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 its statistics were reset (e.g. dpif_flow_put() with
211 * DPIF_FP_ZERO_STATS).
213 uint64_t packet_count; /* Number of packets received. */
214 uint64_t byte_count; /* Number of bytes received. */
216 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
217 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
219 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
220 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
221 long long int rs_used; /* Used time pushed to resubmit children. */
223 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
225 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
226 struct list list_node; /* In owning rule's 'facets' list. */
227 struct rule_dpif *rule; /* Owning rule. */
228 struct flow flow; /* Exact-match flow. */
229 bool installed; /* Installed in datapath? */
230 bool may_install; /* True ordinarily; false if actions must
231 * be reassessed for every packet. */
232 size_t actions_len; /* Number of bytes in actions[]. */
233 struct nlattr *actions; /* Datapath actions. */
234 tag_type tags; /* Tags. */
235 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
238 static struct facet *facet_create(struct rule_dpif *, const struct flow *,
239 const struct ofpbuf *packet);
240 static void facet_remove(struct ofproto_dpif *, struct facet *);
241 static void facet_free(struct facet *);
243 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
244 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
245 const struct flow *);
246 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
248 static void facet_execute(struct ofproto_dpif *, struct facet *,
249 struct ofpbuf *packet);
251 static int facet_put__(struct ofproto_dpif *, struct facet *,
252 const struct nlattr *actions, size_t actions_len,
253 struct dpif_flow_stats *);
254 static void facet_install(struct ofproto_dpif *, struct facet *,
256 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
257 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
259 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
260 const struct ofpbuf *packet);
261 static void facet_update_time(struct ofproto_dpif *, struct facet *,
263 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
264 const struct dpif_flow_stats *);
265 static void facet_reset_counters(struct facet *);
266 static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
267 static void facet_push_stats(struct facet *);
268 static void facet_account(struct ofproto_dpif *, struct facet *);
270 static bool facet_is_controller_flow(struct facet *);
272 static void flow_push_stats(const struct rule_dpif *,
273 struct flow *, uint64_t packets, uint64_t bytes,
280 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
281 struct list bundle_node; /* In struct ofbundle's "ports" list. */
282 struct cfm *cfm; /* Connectivity Fault Management, if any. */
283 tag_type tag; /* Tag associated with this port. */
284 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
285 bool may_enable; /* May be enabled in bonds. */
288 static struct ofport_dpif *
289 ofport_dpif_cast(const struct ofport *ofport)
291 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
292 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
295 static void port_run(struct ofport_dpif *);
296 static void port_wait(struct ofport_dpif *);
297 static int set_cfm(struct ofport *, const struct cfm_settings *);
299 struct dpif_completion {
300 struct list list_node;
301 struct ofoperation *op;
304 struct ofproto_dpif {
313 struct netflow *netflow;
314 struct dpif_sflow *sflow;
315 struct hmap bundles; /* Contains "struct ofbundle"s. */
316 struct mac_learning *ml;
317 struct ofmirror *mirrors[MAX_MIRRORS];
318 bool has_bonded_bundles;
321 struct timer next_expiration;
325 bool need_revalidate;
326 struct tag_set revalidate_set;
328 /* Support for debugging async flow mods. */
329 struct list completions;
331 bool has_bundle_action; /* True when the first bundle action appears. */
334 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
335 * for debugging the asynchronous flow_mod implementation.) */
338 static void ofproto_dpif_unixctl_init(void);
340 static struct ofproto_dpif *
341 ofproto_dpif_cast(const struct ofproto *ofproto)
343 assert(ofproto->ofproto_class == &ofproto_dpif_class);
344 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
347 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
349 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
352 /* Packet processing. */
353 static void update_learning_table(struct ofproto_dpif *,
354 const struct flow *, int vlan,
356 static bool is_admissible(struct ofproto_dpif *, const struct flow *,
357 bool have_packet, tag_type *, int *vlanp,
358 struct ofbundle **in_bundlep);
359 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
361 /* Flow expiration. */
362 static int expire(struct ofproto_dpif *);
365 static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
366 const struct ofpbuf *packet);
368 /* Global variables. */
369 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
371 /* Factory functions. */
374 enumerate_types(struct sset *types)
376 dp_enumerate_types(types);
380 enumerate_names(const char *type, struct sset *names)
382 return dp_enumerate_names(type, names);
386 del(const char *type, const char *name)
391 error = dpif_open(name, type, &dpif);
393 error = dpif_delete(dpif);
399 /* Basic life-cycle. */
401 static struct ofproto *
404 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
409 dealloc(struct ofproto *ofproto_)
411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
416 construct(struct ofproto *ofproto_, int *n_tablesp)
418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
419 const char *name = ofproto->up.name;
423 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
425 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
429 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
430 ofproto->n_matches = 0;
432 error = dpif_recv_set_mask(ofproto->dpif,
433 ((1u << DPIF_UC_MISS) |
434 (1u << DPIF_UC_ACTION) |
435 (1u << DPIF_UC_SAMPLE)));
437 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
438 dpif_close(ofproto->dpif);
441 dpif_flow_flush(ofproto->dpif);
442 dpif_recv_purge(ofproto->dpif);
444 ofproto->netflow = NULL;
445 ofproto->sflow = NULL;
446 hmap_init(&ofproto->bundles);
447 ofproto->ml = mac_learning_create();
448 for (i = 0; i < MAX_MIRRORS; i++) {
449 ofproto->mirrors[i] = NULL;
451 ofproto->has_bonded_bundles = false;
453 timer_set_duration(&ofproto->next_expiration, 1000);
455 hmap_init(&ofproto->facets);
456 ofproto->need_revalidate = false;
457 tag_set_init(&ofproto->revalidate_set);
459 list_init(&ofproto->completions);
461 ofproto_dpif_unixctl_init();
463 ofproto->has_bundle_action = false;
470 complete_operations(struct ofproto_dpif *ofproto)
472 struct dpif_completion *c, *next;
474 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
475 ofoperation_complete(c->op, 0);
476 list_remove(&c->list_node);
482 destruct(struct ofproto *ofproto_)
484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
485 struct rule_dpif *rule, *next_rule;
486 struct classifier *table;
489 complete_operations(ofproto);
491 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
492 struct cls_cursor cursor;
494 cls_cursor_init(&cursor, table, NULL);
495 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
496 ofproto_rule_destroy(&rule->up);
500 for (i = 0; i < MAX_MIRRORS; i++) {
501 mirror_destroy(ofproto->mirrors[i]);
504 netflow_destroy(ofproto->netflow);
505 dpif_sflow_destroy(ofproto->sflow);
506 hmap_destroy(&ofproto->bundles);
507 mac_learning_destroy(ofproto->ml);
509 hmap_destroy(&ofproto->facets);
511 dpif_close(ofproto->dpif);
515 run(struct ofproto *ofproto_)
517 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
518 struct ofport_dpif *ofport;
519 struct ofbundle *bundle;
523 complete_operations(ofproto);
525 dpif_run(ofproto->dpif);
527 for (i = 0; i < 50; i++) {
528 struct dpif_upcall packet;
531 error = dpif_recv(ofproto->dpif, &packet);
533 if (error == ENODEV) {
534 /* Datapath destroyed. */
540 handle_upcall(ofproto, &packet);
543 if (timer_expired(&ofproto->next_expiration)) {
544 int delay = expire(ofproto);
545 timer_set_duration(&ofproto->next_expiration, delay);
548 if (ofproto->netflow) {
549 netflow_run(ofproto->netflow);
551 if (ofproto->sflow) {
552 dpif_sflow_run(ofproto->sflow);
555 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
558 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
562 /* Now revalidate if there's anything to do. */
563 if (ofproto->need_revalidate
564 || !tag_set_is_empty(&ofproto->revalidate_set)) {
565 struct tag_set revalidate_set = ofproto->revalidate_set;
566 bool revalidate_all = ofproto->need_revalidate;
567 struct facet *facet, *next;
569 /* Clear the revalidation flags. */
570 tag_set_init(&ofproto->revalidate_set);
571 ofproto->need_revalidate = false;
573 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
575 || tag_set_intersects(&revalidate_set, facet->tags)) {
576 facet_revalidate(ofproto, facet);
585 wait(struct ofproto *ofproto_)
587 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
588 struct ofport_dpif *ofport;
589 struct ofbundle *bundle;
591 if (!clogged && !list_is_empty(&ofproto->completions)) {
592 poll_immediate_wake();
595 dpif_wait(ofproto->dpif);
596 dpif_recv_wait(ofproto->dpif);
597 if (ofproto->sflow) {
598 dpif_sflow_wait(ofproto->sflow);
600 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
601 poll_immediate_wake();
603 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
606 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
609 if (ofproto->need_revalidate) {
610 /* Shouldn't happen, but if it does just go around again. */
611 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
612 poll_immediate_wake();
614 timer_wait(&ofproto->next_expiration);
619 flush(struct ofproto *ofproto_)
621 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
622 struct facet *facet, *next_facet;
624 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
625 /* Mark the facet as not installed so that facet_remove() doesn't
626 * bother trying to uninstall it. There is no point in uninstalling it
627 * individually since we are about to blow away all the facets with
628 * dpif_flow_flush(). */
629 facet->installed = false;
630 facet->dp_packet_count = 0;
631 facet->dp_byte_count = 0;
632 facet_remove(ofproto, facet);
634 dpif_flow_flush(ofproto->dpif);
638 get_features(struct ofproto *ofproto_ OVS_UNUSED,
639 bool *arp_match_ip, uint32_t *actions)
641 *arp_match_ip = true;
642 *actions = ((1u << OFPAT_OUTPUT) |
643 (1u << OFPAT_SET_VLAN_VID) |
644 (1u << OFPAT_SET_VLAN_PCP) |
645 (1u << OFPAT_STRIP_VLAN) |
646 (1u << OFPAT_SET_DL_SRC) |
647 (1u << OFPAT_SET_DL_DST) |
648 (1u << OFPAT_SET_NW_SRC) |
649 (1u << OFPAT_SET_NW_DST) |
650 (1u << OFPAT_SET_NW_TOS) |
651 (1u << OFPAT_SET_TP_SRC) |
652 (1u << OFPAT_SET_TP_DST) |
653 (1u << OFPAT_ENQUEUE));
657 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
659 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
662 strcpy(ots->name, "classifier");
664 dpif_get_dp_stats(ofproto->dpif, &s);
665 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
666 put_32aligned_be64(&ots->matched_count,
667 htonll(s.n_hit + ofproto->n_matches));
671 set_netflow(struct ofproto *ofproto_,
672 const struct netflow_options *netflow_options)
674 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
676 if (netflow_options) {
677 if (!ofproto->netflow) {
678 ofproto->netflow = netflow_create();
680 return netflow_set_options(ofproto->netflow, netflow_options);
682 netflow_destroy(ofproto->netflow);
683 ofproto->netflow = NULL;
688 static struct ofport *
691 struct ofport_dpif *port = xmalloc(sizeof *port);
696 port_dealloc(struct ofport *port_)
698 struct ofport_dpif *port = ofport_dpif_cast(port_);
703 port_construct(struct ofport *port_)
705 struct ofport_dpif *port = ofport_dpif_cast(port_);
706 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
708 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
711 port->tag = tag_create_random();
712 port->may_enable = true;
714 if (ofproto->sflow) {
715 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
716 netdev_get_name(port->up.netdev));
723 port_destruct(struct ofport *port_)
725 struct ofport_dpif *port = ofport_dpif_cast(port_);
726 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
728 bundle_remove(port_);
729 set_cfm(port_, NULL);
730 if (ofproto->sflow) {
731 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
736 port_modified(struct ofport *port_)
738 struct ofport_dpif *port = ofport_dpif_cast(port_);
740 if (port->bundle && port->bundle->bond) {
741 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
746 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
748 struct ofport_dpif *port = ofport_dpif_cast(port_);
749 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
750 ovs_be32 changed = old_config ^ port->up.opp.config;
752 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
753 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
754 ofproto->need_revalidate = true;
759 set_sflow(struct ofproto *ofproto_,
760 const struct ofproto_sflow_options *sflow_options)
762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
763 struct dpif_sflow *ds = ofproto->sflow;
766 struct ofport_dpif *ofport;
768 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
769 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
770 dpif_sflow_add_port(ds, ofport->odp_port,
771 netdev_get_name(ofport->up.netdev));
774 dpif_sflow_set_options(ds, sflow_options);
776 dpif_sflow_destroy(ds);
777 ofproto->sflow = NULL;
783 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
785 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
792 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
795 if (cfm_configure(ofport->cfm, s)) {
801 cfm_destroy(ofport->cfm);
807 get_cfm_fault(const struct ofport *ofport_)
809 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
811 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
816 /* Expires all MAC learning entries associated with 'port' and forces ofproto
817 * to revalidate every flow. */
819 bundle_flush_macs(struct ofbundle *bundle)
821 struct ofproto_dpif *ofproto = bundle->ofproto;
822 struct mac_learning *ml = ofproto->ml;
823 struct mac_entry *mac, *next_mac;
825 ofproto->need_revalidate = true;
826 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
827 if (mac->port.p == bundle) {
828 mac_learning_expire(ml, mac);
833 static struct ofbundle *
834 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
836 struct ofbundle *bundle;
838 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
840 if (bundle->aux == aux) {
847 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
848 * ones that are found to 'bundles'. */
850 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
851 void **auxes, size_t n_auxes,
852 struct hmapx *bundles)
857 for (i = 0; i < n_auxes; i++) {
858 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
860 hmapx_add(bundles, bundle);
866 bundle_del_port(struct ofport_dpif *port)
868 struct ofbundle *bundle = port->bundle;
870 bundle->ofproto->need_revalidate = true;
872 list_remove(&port->bundle_node);
876 lacp_slave_unregister(bundle->lacp, port);
879 bond_slave_unregister(bundle->bond, port);
882 bundle->floodable = true;
883 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
884 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
885 bundle->floodable = false;
891 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
892 struct lacp_slave_settings *lacp,
893 uint32_t bond_stable_id)
895 struct ofport_dpif *port;
897 port = get_ofp_port(bundle->ofproto, ofp_port);
902 if (port->bundle != bundle) {
903 bundle->ofproto->need_revalidate = true;
905 bundle_del_port(port);
908 port->bundle = bundle;
909 list_push_back(&bundle->ports, &port->bundle_node);
910 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
911 bundle->floodable = false;
915 lacp_slave_register(bundle->lacp, port, lacp);
918 port->bond_stable_id = bond_stable_id;
924 bundle_destroy(struct ofbundle *bundle)
926 struct ofproto_dpif *ofproto;
927 struct ofport_dpif *port, *next_port;
934 ofproto = bundle->ofproto;
935 for (i = 0; i < MAX_MIRRORS; i++) {
936 struct ofmirror *m = ofproto->mirrors[i];
938 if (m->out == bundle) {
940 } else if (hmapx_find_and_delete(&m->srcs, bundle)
941 || hmapx_find_and_delete(&m->dsts, bundle)) {
942 ofproto->need_revalidate = true;
947 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
948 bundle_del_port(port);
951 bundle_flush_macs(bundle);
952 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
954 free(bundle->trunks);
955 lacp_destroy(bundle->lacp);
956 bond_destroy(bundle->bond);
961 bundle_set(struct ofproto *ofproto_, void *aux,
962 const struct ofproto_bundle_settings *s)
964 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
965 bool need_flush = false;
966 const unsigned long *trunks;
967 struct ofport_dpif *port;
968 struct ofbundle *bundle;
973 bundle_destroy(bundle_lookup(ofproto, aux));
977 assert(s->n_slaves == 1 || s->bond != NULL);
978 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
980 bundle = bundle_lookup(ofproto, aux);
982 bundle = xmalloc(sizeof *bundle);
984 bundle->ofproto = ofproto;
985 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
986 hash_pointer(aux, 0));
990 list_init(&bundle->ports);
992 bundle->trunks = NULL;
996 bundle->floodable = true;
998 bundle->src_mirrors = 0;
999 bundle->dst_mirrors = 0;
1000 bundle->mirror_out = 0;
1003 if (!bundle->name || strcmp(s->name, bundle->name)) {
1005 bundle->name = xstrdup(s->name);
1010 if (!bundle->lacp) {
1011 bundle->lacp = lacp_create();
1013 lacp_configure(bundle->lacp, s->lacp);
1015 lacp_destroy(bundle->lacp);
1016 bundle->lacp = NULL;
1019 /* Update set of ports. */
1021 for (i = 0; i < s->n_slaves; i++) {
1022 if (!bundle_add_port(bundle, s->slaves[i],
1023 s->lacp ? &s->lacp_slaves[i] : NULL,
1024 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1028 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1029 struct ofport_dpif *next_port;
1031 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1032 for (i = 0; i < s->n_slaves; i++) {
1033 if (s->slaves[i] == port->up.ofp_port) {
1038 bundle_del_port(port);
1042 assert(list_size(&bundle->ports) <= s->n_slaves);
1044 if (list_is_empty(&bundle->ports)) {
1045 bundle_destroy(bundle);
1050 if (s->vlan != bundle->vlan) {
1051 bundle->vlan = s->vlan;
1055 /* Get trunked VLANs. */
1056 trunks = s->vlan == -1 ? NULL : s->trunks;
1057 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1058 free(bundle->trunks);
1059 bundle->trunks = vlan_bitmap_clone(trunks);
1064 if (!list_is_short(&bundle->ports)) {
1065 bundle->ofproto->has_bonded_bundles = true;
1067 if (bond_reconfigure(bundle->bond, s->bond)) {
1068 ofproto->need_revalidate = true;
1071 bundle->bond = bond_create(s->bond);
1072 ofproto->need_revalidate = true;
1075 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1076 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1080 bond_destroy(bundle->bond);
1081 bundle->bond = NULL;
1084 /* If we changed something that would affect MAC learning, un-learn
1085 * everything on this port and force flow revalidation. */
1087 bundle_flush_macs(bundle);
1094 bundle_remove(struct ofport *port_)
1096 struct ofport_dpif *port = ofport_dpif_cast(port_);
1097 struct ofbundle *bundle = port->bundle;
1100 bundle_del_port(port);
1101 if (list_is_empty(&bundle->ports)) {
1102 bundle_destroy(bundle);
1103 } else if (list_is_short(&bundle->ports)) {
1104 bond_destroy(bundle->bond);
1105 bundle->bond = NULL;
1111 send_pdu_cb(void *port_, const struct lacp_pdu *pdu)
1113 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1114 struct ofport_dpif *port = port_;
1115 uint8_t ea[ETH_ADDR_LEN];
1118 error = netdev_get_etheraddr(port->up.netdev, ea);
1120 struct lacp_pdu *packet_pdu;
1121 struct ofpbuf packet;
1123 ofpbuf_init(&packet, 0);
1124 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1125 sizeof *packet_pdu);
1127 error = netdev_send(port->up.netdev, &packet);
1129 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1130 "(%s)", port->bundle->name,
1131 netdev_get_name(port->up.netdev), strerror(error));
1133 ofpbuf_uninit(&packet);
1135 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1136 "%s (%s)", port->bundle->name,
1137 netdev_get_name(port->up.netdev), strerror(error));
1142 bundle_send_learning_packets(struct ofbundle *bundle)
1144 struct ofproto_dpif *ofproto = bundle->ofproto;
1145 int error, n_packets, n_errors;
1146 struct mac_entry *e;
1148 error = n_packets = n_errors = 0;
1149 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1150 if (e->port.p != bundle) {
1151 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1161 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1162 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1163 "packets, last error was: %s",
1164 bundle->name, n_errors, n_packets, strerror(error));
1166 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1167 bundle->name, n_packets);
1172 bundle_run(struct ofbundle *bundle)
1175 lacp_run(bundle->lacp, send_pdu_cb);
1178 struct ofport_dpif *port;
1180 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1181 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1184 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1185 lacp_negotiated(bundle->lacp));
1186 if (bond_should_send_learning_packets(bundle->bond)) {
1187 bundle_send_learning_packets(bundle);
1193 bundle_wait(struct ofbundle *bundle)
1196 lacp_wait(bundle->lacp);
1199 bond_wait(bundle->bond);
1206 mirror_scan(struct ofproto_dpif *ofproto)
1210 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1211 if (!ofproto->mirrors[idx]) {
1218 static struct ofmirror *
1219 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1223 for (i = 0; i < MAX_MIRRORS; i++) {
1224 struct ofmirror *mirror = ofproto->mirrors[i];
1225 if (mirror && mirror->aux == aux) {
1234 mirror_set(struct ofproto *ofproto_, void *aux,
1235 const struct ofproto_mirror_settings *s)
1237 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1238 mirror_mask_t mirror_bit;
1239 struct ofbundle *bundle;
1240 struct ofmirror *mirror;
1241 struct ofbundle *out;
1242 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1243 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1246 mirror = mirror_lookup(ofproto, aux);
1248 mirror_destroy(mirror);
1254 idx = mirror_scan(ofproto);
1256 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1258 ofproto->up.name, MAX_MIRRORS, s->name);
1262 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1263 mirror->ofproto = ofproto;
1266 mirror->out_vlan = -1;
1267 mirror->name = NULL;
1270 if (!mirror->name || strcmp(s->name, mirror->name)) {
1272 mirror->name = xstrdup(s->name);
1275 /* Get the new configuration. */
1276 if (s->out_bundle) {
1277 out = bundle_lookup(ofproto, s->out_bundle);
1279 mirror_destroy(mirror);
1285 out_vlan = s->out_vlan;
1287 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1288 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1290 /* If the configuration has not changed, do nothing. */
1291 if (hmapx_equals(&srcs, &mirror->srcs)
1292 && hmapx_equals(&dsts, &mirror->dsts)
1293 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1294 && mirror->out == out
1295 && mirror->out_vlan == out_vlan)
1297 hmapx_destroy(&srcs);
1298 hmapx_destroy(&dsts);
1302 hmapx_swap(&srcs, &mirror->srcs);
1303 hmapx_destroy(&srcs);
1305 hmapx_swap(&dsts, &mirror->dsts);
1306 hmapx_destroy(&dsts);
1308 free(mirror->vlans);
1309 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1312 mirror->out_vlan = out_vlan;
1314 /* Update bundles. */
1315 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1316 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1317 if (hmapx_contains(&mirror->srcs, bundle)) {
1318 bundle->src_mirrors |= mirror_bit;
1320 bundle->src_mirrors &= ~mirror_bit;
1323 if (hmapx_contains(&mirror->dsts, bundle)) {
1324 bundle->dst_mirrors |= mirror_bit;
1326 bundle->dst_mirrors &= ~mirror_bit;
1329 if (mirror->out == bundle) {
1330 bundle->mirror_out |= mirror_bit;
1332 bundle->mirror_out &= ~mirror_bit;
1336 ofproto->need_revalidate = true;
1337 mac_learning_flush(ofproto->ml);
1343 mirror_destroy(struct ofmirror *mirror)
1345 struct ofproto_dpif *ofproto;
1346 mirror_mask_t mirror_bit;
1347 struct ofbundle *bundle;
1353 ofproto = mirror->ofproto;
1354 ofproto->need_revalidate = true;
1355 mac_learning_flush(ofproto->ml);
1357 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1358 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1359 bundle->src_mirrors &= ~mirror_bit;
1360 bundle->dst_mirrors &= ~mirror_bit;
1361 bundle->mirror_out &= ~mirror_bit;
1364 hmapx_destroy(&mirror->srcs);
1365 hmapx_destroy(&mirror->dsts);
1366 free(mirror->vlans);
1368 ofproto->mirrors[mirror->idx] = NULL;
1374 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1376 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1377 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1378 ofproto->need_revalidate = true;
1379 mac_learning_flush(ofproto->ml);
1385 is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
1387 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1388 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1389 return bundle && bundle->mirror_out != 0;
1393 forward_bpdu_changed(struct ofproto *ofproto_)
1395 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1396 /* Revalidate cached flows whenever forward_bpdu option changes. */
1397 ofproto->need_revalidate = true;
1402 static struct ofport_dpif *
1403 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1405 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1406 return ofport ? ofport_dpif_cast(ofport) : NULL;
1409 static struct ofport_dpif *
1410 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1412 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1416 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1417 struct dpif_port *dpif_port)
1419 ofproto_port->name = dpif_port->name;
1420 ofproto_port->type = dpif_port->type;
1421 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1425 port_run(struct ofport_dpif *ofport)
1427 bool enable = netdev_get_carrier(ofport->up.netdev);
1430 cfm_run(ofport->cfm);
1432 if (cfm_should_send_ccm(ofport->cfm)) {
1433 struct ofpbuf packet;
1435 ofpbuf_init(&packet, 0);
1436 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1437 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1438 ofport->odp_port, &packet);
1439 ofpbuf_uninit(&packet);
1442 enable = enable && !cfm_get_fault(ofport->cfm);
1445 if (ofport->bundle) {
1446 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1449 if (ofport->may_enable != enable) {
1450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1452 if (ofproto->has_bundle_action) {
1453 ofproto->need_revalidate = true;
1457 ofport->may_enable = enable;
1461 port_wait(struct ofport_dpif *ofport)
1464 cfm_wait(ofport->cfm);
1469 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1470 struct ofproto_port *ofproto_port)
1472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1473 struct dpif_port dpif_port;
1476 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1478 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1484 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1486 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1490 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1492 *ofp_portp = odp_port_to_ofp_port(odp_port);
1498 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1500 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1503 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1505 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1507 /* The caller is going to close ofport->up.netdev. If this is a
1508 * bonded port, then the bond is using that netdev, so remove it
1509 * from the bond. The client will need to reconfigure everything
1510 * after deleting ports, so then the slave will get re-added. */
1511 bundle_remove(&ofport->up);
1517 struct port_dump_state {
1518 struct dpif_port_dump dump;
1523 port_dump_start(const struct ofproto *ofproto_, void **statep)
1525 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1526 struct port_dump_state *state;
1528 *statep = state = xmalloc(sizeof *state);
1529 dpif_port_dump_start(&state->dump, ofproto->dpif);
1530 state->done = false;
1535 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1536 struct ofproto_port *port)
1538 struct port_dump_state *state = state_;
1539 struct dpif_port dpif_port;
1541 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1542 ofproto_port_from_dpif_port(port, &dpif_port);
1545 int error = dpif_port_dump_done(&state->dump);
1547 return error ? error : EOF;
1552 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1554 struct port_dump_state *state = state_;
1557 dpif_port_dump_done(&state->dump);
1564 port_poll(const struct ofproto *ofproto_, char **devnamep)
1566 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1567 return dpif_port_poll(ofproto->dpif, devnamep);
1571 port_poll_wait(const struct ofproto *ofproto_)
1573 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1574 dpif_port_poll_wait(ofproto->dpif);
1578 port_is_lacp_current(const struct ofport *ofport_)
1580 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1581 return (ofport->bundle && ofport->bundle->lacp
1582 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
1586 /* Upcall handling. */
1588 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
1589 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
1590 * their individual configurations.
1592 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
1593 * Otherwise, ownership is transferred to this function. */
1595 send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
1596 const struct flow *flow, bool clone)
1598 struct ofputil_packet_in pin;
1600 pin.packet = upcall->packet;
1601 pin.in_port = flow->in_port;
1602 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
1603 pin.buffer_id = 0; /* not yet known */
1604 pin.send_len = upcall->userdata;
1605 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
1606 clone ? NULL : upcall->packet);
1610 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
1611 const struct ofpbuf *packet)
1613 if (cfm_should_process_flow(flow)) {
1614 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
1615 if (packet && ofport && ofport->cfm) {
1616 cfm_process_heartbeat(ofport->cfm, packet);
1619 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
1620 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
1621 if (packet && port && port->bundle && port->bundle->lacp) {
1622 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
1624 lacp_process_pdu(port->bundle->lacp, port, pdu);
1633 handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1635 struct facet *facet;
1638 /* Obtain in_port and tun_id, at least. */
1639 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1641 /* Set header pointers in 'flow'. */
1642 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
1644 /* Handle 802.1ag and LACP. */
1645 if (process_special(ofproto, &flow, upcall->packet)) {
1646 ofpbuf_delete(upcall->packet);
1647 ofproto->n_matches++;
1651 /* Check with in-band control to see if this packet should be sent
1652 * to the local port regardless of the flow table. */
1653 if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
1654 send_packet(ofproto, ODPP_LOCAL, upcall->packet);
1657 facet = facet_lookup_valid(ofproto, &flow);
1659 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow, 0);
1661 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
1662 struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
1664 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
1665 COVERAGE_INC(ofproto_dpif_no_packet_in);
1666 /* XXX install 'drop' flow entry */
1667 ofpbuf_delete(upcall->packet);
1671 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
1675 send_packet_in(ofproto, upcall, &flow, false);
1679 facet = facet_create(rule, &flow, upcall->packet);
1680 } else if (!facet->may_install) {
1681 /* The facet is not installable, that is, we need to process every
1682 * packet, so process the current packet's actions into 'facet'. */
1683 facet_make_actions(ofproto, facet, upcall->packet);
1686 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
1688 * Extra-special case for fail-open mode.
1690 * We are in fail-open mode and the packet matched the fail-open rule,
1691 * but we are connected to a controller too. We should send the packet
1692 * up to the controller in the hope that it will try to set up a flow
1693 * and thereby allow us to exit fail-open.
1695 * See the top-level comment in fail-open.c for more information.
1697 send_packet_in(ofproto, upcall, &flow, true);
1700 facet_execute(ofproto, facet, upcall->packet);
1701 facet_install(ofproto, facet, false);
1702 ofproto->n_matches++;
1706 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1710 switch (upcall->type) {
1711 case DPIF_UC_ACTION:
1712 COVERAGE_INC(ofproto_dpif_ctlr_action);
1713 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1714 send_packet_in(ofproto, upcall, &flow, false);
1717 case DPIF_UC_SAMPLE:
1718 if (ofproto->sflow) {
1719 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1720 dpif_sflow_received(ofproto->sflow, upcall, &flow);
1722 ofpbuf_delete(upcall->packet);
1726 handle_miss_upcall(ofproto, upcall);
1729 case DPIF_N_UC_TYPES:
1731 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
1736 /* Flow expiration. */
1738 static int facet_max_idle(const struct ofproto_dpif *);
1739 static void update_stats(struct ofproto_dpif *);
1740 static void rule_expire(struct rule_dpif *);
1741 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
1743 /* This function is called periodically by run(). Its job is to collect
1744 * updates for the flows that have been installed into the datapath, most
1745 * importantly when they last were used, and then use that information to
1746 * expire flows that have not been used recently.
1748 * Returns the number of milliseconds after which it should be called again. */
1750 expire(struct ofproto_dpif *ofproto)
1752 struct rule_dpif *rule, *next_rule;
1753 struct classifier *table;
1756 /* Update stats for each flow in the datapath. */
1757 update_stats(ofproto);
1759 /* Expire facets that have been idle too long. */
1760 dp_max_idle = facet_max_idle(ofproto);
1761 expire_facets(ofproto, dp_max_idle);
1763 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
1764 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1765 struct cls_cursor cursor;
1767 cls_cursor_init(&cursor, table, NULL);
1768 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1773 /* All outstanding data in existing flows has been accounted, so it's a
1774 * good time to do bond rebalancing. */
1775 if (ofproto->has_bonded_bundles) {
1776 struct ofbundle *bundle;
1778 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1780 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
1785 return MIN(dp_max_idle, 1000);
1788 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
1790 * This function also pushes statistics updates to rules which each facet
1791 * resubmits into. Generally these statistics will be accurate. However, if a
1792 * facet changes the rule it resubmits into at some time in between
1793 * update_stats() runs, it is possible that statistics accrued to the
1794 * old rule will be incorrectly attributed to the new rule. This could be
1795 * avoided by calling update_stats() whenever rules are created or
1796 * deleted. However, the performance impact of making so many calls to the
1797 * datapath do not justify the benefit of having perfectly accurate statistics.
1800 update_stats(struct ofproto_dpif *p)
1802 const struct dpif_flow_stats *stats;
1803 struct dpif_flow_dump dump;
1804 const struct nlattr *key;
1807 dpif_flow_dump_start(&dump, p->dpif);
1808 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
1809 struct facet *facet;
1812 if (odp_flow_key_to_flow(key, key_len, &flow)) {
1816 odp_flow_key_format(key, key_len, &s);
1817 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
1823 facet = facet_find(p, &flow);
1825 if (facet && facet->installed) {
1827 if (stats->n_packets >= facet->dp_packet_count) {
1828 uint64_t extra = stats->n_packets - facet->dp_packet_count;
1829 facet->packet_count += extra;
1831 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
1834 if (stats->n_bytes >= facet->dp_byte_count) {
1835 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
1837 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
1840 facet->dp_packet_count = stats->n_packets;
1841 facet->dp_byte_count = stats->n_bytes;
1843 facet_update_time(p, facet, stats->used);
1844 facet_account(p, facet);
1845 facet_push_stats(facet);
1847 /* There's a flow in the datapath that we know nothing about.
1849 COVERAGE_INC(facet_unexpected);
1850 dpif_flow_del(p->dpif, key, key_len, NULL);
1853 dpif_flow_dump_done(&dump);
1856 /* Calculates and returns the number of milliseconds of idle time after which
1857 * facets should expire from the datapath and we should fold their statistics
1858 * into their parent rules in userspace. */
1860 facet_max_idle(const struct ofproto_dpif *ofproto)
1863 * Idle time histogram.
1865 * Most of the time a switch has a relatively small number of facets. When
1866 * this is the case we might as well keep statistics for all of them in
1867 * userspace and to cache them in the kernel datapath for performance as
1870 * As the number of facets increases, the memory required to maintain
1871 * statistics about them in userspace and in the kernel becomes
1872 * significant. However, with a large number of facets it is likely that
1873 * only a few of them are "heavy hitters" that consume a large amount of
1874 * bandwidth. At this point, only heavy hitters are worth caching in the
1875 * kernel and maintaining in userspaces; other facets we can discard.
1877 * The technique used to compute the idle time is to build a histogram with
1878 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
1879 * that is installed in the kernel gets dropped in the appropriate bucket.
1880 * After the histogram has been built, we compute the cutoff so that only
1881 * the most-recently-used 1% of facets (but at least
1882 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
1883 * the most-recently-used bucket of facets is kept, so actually an
1884 * arbitrary number of facets can be kept in any given expiration run
1885 * (though the next run will delete most of those unless they receive
1888 * This requires a second pass through the facets, in addition to the pass
1889 * made by update_stats(), because the former function never looks
1890 * at uninstallable facets.
1892 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
1893 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
1894 int buckets[N_BUCKETS] = { 0 };
1895 int total, subtotal, bucket;
1896 struct facet *facet;
1900 total = hmap_count(&ofproto->facets);
1901 if (total <= ofproto->up.flow_eviction_threshold) {
1902 return N_BUCKETS * BUCKET_WIDTH;
1905 /* Build histogram. */
1907 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1908 long long int idle = now - facet->used;
1909 int bucket = (idle <= 0 ? 0
1910 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
1911 : (unsigned int) idle / BUCKET_WIDTH);
1915 /* Find the first bucket whose flows should be expired. */
1916 subtotal = bucket = 0;
1918 subtotal += buckets[bucket++];
1919 } while (bucket < N_BUCKETS &&
1920 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
1922 if (VLOG_IS_DBG_ENABLED()) {
1926 ds_put_cstr(&s, "keep");
1927 for (i = 0; i < N_BUCKETS; i++) {
1929 ds_put_cstr(&s, ", drop");
1932 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
1935 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
1939 return bucket * BUCKET_WIDTH;
1943 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
1945 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
1946 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
1947 struct ofexpired expired;
1949 if (facet->installed) {
1950 struct dpif_flow_stats stats;
1952 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
1954 facet_update_stats(ofproto, facet, &stats);
1957 expired.flow = facet->flow;
1958 expired.packet_count = facet->packet_count;
1959 expired.byte_count = facet->byte_count;
1960 expired.used = facet->used;
1961 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1966 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
1968 long long int cutoff = time_msec() - dp_max_idle;
1969 struct facet *facet, *next_facet;
1971 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1972 facet_active_timeout(ofproto, facet);
1973 if (facet->used < cutoff) {
1974 facet_remove(ofproto, facet);
1979 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
1980 * then delete it entirely. */
1982 rule_expire(struct rule_dpif *rule)
1984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
1985 struct facet *facet, *next_facet;
1989 /* Has 'rule' expired? */
1991 if (rule->up.hard_timeout
1992 && now > rule->up.created + rule->up.hard_timeout * 1000) {
1993 reason = OFPRR_HARD_TIMEOUT;
1994 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
1995 && now > rule->used + rule->up.idle_timeout * 1000) {
1996 reason = OFPRR_IDLE_TIMEOUT;
2001 COVERAGE_INC(ofproto_dpif_expired);
2003 /* Update stats. (This is a no-op if the rule expired due to an idle
2004 * timeout, because that only happens when the rule has no facets left.) */
2005 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2006 facet_remove(ofproto, facet);
2009 /* Get rid of the rule. */
2010 ofproto_rule_expire(&rule->up, reason);
2015 /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
2016 * example 'packet' within that flow.
2018 * The caller must already have determined that no facet with an identical
2019 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2020 * the ofproto's classifier table. */
2021 static struct facet *
2022 facet_create(struct rule_dpif *rule, const struct flow *flow,
2023 const struct ofpbuf *packet)
2025 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2026 struct facet *facet;
2028 facet = xzalloc(sizeof *facet);
2029 facet->used = time_msec();
2030 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2031 list_push_back(&rule->facets, &facet->list_node);
2033 facet->flow = *flow;
2034 netflow_flow_init(&facet->nf_flow);
2035 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2037 facet_make_actions(ofproto, facet, packet);
2043 facet_free(struct facet *facet)
2045 free(facet->actions);
2049 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2050 * 'packet', which arrived on 'in_port'.
2052 * Takes ownership of 'packet'. */
2054 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2055 const struct nlattr *odp_actions, size_t actions_len,
2056 struct ofpbuf *packet)
2058 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2059 && odp_actions->nla_type == ODP_ACTION_ATTR_USERSPACE) {
2060 /* As an optimization, avoid a round-trip from userspace to kernel to
2061 * userspace. This also avoids possibly filling up kernel packet
2062 * buffers along the way. */
2063 struct dpif_upcall upcall;
2065 upcall.type = DPIF_UC_ACTION;
2066 upcall.packet = packet;
2069 upcall.userdata = nl_attr_get_u64(odp_actions);
2070 upcall.sample_pool = 0;
2071 upcall.actions = NULL;
2072 upcall.actions_len = 0;
2074 send_packet_in(ofproto, &upcall, flow, false);
2078 struct odputil_keybuf keybuf;
2082 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2083 odp_flow_key_from_flow(&key, flow);
2085 error = dpif_execute(ofproto->dpif, key.data, key.size,
2086 odp_actions, actions_len, packet);
2088 ofpbuf_delete(packet);
2093 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2094 * statistics appropriately. 'packet' must have at least sizeof(struct
2095 * ofp_packet_in) bytes of headroom.
2097 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2098 * applying flow_extract() to 'packet' would yield the same flow as
2101 * 'facet' must have accurately composed ODP actions; that is, it must not be
2102 * in need of revalidation.
2104 * Takes ownership of 'packet'. */
2106 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2107 struct ofpbuf *packet)
2109 struct dpif_flow_stats stats;
2111 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2113 flow_extract_stats(&facet->flow, packet, &stats);
2114 stats.used = time_msec();
2115 if (execute_odp_actions(ofproto, &facet->flow,
2116 facet->actions, facet->actions_len, packet)) {
2117 facet_update_stats(ofproto, facet, &stats);
2121 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2123 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2124 * rule's statistics, via facet_uninstall().
2126 * - Removes 'facet' from its rule and from ofproto->facets.
2129 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2131 facet_uninstall(ofproto, facet);
2132 facet_flush_stats(ofproto, facet);
2133 hmap_remove(&ofproto->facets, &facet->hmap_node);
2134 list_remove(&facet->list_node);
2138 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2140 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2141 const struct ofpbuf *packet)
2143 const struct rule_dpif *rule = facet->rule;
2144 struct ofpbuf *odp_actions;
2145 struct action_xlate_ctx ctx;
2147 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2148 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2149 facet->tags = ctx.tags;
2150 facet->may_install = ctx.may_set_up_flow;
2151 facet->nf_flow.output_iface = ctx.nf_output_iface;
2153 if (facet->actions_len != odp_actions->size
2154 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2155 free(facet->actions);
2156 facet->actions_len = odp_actions->size;
2157 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2160 ofpbuf_delete(odp_actions);
2163 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2164 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2165 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2166 * since 'facet' was last updated.
2168 * Returns 0 if successful, otherwise a positive errno value.*/
2170 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2171 const struct nlattr *actions, size_t actions_len,
2172 struct dpif_flow_stats *stats)
2174 struct odputil_keybuf keybuf;
2175 enum dpif_flow_put_flags flags;
2179 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2181 flags |= DPIF_FP_ZERO_STATS;
2184 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2185 odp_flow_key_from_flow(&key, &facet->flow);
2187 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2188 actions, actions_len, stats);
2191 facet_reset_dp_stats(facet, stats);
2197 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2198 * 'zero_stats' is true, clears any existing statistics from the datapath for
2201 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2203 struct dpif_flow_stats stats;
2205 if (facet->may_install
2206 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2207 zero_stats ? &stats : NULL)) {
2208 facet->installed = true;
2213 vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci)
2215 return vlan_tci != htons(0) ? vlan_tci_to_vid(vlan_tci) : OFP_VLAN_NONE;
2219 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
2222 struct ofbundle *in_bundle;
2223 const struct nlattr *a;
2229 if (facet->byte_count <= facet->accounted_bytes) {
2232 n_bytes = facet->byte_count - facet->accounted_bytes;
2233 facet->accounted_bytes = facet->byte_count;
2235 /* Test that 'tags' is nonzero to ensure that only flows that include an
2236 * OFPP_NORMAL action are used for learning and bond slave rebalancing.
2237 * This works because OFPP_NORMAL always sets a nonzero tag value.
2239 * Feed information from the active flows back into the learning table to
2240 * ensure that table is always in sync with what is actually flowing
2241 * through the datapath. */
2243 || !is_admissible(ofproto, &facet->flow, false, &dummy,
2244 &vlan, &in_bundle)) {
2248 update_learning_table(ofproto, &facet->flow, vlan, in_bundle);
2250 if (!ofproto->has_bonded_bundles) {
2254 /* This loop feeds byte counters to bond_account() for rebalancing to use
2255 * as a basis. We also need to track the actual VLAN on which the packet
2256 * is going to be sent to ensure that it matches the one passed to
2257 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2259 vlan_tci = facet->flow.vlan_tci;
2260 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2261 struct ofport_dpif *port;
2263 switch (nl_attr_type(a)) {
2264 case ODP_ACTION_ATTR_OUTPUT:
2265 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2266 if (port && port->bundle && port->bundle->bond) {
2267 bond_account(port->bundle->bond, &facet->flow,
2268 vlan_tci_to_openflow_vlan(vlan_tci), n_bytes);
2272 case ODP_ACTION_ATTR_STRIP_VLAN:
2273 vlan_tci = htons(0);
2276 case ODP_ACTION_ATTR_SET_DL_TCI:
2277 vlan_tci = nl_attr_get_be16(a);
2283 /* If 'rule' is installed in the datapath, uninstalls it. */
2285 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2287 if (facet->installed) {
2288 struct odputil_keybuf keybuf;
2289 struct dpif_flow_stats stats;
2293 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2294 odp_flow_key_from_flow(&key, &facet->flow);
2296 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2297 facet_reset_dp_stats(facet, &stats);
2299 facet_update_stats(p, facet, &stats);
2301 facet->installed = false;
2303 assert(facet->dp_packet_count == 0);
2304 assert(facet->dp_byte_count == 0);
2308 /* Returns true if the only action for 'facet' is to send to the controller.
2309 * (We don't report NetFlow expiration messages for such facets because they
2310 * are just part of the control logic for the network, not real traffic). */
2312 facet_is_controller_flow(struct facet *facet)
2315 && facet->rule->up.n_actions == 1
2316 && action_outputs_to_port(&facet->rule->up.actions[0],
2317 htons(OFPP_CONTROLLER)));
2320 /* Resets 'facet''s datapath statistics counters. This should be called when
2321 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2322 * it should contain the statistics returned by dpif when 'facet' was reset in
2323 * the datapath. 'stats' will be modified to only included statistics new
2324 * since 'facet' was last updated. */
2326 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2328 if (stats && facet->dp_packet_count <= stats->n_packets
2329 && facet->dp_byte_count <= stats->n_bytes) {
2330 stats->n_packets -= facet->dp_packet_count;
2331 stats->n_bytes -= facet->dp_byte_count;
2334 facet->dp_packet_count = 0;
2335 facet->dp_byte_count = 0;
2338 /* Folds all of 'facet''s statistics into its rule. Also updates the
2339 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2340 * 'facet''s statistics in the datapath should have been zeroed and folded into
2341 * its packet and byte counts before this function is called. */
2343 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2345 assert(!facet->dp_byte_count);
2346 assert(!facet->dp_packet_count);
2348 facet_push_stats(facet);
2349 facet_account(ofproto, facet);
2351 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2352 struct ofexpired expired;
2353 expired.flow = facet->flow;
2354 expired.packet_count = facet->packet_count;
2355 expired.byte_count = facet->byte_count;
2356 expired.used = facet->used;
2357 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2360 facet->rule->packet_count += facet->packet_count;
2361 facet->rule->byte_count += facet->byte_count;
2363 /* Reset counters to prevent double counting if 'facet' ever gets
2365 facet_reset_counters(facet);
2367 netflow_flow_clear(&facet->nf_flow);
2370 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2371 * Returns it if found, otherwise a null pointer.
2373 * The returned facet might need revalidation; use facet_lookup_valid()
2374 * instead if that is important. */
2375 static struct facet *
2376 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
2378 struct facet *facet;
2380 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2382 if (flow_equal(flow, &facet->flow)) {
2390 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2391 * Returns it if found, otherwise a null pointer.
2393 * The returned facet is guaranteed to be valid. */
2394 static struct facet *
2395 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
2397 struct facet *facet = facet_find(ofproto, flow);
2399 /* The facet we found might not be valid, since we could be in need of
2400 * revalidation. If it is not valid, don't return it. */
2402 && ofproto->need_revalidate
2403 && !facet_revalidate(ofproto, facet)) {
2404 COVERAGE_INC(facet_invalidated);
2411 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2413 * - If the rule found is different from 'facet''s current rule, moves
2414 * 'facet' to the new rule and recompiles its actions.
2416 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2417 * where it is and recompiles its actions anyway.
2419 * - If there is none, destroys 'facet'.
2421 * Returns true if 'facet' still exists, false if it has been destroyed. */
2423 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
2425 struct action_xlate_ctx ctx;
2426 struct ofpbuf *odp_actions;
2427 struct rule_dpif *new_rule;
2428 bool actions_changed;
2430 COVERAGE_INC(facet_revalidate);
2432 /* Determine the new rule. */
2433 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
2435 /* No new rule, so delete the facet. */
2436 facet_remove(ofproto, facet);
2440 /* Calculate new ODP actions.
2442 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2443 * emit a NetFlow expiration and, if so, we need to have the old state
2444 * around to properly compose it. */
2445 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2446 odp_actions = xlate_actions(&ctx,
2447 new_rule->up.actions, new_rule->up.n_actions);
2448 actions_changed = (facet->actions_len != odp_actions->size
2449 || memcmp(facet->actions, odp_actions->data,
2450 facet->actions_len));
2452 /* If the ODP actions changed or the installability changed, then we need
2453 * to talk to the datapath. */
2454 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2455 if (ctx.may_set_up_flow) {
2456 struct dpif_flow_stats stats;
2458 facet_put__(ofproto, facet,
2459 odp_actions->data, odp_actions->size, &stats);
2460 facet_update_stats(ofproto, facet, &stats);
2462 facet_uninstall(ofproto, facet);
2465 /* The datapath flow is gone or has zeroed stats, so push stats out of
2466 * 'facet' into 'rule'. */
2467 facet_flush_stats(ofproto, facet);
2470 /* Update 'facet' now that we've taken care of all the old state. */
2471 facet->tags = ctx.tags;
2472 facet->nf_flow.output_iface = ctx.nf_output_iface;
2473 facet->may_install = ctx.may_set_up_flow;
2474 if (actions_changed) {
2475 free(facet->actions);
2476 facet->actions_len = odp_actions->size;
2477 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2479 if (facet->rule != new_rule) {
2480 COVERAGE_INC(facet_changed_rule);
2481 list_remove(&facet->list_node);
2482 list_push_back(&new_rule->facets, &facet->list_node);
2483 facet->rule = new_rule;
2484 facet->used = new_rule->up.created;
2485 facet->rs_used = facet->used;
2488 ofpbuf_delete(odp_actions);
2493 /* Updates 'facet''s used time. Caller is responsible for calling
2494 * facet_push_stats() to update the flows which 'facet' resubmits into. */
2496 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
2499 if (used > facet->used) {
2501 if (used > facet->rule->used) {
2502 facet->rule->used = used;
2504 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
2508 /* Folds the statistics from 'stats' into the counters in 'facet'.
2510 * Because of the meaning of a facet's counters, it only makes sense to do this
2511 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
2512 * packet that was sent by hand or if it represents statistics that have been
2513 * cleared out of the datapath. */
2515 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
2516 const struct dpif_flow_stats *stats)
2518 if (stats->n_packets || stats->used > facet->used) {
2519 facet_update_time(ofproto, facet, stats->used);
2520 facet->packet_count += stats->n_packets;
2521 facet->byte_count += stats->n_bytes;
2522 facet_push_stats(facet);
2523 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
2528 facet_reset_counters(struct facet *facet)
2530 facet->packet_count = 0;
2531 facet->byte_count = 0;
2532 facet->rs_packet_count = 0;
2533 facet->rs_byte_count = 0;
2534 facet->accounted_bytes = 0;
2538 facet_push_stats(struct facet *facet)
2540 uint64_t rs_packets, rs_bytes;
2542 assert(facet->packet_count >= facet->rs_packet_count);
2543 assert(facet->byte_count >= facet->rs_byte_count);
2544 assert(facet->used >= facet->rs_used);
2546 rs_packets = facet->packet_count - facet->rs_packet_count;
2547 rs_bytes = facet->byte_count - facet->rs_byte_count;
2549 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
2550 facet->rs_packet_count = facet->packet_count;
2551 facet->rs_byte_count = facet->byte_count;
2552 facet->rs_used = facet->used;
2554 flow_push_stats(facet->rule, &facet->flow,
2555 rs_packets, rs_bytes, facet->used);
2559 struct ofproto_push {
2560 struct action_xlate_ctx ctx;
2567 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
2569 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
2572 rule->packet_count += push->packets;
2573 rule->byte_count += push->bytes;
2574 rule->used = MAX(push->used, rule->used);
2578 /* Pushes flow statistics to the rules which 'flow' resubmits into given
2579 * 'rule''s actions. */
2581 flow_push_stats(const struct rule_dpif *rule,
2582 struct flow *flow, uint64_t packets, uint64_t bytes,
2585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2586 struct ofproto_push push;
2588 push.packets = packets;
2592 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
2593 push.ctx.resubmit_hook = push_resubmit;
2594 ofpbuf_delete(xlate_actions(&push.ctx,
2595 rule->up.actions, rule->up.n_actions));
2600 static struct rule_dpif *
2601 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
2604 return rule_dpif_cast(rule_from_cls_rule(
2605 classifier_lookup(&ofproto->up.tables[table_id],
2610 complete_operation(struct rule_dpif *rule)
2612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2614 ofproto->need_revalidate = true;
2616 struct dpif_completion *c = xmalloc(sizeof *c);
2617 c->op = rule->up.pending;
2618 list_push_back(&ofproto->completions, &c->list_node);
2620 ofoperation_complete(rule->up.pending, 0);
2624 static struct rule *
2627 struct rule_dpif *rule = xmalloc(sizeof *rule);
2632 rule_dealloc(struct rule *rule_)
2634 struct rule_dpif *rule = rule_dpif_cast(rule_);
2639 rule_construct(struct rule *rule_)
2641 struct rule_dpif *rule = rule_dpif_cast(rule_);
2642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2643 struct rule_dpif *victim;
2646 error = validate_actions(rule->up.actions, rule->up.n_actions,
2647 &rule->up.cr.flow, ofproto->max_ports);
2652 rule->used = rule->up.created;
2653 rule->packet_count = 0;
2654 rule->byte_count = 0;
2656 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
2657 if (victim && !list_is_empty(&victim->facets)) {
2658 struct facet *facet;
2660 rule->facets = victim->facets;
2661 list_moved(&rule->facets);
2662 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2663 /* XXX: We're only clearing our local counters here. It's possible
2664 * that quite a few packets are unaccounted for in the datapath
2665 * statistics. These will be accounted to the new rule instead of
2666 * cleared as required. This could be fixed by clearing out the
2667 * datapath statistics for this facet, but currently it doesn't
2669 facet_reset_counters(facet);
2673 /* Must avoid list_moved() in this case. */
2674 list_init(&rule->facets);
2677 complete_operation(rule);
2682 rule_destruct(struct rule *rule_)
2684 struct rule_dpif *rule = rule_dpif_cast(rule_);
2685 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2686 struct facet *facet, *next_facet;
2688 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2689 facet_revalidate(ofproto, facet);
2692 complete_operation(rule);
2696 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
2698 struct rule_dpif *rule = rule_dpif_cast(rule_);
2699 struct facet *facet;
2701 /* Start from historical data for 'rule' itself that are no longer tracked
2702 * in facets. This counts, for example, facets that have expired. */
2703 *packets = rule->packet_count;
2704 *bytes = rule->byte_count;
2706 /* Add any statistics that are tracked by facets. This includes
2707 * statistical data recently updated by ofproto_update_stats() as well as
2708 * stats for packets that were executed "by hand" via dpif_execute(). */
2709 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2710 *packets += facet->packet_count;
2711 *bytes += facet->byte_count;
2716 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
2718 struct rule_dpif *rule = rule_dpif_cast(rule_);
2719 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2720 struct action_xlate_ctx ctx;
2721 struct ofpbuf *odp_actions;
2722 struct facet *facet;
2725 /* First look for a related facet. If we find one, account it to that. */
2726 facet = facet_lookup_valid(ofproto, flow);
2727 if (facet && facet->rule == rule) {
2728 facet_execute(ofproto, facet, packet);
2732 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2733 * create a new facet for it and use that. */
2734 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
2735 facet = facet_create(rule, flow, packet);
2736 facet_execute(ofproto, facet, packet);
2737 facet_install(ofproto, facet, true);
2741 /* We can't account anything to a facet. If we were to try, then that
2742 * facet would have a non-matching rule, busting our invariants. */
2743 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
2744 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2745 size = packet->size;
2746 if (execute_odp_actions(ofproto, flow, odp_actions->data,
2747 odp_actions->size, packet)) {
2748 rule->used = time_msec();
2749 rule->packet_count++;
2750 rule->byte_count += size;
2751 flow_push_stats(rule, flow, 1, size, rule->used);
2753 ofpbuf_delete(odp_actions);
2759 rule_modify_actions(struct rule *rule_)
2761 struct rule_dpif *rule = rule_dpif_cast(rule_);
2762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2765 error = validate_actions(rule->up.actions, rule->up.n_actions,
2766 &rule->up.cr.flow, ofproto->max_ports);
2768 ofoperation_complete(rule->up.pending, error);
2772 complete_operation(rule);
2775 /* Sends 'packet' out of port 'odp_port' within 'p'.
2776 * Returns 0 if successful, otherwise a positive errno value. */
2778 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
2779 const struct ofpbuf *packet)
2781 struct ofpbuf key, odp_actions;
2782 struct odputil_keybuf keybuf;
2786 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
2787 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2788 odp_flow_key_from_flow(&key, &flow);
2790 ofpbuf_init(&odp_actions, 32);
2791 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2792 error = dpif_execute(ofproto->dpif,
2794 odp_actions.data, odp_actions.size,
2796 ofpbuf_uninit(&odp_actions);
2799 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
2800 ofproto->up.name, odp_port, strerror(error));
2805 /* OpenFlow to ODP action translation. */
2807 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2808 struct action_xlate_ctx *ctx);
2809 static void xlate_normal(struct action_xlate_ctx *);
2812 commit_odp_actions(struct action_xlate_ctx *ctx)
2814 const struct flow *flow = &ctx->flow;
2815 struct flow *base = &ctx->base_flow;
2816 struct ofpbuf *odp_actions = ctx->odp_actions;
2818 if (base->tun_id != flow->tun_id) {
2819 nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id);
2820 base->tun_id = flow->tun_id;
2823 if (base->nw_src != flow->nw_src) {
2824 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src);
2825 base->nw_src = flow->nw_src;
2828 if (base->nw_dst != flow->nw_dst) {
2829 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst);
2830 base->nw_dst = flow->nw_dst;
2833 if (base->nw_tos != flow->nw_tos) {
2834 nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos);
2835 base->nw_tos = flow->nw_tos;
2838 if (base->vlan_tci != flow->vlan_tci) {
2839 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
2840 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2842 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2843 flow->vlan_tci & ~htons(VLAN_CFI));
2845 base->vlan_tci = flow->vlan_tci;
2848 if (base->tp_src != flow->tp_src) {
2849 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src);
2850 base->tp_src = flow->tp_src;
2853 if (base->tp_dst != flow->tp_dst) {
2854 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst);
2855 base->tp_dst = flow->tp_dst;
2858 if (!eth_addr_equals(base->dl_src, flow->dl_src)) {
2859 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2860 flow->dl_src, ETH_ADDR_LEN);
2861 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
2864 if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) {
2865 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2866 flow->dl_dst, ETH_ADDR_LEN);
2867 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
2870 if (ctx->base_priority != ctx->priority) {
2871 if (ctx->priority) {
2872 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY,
2875 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2877 ctx->base_priority = ctx->priority;
2882 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
2884 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
2885 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
2888 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
2889 /* Forwarding disabled on port. */
2894 * We don't have an ofport record for this port, but it doesn't hurt to
2895 * allow forwarding to it anyhow. Maybe such a port will appear later
2896 * and we're pre-populating the flow table.
2900 commit_odp_actions(ctx);
2901 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2902 ctx->nf_output_iface = ofp_port;
2906 xlate_table_action(struct action_xlate_ctx *ctx,
2907 uint16_t in_port, uint8_t table_id)
2909 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2910 struct rule_dpif *rule;
2911 uint16_t old_in_port;
2912 uint8_t old_table_id;
2914 old_table_id = ctx->table_id;
2915 ctx->table_id = table_id;
2917 /* Look up a flow with 'in_port' as the input port. Then restore the
2918 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2919 * have surprising behavior). */
2920 old_in_port = ctx->flow.in_port;
2921 ctx->flow.in_port = in_port;
2922 rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow, table_id);
2923 ctx->flow.in_port = old_in_port;
2925 if (ctx->resubmit_hook) {
2926 ctx->resubmit_hook(ctx, rule);
2931 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
2935 ctx->table_id = old_table_id;
2937 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2939 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
2940 MAX_RESUBMIT_RECURSION);
2945 xlate_resubmit_table(struct action_xlate_ctx *ctx,
2946 const struct nx_action_resubmit *nar)
2951 in_port = (nar->in_port == htons(OFPP_IN_PORT)
2953 : ntohs(nar->in_port));
2954 table_id = nar->table == 255 ? ctx->table_id : nar->table;
2956 xlate_table_action(ctx, in_port, table_id);
2960 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
2962 struct ofport_dpif *ofport;
2964 commit_odp_actions(ctx);
2965 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
2966 uint16_t ofp_port = ofport->up.ofp_port;
2967 if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) {
2968 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT,
2973 ctx->nf_output_iface = NF_OUT_FLOOD;
2977 xlate_output_action__(struct action_xlate_ctx *ctx,
2978 uint16_t port, uint16_t max_len)
2980 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2982 ctx->nf_output_iface = NF_OUT_DROP;
2986 add_output_action(ctx, ctx->flow.in_port);
2989 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
2995 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
2998 flood_packets(ctx, htonl(0));
3000 case OFPP_CONTROLLER:
3001 commit_odp_actions(ctx);
3002 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_USERSPACE, max_len);
3005 add_output_action(ctx, OFPP_LOCAL);
3010 if (port != ctx->flow.in_port) {
3011 add_output_action(ctx, port);
3016 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3017 ctx->nf_output_iface = NF_OUT_FLOOD;
3018 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3019 ctx->nf_output_iface = prev_nf_output_iface;
3020 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3021 ctx->nf_output_iface != NF_OUT_FLOOD) {
3022 ctx->nf_output_iface = NF_OUT_MULTI;
3027 xlate_output_reg_action(struct action_xlate_ctx *ctx,
3028 const struct nx_action_output_reg *naor)
3032 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
3034 if (ofp_port <= UINT16_MAX) {
3035 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
3040 xlate_output_action(struct action_xlate_ctx *ctx,
3041 const struct ofp_action_output *oao)
3043 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3047 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3048 const struct ofp_action_enqueue *oae)
3050 uint16_t ofp_port, odp_port;
3051 uint32_t ctx_priority, priority;
3054 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3057 /* Fall back to ordinary output action. */
3058 xlate_output_action__(ctx, ntohs(oae->port), 0);
3062 /* Figure out ODP output port. */
3063 ofp_port = ntohs(oae->port);
3064 if (ofp_port == OFPP_IN_PORT) {
3065 ofp_port = ctx->flow.in_port;
3067 odp_port = ofp_port_to_odp_port(ofp_port);
3069 /* Add ODP actions. */
3070 ctx_priority = ctx->priority;
3071 ctx->priority = priority;
3072 add_output_action(ctx, odp_port);
3073 ctx->priority = ctx_priority;
3075 /* Update NetFlow output port. */
3076 if (ctx->nf_output_iface == NF_OUT_DROP) {
3077 ctx->nf_output_iface = odp_port;
3078 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3079 ctx->nf_output_iface = NF_OUT_MULTI;
3084 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3085 const struct nx_action_set_queue *nasq)
3090 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3093 /* Couldn't translate queue to a priority, so ignore. A warning
3094 * has already been logged. */
3098 ctx->priority = priority;
3101 struct xlate_reg_state {
3107 xlate_autopath(struct action_xlate_ctx *ctx,
3108 const struct nx_action_autopath *naa)
3110 uint16_t ofp_port = ntohl(naa->id);
3111 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
3113 if (!port || !port->bundle) {
3114 ofp_port = OFPP_NONE;
3115 } else if (port->bundle->bond) {
3116 /* Autopath does not support VLAN hashing. */
3117 struct ofport_dpif *slave = bond_choose_output_slave(
3118 port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags);
3120 ofp_port = slave->up.ofp_port;
3123 autopath_execute(naa, &ctx->flow, ofp_port);
3127 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
3129 struct ofproto_dpif *ofproto = ofproto_;
3130 struct ofport_dpif *port;
3140 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3143 port = get_ofp_port(ofproto, ofp_port);
3144 return port ? port->may_enable : false;
3149 do_xlate_actions(const union ofp_action *in, size_t n_in,
3150 struct action_xlate_ctx *ctx)
3152 const struct ofport_dpif *port;
3153 const union ofp_action *ia;
3156 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
3158 && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3159 port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3160 ? htonl(OFPPC_NO_RECV_STP)
3161 : htonl(OFPPC_NO_RECV))) {
3162 /* Drop this flow. */
3166 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
3167 const struct ofp_action_dl_addr *oada;
3168 const struct nx_action_resubmit *nar;
3169 const struct nx_action_set_tunnel *nast;
3170 const struct nx_action_set_queue *nasq;
3171 const struct nx_action_multipath *nam;
3172 const struct nx_action_autopath *naa;
3173 const struct nx_action_bundle *nab;
3174 const struct nx_action_output_reg *naor;
3175 enum ofputil_action_code code;
3178 code = ofputil_decode_action_unsafe(ia);
3180 case OFPUTIL_OFPAT_OUTPUT:
3181 xlate_output_action(ctx, &ia->output);
3184 case OFPUTIL_OFPAT_SET_VLAN_VID:
3185 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3186 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3189 case OFPUTIL_OFPAT_SET_VLAN_PCP:
3190 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3191 ctx->flow.vlan_tci |= htons(
3192 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3195 case OFPUTIL_OFPAT_STRIP_VLAN:
3196 ctx->flow.vlan_tci = htons(0);
3199 case OFPUTIL_OFPAT_SET_DL_SRC:
3200 oada = ((struct ofp_action_dl_addr *) ia);
3201 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3204 case OFPUTIL_OFPAT_SET_DL_DST:
3205 oada = ((struct ofp_action_dl_addr *) ia);
3206 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3209 case OFPUTIL_OFPAT_SET_NW_SRC:
3210 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3213 case OFPUTIL_OFPAT_SET_NW_DST:
3214 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3217 case OFPUTIL_OFPAT_SET_NW_TOS:
3218 ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK;
3221 case OFPUTIL_OFPAT_SET_TP_SRC:
3222 ctx->flow.tp_src = ia->tp_port.tp_port;
3225 case OFPUTIL_OFPAT_SET_TP_DST:
3226 ctx->flow.tp_dst = ia->tp_port.tp_port;
3229 case OFPUTIL_OFPAT_ENQUEUE:
3230 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3233 case OFPUTIL_NXAST_RESUBMIT:
3234 nar = (const struct nx_action_resubmit *) ia;
3235 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
3238 case OFPUTIL_NXAST_RESUBMIT_TABLE:
3239 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
3242 case OFPUTIL_NXAST_SET_TUNNEL:
3243 nast = (const struct nx_action_set_tunnel *) ia;
3244 tun_id = htonll(ntohl(nast->tun_id));
3245 ctx->flow.tun_id = tun_id;
3248 case OFPUTIL_NXAST_SET_QUEUE:
3249 nasq = (const struct nx_action_set_queue *) ia;
3250 xlate_set_queue_action(ctx, nasq);
3253 case OFPUTIL_NXAST_POP_QUEUE:
3257 case OFPUTIL_NXAST_REG_MOVE:
3258 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
3262 case OFPUTIL_NXAST_REG_LOAD:
3263 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
3267 case OFPUTIL_NXAST_NOTE:
3268 /* Nothing to do. */
3271 case OFPUTIL_NXAST_SET_TUNNEL64:
3272 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
3273 ctx->flow.tun_id = tun_id;
3276 case OFPUTIL_NXAST_MULTIPATH:
3277 nam = (const struct nx_action_multipath *) ia;
3278 multipath_execute(nam, &ctx->flow);
3281 case OFPUTIL_NXAST_AUTOPATH:
3282 naa = (const struct nx_action_autopath *) ia;
3283 xlate_autopath(ctx, naa);
3286 case OFPUTIL_NXAST_BUNDLE:
3287 ctx->ofproto->has_bundle_action = true;
3288 nab = (const struct nx_action_bundle *) ia;
3289 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
3294 case OFPUTIL_NXAST_BUNDLE_LOAD:
3295 ctx->ofproto->has_bundle_action = true;
3296 nab = (const struct nx_action_bundle *) ia;
3297 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
3301 case OFPUTIL_NXAST_OUTPUT_REG:
3302 naor = (const struct nx_action_output_reg *) ia;
3303 xlate_output_reg_action(ctx, naor);
3310 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3311 struct ofproto_dpif *ofproto, const struct flow *flow,
3312 const struct ofpbuf *packet)
3314 ctx->ofproto = ofproto;
3316 ctx->packet = packet;
3317 ctx->resubmit_hook = NULL;
3320 static struct ofpbuf *
3321 xlate_actions(struct action_xlate_ctx *ctx,
3322 const union ofp_action *in, size_t n_in)
3324 COVERAGE_INC(ofproto_dpif_xlate);
3326 ctx->odp_actions = ofpbuf_new(512);
3328 ctx->may_set_up_flow = true;
3329 ctx->nf_output_iface = NF_OUT_DROP;
3332 ctx->base_priority = 0;
3333 ctx->base_flow = ctx->flow;
3336 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
3337 ctx->may_set_up_flow = false;
3339 do_xlate_actions(in, n_in, ctx);
3342 /* Check with in-band control to see if we're allowed to set up this
3344 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
3345 ctx->odp_actions->data,
3346 ctx->odp_actions->size)) {
3347 ctx->may_set_up_flow = false;
3350 return ctx->odp_actions;
3353 /* OFPP_NORMAL implementation. */
3356 struct ofport_dpif *port;
3361 struct dst builtin[32];
3363 size_t n, allocated;
3366 static void dst_set_init(struct dst_set *);
3367 static void dst_set_add(struct dst_set *, const struct dst *);
3368 static void dst_set_free(struct dst_set *);
3370 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
3373 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
3374 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
3376 dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE
3377 : in_bundle->vlan >= 0 ? in_bundle->vlan
3378 : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE
3379 : vlan_tci_to_vid(ctx->flow.vlan_tci));
3381 dst->port = (!out_bundle->bond
3382 ? ofbundle_get_a_port(out_bundle)
3383 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
3384 dst->vlan, &ctx->tags));
3386 return dst->port != NULL;
3390 mirror_mask_ffs(mirror_mask_t mask)
3392 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
3397 dst_set_init(struct dst_set *set)
3399 set->dsts = set->builtin;
3401 set->allocated = ARRAY_SIZE(set->builtin);
3405 dst_set_add(struct dst_set *set, const struct dst *dst)
3407 if (set->n >= set->allocated) {
3408 size_t new_allocated;
3409 struct dst *new_dsts;
3411 new_allocated = set->allocated * 2;
3412 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
3413 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
3417 set->dsts = new_dsts;
3418 set->allocated = new_allocated;
3420 set->dsts[set->n++] = *dst;
3424 dst_set_free(struct dst_set *set)
3426 if (set->dsts != set->builtin) {
3432 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
3435 for (i = 0; i < set->n; i++) {
3436 if (set->dsts[i].vlan == test->vlan
3437 && set->dsts[i].port == test->port) {
3445 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
3447 return (bundle->vlan < 0
3448 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
3452 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
3454 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
3457 /* Returns an arbitrary interface within 'bundle'. */
3458 static struct ofport_dpif *
3459 ofbundle_get_a_port(const struct ofbundle *bundle)
3461 return CONTAINER_OF(list_front(&bundle->ports),
3462 struct ofport_dpif, bundle_node);
3466 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
3467 const struct ofbundle *in_bundle,
3468 const struct ofbundle *out_bundle, struct dst_set *set)
3472 if (out_bundle == OFBUNDLE_FLOOD) {
3473 struct ofbundle *bundle;
3475 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
3476 if (bundle != in_bundle
3477 && ofbundle_includes_vlan(bundle, vlan)
3478 && bundle->floodable
3479 && !bundle->mirror_out
3480 && set_dst(ctx, &dst, in_bundle, bundle)) {
3481 dst_set_add(set, &dst);
3484 ctx->nf_output_iface = NF_OUT_FLOOD;
3485 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
3486 dst_set_add(set, &dst);
3487 ctx->nf_output_iface = dst.port->odp_port;
3492 vlan_is_mirrored(const struct ofmirror *m, int vlan)
3494 return !m->vlans || bitmap_is_set(m->vlans, vlan);
3497 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
3498 * to a VLAN. In general most packets may be mirrored but we want to drop
3499 * protocols that may confuse switches. */
3501 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
3503 /* If you change this function's behavior, please update corresponding
3504 * documentation in vswitch.xml at the same time. */
3505 if (dst[0] != 0x01) {
3506 /* All the currently banned MACs happen to start with 01 currently, so
3507 * this is a quick way to eliminate most of the good ones. */
3509 if (eth_addr_is_reserved(dst)) {
3510 /* Drop STP, IEEE pause frames, and other reserved protocols
3511 * (01-80-c2-00-00-0x). */
3515 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
3517 if ((dst[3] & 0xfe) == 0xcc &&
3518 (dst[4] & 0xfe) == 0xcc &&
3519 (dst[5] & 0xfe) == 0xcc) {
3520 /* Drop the following protocols plus others following the same
3523 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
3524 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
3525 STP Uplink Fast (01-00-0c-cd-cd-cd) */
3529 if (!(dst[3] | dst[4] | dst[5])) {
3530 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
3539 compose_mirror_dsts(struct action_xlate_ctx *ctx,
3540 uint16_t vlan, const struct ofbundle *in_bundle,
3541 struct dst_set *set)
3543 struct ofproto_dpif *ofproto = ctx->ofproto;
3544 mirror_mask_t mirrors;
3548 mirrors = in_bundle->src_mirrors;
3549 for (i = 0; i < set->n; i++) {
3550 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
3557 flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3558 if (flow_vlan == 0) {
3559 flow_vlan = OFP_VLAN_NONE;
3563 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
3564 if (vlan_is_mirrored(m, vlan)) {
3568 if (set_dst(ctx, &dst, in_bundle, m->out)
3569 && !dst_is_duplicate(set, &dst)) {
3570 dst_set_add(set, &dst);
3572 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
3573 struct ofbundle *bundle;
3575 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3576 if (ofbundle_includes_vlan(bundle, m->out_vlan)
3577 && set_dst(ctx, &dst, in_bundle, bundle))
3579 if (bundle->vlan < 0) {
3580 dst.vlan = m->out_vlan;
3582 if (dst_is_duplicate(set, &dst)) {
3586 /* Use the vlan tag on the original flow instead of
3587 * the one passed in the vlan parameter. This ensures
3588 * that we compare the vlan from before any implicit
3589 * tagging tags place. This is necessary because
3590 * dst->vlan is the final vlan, after removing implicit
3592 if (bundle == in_bundle && dst.vlan == flow_vlan) {
3593 /* Don't send out input port on same VLAN. */
3596 dst_set_add(set, &dst);
3601 mirrors &= mirrors - 1;
3606 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
3607 const struct ofbundle *in_bundle,
3608 const struct ofbundle *out_bundle)
3610 uint16_t initial_vlan, cur_vlan;
3611 const struct dst *dst;
3615 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
3616 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
3618 /* Output all the packets we can without having to change the VLAN. */
3619 initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3620 if (initial_vlan == 0) {
3621 initial_vlan = OFP_VLAN_NONE;
3623 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3624 if (dst->vlan != initial_vlan) {
3627 nl_msg_put_u32(ctx->odp_actions,
3628 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3631 /* Then output the rest. */
3632 cur_vlan = initial_vlan;
3633 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3634 if (dst->vlan == initial_vlan) {
3637 if (dst->vlan != cur_vlan) {
3638 if (dst->vlan == OFP_VLAN_NONE) {
3639 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
3642 tci = htons(dst->vlan & VLAN_VID_MASK);
3643 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
3644 nl_msg_put_be16(ctx->odp_actions,
3645 ODP_ACTION_ATTR_SET_DL_TCI, tci);
3647 cur_vlan = dst->vlan;
3649 nl_msg_put_u32(ctx->odp_actions,
3650 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3656 /* Returns the effective vlan of a packet, taking into account both the
3657 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
3658 * the packet is untagged and -1 indicates it has an invalid header and
3659 * should be dropped. */
3661 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
3662 struct ofbundle *in_bundle, bool have_packet)
3664 int vlan = vlan_tci_to_vid(flow->vlan_tci);
3665 if (in_bundle->vlan >= 0) {
3668 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3669 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3670 "packet received on port %s configured with "
3671 "implicit VLAN %"PRIu16,
3672 ofproto->up.name, vlan,
3673 in_bundle->name, in_bundle->vlan);
3677 vlan = in_bundle->vlan;
3679 if (!ofbundle_includes_vlan(in_bundle, vlan)) {
3681 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3682 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3683 "packet received on port %s not configured for "
3685 ofproto->up.name, vlan, in_bundle->name, vlan);
3694 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
3695 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
3696 * indicate this; newer upstream kernels use gratuitous ARP requests. */
3698 is_gratuitous_arp(const struct flow *flow)
3700 return (flow->dl_type == htons(ETH_TYPE_ARP)
3701 && eth_addr_is_broadcast(flow->dl_dst)
3702 && (flow->nw_proto == ARP_OP_REPLY
3703 || (flow->nw_proto == ARP_OP_REQUEST
3704 && flow->nw_src == flow->nw_dst)));
3708 update_learning_table(struct ofproto_dpif *ofproto,
3709 const struct flow *flow, int vlan,
3710 struct ofbundle *in_bundle)
3712 struct mac_entry *mac;
3714 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
3718 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
3719 if (is_gratuitous_arp(flow)) {
3720 /* We don't want to learn from gratuitous ARP packets that are
3721 * reflected back over bond slaves so we lock the learning table. */
3722 if (!in_bundle->bond) {
3723 mac_entry_set_grat_arp_lock(mac);
3724 } else if (mac_entry_is_grat_arp_locked(mac)) {
3729 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
3730 /* The log messages here could actually be useful in debugging,
3731 * so keep the rate limit relatively high. */
3732 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3733 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
3734 "on port %s in VLAN %d",
3735 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
3736 in_bundle->name, vlan);
3738 mac->port.p = in_bundle;
3739 tag_set_add(&ofproto->revalidate_set,
3740 mac_learning_changed(ofproto->ml, mac));
3744 /* Determines whether packets in 'flow' within 'br' should be forwarded or
3745 * dropped. Returns true if they may be forwarded, false if they should be
3748 * If 'have_packet' is true, it indicates that the caller is processing a
3749 * received packet. If 'have_packet' is false, then the caller is just
3750 * revalidating an existing flow because configuration has changed. Either
3751 * way, 'have_packet' only affects logging (there is no point in logging errors
3752 * during revalidation).
3754 * Sets '*in_portp' to the input port. This will be a null pointer if
3755 * flow->in_port does not designate a known input port (in which case
3756 * is_admissible() returns false).
3758 * When returning true, sets '*vlanp' to the effective VLAN of the input
3759 * packet, as returned by flow_get_vlan().
3761 * May also add tags to '*tags', although the current implementation only does
3762 * so in one special case.
3765 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
3767 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
3769 struct ofport_dpif *in_port;
3770 struct ofbundle *in_bundle;
3773 /* Find the port and bundle for the received packet. */
3774 in_port = get_ofp_port(ofproto, flow->in_port);
3775 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
3776 if (!in_port || !in_bundle) {
3777 /* No interface? Something fishy... */
3779 /* Odd. A few possible reasons here:
3781 * - We deleted a port but there are still a few packets queued up
3784 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
3785 * we don't know about.
3787 * - Packet arrived on the local port but the local port is not
3790 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3792 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3794 ofproto->up.name, flow->in_port);
3798 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
3803 /* Drop frames for reserved multicast addresses
3804 * only if forward_bpdu option is absent. */
3805 if (eth_addr_is_reserved(flow->dl_dst) &&
3806 !ofproto->up.forward_bpdu) {
3810 /* Drop frames on bundles reserved for mirroring. */
3811 if (in_bundle->mirror_out) {
3813 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3814 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3815 "%s, which is reserved exclusively for mirroring",
3816 ofproto->up.name, in_bundle->name);
3821 if (in_bundle->bond) {
3822 struct mac_entry *mac;
3824 switch (bond_check_admissibility(in_bundle->bond, in_port,
3825 flow->dl_dst, tags)) {
3832 case BV_DROP_IF_MOVED:
3833 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
3834 if (mac && mac->port.p != in_bundle &&
3835 (!is_gratuitous_arp(flow)
3836 || mac_entry_is_grat_arp_locked(mac))) {
3847 xlate_normal(struct action_xlate_ctx *ctx)
3849 struct ofbundle *in_bundle;
3850 struct ofbundle *out_bundle;
3851 struct mac_entry *mac;
3854 /* Check whether we should drop packets in this flow. */
3855 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
3856 &ctx->tags, &vlan, &in_bundle)) {
3861 /* Learn source MAC (but don't try to learn from revalidation). */
3863 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
3866 /* Determine output bundle. */
3867 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
3870 out_bundle = mac->port.p;
3871 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
3872 /* If we are revalidating but don't have a learning entry then eject
3873 * the flow. Installing a flow that floods packets opens up a window
3874 * of time where we could learn from a packet reflected on a bond and
3875 * blackhole packets before the learning table is updated to reflect
3876 * the correct port. */
3877 ctx->may_set_up_flow = false;
3880 out_bundle = OFBUNDLE_FLOOD;
3883 /* Don't send packets out their input bundles. */
3884 if (in_bundle == out_bundle) {
3890 compose_actions(ctx, vlan, in_bundle, out_bundle);
3895 get_drop_frags(struct ofproto *ofproto_)
3897 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3900 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
3905 set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
3907 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3909 dpif_set_drop_frags(ofproto->dpif, drop_frags);
3913 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3914 const struct flow *flow,
3915 const union ofp_action *ofp_actions, size_t n_ofp_actions)
3917 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3920 error = validate_actions(ofp_actions, n_ofp_actions, flow,
3921 ofproto->max_ports);
3923 struct odputil_keybuf keybuf;
3924 struct action_xlate_ctx ctx;
3925 struct ofpbuf *odp_actions;
3928 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3929 odp_flow_key_from_flow(&key, flow);
3931 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3932 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3933 dpif_execute(ofproto->dpif, key.data, key.size,
3934 odp_actions->data, odp_actions->size, packet);
3935 ofpbuf_delete(odp_actions);
3941 get_netflow_ids(const struct ofproto *ofproto_,
3942 uint8_t *engine_type, uint8_t *engine_id)
3944 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3946 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3949 static struct ofproto_dpif *
3950 ofproto_dpif_lookup(const char *name)
3952 struct ofproto *ofproto = ofproto_lookup(name);
3953 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
3954 ? ofproto_dpif_cast(ofproto)
3959 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
3960 const char *args, void *aux OVS_UNUSED)
3962 struct ds ds = DS_EMPTY_INITIALIZER;
3963 const struct ofproto_dpif *ofproto;
3964 const struct mac_entry *e;
3966 ofproto = ofproto_dpif_lookup(args);
3968 unixctl_command_reply(conn, 501, "no such bridge");
3972 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3973 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3974 struct ofbundle *bundle = e->port.p;
3975 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
3976 ofbundle_get_a_port(bundle)->odp_port,
3977 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
3979 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3983 struct ofproto_trace {
3984 struct action_xlate_ctx ctx;
3990 trace_format_rule(struct ds *result, uint8_t table_id, int level,
3991 const struct rule_dpif *rule)
3993 ds_put_char_multiple(result, '\t', level);
3995 ds_put_cstr(result, "No match\n");
3999 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4000 table_id, ntohll(rule->up.flow_cookie));
4001 cls_rule_format(&rule->up.cr, result);
4002 ds_put_char(result, '\n');
4004 ds_put_char_multiple(result, '\t', level);
4005 ds_put_cstr(result, "OpenFlow ");
4006 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
4007 ds_put_char(result, '\n');
4011 trace_format_flow(struct ds *result, int level, const char *title,
4012 struct ofproto_trace *trace)
4014 ds_put_char_multiple(result, '\t', level);
4015 ds_put_format(result, "%s: ", title);
4016 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4017 ds_put_cstr(result, "unchanged");
4019 flow_format(result, &trace->ctx.flow);
4020 trace->flow = trace->ctx.flow;
4022 ds_put_char(result, '\n');
4026 trace_format_regs(struct ds *result, int level, const char *title,
4027 struct ofproto_trace *trace)
4031 ds_put_char_multiple(result, '\t', level);
4032 ds_put_format(result, "%s:", title);
4033 for (i = 0; i < FLOW_N_REGS; i++) {
4034 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
4036 ds_put_char(result, '\n');
4040 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
4042 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4043 struct ds *result = trace->result;
4045 ds_put_char(result, '\n');
4046 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4047 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
4048 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
4052 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4053 void *aux OVS_UNUSED)
4055 char *dpname, *arg1, *arg2, *arg3;
4056 char *args = xstrdup(args_);
4057 char *save_ptr = NULL;
4058 struct ofproto_dpif *ofproto;
4059 struct ofpbuf odp_key;
4060 struct ofpbuf *packet;
4061 struct rule_dpif *rule;
4067 ofpbuf_init(&odp_key, 0);
4070 dpname = strtok_r(args, " ", &save_ptr);
4071 arg1 = strtok_r(NULL, " ", &save_ptr);
4072 arg2 = strtok_r(NULL, " ", &save_ptr);
4073 arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
4074 if (dpname && arg1 && !arg2 && !arg3) {
4075 /* ofproto/trace dpname flow */
4078 /* Convert string to ODP key. */
4079 ofpbuf_init(&odp_key, 0);
4080 error = odp_flow_key_from_string(arg1, &odp_key);
4082 unixctl_command_reply(conn, 501, "Bad flow syntax");
4086 /* Convert odp_key to flow. */
4087 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
4089 unixctl_command_reply(conn, 501, "Invalid flow");
4092 } else if (dpname && arg1 && arg2 && arg3) {
4093 /* ofproto/trace dpname tun_id in_port packet */
4097 tun_id = htonll(strtoull(arg1, NULL, 0));
4098 in_port = ofp_port_to_odp_port(atoi(arg2));
4100 packet = ofpbuf_new(strlen(args) / 2);
4101 arg3 = ofpbuf_put_hex(packet, arg3, NULL);
4102 arg3 += strspn(arg3, " ");
4103 if (*arg3 != '\0') {
4104 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4107 if (packet->size < ETH_HEADER_LEN) {
4108 unixctl_command_reply(conn, 501,
4109 "Packet data too short for Ethernet");
4113 ds_put_cstr(&result, "Packet: ");
4114 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
4115 ds_put_cstr(&result, s);
4118 flow_extract(packet, tun_id, in_port, &flow);
4120 unixctl_command_reply(conn, 501, "Bad command syntax");
4124 ofproto = ofproto_dpif_lookup(dpname);
4126 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4131 ds_put_cstr(&result, "Flow: ");
4132 flow_format(&result, &flow);
4133 ds_put_char(&result, '\n');
4135 rule = rule_dpif_lookup(ofproto, &flow, 0);
4136 trace_format_rule(&result, 0, 0, rule);
4138 struct ofproto_trace trace;
4139 struct ofpbuf *odp_actions;
4141 trace.result = &result;
4143 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
4144 trace.ctx.resubmit_hook = trace_resubmit;
4145 odp_actions = xlate_actions(&trace.ctx,
4146 rule->up.actions, rule->up.n_actions);
4148 ds_put_char(&result, '\n');
4149 trace_format_flow(&result, 0, "Final flow", &trace);
4150 ds_put_cstr(&result, "Datapath actions: ");
4151 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4152 ofpbuf_delete(odp_actions);
4154 if (!trace.ctx.may_set_up_flow) {
4156 ds_put_cstr(&result, "\nThis flow is not cachable.");
4158 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
4159 "for complete actions, please supply a packet.");
4164 unixctl_command_reply(conn, 200, ds_cstr(&result));
4167 ds_destroy(&result);
4168 ofpbuf_delete(packet);
4169 ofpbuf_uninit(&odp_key);
4174 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
4175 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4178 unixctl_command_reply(conn, 200, NULL);
4182 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
4183 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4186 unixctl_command_reply(conn, 200, NULL);
4190 ofproto_dpif_unixctl_init(void)
4192 static bool registered;
4198 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4199 unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
4201 unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL);
4202 unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL);
4205 const struct ofproto_class ofproto_dpif_class = {
4232 port_is_lacp_current,
4233 NULL, /* rule_choose_table */
4240 rule_modify_actions,
4253 is_mirror_output_bundle,
4254 forward_bpdu_changed,