp_destroy(bundle->lacp);
1038 bundle->lacp = NULL;
1041 /* Update set of ports. */
1043 for (i = 0; i < s->n_slaves; i++) {
1044 if (!bundle_add_port(bundle, s->slaves[i],
1045 s->lacp ? &s->lacp_slaves[i] : NULL,
1046 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1050 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1051 struct ofport_dpif *next_port;
1053 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1054 for (i = 0; i < s->n_slaves; i++) {
1055 if (s->slaves[i] == port->up.ofp_port) {
1060 bundle_del_port(port);
1064 assert(list_size(&bundle->ports) <= s->n_slaves);
1066 if (list_is_empty(&bundle->ports)) {
1067 bundle_destroy(bundle);
1072 if (s->vlan != bundle->vlan) {
1073 bundle->vlan = s->vlan;
1077 /* Get trunked VLANs. */
1078 trunks = s->vlan == -1 ? s->trunks : NULL;
1079 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1080 free(bundle->trunks);
1081 bundle->trunks = vlan_bitmap_clone(trunks);
1086 if (!list_is_short(&bundle->ports)) {
1087 bundle->ofproto->has_bonded_bundles = true;
1089 if (bond_reconfigure(bundle->bond, s->bond)) {
1090 ofproto->need_revalidate = true;
1093 bundle->bond = bond_create(s->bond);
1094 ofproto->need_revalidate = true;
1097 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1098 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1102 bond_destroy(bundle->bond);
1103 bundle->bond = NULL;
1106 /* If we changed something that would affect MAC learning, un-learn
1107 * everything on this port and force flow revalidation. */
1109 bundle_flush_macs(bundle);
1116 bundle_remove(struct ofport *port_)
1118 struct ofport_dpif *port = ofport_dpif_cast(port_);
1119 struct ofbundle *bundle = port->bundle;
1122 bundle_del_port(port);
1123 if (list_is_empty(&bundle->ports)) {
1124 bundle_destroy(bundle);
1125 } else if (list_is_short(&bundle->ports)) {
1126 bond_destroy(bundle->bond);
1127 bundle->bond = NULL;
1133 send_pdu_cb(void *port_, const struct lacp_pdu *pdu)
1135 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1136 struct ofport_dpif *port = port_;
1137 uint8_t ea[ETH_ADDR_LEN];
1140 error = netdev_get_etheraddr(port->up.netdev, ea);
1142 struct lacp_pdu *packet_pdu;
1143 struct ofpbuf packet;
1145 ofpbuf_init(&packet, 0);
1146 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1147 sizeof *packet_pdu);
1149 error = netdev_send(port->up.netdev, &packet);
1151 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1152 "(%s)", port->bundle->name,
1153 netdev_get_name(port->up.netdev), strerror(error));
1155 ofpbuf_uninit(&packet);
1157 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1158 "%s (%s)", port->bundle->name,
1159 netdev_get_name(port->up.netdev), strerror(error));
1164 bundle_send_learning_packets(struct ofbundle *bundle)
1166 struct ofproto_dpif *ofproto = bundle->ofproto;
1167 int error, n_packets, n_errors;
1168 struct mac_entry *e;
1170 error = n_packets = n_errors = 0;
1171 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1172 if (e->port.p != bundle) {
1173 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1183 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1184 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1185 "packets, last error was: %s",
1186 bundle->name, n_errors, n_packets, strerror(error));
1188 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1189 bundle->name, n_packets);
1194 bundle_run(struct ofbundle *bundle)
1197 lacp_run(bundle->lacp, send_pdu_cb);
1200 struct ofport_dpif *port;
1202 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1203 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1206 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1207 lacp_negotiated(bundle->lacp));
1208 if (bond_should_send_learning_packets(bundle->bond)) {
1209 bundle_send_learning_packets(bundle);
1215 bundle_wait(struct ofbundle *bundle)
1218 lacp_wait(bundle->lacp);
1221 bond_wait(bundle->bond);
1228 mirror_scan(struct ofproto_dpif *ofproto)
1232 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1233 if (!ofproto->mirrors[idx]) {
1240 static struct ofmirror *
1241 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1245 for (i = 0; i < MAX_MIRRORS; i++) {
1246 struct ofmirror *mirror = ofproto->mirrors[i];
1247 if (mirror && mirror->aux == aux) {
1256 mirror_set(struct ofproto *ofproto_, void *aux,
1257 const struct ofproto_mirror_settings *s)
1259 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1260 mirror_mask_t mirror_bit;
1261 struct ofbundle *bundle;
1262 struct ofmirror *mirror;
1263 struct ofbundle *out;
1264 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1265 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1268 mirror = mirror_lookup(ofproto, aux);
1270 mirror_destroy(mirror);
1276 idx = mirror_scan(ofproto);
1278 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1280 ofproto->up.name, MAX_MIRRORS, s->name);
1284 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1285 mirror->ofproto = ofproto;
1288 mirror->out_vlan = -1;
1289 mirror->name = NULL;
1292 if (!mirror->name || strcmp(s->name, mirror->name)) {
1294 mirror->name = xstrdup(s->name);
1297 /* Get the new configuration. */
1298 if (s->out_bundle) {
1299 out = bundle_lookup(ofproto, s->out_bundle);
1301 mirror_destroy(mirror);
1307 out_vlan = s->out_vlan;
1309 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1310 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1312 /* If the configuration has not changed, do nothing. */
1313 if (hmapx_equals(&srcs, &mirror->srcs)
1314 && hmapx_equals(&dsts, &mirror->dsts)
1315 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1316 && mirror->out == out
1317 && mirror->out_vlan == out_vlan)
1319 hmapx_destroy(&srcs);
1320 hmapx_destroy(&dsts);
1324 hmapx_swap(&srcs, &mirror->srcs);
1325 hmapx_destroy(&srcs);
1327 hmapx_swap(&dsts, &mirror->dsts);
1328 hmapx_destroy(&dsts);
1330 free(mirror->vlans);
1331 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1334 mirror->out_vlan = out_vlan;
1336 /* Update bundles. */
1337 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1338 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1339 if (hmapx_contains(&mirror->srcs, bundle)) {
1340 bundle->src_mirrors |= mirror_bit;
1342 bundle->src_mirrors &= ~mirror_bit;
1345 if (hmapx_contains(&mirror->dsts, bundle)) {
1346 bundle->dst_mirrors |= mirror_bit;
1348 bundle->dst_mirrors &= ~mirror_bit;
1351 if (mirror->out == bundle) {
1352 bundle->mirror_out |= mirror_bit;
1354 bundle->mirror_out &= ~mirror_bit;
1358 ofproto->need_revalidate = true;
1359 mac_learning_flush(ofproto->ml);
1365 mirror_destroy(struct ofmirror *mirror)
1367 struct ofproto_dpif *ofproto;
1368 mirror_mask_t mirror_bit;
1369 struct ofbundle *bundle;
1375 ofproto = mirror->ofproto;
1376 ofproto->need_revalidate = true;
1377 mac_learning_flush(ofproto->ml);
1379 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1380 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1381 bundle->src_mirrors &= ~mirror_bit;
1382 bundle->dst_mirrors &= ~mirror_bit;
1383 bundle->mirror_out &= ~mirror_bit;
1386 hmapx_destroy(&mirror->srcs);
1387 hmapx_destroy(&mirror->dsts);
1388 free(mirror->vlans);
1390 ofproto->mirrors[mirror->idx] = NULL;
1396 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1398 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1399 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1400 ofproto->need_revalidate = true;
1401 mac_learning_flush(ofproto->ml);
1407 is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
1409 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1410 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1411 return bundle && bundle->mirror_out != 0;
1415 forward_bpdu_changed(struct ofproto *ofproto_)
1417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1418 /* Revalidate cached flows whenever forward_bpdu option changes. */
1419 ofproto->need_revalidate = true;
1424 static struct ofport_dpif *
1425 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1427 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1428 return ofport ? ofport_dpif_cast(ofport) : NULL;
1431 static struct ofport_dpif *
1432 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1434 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1438 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1439 struct dpif_port *dpif_port)
1441 ofproto_port->name = dpif_port->name;
1442 ofproto_port->type = dpif_port->type;
1443 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1447 port_run(struct ofport_dpif *ofport)
1449 bool enable = netdev_get_carrier(ofport->up.netdev);
1452 cfm_run(ofport->cfm);
1454 if (cfm_should_send_ccm(ofport->cfm)) {
1455 struct ofpbuf packet;
1457 ofpbuf_init(&packet, 0);
1458 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1459 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1460 ofport->odp_port, &packet);
1461 ofpbuf_uninit(&packet);
1464 enable = enable && !cfm_get_fault(ofport->cfm);
1467 if (ofport->bundle) {
1468 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1471 if (ofport->may_enable != enable) {
1472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1474 if (ofproto->has_bundle_action) {
1475 ofproto->need_revalidate = true;
1479 ofport->may_enable = enable;
1483 port_wait(struct ofport_dpif *ofport)
1486 cfm_wait(ofport->cfm);
1491 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1492 struct ofproto_port *ofproto_port)
1494 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1495 struct dpif_port dpif_port;
1498 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1500 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1506 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1512 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1514 *ofp_portp = odp_port_to_ofp_port(odp_port);
1520 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1522 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1525 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1527 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1529 /* The caller is going to close ofport->up.netdev. If this is a
1530 * bonded port, then the bond is using that netdev, so remove it
1531 * from the bond. The client will need to reconfigure everything
1532 * after deleting ports, so then the slave will get re-added. */
1533 bundle_remove(&ofport->up);
1539 struct port_dump_state {
1540 struct dpif_port_dump dump;
1545 port_dump_start(const struct ofproto *ofproto_, void **statep)
1547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1548 struct port_dump_state *state;
1550 *statep = state = xmalloc(sizeof *state);
1551 dpif_port_dump_start(&state->dump, ofproto->dpif);
1552 state->done = false;
1557 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1558 struct ofproto_port *port)
1560 struct port_dump_state *state = state_;
1561 struct dpif_port dpif_port;
1563 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1564 ofproto_port_from_dpif_port(port, &dpif_port);
1567 int error = dpif_port_dump_done(&state->dump);
1569 return error ? error : EOF;
1574 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1576 struct port_dump_state *state = state_;
1579 dpif_port_dump_done(&state->dump);
1586 port_poll(const struct ofproto *ofproto_, char **devnamep)
1588 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1589 return dpif_port_poll(ofproto->dpif, devnamep);
1593 port_poll_wait(const struct ofproto *ofproto_)
1595 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1596 dpif_port_poll_wait(ofproto->dpif);
1600 port_is_lacp_current(const struct ofport *ofport_)
1602 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1603 return (ofport->bundle && ofport->bundle->lacp
1604 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
1608 /* Upcall handling. */
1610 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
1611 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
1612 * their individual configurations.
1614 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
1615 * Otherwise, ownership is transferred to this function. */
1617 send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
1618 const struct flow *flow, bool clone)
1620 struct ofputil_packet_in pin;
1622 pin.packet = upcall->packet;
1623 pin.in_port = flow->in_port;
1624 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
1625 pin.buffer_id = 0; /* not yet known */
1626 pin.send_len = upcall->userdata;
1627 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
1628 clone ? NULL : upcall->packet);
1632 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
1633 const struct ofpbuf *packet)
1635 if (cfm_should_process_flow(flow)) {
1636 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
1637 if (packet && ofport && ofport->cfm) {
1638 cfm_process_heartbeat(ofport->cfm, packet);
1641 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
1642 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
1643 if (packet && port && port->bundle && port->bundle->lacp) {
1644 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
1646 lacp_process_pdu(port->bundle->lacp, port, pdu);
1655 handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1657 struct facet *facet;
1660 /* Obtain in_port and tun_id, at least. */
1661 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1663 /* Set header pointers in 'flow'. */
1664 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
1666 /* Handle 802.1ag and LACP. */
1667 if (process_special(ofproto, &flow, upcall->packet)) {
1668 ofpbuf_delete(upcall->packet);
1669 ofproto->n_matches++;
1673 /* Check with in-band control to see if this packet should be sent
1674 * to the local port regardless of the flow table. */
1675 if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
1676 send_packet(ofproto, ODPP_LOCAL, upcall->packet);
1679 facet = facet_lookup_valid(ofproto, &flow);
1681 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow);
1683 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
1684 struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
1686 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
1687 COVERAGE_INC(ofproto_dpif_no_packet_in);
1688 /* XXX install 'drop' flow entry */
1689 ofpbuf_delete(upcall->packet);
1693 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
1697 send_packet_in(ofproto, upcall, &flow, false);
1701 facet = facet_create(rule, &flow, upcall->packet);
1702 } else if (!facet->may_install) {
1703 /* The facet is not installable, that is, we need to process every
1704 * packet, so process the current packet's actions into 'facet'. */
1705 facet_make_actions(ofproto, facet, upcall->packet);
1708 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
1710 * Extra-special case for fail-open mode.
1712 * We are in fail-open mode and the packet matched the fail-open rule,
1713 * but we are connected to a controller too. We should send the packet
1714 * up to the controller in the hope that it will try to set up a flow
1715 * and thereby allow us to exit fail-open.
1717 * See the top-level comment in fail-open.c for more information.
1719 send_packet_in(ofproto, upcall, &flow, true);
1722 facet_execute(ofproto, facet, upcall->packet);
1723 facet_install(ofproto, facet, false);
1724 ofproto->n_matches++;
1728 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1732 switch (upcall->type) {
1733 case DPIF_UC_ACTION:
1734 COVERAGE_INC(ofproto_dpif_ctlr_action);
1735 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1736 send_packet_in(ofproto, upcall, &flow, false);
1739 case DPIF_UC_SAMPLE:
1740 if (ofproto->sflow) {
1741 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1742 dpif_sflow_received(ofproto->sflow, upcall, &flow);
1744 ofpbuf_delete(upcall->packet);
1748 handle_miss_upcall(ofproto, upcall);
1751 case DPIF_N_UC_TYPES:
1753 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
1758 /* Flow expiration. */
1760 static int facet_max_idle(const struct ofproto_dpif *);
1761 static void update_stats(struct ofproto_dpif *);
1762 static void rule_expire(struct rule_dpif *);
1763 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
1765 /* This function is called periodically by run(). Its job is to collect
1766 * updates for the flows that have been installed into the datapath, most
1767 * importantly when they last were used, and then use that information to
1768 * expire flows that have not been used recently.
1770 * Returns the number of milliseconds after which it should be called again. */
1772 expire(struct ofproto_dpif *ofproto)
1774 struct rule_dpif *rule, *next_rule;
1775 struct cls_cursor cursor;
1778 /* Update stats for each flow in the datapath. */
1779 update_stats(ofproto);
1781 /* Expire facets that have been idle too long. */
1782 dp_max_idle = facet_max_idle(ofproto);
1783 expire_facets(ofproto, dp_max_idle);
1785 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
1786 cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
1787 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1791 /* All outstanding data in existing flows has been accounted, so it's a
1792 * good time to do bond rebalancing. */
1793 if (ofproto->has_bonded_bundles) {
1794 struct ofbundle *bundle;
1796 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1798 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
1803 return MIN(dp_max_idle, 1000);
1806 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
1808 * This function also pushes statistics updates to rules which each facet
1809 * resubmits into. Generally these statistics will be accurate. However, if a
1810 * facet changes the rule it resubmits into at some time in between
1811 * update_stats() runs, it is possible that statistics accrued to the
1812 * old rule will be incorrectly attributed to the new rule. This could be
1813 * avoided by calling update_stats() whenever rules are created or
1814 * deleted. However, the performance impact of making so many calls to the
1815 * datapath do not justify the benefit of having perfectly accurate statistics.
1818 update_stats(struct ofproto_dpif *p)
1820 const struct dpif_flow_stats *stats;
1821 struct dpif_flow_dump dump;
1822 const struct nlattr *key;
1825 dpif_flow_dump_start(&dump, p->dpif);
1826 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
1827 struct facet *facet;
1830 if (odp_flow_key_to_flow(key, key_len, &flow)) {
1834 odp_flow_key_format(key, key_len, &s);
1835 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
1841 facet = facet_find(p, &flow);
1843 if (facet && facet->installed) {
1845 if (stats->n_packets >= facet->dp_packet_count) {
1846 uint64_t extra = stats->n_packets - facet->dp_packet_count;
1847 facet->packet_count += extra;
1849 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
1852 if (stats->n_bytes >= facet->dp_byte_count) {
1853 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
1855 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
1858 facet->dp_packet_count = stats->n_packets;
1859 facet->dp_byte_count = stats->n_bytes;
1861 facet_update_time(p, facet, stats->used);
1862 facet_account(p, facet, stats->n_bytes);
1863 facet_push_stats(facet);
1865 /* There's a flow in the datapath that we know nothing about.
1867 COVERAGE_INC(facet_unexpected);
1868 dpif_flow_del(p->dpif, key, key_len, NULL);
1871 dpif_flow_dump_done(&dump);
1874 /* Calculates and returns the number of milliseconds of idle time after which
1875 * facets should expire from the datapath and we should fold their statistics
1876 * into their parent rules in userspace. */
1878 facet_max_idle(const struct ofproto_dpif *ofproto)
1881 * Idle time histogram.
1883 * Most of the time a switch has a relatively small number of facets. When
1884 * this is the case we might as well keep statistics for all of them in
1885 * userspace and to cache them in the kernel datapath for performance as
1888 * As the number of facets increases, the memory required to maintain
1889 * statistics about them in userspace and in the kernel becomes
1890 * significant. However, with a large number of facets it is likely that
1891 * only a few of them are "heavy hitters" that consume a large amount of
1892 * bandwidth. At this point, only heavy hitters are worth caching in the
1893 * kernel and maintaining in userspaces; other facets we can discard.
1895 * The technique used to compute the idle time is to build a histogram with
1896 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
1897 * that is installed in the kernel gets dropped in the appropriate bucket.
1898 * After the histogram has been built, we compute the cutoff so that only
1899 * the most-recently-used 1% of facets (but at least
1900 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
1901 * the most-recently-used bucket of facets is kept, so actually an
1902 * arbitrary number of facets can be kept in any given expiration run
1903 * (though the next run will delete most of those unless they receive
1906 * This requires a second pass through the facets, in addition to the pass
1907 * made by update_stats(), because the former function never looks
1908 * at uninstallable facets.
1910 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
1911 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
1912 int buckets[N_BUCKETS] = { 0 };
1913 int total, subtotal, bucket;
1914 struct facet *facet;
1918 total = hmap_count(&ofproto->facets);
1919 if (total <= ofproto->up.flow_eviction_threshold) {
1920 return N_BUCKETS * BUCKET_WIDTH;
1923 /* Build histogram. */
1925 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1926 long long int idle = now - facet->used;
1927 int bucket = (idle <= 0 ? 0
1928 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
1929 : (unsigned int) idle / BUCKET_WIDTH);
1933 /* Find the first bucket whose flows should be expired. */
1934 subtotal = bucket = 0;
1936 subtotal += buckets[bucket++];
1937 } while (bucket < N_BUCKETS &&
1938 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
1940 if (VLOG_IS_DBG_ENABLED()) {
1944 ds_put_cstr(&s, "keep");
1945 for (i = 0; i < N_BUCKETS; i++) {
1947 ds_put_cstr(&s, ", drop");
1950 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
1953 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
1957 return bucket * BUCKET_WIDTH;
1961 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
1963 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
1964 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
1965 struct ofexpired expired;
1967 if (facet->installed) {
1968 struct dpif_flow_stats stats;
1970 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
1972 facet_update_stats(ofproto, facet, &stats);
1975 expired.flow = facet->flow;
1976 expired.packet_count = facet->packet_count;
1977 expired.byte_count = facet->byte_count;
1978 expired.used = facet->used;
1979 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1984 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
1986 long long int cutoff = time_msec() - dp_max_idle;
1987 struct facet *facet, *next_facet;
1989 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1990 facet_active_timeout(ofproto, facet);
1991 if (facet->used < cutoff) {
1992 facet_remove(ofproto, facet);
1997 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
1998 * then delete it entirely. */
2000 rule_expire(struct rule_dpif *rule)
2002 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2003 struct facet *facet, *next_facet;
2007 /* Has 'rule' expired? */
2009 if (rule->up.hard_timeout
2010 && now > rule->up.created + rule->up.hard_timeout * 1000) {
2011 reason = OFPRR_HARD_TIMEOUT;
2012 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2013 && now > rule->used + rule->up.idle_timeout * 1000) {
2014 reason = OFPRR_IDLE_TIMEOUT;
2019 COVERAGE_INC(ofproto_dpif_expired);
2021 /* Update stats. (This is a no-op if the rule expired due to an idle
2022 * timeout, because that only happens when the rule has no facets left.) */
2023 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2024 facet_remove(ofproto, facet);
2027 /* Get rid of the rule. */
2028 ofproto_rule_expire(&rule->up, reason);
2033 /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
2034 * example 'packet' within that flow.
2036 * The caller must already have determined that no facet with an identical
2037 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2038 * the ofproto's classifier table. */
2039 static struct facet *
2040 facet_create(struct rule_dpif *rule, const struct flow *flow,
2041 const struct ofpbuf *packet)
2043 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2044 struct facet *facet;
2046 facet = xzalloc(sizeof *facet);
2047 facet->used = time_msec();
2048 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2049 list_push_back(&rule->facets, &facet->list_node);
2051 facet->flow = *flow;
2052 netflow_flow_init(&facet->nf_flow);
2053 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2055 facet_make_actions(ofproto, facet, packet);
2061 facet_free(struct facet *facet)
2063 free(facet->actions);
2067 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2068 * 'packet', which arrived on 'in_port'.
2070 * Takes ownership of 'packet'. */
2072 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2073 const struct nlattr *odp_actions, size_t actions_len,
2074 struct ofpbuf *packet)
2076 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2077 && odp_actions->nla_type == ODP_ACTION_ATTR_USERSPACE) {
2078 /* As an optimization, avoid a round-trip from userspace to kernel to
2079 * userspace. This also avoids possibly filling up kernel packet
2080 * buffers along the way. */
2081 struct dpif_upcall upcall;
2083 upcall.type = DPIF_UC_ACTION;
2084 upcall.packet = packet;
2087 upcall.userdata = nl_attr_get_u64(odp_actions);
2088 upcall.sample_pool = 0;
2089 upcall.actions = NULL;
2090 upcall.actions_len = 0;
2092 send_packet_in(ofproto, &upcall, flow, false);
2096 struct odputil_keybuf keybuf;
2100 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2101 odp_flow_key_from_flow(&key, flow);
2103 error = dpif_execute(ofproto->dpif, key.data, key.size,
2104 odp_actions, actions_len, packet);
2106 ofpbuf_delete(packet);
2111 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2112 * statistics appropriately. 'packet' must have at least sizeof(struct
2113 * ofp_packet_in) bytes of headroom.
2115 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2116 * applying flow_extract() to 'packet' would yield the same flow as
2119 * 'facet' must have accurately composed ODP actions; that is, it must not be
2120 * in need of revalidation.
2122 * Takes ownership of 'packet'. */
2124 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2125 struct ofpbuf *packet)
2127 struct dpif_flow_stats stats;
2129 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2131 flow_extract_stats(&facet->flow, packet, &stats);
2132 stats.used = time_msec();
2133 if (execute_odp_actions(ofproto, &facet->flow,
2134 facet->actions, facet->actions_len, packet)) {
2135 facet_update_stats(ofproto, facet, &stats);
2139 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2141 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2142 * rule's statistics, via facet_uninstall().
2144 * - Removes 'facet' from its rule and from ofproto->facets.
2147 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2149 facet_uninstall(ofproto, facet);
2150 facet_flush_stats(ofproto, facet);
2151 hmap_remove(&ofproto->facets, &facet->hmap_node);
2152 list_remove(&facet->list_node);
2156 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2158 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2159 const struct ofpbuf *packet)
2161 const struct rule_dpif *rule = facet->rule;
2162 struct ofpbuf *odp_actions;
2163 struct action_xlate_ctx ctx;
2165 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2166 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2167 facet->tags = ctx.tags;
2168 facet->may_install = ctx.may_set_up_flow;
2169 facet->nf_flow.output_iface = ctx.nf_output_iface;
2171 if (facet->actions_len != odp_actions->size
2172 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2173 free(facet->actions);
2174 facet->actions_len = odp_actions->size;
2175 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2178 ofpbuf_delete(odp_actions);
2181 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2182 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2183 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2184 * since 'facet' was last updated.
2186 * Returns 0 if successful, otherwise a positive errno value.*/
2188 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2189 const struct nlattr *actions, size_t actions_len,
2190 struct dpif_flow_stats *stats)
2192 struct odputil_keybuf keybuf;
2193 enum dpif_flow_put_flags flags;
2197 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2199 flags |= DPIF_FP_ZERO_STATS;
2202 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2203 odp_flow_key_from_flow(&key, &facet->flow);
2205 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2206 actions, actions_len, stats);
2209 facet_reset_dp_stats(facet, stats);
2215 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2216 * 'zero_stats' is true, clears any existing statistics from the datapath for
2219 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2221 struct dpif_flow_stats stats;
2223 if (facet->may_install
2224 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2225 zero_stats ? &stats : NULL)) {
2226 facet->installed = true;
2231 vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci)
2233 return vlan_tci != htons(0) ? vlan_tci_to_vid(vlan_tci) : OFP_VLAN_NONE;
2237 facet_account(struct ofproto_dpif *ofproto,
2238 struct facet *facet, uint64_t extra_bytes)
2240 uint64_t total_bytes, n_bytes;
2241 struct ofbundle *in_bundle;
2242 const struct nlattr *a;
2248 total_bytes = facet->byte_count + extra_bytes;
2249 if (total_bytes <= facet->accounted_bytes) {
2252 n_bytes = total_bytes - facet->accounted_bytes;
2253 facet->accounted_bytes = total_bytes;
2255 /* Test that 'tags' is nonzero to ensure that only flows that include an
2256 * OFPP_NORMAL action are used for learning and bond slave rebalancing.
2257 * This works because OFPP_NORMAL always sets a nonzero tag value.
2259 * Feed information from the active flows back into the learning table to
2260 * ensure that table is always in sync with what is actually flowing
2261 * through the datapath. */
2263 || !is_admissible(ofproto, &facet->flow, false, &dummy,
2264 &vlan, &in_bundle)) {
2268 update_learning_table(ofproto, &facet->flow, vlan, in_bundle);
2270 if (!ofproto->has_bonded_bundles) {
2274 /* This loop feeds byte counters to bond_account() for rebalancing to use
2275 * as a basis. We also need to track the actual VLAN on which the packet
2276 * is going to be sent to ensure that it matches the one passed to
2277 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2279 vlan_tci = facet->flow.vlan_tci;
2280 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2281 struct ofport_dpif *port;
2283 switch (nl_attr_type(a)) {
2284 case ODP_ACTION_ATTR_OUTPUT:
2285 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2286 if (port && port->bundle && port->bundle->bond) {
2287 bond_account(port->bundle->bond, &facet->flow,
2288 vlan_tci_to_openflow_vlan(vlan_tci), n_bytes);
2292 case ODP_ACTION_ATTR_STRIP_VLAN:
2293 vlan_tci = htons(0);
2296 case ODP_ACTION_ATTR_SET_DL_TCI:
2297 vlan_tci = nl_attr_get_be16(a);
2303 /* If 'rule' is installed in the datapath, uninstalls it. */
2305 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2307 if (facet->installed) {
2308 struct odputil_keybuf keybuf;
2309 struct dpif_flow_stats stats;
2313 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2314 odp_flow_key_from_flow(&key, &facet->flow);
2316 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2317 facet_reset_dp_stats(facet, &stats);
2319 facet_update_stats(p, facet, &stats);
2321 facet->installed = false;
2323 assert(facet->dp_packet_count == 0);
2324 assert(facet->dp_byte_count == 0);
2328 /* Returns true if the only action for 'facet' is to send to the controller.
2329 * (We don't report NetFlow expiration messages for such facets because they
2330 * are just part of the control logic for the network, not real traffic). */
2332 facet_is_controller_flow(struct facet *facet)
2335 && facet->rule->up.n_actions == 1
2336 && action_outputs_to_port(&facet->rule->up.actions[0],
2337 htons(OFPP_CONTROLLER)));
2340 /* Resets 'facet''s datapath statistics counters. This should be called when
2341 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2342 * it should contain the statistics returned by dpif when 'facet' was reset in
2343 * the datapath. 'stats' will be modified to only included statistics new
2344 * since 'facet' was last updated. */
2346 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2348 if (stats && facet->dp_packet_count <= stats->n_packets
2349 && facet->dp_byte_count <= stats->n_bytes) {
2350 stats->n_packets -= facet->dp_packet_count;
2351 stats->n_bytes -= facet->dp_byte_count;
2354 facet->dp_packet_count = 0;
2355 facet->dp_byte_count = 0;
2358 /* Folds all of 'facet''s statistics into its rule. Also updates the
2359 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2360 * 'facet''s statistics in the datapath should have been zeroed and folded into
2361 * its packet and byte counts before this function is called. */
2363 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2365 assert(!facet->dp_byte_count);
2366 assert(!facet->dp_packet_count);
2368 facet_push_stats(facet);
2369 facet_account(ofproto, facet, 0);
2371 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2372 struct ofexpired expired;
2373 expired.flow = facet->flow;
2374 expired.packet_count = facet->packet_count;
2375 expired.byte_count = facet->byte_count;
2376 expired.used = facet->used;
2377 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2380 facet->rule->packet_count += facet->packet_count;
2381 facet->rule->byte_count += facet->byte_count;
2383 /* Reset counters to prevent double counting if 'facet' ever gets
2385 facet->packet_count = 0;
2386 facet->byte_count = 0;
2387 facet->rs_packet_count = 0;
2388 facet->rs_byte_count = 0;
2389 facet->accounted_bytes = 0;
2391 netflow_flow_clear(&facet->nf_flow);
2394 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2395 * Returns it if found, otherwise a null pointer.
2397 * The returned facet might need revalidation; use facet_lookup_valid()
2398 * instead if that is important. */
2399 static struct facet *
2400 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
2402 struct facet *facet;
2404 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2406 if (flow_equal(flow, &facet->flow)) {
2414 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2415 * Returns it if found, otherwise a null pointer.
2417 * The returned facet is guaranteed to be valid. */
2418 static struct facet *
2419 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
2421 struct facet *facet = facet_find(ofproto, flow);
2423 /* The facet we found might not be valid, since we could be in need of
2424 * revalidation. If it is not valid, don't return it. */
2426 && ofproto->need_revalidate
2427 && !facet_revalidate(ofproto, facet)) {
2428 COVERAGE_INC(facet_invalidated);
2435 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2437 * - If the rule found is different from 'facet''s current rule, moves
2438 * 'facet' to the new rule and recompiles its actions.
2440 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2441 * where it is and recompiles its actions anyway.
2443 * - If there is none, destroys 'facet'.
2445 * Returns true if 'facet' still exists, false if it has been destroyed. */
2447 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
2449 struct action_xlate_ctx ctx;
2450 struct ofpbuf *odp_actions;
2451 struct rule_dpif *new_rule;
2452 bool actions_changed;
2454 COVERAGE_INC(facet_revalidate);
2456 /* Determine the new rule. */
2457 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
2459 /* No new rule, so delete the facet. */
2460 facet_remove(ofproto, facet);
2464 /* Calculate new ODP actions.
2466 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2467 * emit a NetFlow expiration and, if so, we need to have the old state
2468 * around to properly compose it. */
2469 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2470 odp_actions = xlate_actions(&ctx,
2471 new_rule->up.actions, new_rule->up.n_actions);
2472 actions_changed = (facet->actions_len != odp_actions->size
2473 || memcmp(facet->actions, odp_actions->data,
2474 facet->actions_len));
2476 /* If the ODP actions changed or the installability changed, then we need
2477 * to talk to the datapath. */
2478 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2479 if (ctx.may_set_up_flow) {
2480 struct dpif_flow_stats stats;
2482 facet_put__(ofproto, facet,
2483 odp_actions->data, odp_actions->size, &stats);
2484 facet_update_stats(ofproto, facet, &stats);
2486 facet_uninstall(ofproto, facet);
2489 /* The datapath flow is gone or has zeroed stats, so push stats out of
2490 * 'facet' into 'rule'. */
2491 facet_flush_stats(ofproto, facet);
2494 /* Update 'facet' now that we've taken care of all the old state. */
2495 facet->tags = ctx.tags;
2496 facet->nf_flow.output_iface = ctx.nf_output_iface;
2497 facet->may_install = ctx.may_set_up_flow;
2498 if (actions_changed) {
2499 free(facet->actions);
2500 facet->actions_len = odp_actions->size;
2501 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2503 if (facet->rule != new_rule) {
2504 COVERAGE_INC(facet_changed_rule);
2505 list_remove(&facet->list_node);
2506 list_push_back(&new_rule->facets, &facet->list_node);
2507 facet->rule = new_rule;
2508 facet->used = new_rule->up.created;
2509 facet->rs_used = facet->used;
2512 ofpbuf_delete(odp_actions);
2517 /* Updates 'facet''s used time. Caller is responsible for calling
2518 * facet_push_stats() to update the flows which 'facet' resubmits into. */
2520 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
2523 if (used > facet->used) {
2525 if (used > facet->rule->used) {
2526 facet->rule->used = used;
2528 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
2532 /* Folds the statistics from 'stats' into the counters in 'facet'.
2534 * Because of the meaning of a facet's counters, it only makes sense to do this
2535 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
2536 * packet that was sent by hand or if it represents statistics that have been
2537 * cleared out of the datapath. */
2539 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
2540 const struct dpif_flow_stats *stats)
2542 if (stats->n_packets || stats->used > facet->used) {
2543 facet_update_time(ofproto, facet, stats->used);
2544 facet->packet_count += stats->n_packets;
2545 facet->byte_count += stats->n_bytes;
2546 facet_push_stats(facet);
2547 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
2552 facet_push_stats(struct facet *facet)
2554 uint64_t rs_packets, rs_bytes;
2556 assert(facet->packet_count >= facet->rs_packet_count);
2557 assert(facet->byte_count >= facet->rs_byte_count);
2558 assert(facet->used >= facet->rs_used);
2560 rs_packets = facet->packet_count - facet->rs_packet_count;
2561 rs_bytes = facet->byte_count - facet->rs_byte_count;
2563 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
2564 facet->rs_packet_count = facet->packet_count;
2565 facet->rs_byte_count = facet->byte_count;
2566 facet->rs_used = facet->used;
2568 flow_push_stats(facet->rule, &facet->flow,
2569 rs_packets, rs_bytes, facet->used);
2573 struct ofproto_push {
2574 struct action_xlate_ctx ctx;
2581 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
2583 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
2586 rule->packet_count += push->packets;
2587 rule->byte_count += push->bytes;
2588 rule->used = MAX(push->used, rule->used);
2592 /* Pushes flow statistics to the rules which 'flow' resubmits into given
2593 * 'rule''s actions. */
2595 flow_push_stats(const struct rule_dpif *rule,
2596 struct flow *flow, uint64_t packets, uint64_t bytes,
2599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2600 struct ofproto_push push;
2602 push.packets = packets;
2606 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
2607 push.ctx.resubmit_hook = push_resubmit;
2608 ofpbuf_delete(xlate_actions(&push.ctx,
2609 rule->up.actions, rule->up.n_actions));
2614 static struct rule_dpif *
2615 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
2617 return rule_dpif_cast(rule_from_cls_rule(
2618 classifier_lookup(&ofproto->up.tables[0],
2623 complete_operation(struct rule_dpif *rule)
2625 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2627 ofproto->need_revalidate = true;
2629 struct dpif_completion *c = xmalloc(sizeof *c);
2630 c->op = rule->up.pending;
2631 list_push_back(&ofproto->completions, &c->list_node);
2633 ofoperation_complete(rule->up.pending, 0);
2637 static struct rule *
2640 struct rule_dpif *rule = xmalloc(sizeof *rule);
2645 rule_dealloc(struct rule *rule_)
2647 struct rule_dpif *rule = rule_dpif_cast(rule_);
2652 rule_construct(struct rule *rule_)
2654 struct rule_dpif *rule = rule_dpif_cast(rule_);
2655 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2656 struct rule_dpif *victim;
2659 error = validate_actions(rule->up.actions, rule->up.n_actions,
2660 &rule->up.cr.flow, ofproto->max_ports);
2665 rule->used = rule->up.created;
2666 rule->packet_count = 0;
2667 rule->byte_count = 0;
2669 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
2670 if (victim && !list_is_empty(&victim->facets)) {
2671 struct facet *facet;
2673 rule->facets = victim->facets;
2674 list_moved(&rule->facets);
2675 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2679 /* Must avoid list_moved() in this case. */
2680 list_init(&rule->facets);
2683 complete_operation(rule);
2688 rule_destruct(struct rule *rule_)
2690 struct rule_dpif *rule = rule_dpif_cast(rule_);
2691 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2692 struct facet *facet, *next_facet;
2694 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2695 facet_revalidate(ofproto, facet);
2698 complete_operation(rule);
2702 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
2704 struct rule_dpif *rule = rule_dpif_cast(rule_);
2705 struct facet *facet;
2707 /* Start from historical data for 'rule' itself that are no longer tracked
2708 * in facets. This counts, for example, facets that have expired. */
2709 *packets = rule->packet_count;
2710 *bytes = rule->byte_count;
2712 /* Add any statistics that are tracked by facets. This includes
2713 * statistical data recently updated by ofproto_update_stats() as well as
2714 * stats for packets that were executed "by hand" via dpif_execute(). */
2715 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2716 *packets += facet->packet_count;
2717 *bytes += facet->byte_count;
2722 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
2724 struct rule_dpif *rule = rule_dpif_cast(rule_);
2725 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2726 struct action_xlate_ctx ctx;
2727 struct ofpbuf *odp_actions;
2728 struct facet *facet;
2731 /* First look for a related facet. If we find one, account it to that. */
2732 facet = facet_lookup_valid(ofproto, flow);
2733 if (facet && facet->rule == rule) {
2734 facet_execute(ofproto, facet, packet);
2738 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2739 * create a new facet for it and use that. */
2740 if (rule_dpif_lookup(ofproto, flow) == rule) {
2741 facet = facet_create(rule, flow, packet);
2742 facet_execute(ofproto, facet, packet);
2743 facet_install(ofproto, facet, true);
2747 /* We can't account anything to a facet. If we were to try, then that
2748 * facet would have a non-matching rule, busting our invariants. */
2749 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
2750 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2751 size = packet->size;
2752 if (execute_odp_actions(ofproto, flow, odp_actions->data,
2753 odp_actions->size, packet)) {
2754 rule->used = time_msec();
2755 rule->packet_count++;
2756 rule->byte_count += size;
2757 flow_push_stats(rule, flow, 1, size, rule->used);
2759 ofpbuf_delete(odp_actions);
2765 rule_modify_actions(struct rule *rule_)
2767 struct rule_dpif *rule = rule_dpif_cast(rule_);
2768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2771 error = validate_actions(rule->up.actions, rule->up.n_actions,
2772 &rule->up.cr.flow, ofproto->max_ports);
2774 ofoperation_complete(rule->up.pending, error);
2778 complete_operation(rule);
2781 /* Sends 'packet' out of port 'odp_port' within 'p'.
2782 * Returns 0 if successful, otherwise a positive errno value. */
2784 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
2785 const struct ofpbuf *packet)
2787 struct ofpbuf key, odp_actions;
2788 struct odputil_keybuf keybuf;
2792 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
2793 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2794 odp_flow_key_from_flow(&key, &flow);
2796 ofpbuf_init(&odp_actions, 32);
2797 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2798 error = dpif_execute(ofproto->dpif,
2800 odp_actions.data, odp_actions.size,
2802 ofpbuf_uninit(&odp_actions);
2805 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
2806 ofproto->up.name, odp_port, strerror(error));
2811 /* OpenFlow to ODP action translation. */
2813 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2814 struct action_xlate_ctx *ctx);
2815 static void xlate_normal(struct action_xlate_ctx *);
2818 commit_vlan_tci(struct action_xlate_ctx *ctx, ovs_be16 vlan_tci)
2820 struct flow *base = &ctx->base_flow;
2821 struct ofpbuf *odp_actions = ctx->odp_actions;
2823 if (base->vlan_tci != vlan_tci) {
2824 if (!(vlan_tci & htons(VLAN_CFI))) {
2825 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2827 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2828 vlan_tci & ~htons(VLAN_CFI));
2830 base->vlan_tci = vlan_tci;
2835 commit_odp_actions(struct action_xlate_ctx *ctx)
2837 const struct flow *flow = &ctx->flow;
2838 struct flow *base = &ctx->base_flow;
2839 struct ofpbuf *odp_actions = ctx->odp_actions;
2841 if (base->tun_id != flow->tun_id) {
2842 nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id);
2843 base->tun_id = flow->tun_id;
2846 if (base->nw_src != flow->nw_src) {
2847 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src);
2848 base->nw_src = flow->nw_src;
2851 if (base->nw_dst != flow->nw_dst) {
2852 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst);
2853 base->nw_dst = flow->nw_dst;
2856 if (base->nw_tos != flow->nw_tos) {
2857 nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos);
2858 base->nw_tos = flow->nw_tos;
2861 commit_vlan_tci(ctx, flow->vlan_tci);
2863 if (base->tp_src != flow->tp_src) {
2864 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src);
2865 base->tp_src = flow->tp_src;
2868 if (base->tp_dst != flow->tp_dst) {
2869 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst);
2870 base->tp_dst = flow->tp_dst;
2873 if (!eth_addr_equals(base->dl_src, flow->dl_src)) {
2874 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2875 flow->dl_src, ETH_ADDR_LEN);
2876 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
2879 if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) {
2880 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2881 flow->dl_dst, ETH_ADDR_LEN);
2882 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
2885 if (ctx->base_priority != ctx->priority) {
2886 if (ctx->priority) {
2887 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY,
2890 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2892 ctx->base_priority = ctx->priority;
2897 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
2899 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
2900 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
2903 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
2904 /* Forwarding disabled on port. */
2909 * We don't have an ofport record for this port, but it doesn't hurt to
2910 * allow forwarding to it anyhow. Maybe such a port will appear later
2911 * and we're pre-populating the flow table.
2915 commit_odp_actions(ctx);
2916 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2917 ctx->nf_output_iface = ofp_port;
2921 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2923 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2924 struct rule_dpif *rule;
2925 uint16_t old_in_port;
2927 /* Look up a flow with 'in_port' as the input port. Then restore the
2928 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2929 * have surprising behavior). */
2930 old_in_port = ctx->flow.in_port;
2931 ctx->flow.in_port = in_port;
2932 rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow);
2933 ctx->flow.in_port = old_in_port;
2935 if (ctx->resubmit_hook) {
2936 ctx->resubmit_hook(ctx, rule);
2941 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
2945 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2947 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2948 MAX_RESUBMIT_RECURSION);
2953 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
2955 struct ofport_dpif *ofport;
2957 commit_odp_actions(ctx);
2958 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
2959 uint16_t ofp_port = ofport->up.ofp_port;
2960 if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) {
2961 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT,
2966 ctx->nf_output_iface = NF_OUT_FLOOD;
2970 xlate_output_action__(struct action_xlate_ctx *ctx,
2971 uint16_t port, uint16_t max_len)
2973 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2975 ctx->nf_output_iface = NF_OUT_DROP;
2979 add_output_action(ctx, ctx->flow.in_port);
2982 xlate_table_action(ctx, ctx->flow.in_port);
2988 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
2991 flood_packets(ctx, htonl(0));
2993 case OFPP_CONTROLLER:
2994 commit_odp_actions(ctx);
2995 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_USERSPACE, max_len);
2998 add_output_action(ctx, OFPP_LOCAL);
3003 if (port != ctx->flow.in_port) {
3004 add_output_action(ctx, port);
3009 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3010 ctx->nf_output_iface = NF_OUT_FLOOD;
3011 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3012 ctx->nf_output_iface = prev_nf_output_iface;
3013 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3014 ctx->nf_output_iface != NF_OUT_FLOOD) {
3015 ctx->nf_output_iface = NF_OUT_MULTI;
3020 xlate_output_action(struct action_xlate_ctx *ctx,
3021 const struct ofp_action_output *oao)
3023 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3027 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3028 const struct ofp_action_enqueue *oae)
3030 uint16_t ofp_port, odp_port;
3031 uint32_t ctx_priority, priority;
3034 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3037 /* Fall back to ordinary output action. */
3038 xlate_output_action__(ctx, ntohs(oae->port), 0);
3042 /* Figure out ODP output port. */
3043 ofp_port = ntohs(oae->port);
3044 if (ofp_port == OFPP_IN_PORT) {
3045 ofp_port = ctx->flow.in_port;
3046 } else if (ofp_port == ctx->flow.in_port) {
3049 odp_port = ofp_port_to_odp_port(ofp_port);
3051 /* Add ODP actions. */
3052 ctx_priority = ctx->priority;
3053 ctx->priority = priority;
3054 add_output_action(ctx, odp_port);
3055 ctx->priority = ctx_priority;
3057 /* Update NetFlow output port. */
3058 if (ctx->nf_output_iface == NF_OUT_DROP) {
3059 ctx->nf_output_iface = odp_port;
3060 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3061 ctx->nf_output_iface = NF_OUT_MULTI;
3066 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3067 const struct nx_action_set_queue *nasq)
3072 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3075 /* Couldn't translate queue to a priority, so ignore. A warning
3076 * has already been logged. */
3080 ctx->priority = priority;
3083 struct xlate_reg_state {
3089 xlate_autopath(struct action_xlate_ctx *ctx,
3090 const struct nx_action_autopath *naa)
3092 uint16_t ofp_port = ntohl(naa->id);
3093 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
3095 if (!port || !port->bundle) {
3096 ofp_port = OFPP_NONE;
3097 } else if (port->bundle->bond) {
3098 /* Autopath does not support VLAN hashing. */
3099 struct ofport_dpif *slave = bond_choose_output_slave(
3100 port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags);
3102 ofp_port = slave->up.ofp_port;
3105 autopath_execute(naa, &ctx->flow, ofp_port);
3109 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
3111 struct ofproto_dpif *ofproto = ofproto_;
3112 struct ofport_dpif *port;
3122 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3125 port = get_ofp_port(ofproto, ofp_port);
3126 return port ? port->may_enable : false;
3131 do_xlate_actions(const union ofp_action *in, size_t n_in,
3132 struct action_xlate_ctx *ctx)
3134 const struct ofport_dpif *port;
3135 const union ofp_action *ia;
3138 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
3140 && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3141 port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3142 ? htonl(OFPPC_NO_RECV_STP)
3143 : htonl(OFPPC_NO_RECV))) {
3144 /* Drop this flow. */
3148 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
3149 const struct ofp_action_dl_addr *oada;
3150 const struct nx_action_resubmit *nar;
3151 const struct nx_action_set_tunnel *nast;
3152 const struct nx_action_set_queue *nasq;
3153 const struct nx_action_multipath *nam;
3154 const struct nx_action_autopath *naa;
3155 const struct nx_action_bundle *nab;
3156 enum ofputil_action_code code;
3159 code = ofputil_decode_action_unsafe(ia);
3161 case OFPUTIL_OFPAT_OUTPUT:
3162 xlate_output_action(ctx, &ia->output);
3165 case OFPUTIL_OFPAT_SET_VLAN_VID:
3166 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3167 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3170 case OFPUTIL_OFPAT_SET_VLAN_PCP:
3171 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3172 ctx->flow.vlan_tci |= htons(
3173 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3176 case OFPUTIL_OFPAT_STRIP_VLAN:
3177 ctx->flow.vlan_tci = htons(0);
3180 case OFPUTIL_OFPAT_SET_DL_SRC:
3181 oada = ((struct ofp_action_dl_addr *) ia);
3182 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3185 case OFPUTIL_OFPAT_SET_DL_DST:
3186 oada = ((struct ofp_action_dl_addr *) ia);
3187 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3190 case OFPUTIL_OFPAT_SET_NW_SRC:
3191 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3194 case OFPUTIL_OFPAT_SET_NW_DST:
3195 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3198 case OFPUTIL_OFPAT_SET_NW_TOS:
3199 ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK;
3202 case OFPUTIL_OFPAT_SET_TP_SRC:
3203 ctx->flow.tp_src = ia->tp_port.tp_port;
3206 case OFPUTIL_OFPAT_SET_TP_DST:
3207 ctx->flow.tp_dst = ia->tp_port.tp_port;
3210 case OFPUTIL_OFPAT_ENQUEUE:
3211 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3214 case OFPUTIL_NXAST_RESUBMIT:
3215 nar = (const struct nx_action_resubmit *) ia;
3216 xlate_table_action(ctx, ntohs(nar->in_port));
3219 case OFPUTIL_NXAST_SET_TUNNEL:
3220 nast = (const struct nx_action_set_tunnel *) ia;
3221 tun_id = htonll(ntohl(nast->tun_id));
3222 ctx->flow.tun_id = tun_id;
3225 case OFPUTIL_NXAST_SET_QUEUE:
3226 nasq = (const struct nx_action_set_queue *) ia;
3227 xlate_set_queue_action(ctx, nasq);
3230 case OFPUTIL_NXAST_POP_QUEUE:
3234 case OFPUTIL_NXAST_REG_MOVE:
3235 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
3239 case OFPUTIL_NXAST_REG_LOAD:
3240 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
3244 case OFPUTIL_NXAST_NOTE:
3245 /* Nothing to do. */
3248 case OFPUTIL_NXAST_SET_TUNNEL64:
3249 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
3250 ctx->flow.tun_id = tun_id;
3253 case OFPUTIL_NXAST_MULTIPATH:
3254 nam = (const struct nx_action_multipath *) ia;
3255 multipath_execute(nam, &ctx->flow);
3258 case OFPUTIL_NXAST_AUTOPATH:
3259 naa = (const struct nx_action_autopath *) ia;
3260 xlate_autopath(ctx, naa);
3263 case OFPUTIL_NXAST_BUNDLE:
3264 ctx->ofproto->has_bundle_action = true;
3265 nab = (const struct nx_action_bundle *) ia;
3266 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
3271 case OFPUTIL_NXAST_BUNDLE_LOAD:
3272 ctx->ofproto->has_bundle_action = true;
3273 nab = (const struct nx_action_bundle *) ia;
3274 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
3282 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3283 struct ofproto_dpif *ofproto, const struct flow *flow,
3284 const struct ofpbuf *packet)
3286 ctx->ofproto = ofproto;
3288 ctx->packet = packet;
3289 ctx->resubmit_hook = NULL;
3292 static struct ofpbuf *
3293 xlate_actions(struct action_xlate_ctx *ctx,
3294 const union ofp_action *in, size_t n_in)
3296 COVERAGE_INC(ofproto_dpif_xlate);
3298 ctx->odp_actions = ofpbuf_new(512);
3300 ctx->may_set_up_flow = true;
3301 ctx->nf_output_iface = NF_OUT_DROP;
3304 ctx->base_priority = 0;
3305 ctx->base_flow = ctx->flow;
3306 ctx->base_flow.tun_id = 0;
3308 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
3309 ctx->may_set_up_flow = false;
3311 do_xlate_actions(in, n_in, ctx);
3314 /* Check with in-band control to see if we're allowed to set up this
3316 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
3317 ctx->odp_actions->data,
3318 ctx->odp_actions->size)) {
3319 ctx->may_set_up_flow = false;
3322 return ctx->odp_actions;
3325 /* OFPP_NORMAL implementation. */
3328 struct ofport_dpif *port;
3333 struct dst builtin[32];
3335 size_t n, allocated;
3338 static void dst_set_init(struct dst_set *);
3339 static void dst_set_add(struct dst_set *, const struct dst *);
3340 static void dst_set_free(struct dst_set *);
3342 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
3345 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
3346 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
3348 dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE
3349 : in_bundle->vlan >= 0 ? in_bundle->vlan
3350 : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE
3351 : vlan_tci_to_vid(ctx->flow.vlan_tci));
3353 dst->port = (!out_bundle->bond
3354 ? ofbundle_get_a_port(out_bundle)
3355 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
3356 dst->vlan, &ctx->tags));
3358 return dst->port != NULL;
3362 mirror_mask_ffs(mirror_mask_t mask)
3364 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
3369 dst_set_init(struct dst_set *set)
3371 set->dsts = set->builtin;
3373 set->allocated = ARRAY_SIZE(set->builtin);
3377 dst_set_add(struct dst_set *set, const struct dst *dst)
3379 if (set->n >= set->allocated) {
3380 size_t new_allocated;
3381 struct dst *new_dsts;
3383 new_allocated = set->allocated * 2;
3384 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
3385 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
3389 set->dsts = new_dsts;
3390 set->allocated = new_allocated;
3392 set->dsts[set->n++] = *dst;
3396 dst_set_free(struct dst_set *set)
3398 if (set->dsts != set->builtin) {
3404 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
3407 for (i = 0; i < set->n; i++) {
3408 if (set->dsts[i].vlan == test->vlan
3409 && set->dsts[i].port == test->port) {
3417 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
3419 return (bundle->vlan < 0
3420 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
3424 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
3426 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
3429 /* Returns an arbitrary interface within 'bundle'. */
3430 static struct ofport_dpif *
3431 ofbundle_get_a_port(const struct ofbundle *bundle)
3433 return CONTAINER_OF(list_front(&bundle->ports),
3434 struct ofport_dpif, bundle_node);
3438 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
3439 const struct ofbundle *in_bundle,
3440 const struct ofbundle *out_bundle, struct dst_set *set)
3444 if (out_bundle == OFBUNDLE_FLOOD) {
3445 struct ofbundle *bundle;
3447 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
3448 if (bundle != in_bundle
3449 && ofbundle_includes_vlan(bundle, vlan)
3450 && bundle->floodable
3451 && !bundle->mirror_out
3452 && set_dst(ctx, &dst, in_bundle, bundle)) {
3453 dst_set_add(set, &dst);
3456 ctx->nf_output_iface = NF_OUT_FLOOD;
3457 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
3458 dst_set_add(set, &dst);
3459 ctx->nf_output_iface = dst.port->odp_port;
3464 vlan_is_mirrored(const struct ofmirror *m, int vlan)
3466 return !m->vlans || bitmap_is_set(m->vlans, vlan);
3469 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
3470 * to a VLAN. In general most packets may be mirrored but we want to drop
3471 * protocols that may confuse switches. */
3473 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
3475 /* If you change this function's behavior, please update corresponding
3476 * documentation in vswitch.xml at the same time. */
3477 if (dst[0] != 0x01) {
3478 /* All the currently banned MACs happen to start with 01 currently, so
3479 * this is a quick way to eliminate most of the good ones. */
3481 if (eth_addr_is_reserved(dst)) {
3482 /* Drop STP, IEEE pause frames, and other reserved protocols
3483 * (01-80-c2-00-00-0x). */
3487 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
3489 if ((dst[3] & 0xfe) == 0xcc &&
3490 (dst[4] & 0xfe) == 0xcc &&
3491 (dst[5] & 0xfe) == 0xcc) {
3492 /* Drop the following protocols plus others following the same
3495 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
3496 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
3497 STP Uplink Fast (01-00-0c-cd-cd-cd) */
3501 if (!(dst[3] | dst[4] | dst[5])) {
3502 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
3511 compose_mirror_dsts(struct action_xlate_ctx *ctx,
3512 uint16_t vlan, const struct ofbundle *in_bundle,
3513 struct dst_set *set)
3515 struct ofproto_dpif *ofproto = ctx->ofproto;
3516 mirror_mask_t mirrors;
3520 mirrors = in_bundle->src_mirrors;
3521 for (i = 0; i < set->n; i++) {
3522 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
3529 flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3530 if (flow_vlan == 0) {
3531 flow_vlan = OFP_VLAN_NONE;
3535 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
3536 if (vlan_is_mirrored(m, vlan)) {
3540 if (set_dst(ctx, &dst, in_bundle, m->out)
3541 && !dst_is_duplicate(set, &dst)) {
3542 dst_set_add(set, &dst);
3544 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
3545 struct ofbundle *bundle;
3547 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3548 if (ofbundle_includes_vlan(bundle, m->out_vlan)
3549 && set_dst(ctx, &dst, in_bundle, bundle))
3551 if (bundle->vlan < 0) {
3552 dst.vlan = m->out_vlan;
3554 if (dst_is_duplicate(set, &dst)) {
3558 /* Use the vlan tag on the original flow instead of
3559 * the one passed in the vlan parameter. This ensures
3560 * that we compare the vlan from before any implicit
3561 * tagging tags place. This is necessary because
3562 * dst->vlan is the final vlan, after removing implicit
3564 if (bundle == in_bundle && dst.vlan == flow_vlan) {
3565 /* Don't send out input port on same VLAN. */
3568 dst_set_add(set, &dst);
3573 mirrors &= mirrors - 1;
3578 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
3579 const struct ofbundle *in_bundle,
3580 const struct ofbundle *out_bundle)
3582 uint16_t initial_vlan, cur_vlan;
3583 const struct dst *dst;
3587 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
3588 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
3593 /* Output all the packets we can without having to change the VLAN. */
3594 commit_odp_actions(ctx);
3595 initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3596 if (initial_vlan == 0) {
3597 initial_vlan = OFP_VLAN_NONE;
3599 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3600 if (dst->vlan != initial_vlan) {
3603 nl_msg_put_u32(ctx->odp_actions,
3604 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3607 /* Then output the rest. */
3608 cur_vlan = initial_vlan;
3609 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3610 if (dst->vlan == initial_vlan) {
3613 if (dst->vlan != cur_vlan) {
3616 tci = htons(dst->vlan == OFP_VLAN_NONE ? 0 : dst->vlan);
3617 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
3619 tci |= htons(VLAN_CFI);
3621 commit_vlan_tci(ctx, tci);
3623 cur_vlan = dst->vlan;
3625 nl_msg_put_u32(ctx->odp_actions,
3626 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3632 /* Returns the effective vlan of a packet, taking into account both the
3633 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
3634 * the packet is untagged and -1 indicates it has an invalid header and
3635 * should be dropped. */
3637 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
3638 struct ofbundle *in_bundle, bool have_packet)
3640 int vlan = vlan_tci_to_vid(flow->vlan_tci);
3641 if (in_bundle->vlan >= 0) {
3644 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3645 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3646 "packet received on port %s configured with "
3647 "implicit VLAN %"PRIu16,
3648 ofproto->up.name, vlan,
3649 in_bundle->name, in_bundle->vlan);
3653 vlan = in_bundle->vlan;
3655 if (!ofbundle_includes_vlan(in_bundle, vlan)) {
3657 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3658 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3659 "packet received on port %s not configured for "
3661 ofproto->up.name, vlan, in_bundle->name, vlan);
3670 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
3671 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
3672 * indicate this; newer upstream kernels use gratuitous ARP requests. */
3674 is_gratuitous_arp(const struct flow *flow)
3676 return (flow->dl_type == htons(ETH_TYPE_ARP)
3677 && eth_addr_is_broadcast(flow->dl_dst)
3678 && (flow->nw_proto == ARP_OP_REPLY
3679 || (flow->nw_proto == ARP_OP_REQUEST
3680 && flow->nw_src == flow->nw_dst)));
3684 update_learning_table(struct ofproto_dpif *ofproto,
3685 const struct flow *flow, int vlan,
3686 struct ofbundle *in_bundle)
3688 struct mac_entry *mac;
3690 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
3694 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
3695 if (is_gratuitous_arp(flow)) {
3696 /* We don't want to learn from gratuitous ARP packets that are
3697 * reflected back over bond slaves so we lock the learning table. */
3698 if (!in_bundle->bond) {
3699 mac_entry_set_grat_arp_lock(mac);
3700 } else if (mac_entry_is_grat_arp_locked(mac)) {
3705 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
3706 /* The log messages here could actually be useful in debugging,
3707 * so keep the rate limit relatively high. */
3708 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3709 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
3710 "on port %s in VLAN %d",
3711 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
3712 in_bundle->name, vlan);
3714 mac->port.p = in_bundle;
3715 tag_set_add(&ofproto->revalidate_set,
3716 mac_learning_changed(ofproto->ml, mac));
3720 /* Determines whether packets in 'flow' within 'br' should be forwarded or
3721 * dropped. Returns true if they may be forwarded, false if they should be
3724 * If 'have_packet' is true, it indicates that the caller is processing a
3725 * received packet. If 'have_packet' is false, then the caller is just
3726 * revalidating an existing flow because configuration has changed. Either
3727 * way, 'have_packet' only affects logging (there is no point in logging errors
3728 * during revalidation).
3730 * Sets '*in_portp' to the input port. This will be a null pointer if
3731 * flow->in_port does not designate a known input port (in which case
3732 * is_admissible() returns false).
3734 * When returning true, sets '*vlanp' to the effective VLAN of the input
3735 * packet, as returned by flow_get_vlan().
3737 * May also add tags to '*tags', although the current implementation only does
3738 * so in one special case.
3741 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
3743 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
3745 struct ofport_dpif *in_port;
3746 struct ofbundle *in_bundle;
3749 /* Find the port and bundle for the received packet. */
3750 in_port = get_ofp_port(ofproto, flow->in_port);
3751 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
3752 if (!in_port || !in_bundle) {
3753 /* No interface? Something fishy... */
3755 /* Odd. A few possible reasons here:
3757 * - We deleted a port but there are still a few packets queued up
3760 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
3761 * we don't know about.
3763 * - Packet arrived on the local port but the local port is not
3766 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3768 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3770 ofproto->up.name, flow->in_port);
3774 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
3779 /* Drop frames for reserved multicast addresses
3780 * only if forward_bpdu option is absent. */
3781 if (eth_addr_is_reserved(flow->dl_dst) &&
3782 !ofproto->up.forward_bpdu) {
3786 /* Drop frames on bundles reserved for mirroring. */
3787 if (in_bundle->mirror_out) {
3789 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3790 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3791 "%s, which is reserved exclusively for mirroring",
3792 ofproto->up.name, in_bundle->name);
3797 if (in_bundle->bond) {
3798 struct mac_entry *mac;
3800 switch (bond_check_admissibility(in_bundle->bond, in_port,
3801 flow->dl_dst, tags)) {
3808 case BV_DROP_IF_MOVED:
3809 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
3810 if (mac && mac->port.p != in_bundle &&
3811 (!is_gratuitous_arp(flow)
3812 || mac_entry_is_grat_arp_locked(mac))) {
3823 xlate_normal(struct action_xlate_ctx *ctx)
3825 struct ofbundle *in_bundle;
3826 struct ofbundle *out_bundle;
3827 struct mac_entry *mac;
3830 /* Check whether we should drop packets in this flow. */
3831 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
3832 &ctx->tags, &vlan, &in_bundle)) {
3837 /* Learn source MAC (but don't try to learn from revalidation). */
3839 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
3842 /* Determine output bundle. */
3843 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
3846 out_bundle = mac->port.p;
3847 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
3848 /* If we are revalidating but don't have a learning entry then eject
3849 * the flow. Installing a flow that floods packets opens up a window
3850 * of time where we could learn from a packet reflected on a bond and
3851 * blackhole packets before the learning table is updated to reflect
3852 * the correct port. */
3853 ctx->may_set_up_flow = false;
3856 out_bundle = OFBUNDLE_FLOOD;
3859 /* Don't send packets out their input bundles. */
3860 if (in_bundle == out_bundle) {
3866 compose_actions(ctx, vlan, in_bundle, out_bundle);
3871 get_drop_frags(struct ofproto *ofproto_)
3873 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3876 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
3881 set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
3883 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3885 dpif_set_drop_frags(ofproto->dpif, drop_frags);
3889 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3890 const struct flow *flow,
3891 const union ofp_action *ofp_actions, size_t n_ofp_actions)
3893 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3896 error = validate_actions(ofp_actions, n_ofp_actions, flow,
3897 ofproto->max_ports);
3899 struct odputil_keybuf keybuf;
3900 struct action_xlate_ctx ctx;
3901 struct ofpbuf *odp_actions;
3904 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3905 odp_flow_key_from_flow(&key, flow);
3907 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3908 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3909 dpif_execute(ofproto->dpif, key.data, key.size,
3910 odp_actions->data, odp_actions->size, packet);
3911 ofpbuf_delete(odp_actions);
3917 get_netflow_ids(const struct ofproto *ofproto_,
3918 uint8_t *engine_type, uint8_t *engine_id)
3920 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3922 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3925 static struct ofproto_dpif *
3926 ofproto_dpif_lookup(const char *name)
3928 struct ofproto *ofproto = ofproto_lookup(name);
3929 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
3930 ? ofproto_dpif_cast(ofproto)
3935 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
3936 const char *args, void *aux OVS_UNUSED)
3938 struct ds ds = DS_EMPTY_INITIALIZER;
3939 const struct ofproto_dpif *ofproto;
3940 const struct mac_entry *e;
3942 ofproto = ofproto_dpif_lookup(args);
3944 unixctl_command_reply(conn, 501, "no such bridge");
3948 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3949 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3950 struct ofbundle *bundle = e->port.p;
3951 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
3952 ofbundle_get_a_port(bundle)->odp_port,
3953 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
3955 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3959 struct ofproto_trace {
3960 struct action_xlate_ctx ctx;
3966 trace_format_rule(struct ds *result, int level, const struct rule *rule)
3968 ds_put_char_multiple(result, '\t', level);
3970 ds_put_cstr(result, "No match\n");
3974 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
3975 ntohll(rule->flow_cookie));
3976 cls_rule_format(&rule->cr, result);
3977 ds_put_char(result, '\n');
3979 ds_put_char_multiple(result, '\t', level);
3980 ds_put_cstr(result, "OpenFlow ");
3981 ofp_print_actions(result, rule->actions, rule->n_actions);
3982 ds_put_char(result, '\n');
3986 trace_format_flow(struct ds *result, int level, const char *title,
3987 struct ofproto_trace *trace)
3989 ds_put_char_multiple(result, '\t', level);
3990 ds_put_format(result, "%s: ", title);
3991 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
3992 ds_put_cstr(result, "unchanged");
3994 flow_format(result, &trace->ctx.flow);
3995 trace->flow = trace->ctx.flow;
3997 ds_put_char(result, '\n');
4001 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
4003 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4004 struct ds *result = trace->result;
4006 ds_put_char(result, '\n');
4007 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4008 trace_format_rule(result, ctx->recurse + 1, &rule->up);
4012 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4013 void *aux OVS_UNUSED)
4015 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4016 char *args = xstrdup(args_);
4017 char *save_ptr = NULL;
4018 struct ofproto_dpif *ofproto;
4019 struct ofpbuf packet;
4020 struct rule_dpif *rule;
4027 ofpbuf_init(&packet, strlen(args) / 2);
4030 dpname = strtok_r(args, " ", &save_ptr);
4031 tun_id_s = strtok_r(NULL, " ", &save_ptr);
4032 in_port_s = strtok_r(NULL, " ", &save_ptr);
4033 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4034 if (!dpname || !in_port_s || !packet_s) {
4035 unixctl_command_reply(conn, 501, "Bad command syntax");
4039 ofproto = ofproto_dpif_lookup(dpname);
4041 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4046 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
4047 in_port = ofp_port_to_odp_port(atoi(in_port_s));
4049 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
4050 packet_s += strspn(packet_s, " ");
4051 if (*packet_s != '\0') {
4052 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4055 if (packet.size < ETH_HEADER_LEN) {
4056 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
4060 ds_put_cstr(&result, "Packet: ");
4061 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
4062 ds_put_cstr(&result, s);
4065 flow_extract(&packet, tun_id, in_port, &flow);
4066 ds_put_cstr(&result, "Flow: ");
4067 flow_format(&result, &flow);
4068 ds_put_char(&result, '\n');
4070 rule = rule_dpif_lookup(ofproto, &flow);
4071 trace_format_rule(&result, 0, &rule->up);
4073 struct ofproto_trace trace;
4074 struct ofpbuf *odp_actions;
4076 trace.result = &result;
4078 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
4079 trace.ctx.resubmit_hook = trace_resubmit;
4080 odp_actions = xlate_actions(&trace.ctx,
4081 rule->up.actions, rule->up.n_actions);
4083 ds_put_char(&result, '\n');
4084 trace_format_flow(&result, 0, "Final flow", &trace);
4085 ds_put_cstr(&result, "Datapath actions: ");
4086 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4087 ofpbuf_delete(odp_actions);
4090 unixctl_command_reply(conn, 200, ds_cstr(&result));
4093 ds_destroy(&result);
4094 ofpbuf_uninit(&packet);
4099 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
4100 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4103 unixctl_command_reply(conn, 200, NULL);
4107 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
4108 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4111 unixctl_command_reply(conn, 200, NULL);
4115 ofproto_dpif_unixctl_init(void)
4117 static bool registered;
4123 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4124 unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
4126 unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL);
4127 unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL);
4130 const struct ofproto_class ofproto_dpif_class = {
4157 port_is_lacp_current,
4158 NULL, /* rule_choose_table */
4165 rule_modify_actions,
4178 is_mirror_output_bundle,
4179 forward_bpdu_changed,