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 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 *,
100 const struct flow *, uint8_t table);
102 #define MAX_MIRRORS 32
103 typedef uint32_t mirror_mask_t;
104 #define MIRROR_MASK_C(X) UINT32_C(X)
105 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
107 struct ofproto_dpif *ofproto; /* Owning ofproto. */
108 size_t idx; /* In ofproto's "mirrors" array. */
109 void *aux; /* Key supplied by ofproto's client. */
110 char *name; /* Identifier for log messages. */
112 /* Selection criteria. */
113 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
114 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
115 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
117 /* Output (mutually exclusive). */
118 struct ofbundle *out; /* Output port or NULL. */
119 int out_vlan; /* Output VLAN or -1. */
122 static void mirror_destroy(struct ofmirror *);
124 /* A group of one or more OpenFlow ports. */
125 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
127 struct ofproto_dpif *ofproto; /* Owning ofproto. */
128 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
129 void *aux; /* Key supplied by ofproto's client. */
130 char *name; /* Identifier for log messages. */
133 struct list ports; /* Contains "struct ofport"s. */
134 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
135 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
136 * NULL if all VLANs are trunked. */
137 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
138 struct bond *bond; /* Nonnull iff more than one port. */
141 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
143 /* Port mirroring info. */
144 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
145 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
146 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
149 static void bundle_remove(struct ofport *);
150 static void bundle_destroy(struct ofbundle *);
151 static void bundle_del_port(struct ofport_dpif *);
152 static void bundle_run(struct ofbundle *);
153 static void bundle_wait(struct ofbundle *);
155 struct action_xlate_ctx {
156 /* action_xlate_ctx_init() initializes these members. */
159 struct ofproto_dpif *ofproto;
161 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
162 * this flow when actions change header fields. */
165 /* The packet corresponding to 'flow', or a null pointer if we are
166 * revalidating without a packet to refer to. */
167 const struct ofpbuf *packet;
169 /* If nonnull, called just before executing a resubmit action.
171 * This is normally null so the client has to set it manually after
172 * calling action_xlate_ctx_init(). */
173 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
175 /* xlate_actions() initializes and uses these members. The client might want
176 * to look at them after it returns. */
178 struct ofpbuf *odp_actions; /* Datapath actions. */
179 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
180 bool may_set_up_flow; /* True ordinarily; false if the actions must
181 * be reassessed for every packet. */
182 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
184 /* xlate_actions() initializes and uses these members, but the client has no
185 * reason to look at them. */
187 int recurse; /* Recursion level, via xlate_table_action. */
188 uint32_t priority; /* Current flow priority. 0 if none. */
189 struct flow base_flow; /* Flow at the last commit. */
190 uint32_t base_priority; /* Priority at the last commit. */
191 uint8_t table_id; /* OpenFlow table ID where flow was found. */
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_, int *n_tablesp)
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_dpif_unixctl_init();
470 ofproto->has_bundle_action = false;
477 complete_operations(struct ofproto_dpif *ofproto)
479 struct dpif_completion *c, *next;
481 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
482 ofoperation_complete(c->op, 0);
483 list_remove(&c->list_node);
489 destruct(struct ofproto *ofproto_)
491 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
492 struct rule_dpif *rule, *next_rule;
493 struct classifier *table;
496 complete_operations(ofproto);
498 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
499 struct cls_cursor cursor;
501 cls_cursor_init(&cursor, table, NULL);
502 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
503 ofproto_rule_destroy(&rule->up);
507 for (i = 0; i < MAX_MIRRORS; i++) {
508 mirror_destroy(ofproto->mirrors[i]);
511 netflow_destroy(ofproto->netflow);
512 dpif_sflow_destroy(ofproto->sflow);
513 hmap_destroy(&ofproto->bundles);
514 mac_learning_destroy(ofproto->ml);
516 hmap_destroy(&ofproto->facets);
518 dpif_close(ofproto->dpif);
522 run(struct ofproto *ofproto_)
524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
525 struct ofport_dpif *ofport;
526 struct ofbundle *bundle;
530 complete_operations(ofproto);
532 dpif_run(ofproto->dpif);
534 for (i = 0; i < 50; i++) {
535 struct dpif_upcall packet;
538 error = dpif_recv(ofproto->dpif, &packet);
540 if (error == ENODEV) {
541 /* Datapath destroyed. */
547 handle_upcall(ofproto, &packet);
550 if (timer_expired(&ofproto->next_expiration)) {
551 int delay = expire(ofproto);
552 timer_set_duration(&ofproto->next_expiration, delay);
555 if (ofproto->netflow) {
556 netflow_run(ofproto->netflow);
558 if (ofproto->sflow) {
559 dpif_sflow_run(ofproto->sflow);
562 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
565 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
569 /* Now revalidate if there's anything to do. */
570 if (ofproto->need_revalidate
571 || !tag_set_is_empty(&ofproto->revalidate_set)) {
572 struct tag_set revalidate_set = ofproto->revalidate_set;
573 bool revalidate_all = ofproto->need_revalidate;
574 struct facet *facet, *next;
576 /* Clear the revalidation flags. */
577 tag_set_init(&ofproto->revalidate_set);
578 ofproto->need_revalidate = false;
580 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
582 || tag_set_intersects(&revalidate_set, facet->tags)) {
583 facet_revalidate(ofproto, facet);
592 wait(struct ofproto *ofproto_)
594 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
595 struct ofport_dpif *ofport;
596 struct ofbundle *bundle;
598 if (!clogged && !list_is_empty(&ofproto->completions)) {
599 poll_immediate_wake();
602 dpif_wait(ofproto->dpif);
603 dpif_recv_wait(ofproto->dpif);
604 if (ofproto->sflow) {
605 dpif_sflow_wait(ofproto->sflow);
607 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
608 poll_immediate_wake();
610 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
613 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
616 if (ofproto->need_revalidate) {
617 /* Shouldn't happen, but if it does just go around again. */
618 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
619 poll_immediate_wake();
621 timer_wait(&ofproto->next_expiration);
626 flush(struct ofproto *ofproto_)
628 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
629 struct facet *facet, *next_facet;
631 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
632 /* Mark the facet as not installed so that facet_remove() doesn't
633 * bother trying to uninstall it. There is no point in uninstalling it
634 * individually since we are about to blow away all the facets with
635 * dpif_flow_flush(). */
636 facet->installed = false;
637 facet->dp_packet_count = 0;
638 facet->dp_byte_count = 0;
639 facet_remove(ofproto, facet);
641 dpif_flow_flush(ofproto->dpif);
645 get_features(struct ofproto *ofproto_ OVS_UNUSED,
646 bool *arp_match_ip, uint32_t *actions)
648 *arp_match_ip = true;
649 *actions = ((1u << OFPAT_OUTPUT) |
650 (1u << OFPAT_SET_VLAN_VID) |
651 (1u << OFPAT_SET_VLAN_PCP) |
652 (1u << OFPAT_STRIP_VLAN) |
653 (1u << OFPAT_SET_DL_SRC) |
654 (1u << OFPAT_SET_DL_DST) |
655 (1u << OFPAT_SET_NW_SRC) |
656 (1u << OFPAT_SET_NW_DST) |
657 (1u << OFPAT_SET_NW_TOS) |
658 (1u << OFPAT_SET_TP_SRC) |
659 (1u << OFPAT_SET_TP_DST) |
660 (1u << OFPAT_ENQUEUE));
664 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
666 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
669 strcpy(ots->name, "classifier");
671 dpif_get_dp_stats(ofproto->dpif, &s);
672 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
673 put_32aligned_be64(&ots->matched_count,
674 htonll(s.n_hit + ofproto->n_matches));
678 set_netflow(struct ofproto *ofproto_,
679 const struct netflow_options *netflow_options)
681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
683 if (netflow_options) {
684 if (!ofproto->netflow) {
685 ofproto->netflow = netflow_create();
687 return netflow_set_options(ofproto->netflow, netflow_options);
689 netflow_destroy(ofproto->netflow);
690 ofproto->netflow = NULL;
695 static struct ofport *
698 struct ofport_dpif *port = xmalloc(sizeof *port);
703 port_dealloc(struct ofport *port_)
705 struct ofport_dpif *port = ofport_dpif_cast(port_);
710 port_construct(struct ofport *port_)
712 struct ofport_dpif *port = ofport_dpif_cast(port_);
713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
715 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
718 port->tag = tag_create_random();
719 port->may_enable = true;
721 if (ofproto->sflow) {
722 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
723 netdev_get_name(port->up.netdev));
730 port_destruct(struct ofport *port_)
732 struct ofport_dpif *port = ofport_dpif_cast(port_);
733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
735 bundle_remove(port_);
736 set_cfm(port_, NULL);
737 if (ofproto->sflow) {
738 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
743 port_modified(struct ofport *port_)
745 struct ofport_dpif *port = ofport_dpif_cast(port_);
747 if (port->bundle && port->bundle->bond) {
748 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
753 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
755 struct ofport_dpif *port = ofport_dpif_cast(port_);
756 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
757 ovs_be32 changed = old_config ^ port->up.opp.config;
759 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
760 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
761 ofproto->need_revalidate = true;
766 set_sflow(struct ofproto *ofproto_,
767 const struct ofproto_sflow_options *sflow_options)
769 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
770 struct dpif_sflow *ds = ofproto->sflow;
773 struct ofport_dpif *ofport;
775 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
776 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
777 dpif_sflow_add_port(ds, ofport->odp_port,
778 netdev_get_name(ofport->up.netdev));
781 dpif_sflow_set_options(ds, sflow_options);
783 dpif_sflow_destroy(ds);
784 ofproto->sflow = NULL;
790 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
792 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
799 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
802 if (cfm_configure(ofport->cfm, s)) {
808 cfm_destroy(ofport->cfm);
814 get_cfm_fault(const struct ofport *ofport_)
816 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
818 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
823 /* Expires all MAC learning entries associated with 'port' and forces ofproto
824 * to revalidate every flow. */
826 bundle_flush_macs(struct ofbundle *bundle)
828 struct ofproto_dpif *ofproto = bundle->ofproto;
829 struct mac_learning *ml = ofproto->ml;
830 struct mac_entry *mac, *next_mac;
832 ofproto->need_revalidate = true;
833 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
834 if (mac->port.p == bundle) {
835 mac_learning_expire(ml, mac);
840 static struct ofbundle *
841 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
843 struct ofbundle *bundle;
845 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
847 if (bundle->aux == aux) {
854 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
855 * ones that are found to 'bundles'. */
857 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
858 void **auxes, size_t n_auxes,
859 struct hmapx *bundles)
864 for (i = 0; i < n_auxes; i++) {
865 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
867 hmapx_add(bundles, bundle);
873 bundle_del_port(struct ofport_dpif *port)
875 struct ofbundle *bundle = port->bundle;
877 bundle->ofproto->need_revalidate = true;
879 list_remove(&port->bundle_node);
883 lacp_slave_unregister(bundle->lacp, port);
886 bond_slave_unregister(bundle->bond, port);
889 bundle->floodable = true;
890 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
891 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
892 bundle->floodable = false;
898 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
899 struct lacp_slave_settings *lacp,
900 uint32_t bond_stable_id)
902 struct ofport_dpif *port;
904 port = get_ofp_port(bundle->ofproto, ofp_port);
909 if (port->bundle != bundle) {
910 bundle->ofproto->need_revalidate = true;
912 bundle_del_port(port);
915 port->bundle = bundle;
916 list_push_back(&bundle->ports, &port->bundle_node);
917 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
918 bundle->floodable = false;
922 lacp_slave_register(bundle->lacp, port, lacp);
925 port->bond_stable_id = bond_stable_id;
931 bundle_destroy(struct ofbundle *bundle)
933 struct ofproto_dpif *ofproto;
934 struct ofport_dpif *port, *next_port;
941 ofproto = bundle->ofproto;
942 for (i = 0; i < MAX_MIRRORS; i++) {
943 struct ofmirror *m = ofproto->mirrors[i];
945 if (m->out == bundle) {
947 } else if (hmapx_find_and_delete(&m->srcs, bundle)
948 || hmapx_find_and_delete(&m->dsts, bundle)) {
949 ofproto->need_revalidate = true;
954 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
955 bundle_del_port(port);
958 bundle_flush_macs(bundle);
959 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
961 free(bundle->trunks);
962 lacp_destroy(bundle->lacp);
963 bond_destroy(bundle->bond);
968 bundle_set(struct ofproto *ofproto_, void *aux,
969 const struct ofproto_bundle_settings *s)
971 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
972 bool need_flush = false;
973 const unsigned long *trunks;
974 struct ofport_dpif *port;
975 struct ofbundle *bundle;
980 bundle_destroy(bundle_lookup(ofproto, aux));
984 assert(s->n_slaves == 1 || s->bond != NULL);
985 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
987 bundle = bundle_lookup(ofproto, aux);
989 bundle = xmalloc(sizeof *bundle);
991 bundle->ofproto = ofproto;
992 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
993 hash_pointer(aux, 0));
997 list_init(&bundle->ports);
999 bundle->trunks = NULL;
1000 bundle->lacp = NULL;
1001 bundle->bond = NULL;
1003 bundle->floodable = true;
1005 bundle->src_mirrors = 0;
1006 bundle->dst_mirrors = 0;
1007 bundle->mirror_out = 0;
1010 if (!bundle->name || strcmp(s->name, bundle->name)) {
1012 bundle->name = xstrdup(s->name);
1017 if (!bundle->lacp) {
1018 bundle->lacp = lacp_create();
1020 lacp_configure(bundle->lacp, s->lacp);
1022 lacp_destroy(bundle->lacp);
1023 bundle->lacp = NULL;
1026 /* Update set of ports. */
1028 for (i = 0; i < s->n_slaves; i++) {
1029 if (!bundle_add_port(bundle, s->slaves[i],
1030 s->lacp ? &s->lacp_slaves[i] : NULL,
1031 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1035 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1036 struct ofport_dpif *next_port;
1038 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1039 for (i = 0; i < s->n_slaves; i++) {
1040 if (s->slaves[i] == port->up.ofp_port) {
1045 bundle_del_port(port);
1049 assert(list_size(&bundle->ports) <= s->n_slaves);
1051 if (list_is_empty(&bundle->ports)) {
1052 bundle_destroy(bundle);
1057 if (s->vlan != bundle->vlan) {
1058 bundle->vlan = s->vlan;
1062 /* Get trunked VLANs. */
1063 trunks = s->vlan == -1 ? NULL : s->trunks;
1064 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1065 free(bundle->trunks);
1066 bundle->trunks = vlan_bitmap_clone(trunks);
1071 if (!list_is_short(&bundle->ports)) {
1072 bundle->ofproto->has_bonded_bundles = true;
1074 if (bond_reconfigure(bundle->bond, s->bond)) {
1075 ofproto->need_revalidate = true;
1078 bundle->bond = bond_create(s->bond);
1079 ofproto->need_revalidate = true;
1082 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1083 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1087 bond_destroy(bundle->bond);
1088 bundle->bond = NULL;
1091 /* If we changed something that would affect MAC learning, un-learn
1092 * everything on this port and force flow revalidation. */
1094 bundle_flush_macs(bundle);
1101 bundle_remove(struct ofport *port_)
1103 struct ofport_dpif *port = ofport_dpif_cast(port_);
1104 struct ofbundle *bundle = port->bundle;
1107 bundle_del_port(port);
1108 if (list_is_empty(&bundle->ports)) {
1109 bundle_destroy(bundle);
1110 } else if (list_is_short(&bundle->ports)) {
1111 bond_destroy(bundle->bond);
1112 bundle->bond = NULL;
1118 send_pdu_cb(void *port_, const struct lacp_pdu *pdu)
1120 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1121 struct ofport_dpif *port = port_;
1122 uint8_t ea[ETH_ADDR_LEN];
1125 error = netdev_get_etheraddr(port->up.netdev, ea);
1127 struct lacp_pdu *packet_pdu;
1128 struct ofpbuf packet;
1130 ofpbuf_init(&packet, 0);
1131 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1132 sizeof *packet_pdu);
1134 error = netdev_send(port->up.netdev, &packet);
1136 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1137 "(%s)", port->bundle->name,
1138 netdev_get_name(port->up.netdev), strerror(error));
1140 ofpbuf_uninit(&packet);
1142 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1143 "%s (%s)", port->bundle->name,
1144 netdev_get_name(port->up.netdev), strerror(error));
1149 bundle_send_learning_packets(struct ofbundle *bundle)
1151 struct ofproto_dpif *ofproto = bundle->ofproto;
1152 int error, n_packets, n_errors;
1153 struct mac_entry *e;
1155 error = n_packets = n_errors = 0;
1156 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1157 if (e->port.p != bundle) {
1158 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1168 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1169 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1170 "packets, last error was: %s",
1171 bundle->name, n_errors, n_packets, strerror(error));
1173 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1174 bundle->name, n_packets);
1179 bundle_run(struct ofbundle *bundle)
1182 lacp_run(bundle->lacp, send_pdu_cb);
1185 struct ofport_dpif *port;
1187 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1188 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1191 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1192 lacp_negotiated(bundle->lacp));
1193 if (bond_should_send_learning_packets(bundle->bond)) {
1194 bundle_send_learning_packets(bundle);
1200 bundle_wait(struct ofbundle *bundle)
1203 lacp_wait(bundle->lacp);
1206 bond_wait(bundle->bond);
1213 mirror_scan(struct ofproto_dpif *ofproto)
1217 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1218 if (!ofproto->mirrors[idx]) {
1225 static struct ofmirror *
1226 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1230 for (i = 0; i < MAX_MIRRORS; i++) {
1231 struct ofmirror *mirror = ofproto->mirrors[i];
1232 if (mirror && mirror->aux == aux) {
1241 mirror_set(struct ofproto *ofproto_, void *aux,
1242 const struct ofproto_mirror_settings *s)
1244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1245 mirror_mask_t mirror_bit;
1246 struct ofbundle *bundle;
1247 struct ofmirror *mirror;
1248 struct ofbundle *out;
1249 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1250 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1253 mirror = mirror_lookup(ofproto, aux);
1255 mirror_destroy(mirror);
1261 idx = mirror_scan(ofproto);
1263 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1265 ofproto->up.name, MAX_MIRRORS, s->name);
1269 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1270 mirror->ofproto = ofproto;
1273 mirror->out_vlan = -1;
1274 mirror->name = NULL;
1277 if (!mirror->name || strcmp(s->name, mirror->name)) {
1279 mirror->name = xstrdup(s->name);
1282 /* Get the new configuration. */
1283 if (s->out_bundle) {
1284 out = bundle_lookup(ofproto, s->out_bundle);
1286 mirror_destroy(mirror);
1292 out_vlan = s->out_vlan;
1294 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1295 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1297 /* If the configuration has not changed, do nothing. */
1298 if (hmapx_equals(&srcs, &mirror->srcs)
1299 && hmapx_equals(&dsts, &mirror->dsts)
1300 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1301 && mirror->out == out
1302 && mirror->out_vlan == out_vlan)
1304 hmapx_destroy(&srcs);
1305 hmapx_destroy(&dsts);
1309 hmapx_swap(&srcs, &mirror->srcs);
1310 hmapx_destroy(&srcs);
1312 hmapx_swap(&dsts, &mirror->dsts);
1313 hmapx_destroy(&dsts);
1315 free(mirror->vlans);
1316 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1319 mirror->out_vlan = out_vlan;
1321 /* Update bundles. */
1322 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1323 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1324 if (hmapx_contains(&mirror->srcs, bundle)) {
1325 bundle->src_mirrors |= mirror_bit;
1327 bundle->src_mirrors &= ~mirror_bit;
1330 if (hmapx_contains(&mirror->dsts, bundle)) {
1331 bundle->dst_mirrors |= mirror_bit;
1333 bundle->dst_mirrors &= ~mirror_bit;
1336 if (mirror->out == bundle) {
1337 bundle->mirror_out |= mirror_bit;
1339 bundle->mirror_out &= ~mirror_bit;
1343 ofproto->need_revalidate = true;
1344 mac_learning_flush(ofproto->ml);
1350 mirror_destroy(struct ofmirror *mirror)
1352 struct ofproto_dpif *ofproto;
1353 mirror_mask_t mirror_bit;
1354 struct ofbundle *bundle;
1360 ofproto = mirror->ofproto;
1361 ofproto->need_revalidate = true;
1362 mac_learning_flush(ofproto->ml);
1364 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1365 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1366 bundle->src_mirrors &= ~mirror_bit;
1367 bundle->dst_mirrors &= ~mirror_bit;
1368 bundle->mirror_out &= ~mirror_bit;
1371 hmapx_destroy(&mirror->srcs);
1372 hmapx_destroy(&mirror->dsts);
1373 free(mirror->vlans);
1375 ofproto->mirrors[mirror->idx] = NULL;
1381 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1383 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1384 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1385 ofproto->need_revalidate = true;
1386 mac_learning_flush(ofproto->ml);
1392 is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
1394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1395 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1396 return bundle && bundle->mirror_out != 0;
1400 forward_bpdu_changed(struct ofproto *ofproto_)
1402 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1403 /* Revalidate cached flows whenever forward_bpdu option changes. */
1404 ofproto->need_revalidate = true;
1409 static struct ofport_dpif *
1410 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1412 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1413 return ofport ? ofport_dpif_cast(ofport) : NULL;
1416 static struct ofport_dpif *
1417 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1419 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1423 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1424 struct dpif_port *dpif_port)
1426 ofproto_port->name = dpif_port->name;
1427 ofproto_port->type = dpif_port->type;
1428 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1432 port_run(struct ofport_dpif *ofport)
1434 bool enable = netdev_get_carrier(ofport->up.netdev);
1437 cfm_run(ofport->cfm);
1439 if (cfm_should_send_ccm(ofport->cfm)) {
1440 struct ofpbuf packet;
1442 ofpbuf_init(&packet, 0);
1443 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1444 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1445 ofport->odp_port, &packet);
1446 ofpbuf_uninit(&packet);
1449 enable = enable && !cfm_get_fault(ofport->cfm);
1452 if (ofport->bundle) {
1453 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1456 if (ofport->may_enable != enable) {
1457 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1459 if (ofproto->has_bundle_action) {
1460 ofproto->need_revalidate = true;
1464 ofport->may_enable = enable;
1468 port_wait(struct ofport_dpif *ofport)
1471 cfm_wait(ofport->cfm);
1476 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1477 struct ofproto_port *ofproto_port)
1479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1480 struct dpif_port dpif_port;
1483 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1485 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1491 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1493 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1497 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1499 *ofp_portp = odp_port_to_ofp_port(odp_port);
1505 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1507 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1510 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1512 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1514 /* The caller is going to close ofport->up.netdev. If this is a
1515 * bonded port, then the bond is using that netdev, so remove it
1516 * from the bond. The client will need to reconfigure everything
1517 * after deleting ports, so then the slave will get re-added. */
1518 bundle_remove(&ofport->up);
1524 struct port_dump_state {
1525 struct dpif_port_dump dump;
1530 port_dump_start(const struct ofproto *ofproto_, void **statep)
1532 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1533 struct port_dump_state *state;
1535 *statep = state = xmalloc(sizeof *state);
1536 dpif_port_dump_start(&state->dump, ofproto->dpif);
1537 state->done = false;
1542 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1543 struct ofproto_port *port)
1545 struct port_dump_state *state = state_;
1546 struct dpif_port dpif_port;
1548 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1549 ofproto_port_from_dpif_port(port, &dpif_port);
1552 int error = dpif_port_dump_done(&state->dump);
1554 return error ? error : EOF;
1559 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1561 struct port_dump_state *state = state_;
1564 dpif_port_dump_done(&state->dump);
1571 port_poll(const struct ofproto *ofproto_, char **devnamep)
1573 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1574 return dpif_port_poll(ofproto->dpif, devnamep);
1578 port_poll_wait(const struct ofproto *ofproto_)
1580 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1581 dpif_port_poll_wait(ofproto->dpif);
1585 port_is_lacp_current(const struct ofport *ofport_)
1587 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1588 return (ofport->bundle && ofport->bundle->lacp
1589 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
1593 /* Upcall handling. */
1595 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
1596 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
1597 * their individual configurations.
1599 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
1600 * Otherwise, ownership is transferred to this function. */
1602 send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
1603 const struct flow *flow, bool clone)
1605 struct ofputil_packet_in pin;
1607 pin.packet = upcall->packet;
1608 pin.in_port = flow->in_port;
1609 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
1610 pin.buffer_id = 0; /* not yet known */
1611 pin.send_len = upcall->userdata;
1612 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
1613 clone ? NULL : upcall->packet);
1617 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
1618 const struct ofpbuf *packet)
1620 if (cfm_should_process_flow(flow)) {
1621 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
1622 if (packet && ofport && ofport->cfm) {
1623 cfm_process_heartbeat(ofport->cfm, packet);
1626 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
1627 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
1628 if (packet && port && port->bundle && port->bundle->lacp) {
1629 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
1631 lacp_process_pdu(port->bundle->lacp, port, pdu);
1640 handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1642 struct facet *facet;
1645 /* Obtain in_port and tun_id, at least. */
1646 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1648 /* Set header pointers in 'flow'. */
1649 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
1651 /* Handle 802.1ag and LACP. */
1652 if (process_special(ofproto, &flow, upcall->packet)) {
1653 ofpbuf_delete(upcall->packet);
1654 ofproto->n_matches++;
1658 /* Check with in-band control to see if this packet should be sent
1659 * to the local port regardless of the flow table. */
1660 if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
1661 send_packet(ofproto, ODPP_LOCAL, upcall->packet);
1664 facet = facet_lookup_valid(ofproto, &flow);
1666 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow, 0);
1668 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
1669 struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
1671 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
1672 COVERAGE_INC(ofproto_dpif_no_packet_in);
1673 /* XXX install 'drop' flow entry */
1674 ofpbuf_delete(upcall->packet);
1678 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
1682 send_packet_in(ofproto, upcall, &flow, false);
1686 facet = facet_create(rule, &flow, upcall->packet);
1687 } else if (!facet->may_install) {
1688 /* The facet is not installable, that is, we need to process every
1689 * packet, so process the current packet's actions into 'facet'. */
1690 facet_make_actions(ofproto, facet, upcall->packet);
1693 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
1695 * Extra-special case for fail-open mode.
1697 * We are in fail-open mode and the packet matched the fail-open rule,
1698 * but we are connected to a controller too. We should send the packet
1699 * up to the controller in the hope that it will try to set up a flow
1700 * and thereby allow us to exit fail-open.
1702 * See the top-level comment in fail-open.c for more information.
1704 send_packet_in(ofproto, upcall, &flow, true);
1707 facet_execute(ofproto, facet, upcall->packet);
1708 facet_install(ofproto, facet, false);
1709 ofproto->n_matches++;
1713 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1717 switch (upcall->type) {
1718 case DPIF_UC_ACTION:
1719 COVERAGE_INC(ofproto_dpif_ctlr_action);
1720 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1721 send_packet_in(ofproto, upcall, &flow, false);
1724 case DPIF_UC_SAMPLE:
1725 if (ofproto->sflow) {
1726 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1727 dpif_sflow_received(ofproto->sflow, upcall, &flow);
1729 ofpbuf_delete(upcall->packet);
1733 handle_miss_upcall(ofproto, upcall);
1736 case DPIF_N_UC_TYPES:
1738 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
1743 /* Flow expiration. */
1745 static int facet_max_idle(const struct ofproto_dpif *);
1746 static void update_stats(struct ofproto_dpif *);
1747 static void rule_expire(struct rule_dpif *);
1748 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
1750 /* This function is called periodically by run(). Its job is to collect
1751 * updates for the flows that have been installed into the datapath, most
1752 * importantly when they last were used, and then use that information to
1753 * expire flows that have not been used recently.
1755 * Returns the number of milliseconds after which it should be called again. */
1757 expire(struct ofproto_dpif *ofproto)
1759 struct rule_dpif *rule, *next_rule;
1760 struct classifier *table;
1763 /* Update stats for each flow in the datapath. */
1764 update_stats(ofproto);
1766 /* Expire facets that have been idle too long. */
1767 dp_max_idle = facet_max_idle(ofproto);
1768 expire_facets(ofproto, dp_max_idle);
1770 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
1771 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1772 struct cls_cursor cursor;
1774 cls_cursor_init(&cursor, table, NULL);
1775 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1780 /* All outstanding data in existing flows has been accounted, so it's a
1781 * good time to do bond rebalancing. */
1782 if (ofproto->has_bonded_bundles) {
1783 struct ofbundle *bundle;
1785 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1787 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
1792 return MIN(dp_max_idle, 1000);
1795 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
1797 * This function also pushes statistics updates to rules which each facet
1798 * resubmits into. Generally these statistics will be accurate. However, if a
1799 * facet changes the rule it resubmits into at some time in between
1800 * update_stats() runs, it is possible that statistics accrued to the
1801 * old rule will be incorrectly attributed to the new rule. This could be
1802 * avoided by calling update_stats() whenever rules are created or
1803 * deleted. However, the performance impact of making so many calls to the
1804 * datapath do not justify the benefit of having perfectly accurate statistics.
1807 update_stats(struct ofproto_dpif *p)
1809 const struct dpif_flow_stats *stats;
1810 struct dpif_flow_dump dump;
1811 const struct nlattr *key;
1814 dpif_flow_dump_start(&dump, p->dpif);
1815 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
1816 struct facet *facet;
1819 if (odp_flow_key_to_flow(key, key_len, &flow)) {
1823 odp_flow_key_format(key, key_len, &s);
1824 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
1830 facet = facet_find(p, &flow);
1832 if (facet && facet->installed) {
1834 if (stats->n_packets >= facet->dp_packet_count) {
1835 uint64_t extra = stats->n_packets - facet->dp_packet_count;
1836 facet->packet_count += extra;
1838 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
1841 if (stats->n_bytes >= facet->dp_byte_count) {
1842 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
1844 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
1847 facet->dp_packet_count = stats->n_packets;
1848 facet->dp_byte_count = stats->n_bytes;
1850 facet_update_time(p, facet, stats->used);
1851 facet_account(p, facet, stats->n_bytes);
1852 facet_push_stats(facet);
1854 /* There's a flow in the datapath that we know nothing about.
1856 COVERAGE_INC(facet_unexpected);
1857 dpif_flow_del(p->dpif, key, key_len, NULL);
1860 dpif_flow_dump_done(&dump);
1863 /* Calculates and returns the number of milliseconds of idle time after which
1864 * facets should expire from the datapath and we should fold their statistics
1865 * into their parent rules in userspace. */
1867 facet_max_idle(const struct ofproto_dpif *ofproto)
1870 * Idle time histogram.
1872 * Most of the time a switch has a relatively small number of facets. When
1873 * this is the case we might as well keep statistics for all of them in
1874 * userspace and to cache them in the kernel datapath for performance as
1877 * As the number of facets increases, the memory required to maintain
1878 * statistics about them in userspace and in the kernel becomes
1879 * significant. However, with a large number of facets it is likely that
1880 * only a few of them are "heavy hitters" that consume a large amount of
1881 * bandwidth. At this point, only heavy hitters are worth caching in the
1882 * kernel and maintaining in userspaces; other facets we can discard.
1884 * The technique used to compute the idle time is to build a histogram with
1885 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
1886 * that is installed in the kernel gets dropped in the appropriate bucket.
1887 * After the histogram has been built, we compute the cutoff so that only
1888 * the most-recently-used 1% of facets (but at least
1889 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
1890 * the most-recently-used bucket of facets is kept, so actually an
1891 * arbitrary number of facets can be kept in any given expiration run
1892 * (though the next run will delete most of those unless they receive
1895 * This requires a second pass through the facets, in addition to the pass
1896 * made by update_stats(), because the former function never looks
1897 * at uninstallable facets.
1899 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
1900 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
1901 int buckets[N_BUCKETS] = { 0 };
1902 int total, subtotal, bucket;
1903 struct facet *facet;
1907 total = hmap_count(&ofproto->facets);
1908 if (total <= ofproto->up.flow_eviction_threshold) {
1909 return N_BUCKETS * BUCKET_WIDTH;
1912 /* Build histogram. */
1914 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1915 long long int idle = now - facet->used;
1916 int bucket = (idle <= 0 ? 0
1917 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
1918 : (unsigned int) idle / BUCKET_WIDTH);
1922 /* Find the first bucket whose flows should be expired. */
1923 subtotal = bucket = 0;
1925 subtotal += buckets[bucket++];
1926 } while (bucket < N_BUCKETS &&
1927 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
1929 if (VLOG_IS_DBG_ENABLED()) {
1933 ds_put_cstr(&s, "keep");
1934 for (i = 0; i < N_BUCKETS; i++) {
1936 ds_put_cstr(&s, ", drop");
1939 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
1942 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
1946 return bucket * BUCKET_WIDTH;
1950 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
1952 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
1953 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
1954 struct ofexpired expired;
1956 if (facet->installed) {
1957 struct dpif_flow_stats stats;
1959 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
1961 facet_update_stats(ofproto, facet, &stats);
1964 expired.flow = facet->flow;
1965 expired.packet_count = facet->packet_count;
1966 expired.byte_count = facet->byte_count;
1967 expired.used = facet->used;
1968 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1973 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
1975 long long int cutoff = time_msec() - dp_max_idle;
1976 struct facet *facet, *next_facet;
1978 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1979 facet_active_timeout(ofproto, facet);
1980 if (facet->used < cutoff) {
1981 facet_remove(ofproto, facet);
1986 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
1987 * then delete it entirely. */
1989 rule_expire(struct rule_dpif *rule)
1991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
1992 struct facet *facet, *next_facet;
1996 /* Has 'rule' expired? */
1998 if (rule->up.hard_timeout
1999 && now > rule->up.created + rule->up.hard_timeout * 1000) {
2000 reason = OFPRR_HARD_TIMEOUT;
2001 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2002 && now > rule->used + rule->up.idle_timeout * 1000) {
2003 reason = OFPRR_IDLE_TIMEOUT;
2008 COVERAGE_INC(ofproto_dpif_expired);
2010 /* Update stats. (This is a no-op if the rule expired due to an idle
2011 * timeout, because that only happens when the rule has no facets left.) */
2012 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2013 facet_remove(ofproto, facet);
2016 /* Get rid of the rule. */
2017 ofproto_rule_expire(&rule->up, reason);
2022 /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
2023 * example 'packet' within that flow.
2025 * The caller must already have determined that no facet with an identical
2026 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2027 * the ofproto's classifier table. */
2028 static struct facet *
2029 facet_create(struct rule_dpif *rule, const struct flow *flow,
2030 const struct ofpbuf *packet)
2032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2033 struct facet *facet;
2035 facet = xzalloc(sizeof *facet);
2036 facet->used = time_msec();
2037 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2038 list_push_back(&rule->facets, &facet->list_node);
2040 facet->flow = *flow;
2041 netflow_flow_init(&facet->nf_flow);
2042 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2044 facet_make_actions(ofproto, facet, packet);
2050 facet_free(struct facet *facet)
2052 free(facet->actions);
2056 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2057 * 'packet', which arrived on 'in_port'.
2059 * Takes ownership of 'packet'. */
2061 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2062 const struct nlattr *odp_actions, size_t actions_len,
2063 struct ofpbuf *packet)
2065 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2066 && odp_actions->nla_type == ODP_ACTION_ATTR_USERSPACE) {
2067 /* As an optimization, avoid a round-trip from userspace to kernel to
2068 * userspace. This also avoids possibly filling up kernel packet
2069 * buffers along the way. */
2070 struct dpif_upcall upcall;
2072 upcall.type = DPIF_UC_ACTION;
2073 upcall.packet = packet;
2076 upcall.userdata = nl_attr_get_u64(odp_actions);
2077 upcall.sample_pool = 0;
2078 upcall.actions = NULL;
2079 upcall.actions_len = 0;
2081 send_packet_in(ofproto, &upcall, flow, false);
2085 struct odputil_keybuf keybuf;
2089 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2090 odp_flow_key_from_flow(&key, flow);
2092 error = dpif_execute(ofproto->dpif, key.data, key.size,
2093 odp_actions, actions_len, packet);
2095 ofpbuf_delete(packet);
2100 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2101 * statistics appropriately. 'packet' must have at least sizeof(struct
2102 * ofp_packet_in) bytes of headroom.
2104 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2105 * applying flow_extract() to 'packet' would yield the same flow as
2108 * 'facet' must have accurately composed ODP actions; that is, it must not be
2109 * in need of revalidation.
2111 * Takes ownership of 'packet'. */
2113 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2114 struct ofpbuf *packet)
2116 struct dpif_flow_stats stats;
2118 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2120 flow_extract_stats(&facet->flow, packet, &stats);
2121 stats.used = time_msec();
2122 if (execute_odp_actions(ofproto, &facet->flow,
2123 facet->actions, facet->actions_len, packet)) {
2124 facet_update_stats(ofproto, facet, &stats);
2128 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2130 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2131 * rule's statistics, via facet_uninstall().
2133 * - Removes 'facet' from its rule and from ofproto->facets.
2136 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2138 facet_uninstall(ofproto, facet);
2139 facet_flush_stats(ofproto, facet);
2140 hmap_remove(&ofproto->facets, &facet->hmap_node);
2141 list_remove(&facet->list_node);
2145 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2147 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2148 const struct ofpbuf *packet)
2150 const struct rule_dpif *rule = facet->rule;
2151 struct ofpbuf *odp_actions;
2152 struct action_xlate_ctx ctx;
2154 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2155 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2156 facet->tags = ctx.tags;
2157 facet->may_install = ctx.may_set_up_flow;
2158 facet->nf_flow.output_iface = ctx.nf_output_iface;
2160 if (facet->actions_len != odp_actions->size
2161 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2162 free(facet->actions);
2163 facet->actions_len = odp_actions->size;
2164 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2167 ofpbuf_delete(odp_actions);
2170 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2171 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2172 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2173 * since 'facet' was last updated.
2175 * Returns 0 if successful, otherwise a positive errno value.*/
2177 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2178 const struct nlattr *actions, size_t actions_len,
2179 struct dpif_flow_stats *stats)
2181 struct odputil_keybuf keybuf;
2182 enum dpif_flow_put_flags flags;
2186 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2188 flags |= DPIF_FP_ZERO_STATS;
2191 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2192 odp_flow_key_from_flow(&key, &facet->flow);
2194 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2195 actions, actions_len, stats);
2198 facet_reset_dp_stats(facet, stats);
2204 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2205 * 'zero_stats' is true, clears any existing statistics from the datapath for
2208 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2210 struct dpif_flow_stats stats;
2212 if (facet->may_install
2213 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2214 zero_stats ? &stats : NULL)) {
2215 facet->installed = true;
2220 vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci)
2222 return vlan_tci != htons(0) ? vlan_tci_to_vid(vlan_tci) : OFP_VLAN_NONE;
2226 facet_account(struct ofproto_dpif *ofproto,
2227 struct facet *facet, uint64_t extra_bytes)
2229 uint64_t total_bytes, n_bytes;
2230 struct ofbundle *in_bundle;
2231 const struct nlattr *a;
2237 total_bytes = facet->byte_count + extra_bytes;
2238 if (total_bytes <= facet->accounted_bytes) {
2241 n_bytes = total_bytes - facet->accounted_bytes;
2242 facet->accounted_bytes = total_bytes;
2244 /* Test that 'tags' is nonzero to ensure that only flows that include an
2245 * OFPP_NORMAL action are used for learning and bond slave rebalancing.
2246 * This works because OFPP_NORMAL always sets a nonzero tag value.
2248 * Feed information from the active flows back into the learning table to
2249 * ensure that table is always in sync with what is actually flowing
2250 * through the datapath. */
2252 || !is_admissible(ofproto, &facet->flow, false, &dummy,
2253 &vlan, &in_bundle)) {
2257 update_learning_table(ofproto, &facet->flow, vlan, in_bundle);
2259 if (!ofproto->has_bonded_bundles) {
2263 /* This loop feeds byte counters to bond_account() for rebalancing to use
2264 * as a basis. We also need to track the actual VLAN on which the packet
2265 * is going to be sent to ensure that it matches the one passed to
2266 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2268 vlan_tci = facet->flow.vlan_tci;
2269 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2270 struct ofport_dpif *port;
2272 switch (nl_attr_type(a)) {
2273 case ODP_ACTION_ATTR_OUTPUT:
2274 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2275 if (port && port->bundle && port->bundle->bond) {
2276 bond_account(port->bundle->bond, &facet->flow,
2277 vlan_tci_to_openflow_vlan(vlan_tci), n_bytes);
2281 case ODP_ACTION_ATTR_STRIP_VLAN:
2282 vlan_tci = htons(0);
2285 case ODP_ACTION_ATTR_SET_DL_TCI:
2286 vlan_tci = nl_attr_get_be16(a);
2292 /* If 'rule' is installed in the datapath, uninstalls it. */
2294 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2296 if (facet->installed) {
2297 struct odputil_keybuf keybuf;
2298 struct dpif_flow_stats stats;
2302 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2303 odp_flow_key_from_flow(&key, &facet->flow);
2305 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2306 facet_reset_dp_stats(facet, &stats);
2308 facet_update_stats(p, facet, &stats);
2310 facet->installed = false;
2312 assert(facet->dp_packet_count == 0);
2313 assert(facet->dp_byte_count == 0);
2317 /* Returns true if the only action for 'facet' is to send to the controller.
2318 * (We don't report NetFlow expiration messages for such facets because they
2319 * are just part of the control logic for the network, not real traffic). */
2321 facet_is_controller_flow(struct facet *facet)
2324 && facet->rule->up.n_actions == 1
2325 && action_outputs_to_port(&facet->rule->up.actions[0],
2326 htons(OFPP_CONTROLLER)));
2329 /* Resets 'facet''s datapath statistics counters. This should be called when
2330 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2331 * it should contain the statistics returned by dpif when 'facet' was reset in
2332 * the datapath. 'stats' will be modified to only included statistics new
2333 * since 'facet' was last updated. */
2335 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2337 if (stats && facet->dp_packet_count <= stats->n_packets
2338 && facet->dp_byte_count <= stats->n_bytes) {
2339 stats->n_packets -= facet->dp_packet_count;
2340 stats->n_bytes -= facet->dp_byte_count;
2343 facet->dp_packet_count = 0;
2344 facet->dp_byte_count = 0;
2347 /* Folds all of 'facet''s statistics into its rule. Also updates the
2348 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2349 * 'facet''s statistics in the datapath should have been zeroed and folded into
2350 * its packet and byte counts before this function is called. */
2352 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2354 assert(!facet->dp_byte_count);
2355 assert(!facet->dp_packet_count);
2357 facet_push_stats(facet);
2358 facet_account(ofproto, facet, 0);
2360 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2361 struct ofexpired expired;
2362 expired.flow = facet->flow;
2363 expired.packet_count = facet->packet_count;
2364 expired.byte_count = facet->byte_count;
2365 expired.used = facet->used;
2366 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2369 facet->rule->packet_count += facet->packet_count;
2370 facet->rule->byte_count += facet->byte_count;
2372 /* Reset counters to prevent double counting if 'facet' ever gets
2374 facet->packet_count = 0;
2375 facet->byte_count = 0;
2376 facet->rs_packet_count = 0;
2377 facet->rs_byte_count = 0;
2378 facet->accounted_bytes = 0;
2380 netflow_flow_clear(&facet->nf_flow);
2383 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2384 * Returns it if found, otherwise a null pointer.
2386 * The returned facet might need revalidation; use facet_lookup_valid()
2387 * instead if that is important. */
2388 static struct facet *
2389 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
2391 struct facet *facet;
2393 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2395 if (flow_equal(flow, &facet->flow)) {
2403 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2404 * Returns it if found, otherwise a null pointer.
2406 * The returned facet is guaranteed to be valid. */
2407 static struct facet *
2408 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
2410 struct facet *facet = facet_find(ofproto, flow);
2412 /* The facet we found might not be valid, since we could be in need of
2413 * revalidation. If it is not valid, don't return it. */
2415 && ofproto->need_revalidate
2416 && !facet_revalidate(ofproto, facet)) {
2417 COVERAGE_INC(facet_invalidated);
2424 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2426 * - If the rule found is different from 'facet''s current rule, moves
2427 * 'facet' to the new rule and recompiles its actions.
2429 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2430 * where it is and recompiles its actions anyway.
2432 * - If there is none, destroys 'facet'.
2434 * Returns true if 'facet' still exists, false if it has been destroyed. */
2436 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
2438 struct action_xlate_ctx ctx;
2439 struct ofpbuf *odp_actions;
2440 struct rule_dpif *new_rule;
2441 bool actions_changed;
2443 COVERAGE_INC(facet_revalidate);
2445 /* Determine the new rule. */
2446 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
2448 /* No new rule, so delete the facet. */
2449 facet_remove(ofproto, facet);
2453 /* Calculate new ODP actions.
2455 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2456 * emit a NetFlow expiration and, if so, we need to have the old state
2457 * around to properly compose it. */
2458 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2459 odp_actions = xlate_actions(&ctx,
2460 new_rule->up.actions, new_rule->up.n_actions);
2461 actions_changed = (facet->actions_len != odp_actions->size
2462 || memcmp(facet->actions, odp_actions->data,
2463 facet->actions_len));
2465 /* If the ODP actions changed or the installability changed, then we need
2466 * to talk to the datapath. */
2467 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2468 if (ctx.may_set_up_flow) {
2469 struct dpif_flow_stats stats;
2471 facet_put__(ofproto, facet,
2472 odp_actions->data, odp_actions->size, &stats);
2473 facet_update_stats(ofproto, facet, &stats);
2475 facet_uninstall(ofproto, facet);
2478 /* The datapath flow is gone or has zeroed stats, so push stats out of
2479 * 'facet' into 'rule'. */
2480 facet_flush_stats(ofproto, facet);
2483 /* Update 'facet' now that we've taken care of all the old state. */
2484 facet->tags = ctx.tags;
2485 facet->nf_flow.output_iface = ctx.nf_output_iface;
2486 facet->may_install = ctx.may_set_up_flow;
2487 if (actions_changed) {
2488 free(facet->actions);
2489 facet->actions_len = odp_actions->size;
2490 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2492 if (facet->rule != new_rule) {
2493 COVERAGE_INC(facet_changed_rule);
2494 list_remove(&facet->list_node);
2495 list_push_back(&new_rule->facets, &facet->list_node);
2496 facet->rule = new_rule;
2497 facet->used = new_rule->up.created;
2498 facet->rs_used = facet->used;
2501 ofpbuf_delete(odp_actions);
2506 /* Updates 'facet''s used time. Caller is responsible for calling
2507 * facet_push_stats() to update the flows which 'facet' resubmits into. */
2509 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
2512 if (used > facet->used) {
2514 if (used > facet->rule->used) {
2515 facet->rule->used = used;
2517 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
2521 /* Folds the statistics from 'stats' into the counters in 'facet'.
2523 * Because of the meaning of a facet's counters, it only makes sense to do this
2524 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
2525 * packet that was sent by hand or if it represents statistics that have been
2526 * cleared out of the datapath. */
2528 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
2529 const struct dpif_flow_stats *stats)
2531 if (stats->n_packets || stats->used > facet->used) {
2532 facet_update_time(ofproto, facet, stats->used);
2533 facet->packet_count += stats->n_packets;
2534 facet->byte_count += stats->n_bytes;
2535 facet_push_stats(facet);
2536 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
2541 facet_push_stats(struct facet *facet)
2543 uint64_t rs_packets, rs_bytes;
2545 assert(facet->packet_count >= facet->rs_packet_count);
2546 assert(facet->byte_count >= facet->rs_byte_count);
2547 assert(facet->used >= facet->rs_used);
2549 rs_packets = facet->packet_count - facet->rs_packet_count;
2550 rs_bytes = facet->byte_count - facet->rs_byte_count;
2552 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
2553 facet->rs_packet_count = facet->packet_count;
2554 facet->rs_byte_count = facet->byte_count;
2555 facet->rs_used = facet->used;
2557 flow_push_stats(facet->rule, &facet->flow,
2558 rs_packets, rs_bytes, facet->used);
2562 struct ofproto_push {
2563 struct action_xlate_ctx ctx;
2570 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
2572 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
2575 rule->packet_count += push->packets;
2576 rule->byte_count += push->bytes;
2577 rule->used = MAX(push->used, rule->used);
2581 /* Pushes flow statistics to the rules which 'flow' resubmits into given
2582 * 'rule''s actions. */
2584 flow_push_stats(const struct rule_dpif *rule,
2585 struct flow *flow, uint64_t packets, uint64_t bytes,
2588 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2589 struct ofproto_push push;
2591 push.packets = packets;
2595 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
2596 push.ctx.resubmit_hook = push_resubmit;
2597 ofpbuf_delete(xlate_actions(&push.ctx,
2598 rule->up.actions, rule->up.n_actions));
2603 static struct rule_dpif *
2604 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
2607 return rule_dpif_cast(rule_from_cls_rule(
2608 classifier_lookup(&ofproto->up.tables[table_id],
2613 complete_operation(struct rule_dpif *rule)
2615 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2617 ofproto->need_revalidate = true;
2619 struct dpif_completion *c = xmalloc(sizeof *c);
2620 c->op = rule->up.pending;
2621 list_push_back(&ofproto->completions, &c->list_node);
2623 ofoperation_complete(rule->up.pending, 0);
2627 static struct rule *
2630 struct rule_dpif *rule = xmalloc(sizeof *rule);
2635 rule_dealloc(struct rule *rule_)
2637 struct rule_dpif *rule = rule_dpif_cast(rule_);
2642 rule_construct(struct rule *rule_)
2644 struct rule_dpif *rule = rule_dpif_cast(rule_);
2645 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2646 struct rule_dpif *victim;
2649 error = validate_actions(rule->up.actions, rule->up.n_actions,
2650 &rule->up.cr.flow, ofproto->max_ports);
2655 rule->used = rule->up.created;
2656 rule->packet_count = 0;
2657 rule->byte_count = 0;
2659 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
2660 if (victim && !list_is_empty(&victim->facets)) {
2661 struct facet *facet;
2663 rule->facets = victim->facets;
2664 list_moved(&rule->facets);
2665 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2669 /* Must avoid list_moved() in this case. */
2670 list_init(&rule->facets);
2673 complete_operation(rule);
2678 rule_destruct(struct rule *rule_)
2680 struct rule_dpif *rule = rule_dpif_cast(rule_);
2681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2682 struct facet *facet, *next_facet;
2684 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2685 facet_revalidate(ofproto, facet);
2688 complete_operation(rule);
2692 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
2694 struct rule_dpif *rule = rule_dpif_cast(rule_);
2695 struct facet *facet;
2697 /* Start from historical data for 'rule' itself that are no longer tracked
2698 * in facets. This counts, for example, facets that have expired. */
2699 *packets = rule->packet_count;
2700 *bytes = rule->byte_count;
2702 /* Add any statistics that are tracked by facets. This includes
2703 * statistical data recently updated by ofproto_update_stats() as well as
2704 * stats for packets that were executed "by hand" via dpif_execute(). */
2705 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2706 *packets += facet->packet_count;
2707 *bytes += facet->byte_count;
2712 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
2714 struct rule_dpif *rule = rule_dpif_cast(rule_);
2715 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2716 struct action_xlate_ctx ctx;
2717 struct ofpbuf *odp_actions;
2718 struct facet *facet;
2721 /* First look for a related facet. If we find one, account it to that. */
2722 facet = facet_lookup_valid(ofproto, flow);
2723 if (facet && facet->rule == rule) {
2724 facet_execute(ofproto, facet, packet);
2728 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2729 * create a new facet for it and use that. */
2730 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
2731 facet = facet_create(rule, flow, packet);
2732 facet_execute(ofproto, facet, packet);
2733 facet_install(ofproto, facet, true);
2737 /* We can't account anything to a facet. If we were to try, then that
2738 * facet would have a non-matching rule, busting our invariants. */
2739 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
2740 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2741 size = packet->size;
2742 if (execute_odp_actions(ofproto, flow, odp_actions->data,
2743 odp_actions->size, packet)) {
2744 rule->used = time_msec();
2745 rule->packet_count++;
2746 rule->byte_count += size;
2747 flow_push_stats(rule, flow, 1, size, rule->used);
2749 ofpbuf_delete(odp_actions);
2755 rule_modify_actions(struct rule *rule_)
2757 struct rule_dpif *rule = rule_dpif_cast(rule_);
2758 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2761 error = validate_actions(rule->up.actions, rule->up.n_actions,
2762 &rule->up.cr.flow, ofproto->max_ports);
2764 ofoperation_complete(rule->up.pending, error);
2768 complete_operation(rule);
2771 /* Sends 'packet' out of port 'odp_port' within 'p'.
2772 * Returns 0 if successful, otherwise a positive errno value. */
2774 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
2775 const struct ofpbuf *packet)
2777 struct ofpbuf key, odp_actions;
2778 struct odputil_keybuf keybuf;
2782 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
2783 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2784 odp_flow_key_from_flow(&key, &flow);
2786 ofpbuf_init(&odp_actions, 32);
2787 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2788 error = dpif_execute(ofproto->dpif,
2790 odp_actions.data, odp_actions.size,
2792 ofpbuf_uninit(&odp_actions);
2795 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
2796 ofproto->up.name, odp_port, strerror(error));
2801 /* OpenFlow to ODP action translation. */
2803 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2804 struct action_xlate_ctx *ctx);
2805 static void xlate_normal(struct action_xlate_ctx *);
2808 commit_odp_actions(struct action_xlate_ctx *ctx)
2810 const struct flow *flow = &ctx->flow;
2811 struct flow *base = &ctx->base_flow;
2812 struct ofpbuf *odp_actions = ctx->odp_actions;
2814 if (base->tun_id != flow->tun_id) {
2815 nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id);
2816 base->tun_id = flow->tun_id;
2819 if (base->nw_src != flow->nw_src) {
2820 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src);
2821 base->nw_src = flow->nw_src;
2824 if (base->nw_dst != flow->nw_dst) {
2825 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst);
2826 base->nw_dst = flow->nw_dst;
2829 if (base->nw_tos != flow->nw_tos) {
2830 nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos);
2831 base->nw_tos = flow->nw_tos;
2834 if (base->vlan_tci != flow->vlan_tci) {
2835 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
2836 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2838 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2839 flow->vlan_tci & ~htons(VLAN_CFI));
2841 base->vlan_tci = flow->vlan_tci;
2844 if (base->tp_src != flow->tp_src) {
2845 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src);
2846 base->tp_src = flow->tp_src;
2849 if (base->tp_dst != flow->tp_dst) {
2850 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst);
2851 base->tp_dst = flow->tp_dst;
2854 if (!eth_addr_equals(base->dl_src, flow->dl_src)) {
2855 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2856 flow->dl_src, ETH_ADDR_LEN);
2857 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
2860 if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) {
2861 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2862 flow->dl_dst, ETH_ADDR_LEN);
2863 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
2866 if (ctx->base_priority != ctx->priority) {
2867 if (ctx->priority) {
2868 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY,
2871 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2873 ctx->base_priority = ctx->priority;
2878 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
2880 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
2881 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
2884 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
2885 /* Forwarding disabled on port. */
2890 * We don't have an ofport record for this port, but it doesn't hurt to
2891 * allow forwarding to it anyhow. Maybe such a port will appear later
2892 * and we're pre-populating the flow table.
2896 commit_odp_actions(ctx);
2897 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2898 ctx->nf_output_iface = ofp_port;
2902 xlate_table_action(struct action_xlate_ctx *ctx,
2903 uint16_t in_port, uint8_t table_id)
2905 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2906 struct rule_dpif *rule;
2907 uint16_t old_in_port;
2908 uint8_t old_table_id;
2910 old_table_id = ctx->table_id;
2911 ctx->table_id = table_id;
2913 /* Look up a flow with 'in_port' as the input port. Then restore the
2914 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2915 * have surprising behavior). */
2916 old_in_port = ctx->flow.in_port;
2917 ctx->flow.in_port = in_port;
2918 rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow, table_id);
2919 ctx->flow.in_port = old_in_port;
2921 if (ctx->resubmit_hook) {
2922 ctx->resubmit_hook(ctx, rule);
2927 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
2931 ctx->table_id = old_table_id;
2933 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2935 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
2936 MAX_RESUBMIT_RECURSION);
2941 xlate_resubmit_table(struct action_xlate_ctx *ctx,
2942 const struct nx_action_resubmit *nar)
2947 in_port = (nar->in_port == htons(OFPP_IN_PORT)
2949 : ntohs(nar->in_port));
2950 table_id = nar->table == 255 ? ctx->table_id : nar->table;
2952 xlate_table_action(ctx, in_port, table_id);
2956 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
2958 struct ofport_dpif *ofport;
2960 commit_odp_actions(ctx);
2961 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
2962 uint16_t ofp_port = ofport->up.ofp_port;
2963 if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) {
2964 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT,
2969 ctx->nf_output_iface = NF_OUT_FLOOD;
2973 xlate_output_action__(struct action_xlate_ctx *ctx,
2974 uint16_t port, uint16_t max_len)
2976 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2978 ctx->nf_output_iface = NF_OUT_DROP;
2982 add_output_action(ctx, ctx->flow.in_port);
2985 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
2991 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
2994 flood_packets(ctx, htonl(0));
2996 case OFPP_CONTROLLER:
2997 commit_odp_actions(ctx);
2998 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_USERSPACE, max_len);
3001 add_output_action(ctx, OFPP_LOCAL);
3006 if (port != ctx->flow.in_port) {
3007 add_output_action(ctx, port);
3012 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3013 ctx->nf_output_iface = NF_OUT_FLOOD;
3014 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3015 ctx->nf_output_iface = prev_nf_output_iface;
3016 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3017 ctx->nf_output_iface != NF_OUT_FLOOD) {
3018 ctx->nf_output_iface = NF_OUT_MULTI;
3023 xlate_output_action(struct action_xlate_ctx *ctx,
3024 const struct ofp_action_output *oao)
3026 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3030 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3031 const struct ofp_action_enqueue *oae)
3033 uint16_t ofp_port, odp_port;
3034 uint32_t ctx_priority, priority;
3037 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3040 /* Fall back to ordinary output action. */
3041 xlate_output_action__(ctx, ntohs(oae->port), 0);
3045 /* Figure out ODP output port. */
3046 ofp_port = ntohs(oae->port);
3047 if (ofp_port == OFPP_IN_PORT) {
3048 ofp_port = ctx->flow.in_port;
3050 odp_port = ofp_port_to_odp_port(ofp_port);
3052 /* Add ODP actions. */
3053 ctx_priority = ctx->priority;
3054 ctx->priority = priority;
3055 add_output_action(ctx, odp_port);
3056 ctx->priority = ctx_priority;
3058 /* Update NetFlow output port. */
3059 if (ctx->nf_output_iface == NF_OUT_DROP) {
3060 ctx->nf_output_iface = odp_port;
3061 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3062 ctx->nf_output_iface = NF_OUT_MULTI;
3067 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3068 const struct nx_action_set_queue *nasq)
3073 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3076 /* Couldn't translate queue to a priority, so ignore. A warning
3077 * has already been logged. */
3081 ctx->priority = priority;
3084 struct xlate_reg_state {
3090 xlate_autopath(struct action_xlate_ctx *ctx,
3091 const struct nx_action_autopath *naa)
3093 uint16_t ofp_port = ntohl(naa->id);
3094 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
3096 if (!port || !port->bundle) {
3097 ofp_port = OFPP_NONE;
3098 } else if (port->bundle->bond) {
3099 /* Autopath does not support VLAN hashing. */
3100 struct ofport_dpif *slave = bond_choose_output_slave(
3101 port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags);
3103 ofp_port = slave->up.ofp_port;
3106 autopath_execute(naa, &ctx->flow, ofp_port);
3110 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
3112 struct ofproto_dpif *ofproto = ofproto_;
3113 struct ofport_dpif *port;
3123 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3126 port = get_ofp_port(ofproto, ofp_port);
3127 return port ? port->may_enable : false;
3132 do_xlate_actions(const union ofp_action *in, size_t n_in,
3133 struct action_xlate_ctx *ctx)
3135 const struct ofport_dpif *port;
3136 const union ofp_action *ia;
3139 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
3141 && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3142 port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3143 ? htonl(OFPPC_NO_RECV_STP)
3144 : htonl(OFPPC_NO_RECV))) {
3145 /* Drop this flow. */
3149 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
3150 const struct ofp_action_dl_addr *oada;
3151 const struct nx_action_resubmit *nar;
3152 const struct nx_action_set_tunnel *nast;
3153 const struct nx_action_set_queue *nasq;
3154 const struct nx_action_multipath *nam;
3155 const struct nx_action_autopath *naa;
3156 const struct nx_action_bundle *nab;
3157 enum ofputil_action_code code;
3160 code = ofputil_decode_action_unsafe(ia);
3162 case OFPUTIL_OFPAT_OUTPUT:
3163 xlate_output_action(ctx, &ia->output);
3166 case OFPUTIL_OFPAT_SET_VLAN_VID:
3167 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3168 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3171 case OFPUTIL_OFPAT_SET_VLAN_PCP:
3172 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3173 ctx->flow.vlan_tci |= htons(
3174 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3177 case OFPUTIL_OFPAT_STRIP_VLAN:
3178 ctx->flow.vlan_tci = htons(0);
3181 case OFPUTIL_OFPAT_SET_DL_SRC:
3182 oada = ((struct ofp_action_dl_addr *) ia);
3183 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3186 case OFPUTIL_OFPAT_SET_DL_DST:
3187 oada = ((struct ofp_action_dl_addr *) ia);
3188 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3191 case OFPUTIL_OFPAT_SET_NW_SRC:
3192 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3195 case OFPUTIL_OFPAT_SET_NW_DST:
3196 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3199 case OFPUTIL_OFPAT_SET_NW_TOS:
3200 ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK;
3203 case OFPUTIL_OFPAT_SET_TP_SRC:
3204 ctx->flow.tp_src = ia->tp_port.tp_port;
3207 case OFPUTIL_OFPAT_SET_TP_DST:
3208 ctx->flow.tp_dst = ia->tp_port.tp_port;
3211 case OFPUTIL_OFPAT_ENQUEUE:
3212 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3215 case OFPUTIL_NXAST_RESUBMIT:
3216 nar = (const struct nx_action_resubmit *) ia;
3217 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
3220 case OFPUTIL_NXAST_RESUBMIT_TABLE:
3221 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
3224 case OFPUTIL_NXAST_SET_TUNNEL:
3225 nast = (const struct nx_action_set_tunnel *) ia;
3226 tun_id = htonll(ntohl(nast->tun_id));
3227 ctx->flow.tun_id = tun_id;
3230 case OFPUTIL_NXAST_SET_QUEUE:
3231 nasq = (const struct nx_action_set_queue *) ia;
3232 xlate_set_queue_action(ctx, nasq);
3235 case OFPUTIL_NXAST_POP_QUEUE:
3239 case OFPUTIL_NXAST_REG_MOVE:
3240 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
3244 case OFPUTIL_NXAST_REG_LOAD:
3245 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
3249 case OFPUTIL_NXAST_NOTE:
3250 /* Nothing to do. */
3253 case OFPUTIL_NXAST_SET_TUNNEL64:
3254 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
3255 ctx->flow.tun_id = tun_id;
3258 case OFPUTIL_NXAST_MULTIPATH:
3259 nam = (const struct nx_action_multipath *) ia;
3260 multipath_execute(nam, &ctx->flow);
3263 case OFPUTIL_NXAST_AUTOPATH:
3264 naa = (const struct nx_action_autopath *) ia;
3265 xlate_autopath(ctx, naa);
3268 case OFPUTIL_NXAST_BUNDLE:
3269 ctx->ofproto->has_bundle_action = true;
3270 nab = (const struct nx_action_bundle *) ia;
3271 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
3276 case OFPUTIL_NXAST_BUNDLE_LOAD:
3277 ctx->ofproto->has_bundle_action = true;
3278 nab = (const struct nx_action_bundle *) ia;
3279 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
3287 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3288 struct ofproto_dpif *ofproto, const struct flow *flow,
3289 const struct ofpbuf *packet)
3291 ctx->ofproto = ofproto;
3293 ctx->packet = packet;
3294 ctx->resubmit_hook = NULL;
3297 static struct ofpbuf *
3298 xlate_actions(struct action_xlate_ctx *ctx,
3299 const union ofp_action *in, size_t n_in)
3301 COVERAGE_INC(ofproto_dpif_xlate);
3303 ctx->odp_actions = ofpbuf_new(512);
3305 ctx->may_set_up_flow = true;
3306 ctx->nf_output_iface = NF_OUT_DROP;
3309 ctx->base_priority = 0;
3310 ctx->base_flow = ctx->flow;
3313 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
3314 ctx->may_set_up_flow = false;
3316 do_xlate_actions(in, n_in, ctx);
3319 /* Check with in-band control to see if we're allowed to set up this
3321 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
3322 ctx->odp_actions->data,
3323 ctx->odp_actions->size)) {
3324 ctx->may_set_up_flow = false;
3327 return ctx->odp_actions;
3330 /* OFPP_NORMAL implementation. */
3333 struct ofport_dpif *port;
3338 struct dst builtin[32];
3340 size_t n, allocated;
3343 static void dst_set_init(struct dst_set *);
3344 static void dst_set_add(struct dst_set *, const struct dst *);
3345 static void dst_set_free(struct dst_set *);
3347 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
3350 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
3351 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
3353 dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE
3354 : in_bundle->vlan >= 0 ? in_bundle->vlan
3355 : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE
3356 : vlan_tci_to_vid(ctx->flow.vlan_tci));
3358 dst->port = (!out_bundle->bond
3359 ? ofbundle_get_a_port(out_bundle)
3360 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
3361 dst->vlan, &ctx->tags));
3363 return dst->port != NULL;
3367 mirror_mask_ffs(mirror_mask_t mask)
3369 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
3374 dst_set_init(struct dst_set *set)
3376 set->dsts = set->builtin;
3378 set->allocated = ARRAY_SIZE(set->builtin);
3382 dst_set_add(struct dst_set *set, const struct dst *dst)
3384 if (set->n >= set->allocated) {
3385 size_t new_allocated;
3386 struct dst *new_dsts;
3388 new_allocated = set->allocated * 2;
3389 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
3390 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
3394 set->dsts = new_dsts;
3395 set->allocated = new_allocated;
3397 set->dsts[set->n++] = *dst;
3401 dst_set_free(struct dst_set *set)
3403 if (set->dsts != set->builtin) {
3409 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
3412 for (i = 0; i < set->n; i++) {
3413 if (set->dsts[i].vlan == test->vlan
3414 && set->dsts[i].port == test->port) {
3422 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
3424 return (bundle->vlan < 0
3425 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
3429 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
3431 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
3434 /* Returns an arbitrary interface within 'bundle'. */
3435 static struct ofport_dpif *
3436 ofbundle_get_a_port(const struct ofbundle *bundle)
3438 return CONTAINER_OF(list_front(&bundle->ports),
3439 struct ofport_dpif, bundle_node);
3443 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
3444 const struct ofbundle *in_bundle,
3445 const struct ofbundle *out_bundle, struct dst_set *set)
3449 if (out_bundle == OFBUNDLE_FLOOD) {
3450 struct ofbundle *bundle;
3452 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
3453 if (bundle != in_bundle
3454 && ofbundle_includes_vlan(bundle, vlan)
3455 && bundle->floodable
3456 && !bundle->mirror_out
3457 && set_dst(ctx, &dst, in_bundle, bundle)) {
3458 dst_set_add(set, &dst);
3461 ctx->nf_output_iface = NF_OUT_FLOOD;
3462 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
3463 dst_set_add(set, &dst);
3464 ctx->nf_output_iface = dst.port->odp_port;
3469 vlan_is_mirrored(const struct ofmirror *m, int vlan)
3471 return !m->vlans || bitmap_is_set(m->vlans, vlan);
3474 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
3475 * to a VLAN. In general most packets may be mirrored but we want to drop
3476 * protocols that may confuse switches. */
3478 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
3480 /* If you change this function's behavior, please update corresponding
3481 * documentation in vswitch.xml at the same time. */
3482 if (dst[0] != 0x01) {
3483 /* All the currently banned MACs happen to start with 01 currently, so
3484 * this is a quick way to eliminate most of the good ones. */
3486 if (eth_addr_is_reserved(dst)) {
3487 /* Drop STP, IEEE pause frames, and other reserved protocols
3488 * (01-80-c2-00-00-0x). */
3492 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
3494 if ((dst[3] & 0xfe) == 0xcc &&
3495 (dst[4] & 0xfe) == 0xcc &&
3496 (dst[5] & 0xfe) == 0xcc) {
3497 /* Drop the following protocols plus others following the same
3500 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
3501 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
3502 STP Uplink Fast (01-00-0c-cd-cd-cd) */
3506 if (!(dst[3] | dst[4] | dst[5])) {
3507 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
3516 compose_mirror_dsts(struct action_xlate_ctx *ctx,
3517 uint16_t vlan, const struct ofbundle *in_bundle,
3518 struct dst_set *set)
3520 struct ofproto_dpif *ofproto = ctx->ofproto;
3521 mirror_mask_t mirrors;
3525 mirrors = in_bundle->src_mirrors;
3526 for (i = 0; i < set->n; i++) {
3527 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
3534 flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3535 if (flow_vlan == 0) {
3536 flow_vlan = OFP_VLAN_NONE;
3540 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
3541 if (vlan_is_mirrored(m, vlan)) {
3545 if (set_dst(ctx, &dst, in_bundle, m->out)
3546 && !dst_is_duplicate(set, &dst)) {
3547 dst_set_add(set, &dst);
3549 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
3550 struct ofbundle *bundle;
3552 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3553 if (ofbundle_includes_vlan(bundle, m->out_vlan)
3554 && set_dst(ctx, &dst, in_bundle, bundle))
3556 if (bundle->vlan < 0) {
3557 dst.vlan = m->out_vlan;
3559 if (dst_is_duplicate(set, &dst)) {
3563 /* Use the vlan tag on the original flow instead of
3564 * the one passed in the vlan parameter. This ensures
3565 * that we compare the vlan from before any implicit
3566 * tagging tags place. This is necessary because
3567 * dst->vlan is the final vlan, after removing implicit
3569 if (bundle == in_bundle && dst.vlan == flow_vlan) {
3570 /* Don't send out input port on same VLAN. */
3573 dst_set_add(set, &dst);
3578 mirrors &= mirrors - 1;
3583 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
3584 const struct ofbundle *in_bundle,
3585 const struct ofbundle *out_bundle)
3587 uint16_t initial_vlan, cur_vlan;
3588 const struct dst *dst;
3592 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
3593 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
3595 /* Output all the packets we can without having to change the VLAN. */
3596 initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3597 if (initial_vlan == 0) {
3598 initial_vlan = OFP_VLAN_NONE;
3600 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3601 if (dst->vlan != initial_vlan) {
3604 nl_msg_put_u32(ctx->odp_actions,
3605 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3608 /* Then output the rest. */
3609 cur_vlan = initial_vlan;
3610 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3611 if (dst->vlan == initial_vlan) {
3614 if (dst->vlan != cur_vlan) {
3615 if (dst->vlan == OFP_VLAN_NONE) {
3616 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
3619 tci = htons(dst->vlan & VLAN_VID_MASK);
3620 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
3621 nl_msg_put_be16(ctx->odp_actions,
3622 ODP_ACTION_ATTR_SET_DL_TCI, tci);
3624 cur_vlan = dst->vlan;
3626 nl_msg_put_u32(ctx->odp_actions,
3627 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3633 /* Returns the effective vlan of a packet, taking into account both the
3634 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
3635 * the packet is untagged and -1 indicates it has an invalid header and
3636 * should be dropped. */
3638 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
3639 struct ofbundle *in_bundle, bool have_packet)
3641 int vlan = vlan_tci_to_vid(flow->vlan_tci);
3642 if (in_bundle->vlan >= 0) {
3645 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3646 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3647 "packet received on port %s configured with "
3648 "implicit VLAN %"PRIu16,
3649 ofproto->up.name, vlan,
3650 in_bundle->name, in_bundle->vlan);
3654 vlan = in_bundle->vlan;
3656 if (!ofbundle_includes_vlan(in_bundle, vlan)) {
3658 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3659 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3660 "packet received on port %s not configured for "
3662 ofproto->up.name, vlan, in_bundle->name, vlan);
3671 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
3672 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
3673 * indicate this; newer upstream kernels use gratuitous ARP requests. */
3675 is_gratuitous_arp(const struct flow *flow)
3677 return (flow->dl_type == htons(ETH_TYPE_ARP)
3678 && eth_addr_is_broadcast(flow->dl_dst)
3679 && (flow->nw_proto == ARP_OP_REPLY
3680 || (flow->nw_proto == ARP_OP_REQUEST
3681 && flow->nw_src == flow->nw_dst)));
3685 update_learning_table(struct ofproto_dpif *ofproto,
3686 const struct flow *flow, int vlan,
3687 struct ofbundle *in_bundle)
3689 struct mac_entry *mac;
3691 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
3695 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
3696 if (is_gratuitous_arp(flow)) {
3697 /* We don't want to learn from gratuitous ARP packets that are
3698 * reflected back over bond slaves so we lock the learning table. */
3699 if (!in_bundle->bond) {
3700 mac_entry_set_grat_arp_lock(mac);
3701 } else if (mac_entry_is_grat_arp_locked(mac)) {
3706 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
3707 /* The log messages here could actually be useful in debugging,
3708 * so keep the rate limit relatively high. */
3709 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3710 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
3711 "on port %s in VLAN %d",
3712 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
3713 in_bundle->name, vlan);
3715 mac->port.p = in_bundle;
3716 tag_set_add(&ofproto->revalidate_set,
3717 mac_learning_changed(ofproto->ml, mac));
3721 /* Determines whether packets in 'flow' within 'br' should be forwarded or
3722 * dropped. Returns true if they may be forwarded, false if they should be
3725 * If 'have_packet' is true, it indicates that the caller is processing a
3726 * received packet. If 'have_packet' is false, then the caller is just
3727 * revalidating an existing flow because configuration has changed. Either
3728 * way, 'have_packet' only affects logging (there is no point in logging errors
3729 * during revalidation).
3731 * Sets '*in_portp' to the input port. This will be a null pointer if
3732 * flow->in_port does not designate a known input port (in which case
3733 * is_admissible() returns false).
3735 * When returning true, sets '*vlanp' to the effective VLAN of the input
3736 * packet, as returned by flow_get_vlan().
3738 * May also add tags to '*tags', although the current implementation only does
3739 * so in one special case.
3742 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
3744 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
3746 struct ofport_dpif *in_port;
3747 struct ofbundle *in_bundle;
3750 /* Find the port and bundle for the received packet. */
3751 in_port = get_ofp_port(ofproto, flow->in_port);
3752 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
3753 if (!in_port || !in_bundle) {
3754 /* No interface? Something fishy... */
3756 /* Odd. A few possible reasons here:
3758 * - We deleted a port but there are still a few packets queued up
3761 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
3762 * we don't know about.
3764 * - Packet arrived on the local port but the local port is not
3767 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3769 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3771 ofproto->up.name, flow->in_port);
3775 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
3780 /* Drop frames for reserved multicast addresses
3781 * only if forward_bpdu option is absent. */
3782 if (eth_addr_is_reserved(flow->dl_dst) &&
3783 !ofproto->up.forward_bpdu) {
3787 /* Drop frames on bundles reserved for mirroring. */
3788 if (in_bundle->mirror_out) {
3790 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3791 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3792 "%s, which is reserved exclusively for mirroring",
3793 ofproto->up.name, in_bundle->name);
3798 if (in_bundle->bond) {
3799 struct mac_entry *mac;
3801 switch (bond_check_admissibility(in_bundle->bond, in_port,
3802 flow->dl_dst, tags)) {
3809 case BV_DROP_IF_MOVED:
3810 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
3811 if (mac && mac->port.p != in_bundle &&
3812 (!is_gratuitous_arp(flow)
3813 || mac_entry_is_grat_arp_locked(mac))) {
3824 xlate_normal(struct action_xlate_ctx *ctx)
3826 struct ofbundle *in_bundle;
3827 struct ofbundle *out_bundle;
3828 struct mac_entry *mac;
3831 /* Check whether we should drop packets in this flow. */
3832 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
3833 &ctx->tags, &vlan, &in_bundle)) {
3838 /* Learn source MAC (but don't try to learn from revalidation). */
3840 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
3843 /* Determine output bundle. */
3844 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
3847 out_bundle = mac->port.p;
3848 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
3849 /* If we are revalidating but don't have a learning entry then eject
3850 * the flow. Installing a flow that floods packets opens up a window
3851 * of time where we could learn from a packet reflected on a bond and
3852 * blackhole packets before the learning table is updated to reflect
3853 * the correct port. */
3854 ctx->may_set_up_flow = false;
3857 out_bundle = OFBUNDLE_FLOOD;
3860 /* Don't send packets out their input bundles. */
3861 if (in_bundle == out_bundle) {
3867 compose_actions(ctx, vlan, in_bundle, out_bundle);
3872 get_drop_frags(struct ofproto *ofproto_)
3874 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3877 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
3882 set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
3884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3886 dpif_set_drop_frags(ofproto->dpif, drop_frags);
3890 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3891 const struct flow *flow,
3892 const union ofp_action *ofp_actions, size_t n_ofp_actions)
3894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3897 error = validate_actions(ofp_actions, n_ofp_actions, flow,
3898 ofproto->max_ports);
3900 struct odputil_keybuf keybuf;
3901 struct action_xlate_ctx ctx;
3902 struct ofpbuf *odp_actions;
3905 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3906 odp_flow_key_from_flow(&key, flow);
3908 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3909 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3910 dpif_execute(ofproto->dpif, key.data, key.size,
3911 odp_actions->data, odp_actions->size, packet);
3912 ofpbuf_delete(odp_actions);
3918 get_netflow_ids(const struct ofproto *ofproto_,
3919 uint8_t *engine_type, uint8_t *engine_id)
3921 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3923 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3926 static struct ofproto_dpif *
3927 ofproto_dpif_lookup(const char *name)
3929 struct ofproto *ofproto = ofproto_lookup(name);
3930 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
3931 ? ofproto_dpif_cast(ofproto)
3936 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
3937 const char *args, void *aux OVS_UNUSED)
3939 struct ds ds = DS_EMPTY_INITIALIZER;
3940 const struct ofproto_dpif *ofproto;
3941 const struct mac_entry *e;
3943 ofproto = ofproto_dpif_lookup(args);
3945 unixctl_command_reply(conn, 501, "no such bridge");
3949 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3950 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3951 struct ofbundle *bundle = e->port.p;
3952 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
3953 ofbundle_get_a_port(bundle)->odp_port,
3954 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
3956 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3960 struct ofproto_trace {
3961 struct action_xlate_ctx ctx;
3967 trace_format_rule(struct ds *result, uint8_t table_id, int level,
3968 const struct rule_dpif *rule)
3970 ds_put_char_multiple(result, '\t', level);
3972 ds_put_cstr(result, "No match\n");
3976 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3977 table_id, ntohll(rule->up.flow_cookie));
3978 cls_rule_format(&rule->up.cr, result);
3979 ds_put_char(result, '\n');
3981 ds_put_char_multiple(result, '\t', level);
3982 ds_put_cstr(result, "OpenFlow ");
3983 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
3984 ds_put_char(result, '\n');
3988 trace_format_flow(struct ds *result, int level, const char *title,
3989 struct ofproto_trace *trace)
3991 ds_put_char_multiple(result, '\t', level);
3992 ds_put_format(result, "%s: ", title);
3993 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
3994 ds_put_cstr(result, "unchanged");
3996 flow_format(result, &trace->ctx.flow);
3997 trace->flow = trace->ctx.flow;
3999 ds_put_char(result, '\n');
4003 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
4005 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4006 struct ds *result = trace->result;
4008 ds_put_char(result, '\n');
4009 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4010 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
4014 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4015 void *aux OVS_UNUSED)
4017 char *dpname, *arg1, *arg2, *arg3;
4018 char *args = xstrdup(args_);
4019 char *save_ptr = NULL;
4020 struct ofproto_dpif *ofproto;
4021 struct ofpbuf odp_key;
4022 struct ofpbuf *packet;
4023 struct rule_dpif *rule;
4029 ofpbuf_init(&odp_key, 0);
4032 dpname = strtok_r(args, " ", &save_ptr);
4033 arg1 = strtok_r(NULL, " ", &save_ptr);
4034 arg2 = strtok_r(NULL, " ", &save_ptr);
4035 arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4036 if (dpname && arg1 && !arg2 && !arg3) {
4037 /* ofproto/trace dpname flow */
4040 /* Convert string to ODP key. */
4041 ofpbuf_init(&odp_key, 0);
4042 error = odp_flow_key_from_string(arg1, &odp_key);
4044 unixctl_command_reply(conn, 501, "Bad flow syntax");
4048 /* Convert odp_key to flow. */
4049 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
4051 unixctl_command_reply(conn, 501, "Invalid flow");
4054 } else if (dpname && arg1 && arg2 && arg3) {
4055 /* ofproto/trace dpname tun_id in_port packet */
4059 tun_id = htonll(strtoull(arg1, NULL, 0));
4060 in_port = ofp_port_to_odp_port(atoi(arg2));
4062 packet = ofpbuf_new(strlen(args) / 2);
4063 arg3 = ofpbuf_put_hex(packet, arg3, NULL);
4064 arg3 += strspn(arg3, " ");
4065 if (*arg3 != '\0') {
4066 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4069 if (packet->size < ETH_HEADER_LEN) {
4070 unixctl_command_reply(conn, 501,
4071 "Packet data too short for Ethernet");
4075 ds_put_cstr(&result, "Packet: ");
4076 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
4077 ds_put_cstr(&result, s);
4080 flow_extract(packet, tun_id, in_port, &flow);
4082 unixctl_command_reply(conn, 501, "Bad command syntax");
4086 ofproto = ofproto_dpif_lookup(dpname);
4088 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4093 ds_put_cstr(&result, "Flow: ");
4094 flow_format(&result, &flow);
4095 ds_put_char(&result, '\n');
4097 rule = rule_dpif_lookup(ofproto, &flow, 0);
4098 trace_format_rule(&result, 0, 0, rule);
4100 struct ofproto_trace trace;
4101 struct ofpbuf *odp_actions;
4103 trace.result = &result;
4105 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
4106 trace.ctx.resubmit_hook = trace_resubmit;
4107 odp_actions = xlate_actions(&trace.ctx,
4108 rule->up.actions, rule->up.n_actions);
4110 ds_put_char(&result, '\n');
4111 trace_format_flow(&result, 0, "Final flow", &trace);
4112 ds_put_cstr(&result, "Datapath actions: ");
4113 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4114 ofpbuf_delete(odp_actions);
4116 if (!trace.ctx.may_set_up_flow) {
4118 ds_put_cstr(&result, "\nThis flow is not cachable.");
4120 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
4121 "for complete actions, please supply a packet.");
4126 unixctl_command_reply(conn, 200, ds_cstr(&result));
4129 ds_destroy(&result);
4130 ofpbuf_delete(packet);
4131 ofpbuf_uninit(&odp_key);
4136 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
4137 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4140 unixctl_command_reply(conn, 200, NULL);
4144 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
4145 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4148 unixctl_command_reply(conn, 200, NULL);
4152 ofproto_dpif_unixctl_init(void)
4154 static bool registered;
4160 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4161 unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
4163 unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL);
4164 unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL);
4167 const struct ofproto_class ofproto_dpif_class = {
4194 port_is_lacp_current,
4195 NULL, /* rule_choose_table */
4202 rule_modify_actions,
4215 is_mirror_output_bundle,
4216 forward_bpdu_changed,