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"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any facet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 #define MAX_MIRRORS 32
110 typedef uint32_t mirror_mask_t;
111 #define MIRROR_MASK_C(X) UINT32_C(X)
112 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
114 struct ofproto_dpif *ofproto; /* Owning ofproto. */
115 size_t idx; /* In ofproto's "mirrors" array. */
116 void *aux; /* Key supplied by ofproto's client. */
117 char *name; /* Identifier for log messages. */
119 /* Selection criteria. */
120 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
121 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
122 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
124 /* Output (mutually exclusive). */
125 struct ofbundle *out; /* Output port or NULL. */
126 int out_vlan; /* Output VLAN or -1. */
129 static void mirror_destroy(struct ofmirror *);
131 /* A group of one or more OpenFlow ports. */
132 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
140 struct list ports; /* Contains "struct ofport"s. */
141 enum port_vlan_mode vlan_mode; /* VLAN mode */
142 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
143 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
144 * NULL if all VLANs are trunked. */
145 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
146 struct bond *bond; /* Nonnull iff more than one port. */
149 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
151 /* Port mirroring info. */
152 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
153 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
154 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
157 static void bundle_remove(struct ofport *);
158 static void bundle_update(struct ofbundle *);
159 static void bundle_destroy(struct ofbundle *);
160 static void bundle_del_port(struct ofport_dpif *);
161 static void bundle_run(struct ofbundle *);
162 static void bundle_wait(struct ofbundle *);
164 static void stp_run(struct ofproto_dpif *ofproto);
165 static void stp_wait(struct ofproto_dpif *ofproto);
167 struct action_xlate_ctx {
168 /* action_xlate_ctx_init() initializes these members. */
171 struct ofproto_dpif *ofproto;
173 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
174 * this flow when actions change header fields. */
177 /* The packet corresponding to 'flow', or a null pointer if we are
178 * revalidating without a packet to refer to. */
179 const struct ofpbuf *packet;
181 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
182 * want to execute them if we are actually processing a packet, or if we
183 * are accounting for packets that the datapath has processed, but not if
184 * we are just revalidating. */
187 /* If nonnull, called just before executing a resubmit action.
189 * This is normally null so the client has to set it manually after
190 * calling action_xlate_ctx_init(). */
191 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
193 /* xlate_actions() initializes and uses these members. The client might want
194 * to look at them after it returns. */
196 struct ofpbuf *odp_actions; /* Datapath actions. */
197 tag_type tags; /* Tags associated with actions. */
198 bool may_set_up_flow; /* True ordinarily; false if the actions must
199 * be reassessed for every packet. */
200 bool has_learn; /* Actions include NXAST_LEARN? */
201 bool has_normal; /* Actions output to OFPP_NORMAL? */
202 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
204 /* xlate_actions() initializes and uses these members, but the client has no
205 * reason to look at them. */
207 int recurse; /* Recursion level, via xlate_table_action. */
208 uint32_t priority; /* Current flow priority. 0 if none. */
209 struct flow base_flow; /* Flow at the last commit. */
210 uint32_t base_priority; /* Priority at the last commit. */
211 uint8_t table_id; /* OpenFlow table ID where flow was found. */
212 uint32_t sflow_n_outputs; /* Number of output ports. */
213 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
214 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
215 bool exit; /* No further actions should be processed. */
218 static void action_xlate_ctx_init(struct action_xlate_ctx *,
219 struct ofproto_dpif *, const struct flow *,
220 const struct ofpbuf *);
221 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
222 const union ofp_action *in, size_t n_in);
224 /* An exact-match instantiation of an OpenFlow flow. */
226 long long int used; /* Time last used; time created if not used. */
230 * - Do include packets and bytes sent "by hand", e.g. with
233 * - Do include packets and bytes that were obtained from the datapath
234 * when its statistics were reset (e.g. dpif_flow_put() with
235 * DPIF_FP_ZERO_STATS).
237 uint64_t packet_count; /* Number of packets received. */
238 uint64_t byte_count; /* Number of bytes received. */
240 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
241 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
243 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
244 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
245 long long int rs_used; /* Used time pushed to resubmit children. */
247 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
249 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
250 struct list list_node; /* In owning rule's 'facets' list. */
251 struct rule_dpif *rule; /* Owning rule. */
252 struct flow flow; /* Exact-match flow. */
253 bool installed; /* Installed in datapath? */
254 bool may_install; /* True ordinarily; false if actions must
255 * be reassessed for every packet. */
256 bool has_learn; /* Actions include NXAST_LEARN? */
257 bool has_normal; /* Actions output to OFPP_NORMAL? */
258 size_t actions_len; /* Number of bytes in actions[]. */
259 struct nlattr *actions; /* Datapath actions. */
260 tag_type tags; /* Tags. */
261 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
264 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
265 static void facet_remove(struct ofproto_dpif *, struct facet *);
266 static void facet_free(struct facet *);
268 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
269 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
270 const struct flow *);
271 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
273 static bool execute_controller_action(struct ofproto_dpif *,
275 const struct nlattr *odp_actions,
277 struct ofpbuf *packet);
278 static void facet_execute(struct ofproto_dpif *, struct facet *,
279 struct ofpbuf *packet);
281 static int facet_put__(struct ofproto_dpif *, struct facet *,
282 const struct nlattr *actions, size_t actions_len,
283 struct dpif_flow_stats *);
284 static void facet_install(struct ofproto_dpif *, struct facet *,
286 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
287 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
289 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
290 const struct ofpbuf *packet);
291 static void facet_update_time(struct ofproto_dpif *, struct facet *,
293 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
294 const struct dpif_flow_stats *);
295 static void facet_reset_counters(struct facet *);
296 static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
297 static void facet_push_stats(struct facet *);
298 static void facet_account(struct ofproto_dpif *, struct facet *);
300 static bool facet_is_controller_flow(struct facet *);
302 static void flow_push_stats(const struct rule_dpif *,
303 struct flow *, uint64_t packets, uint64_t bytes,
306 static uint32_t rule_calculate_tag(const struct flow *,
307 const struct flow_wildcards *,
309 static void rule_invalidate(const struct rule_dpif *);
315 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
316 struct list bundle_node; /* In struct ofbundle's "ports" list. */
317 struct cfm *cfm; /* Connectivity Fault Management, if any. */
318 tag_type tag; /* Tag associated with this port. */
319 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
320 bool may_enable; /* May be enabled in bonds. */
322 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
323 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
324 long long int stp_state_entered;
327 static struct ofport_dpif *
328 ofport_dpif_cast(const struct ofport *ofport)
330 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
331 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
334 static void port_run(struct ofport_dpif *);
335 static void port_wait(struct ofport_dpif *);
336 static int set_cfm(struct ofport *, const struct cfm_settings *);
338 struct dpif_completion {
339 struct list list_node;
340 struct ofoperation *op;
343 /* Extra information about a classifier table.
344 * Currently used just for optimized flow revalidation. */
346 /* If either of these is nonnull, then this table has a form that allows
347 * flows to be tagged to avoid revalidating most flows for the most common
348 * kinds of flow table changes. */
349 struct cls_table *catchall_table; /* Table that wildcards all fields. */
350 struct cls_table *other_table; /* Table with any other wildcard set. */
351 uint32_t basis; /* Keeps each table's tags separate. */
354 struct ofproto_dpif {
363 struct netflow *netflow;
364 struct dpif_sflow *sflow;
365 struct hmap bundles; /* Contains "struct ofbundle"s. */
366 struct mac_learning *ml;
367 struct ofmirror *mirrors[MAX_MIRRORS];
368 bool has_bonded_bundles;
371 struct timer next_expiration;
377 struct table_dpif tables[N_TABLES];
378 bool need_revalidate;
379 struct tag_set revalidate_set;
381 /* Support for debugging async flow mods. */
382 struct list completions;
384 bool has_bundle_action; /* True when the first bundle action appears. */
388 long long int stp_last_tick;
391 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
392 * for debugging the asynchronous flow_mod implementation.) */
395 static void ofproto_dpif_unixctl_init(void);
397 static struct ofproto_dpif *
398 ofproto_dpif_cast(const struct ofproto *ofproto)
400 assert(ofproto->ofproto_class == &ofproto_dpif_class);
401 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
404 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
406 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
409 /* Packet processing. */
410 static void update_learning_table(struct ofproto_dpif *,
411 const struct flow *, int vlan,
413 static bool is_admissible(struct ofproto_dpif *, const struct flow *,
414 bool have_packet, tag_type *, int *vlanp,
415 struct ofbundle **in_bundlep);
418 #define FLOW_MISS_MAX_BATCH 50
419 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
420 static void handle_miss_upcalls(struct ofproto_dpif *,
421 struct dpif_upcall *, size_t n);
423 /* Flow expiration. */
424 static int expire(struct ofproto_dpif *);
427 static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
428 const struct ofpbuf *packet);
430 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
431 const struct flow *, uint32_t odp_port);
432 /* Global variables. */
433 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
435 /* Factory functions. */
438 enumerate_types(struct sset *types)
440 dp_enumerate_types(types);
444 enumerate_names(const char *type, struct sset *names)
446 return dp_enumerate_names(type, names);
450 del(const char *type, const char *name)
455 error = dpif_open(name, type, &dpif);
457 error = dpif_delete(dpif);
463 /* Basic life-cycle. */
465 static struct ofproto *
468 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
473 dealloc(struct ofproto *ofproto_)
475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
480 construct(struct ofproto *ofproto_, int *n_tablesp)
482 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
483 const char *name = ofproto->up.name;
487 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
489 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
493 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
494 ofproto->n_matches = 0;
496 dpif_flow_flush(ofproto->dpif);
497 dpif_recv_purge(ofproto->dpif);
499 error = dpif_recv_set_mask(ofproto->dpif,
500 ((1u << DPIF_UC_MISS) |
501 (1u << DPIF_UC_ACTION)));
503 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
504 dpif_close(ofproto->dpif);
508 ofproto->netflow = NULL;
509 ofproto->sflow = NULL;
511 hmap_init(&ofproto->bundles);
512 ofproto->ml = mac_learning_create();
513 for (i = 0; i < MAX_MIRRORS; i++) {
514 ofproto->mirrors[i] = NULL;
516 ofproto->has_bonded_bundles = false;
518 timer_set_duration(&ofproto->next_expiration, 1000);
520 hmap_init(&ofproto->facets);
522 for (i = 0; i < N_TABLES; i++) {
523 struct table_dpif *table = &ofproto->tables[i];
525 table->catchall_table = NULL;
526 table->other_table = NULL;
527 table->basis = random_uint32();
529 ofproto->need_revalidate = false;
530 tag_set_init(&ofproto->revalidate_set);
532 list_init(&ofproto->completions);
534 ofproto_dpif_unixctl_init();
536 ofproto->has_bundle_action = false;
538 *n_tablesp = N_TABLES;
543 complete_operations(struct ofproto_dpif *ofproto)
545 struct dpif_completion *c, *next;
547 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
548 ofoperation_complete(c->op, 0);
549 list_remove(&c->list_node);
555 destruct(struct ofproto *ofproto_)
557 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
558 struct rule_dpif *rule, *next_rule;
559 struct classifier *table;
562 complete_operations(ofproto);
564 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
565 struct cls_cursor cursor;
567 cls_cursor_init(&cursor, table, NULL);
568 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
569 ofproto_rule_destroy(&rule->up);
573 for (i = 0; i < MAX_MIRRORS; i++) {
574 mirror_destroy(ofproto->mirrors[i]);
577 netflow_destroy(ofproto->netflow);
578 dpif_sflow_destroy(ofproto->sflow);
579 hmap_destroy(&ofproto->bundles);
580 mac_learning_destroy(ofproto->ml);
582 hmap_destroy(&ofproto->facets);
584 dpif_close(ofproto->dpif);
588 run(struct ofproto *ofproto_)
590 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
591 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
592 struct ofport_dpif *ofport;
593 struct ofbundle *bundle;
598 complete_operations(ofproto);
600 dpif_run(ofproto->dpif);
603 for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) {
604 struct dpif_upcall *upcall = &misses[n_misses];
607 error = dpif_recv(ofproto->dpif, upcall);
609 if (error == ENODEV && n_misses == 0) {
615 if (upcall->type == DPIF_UC_MISS) {
616 /* Handle it later. */
619 handle_upcall(ofproto, upcall);
623 handle_miss_upcalls(ofproto, misses, n_misses);
625 if (timer_expired(&ofproto->next_expiration)) {
626 int delay = expire(ofproto);
627 timer_set_duration(&ofproto->next_expiration, delay);
630 if (ofproto->netflow) {
631 netflow_run(ofproto->netflow);
633 if (ofproto->sflow) {
634 dpif_sflow_run(ofproto->sflow);
637 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
640 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
645 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
647 /* Now revalidate if there's anything to do. */
648 if (ofproto->need_revalidate
649 || !tag_set_is_empty(&ofproto->revalidate_set)) {
650 struct tag_set revalidate_set = ofproto->revalidate_set;
651 bool revalidate_all = ofproto->need_revalidate;
652 struct facet *facet, *next;
654 /* Clear the revalidation flags. */
655 tag_set_init(&ofproto->revalidate_set);
656 ofproto->need_revalidate = false;
658 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
660 || tag_set_intersects(&revalidate_set, facet->tags)) {
661 facet_revalidate(ofproto, facet);
670 wait(struct ofproto *ofproto_)
672 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
673 struct ofport_dpif *ofport;
674 struct ofbundle *bundle;
676 if (!clogged && !list_is_empty(&ofproto->completions)) {
677 poll_immediate_wake();
680 dpif_wait(ofproto->dpif);
681 dpif_recv_wait(ofproto->dpif);
682 if (ofproto->sflow) {
683 dpif_sflow_wait(ofproto->sflow);
685 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
686 poll_immediate_wake();
688 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
691 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
694 mac_learning_wait(ofproto->ml);
696 if (ofproto->need_revalidate) {
697 /* Shouldn't happen, but if it does just go around again. */
698 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
699 poll_immediate_wake();
701 timer_wait(&ofproto->next_expiration);
706 flush(struct ofproto *ofproto_)
708 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
709 struct facet *facet, *next_facet;
711 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
712 /* Mark the facet as not installed so that facet_remove() doesn't
713 * bother trying to uninstall it. There is no point in uninstalling it
714 * individually since we are about to blow away all the facets with
715 * dpif_flow_flush(). */
716 facet->installed = false;
717 facet->dp_packet_count = 0;
718 facet->dp_byte_count = 0;
719 facet_remove(ofproto, facet);
721 dpif_flow_flush(ofproto->dpif);
725 get_features(struct ofproto *ofproto_ OVS_UNUSED,
726 bool *arp_match_ip, uint32_t *actions)
728 *arp_match_ip = true;
729 *actions = ((1u << OFPAT_OUTPUT) |
730 (1u << OFPAT_SET_VLAN_VID) |
731 (1u << OFPAT_SET_VLAN_PCP) |
732 (1u << OFPAT_STRIP_VLAN) |
733 (1u << OFPAT_SET_DL_SRC) |
734 (1u << OFPAT_SET_DL_DST) |
735 (1u << OFPAT_SET_NW_SRC) |
736 (1u << OFPAT_SET_NW_DST) |
737 (1u << OFPAT_SET_NW_TOS) |
738 (1u << OFPAT_SET_TP_SRC) |
739 (1u << OFPAT_SET_TP_DST) |
740 (1u << OFPAT_ENQUEUE));
744 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
747 struct dpif_dp_stats s;
749 strcpy(ots->name, "classifier");
751 dpif_get_dp_stats(ofproto->dpif, &s);
752 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
753 put_32aligned_be64(&ots->matched_count,
754 htonll(s.n_hit + ofproto->n_matches));
758 set_netflow(struct ofproto *ofproto_,
759 const struct netflow_options *netflow_options)
761 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
763 if (netflow_options) {
764 if (!ofproto->netflow) {
765 ofproto->netflow = netflow_create();
767 return netflow_set_options(ofproto->netflow, netflow_options);
769 netflow_destroy(ofproto->netflow);
770 ofproto->netflow = NULL;
775 static struct ofport *
778 struct ofport_dpif *port = xmalloc(sizeof *port);
783 port_dealloc(struct ofport *port_)
785 struct ofport_dpif *port = ofport_dpif_cast(port_);
790 port_construct(struct ofport *port_)
792 struct ofport_dpif *port = ofport_dpif_cast(port_);
793 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
795 ofproto->need_revalidate = true;
796 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
799 port->tag = tag_create_random();
800 port->may_enable = true;
801 port->stp_port = NULL;
802 port->stp_state = STP_DISABLED;
804 if (ofproto->sflow) {
805 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
806 netdev_get_name(port->up.netdev));
813 port_destruct(struct ofport *port_)
815 struct ofport_dpif *port = ofport_dpif_cast(port_);
816 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
818 ofproto->need_revalidate = true;
819 bundle_remove(port_);
820 set_cfm(port_, NULL);
821 if (ofproto->sflow) {
822 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
827 port_modified(struct ofport *port_)
829 struct ofport_dpif *port = ofport_dpif_cast(port_);
831 if (port->bundle && port->bundle->bond) {
832 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
837 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
839 struct ofport_dpif *port = ofport_dpif_cast(port_);
840 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
841 ovs_be32 changed = old_config ^ port->up.opp.config;
843 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
844 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
845 ofproto->need_revalidate = true;
847 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
848 bundle_update(port->bundle);
854 set_sflow(struct ofproto *ofproto_,
855 const struct ofproto_sflow_options *sflow_options)
857 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
858 struct dpif_sflow *ds = ofproto->sflow;
862 struct ofport_dpif *ofport;
864 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
865 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
866 dpif_sflow_add_port(ds, ofport->odp_port,
867 netdev_get_name(ofport->up.netdev));
869 ofproto->need_revalidate = true;
871 dpif_sflow_set_options(ds, sflow_options);
874 dpif_sflow_destroy(ds);
875 ofproto->need_revalidate = true;
876 ofproto->sflow = NULL;
883 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
885 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
892 struct ofproto_dpif *ofproto;
894 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
895 ofproto->need_revalidate = true;
896 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
899 if (cfm_configure(ofport->cfm, s)) {
905 cfm_destroy(ofport->cfm);
911 get_cfm_fault(const struct ofport *ofport_)
913 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
915 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
919 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
922 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
925 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
935 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
937 struct ofproto_dpif *ofproto = ofproto_;
938 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
939 struct ofport_dpif *ofport;
941 ofport = stp_port_get_aux(sp);
943 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
944 ofproto->up.name, port_num);
946 struct eth_header *eth = pkt->l2;
948 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
949 if (eth_addr_is_zero(eth->eth_src)) {
950 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
951 "with unknown MAC", ofproto->up.name, port_num);
953 int error = netdev_send(ofport->up.netdev, pkt);
955 VLOG_WARN_RL(&rl, "%s: sending BPDU on port %s failed (%s)",
957 netdev_get_name(ofport->up.netdev),
965 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
967 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
969 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
971 /* Only revalidate flows if the configuration changed. */
972 if (!s != !ofproto->stp) {
973 ofproto->need_revalidate = true;
978 ofproto->stp = stp_create(ofproto_->name, s->system_id,
979 send_bpdu_cb, ofproto);
980 ofproto->stp_last_tick = time_msec();
983 stp_set_bridge_id(ofproto->stp, s->system_id);
984 stp_set_bridge_priority(ofproto->stp, s->priority);
985 stp_set_hello_time(ofproto->stp, s->hello_time);
986 stp_set_max_age(ofproto->stp, s->max_age);
987 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
989 stp_destroy(ofproto->stp);
997 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1003 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1004 s->designated_root = stp_get_designated_root(ofproto->stp);
1005 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1014 update_stp_port_state(struct ofport_dpif *ofport)
1016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1017 enum stp_state state;
1019 /* Figure out new state. */
1020 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1024 if (ofport->stp_state != state) {
1028 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1029 netdev_get_name(ofport->up.netdev),
1030 stp_state_name(ofport->stp_state),
1031 stp_state_name(state));
1032 if (stp_learn_in_state(ofport->stp_state)
1033 != stp_learn_in_state(state)) {
1034 /* xxx Learning action flows should also be flushed. */
1035 mac_learning_flush(ofproto->ml);
1037 fwd_change = stp_forward_in_state(ofport->stp_state)
1038 != stp_forward_in_state(state);
1040 ofproto->need_revalidate = true;
1041 ofport->stp_state = state;
1042 ofport->stp_state_entered = time_msec();
1045 bundle_update(ofport->bundle);
1048 /* Update the STP state bits in the OpenFlow port description. */
1049 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1050 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1051 : state == STP_LEARNING ? OFPPS_STP_LEARN
1052 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1053 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1055 ofproto_port_set_state(&ofport->up, of_state);
1059 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1060 * caller is responsible for assigning STP port numbers and ensuring
1061 * there are no duplicates. */
1063 set_stp_port(struct ofport *ofport_,
1064 const struct ofproto_port_stp_settings *s)
1066 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1068 struct stp_port *sp = ofport->stp_port;
1070 if (!s || !s->enable) {
1072 ofport->stp_port = NULL;
1073 stp_port_disable(sp);
1076 } else if (sp && stp_port_no(sp) != s->port_num
1077 && ofport == stp_port_get_aux(sp)) {
1078 /* The port-id changed, so disable the old one if it's not
1079 * already in use by another port. */
1080 stp_port_disable(sp);
1083 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1084 stp_port_enable(sp);
1086 stp_port_set_aux(sp, ofport);
1087 stp_port_set_priority(sp, s->priority);
1088 stp_port_set_path_cost(sp, s->path_cost);
1090 update_stp_port_state(ofport);
1096 get_stp_port_status(struct ofport *ofport_,
1097 struct ofproto_port_stp_status *s)
1099 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1101 struct stp_port *sp = ofport->stp_port;
1103 if (!ofproto->stp || !sp) {
1109 s->port_id = stp_port_get_id(sp);
1110 s->state = stp_port_get_state(sp);
1111 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1112 s->role = stp_port_get_role(sp);
1118 stp_run(struct ofproto_dpif *ofproto)
1121 long long int now = time_msec();
1122 long long int elapsed = now - ofproto->stp_last_tick;
1123 struct stp_port *sp;
1126 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1127 ofproto->stp_last_tick = now;
1129 while (stp_get_changed_port(ofproto->stp, &sp)) {
1130 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1133 update_stp_port_state(ofport);
1140 stp_wait(struct ofproto_dpif *ofproto)
1143 poll_timer_wait(1000);
1147 /* Returns true if STP should process 'flow'. */
1149 stp_should_process_flow(const struct flow *flow)
1151 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1155 stp_process_packet(const struct ofport_dpif *ofport,
1156 const struct ofpbuf *packet)
1158 struct ofpbuf payload = *packet;
1159 struct eth_header *eth = payload.data;
1160 struct stp_port *sp = ofport->stp_port;
1162 /* Sink packets on ports that have STP disabled when the bridge has
1164 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1168 /* Trim off padding on payload. */
1169 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1170 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1173 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1174 stp_received_bpdu(sp, payload.data, payload.size);
1180 /* Expires all MAC learning entries associated with 'port' and forces ofproto
1181 * to revalidate every flow. */
1183 bundle_flush_macs(struct ofbundle *bundle)
1185 struct ofproto_dpif *ofproto = bundle->ofproto;
1186 struct mac_learning *ml = ofproto->ml;
1187 struct mac_entry *mac, *next_mac;
1189 ofproto->need_revalidate = true;
1190 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1191 if (mac->port.p == bundle) {
1192 mac_learning_expire(ml, mac);
1197 static struct ofbundle *
1198 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1200 struct ofbundle *bundle;
1202 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1203 &ofproto->bundles) {
1204 if (bundle->aux == aux) {
1211 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1212 * ones that are found to 'bundles'. */
1214 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1215 void **auxes, size_t n_auxes,
1216 struct hmapx *bundles)
1220 hmapx_init(bundles);
1221 for (i = 0; i < n_auxes; i++) {
1222 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1224 hmapx_add(bundles, bundle);
1230 bundle_update(struct ofbundle *bundle)
1232 struct ofport_dpif *port;
1234 bundle->floodable = true;
1235 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1236 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1237 || !stp_forward_in_state(port->stp_state)) {
1238 bundle->floodable = false;
1245 bundle_del_port(struct ofport_dpif *port)
1247 struct ofbundle *bundle = port->bundle;
1249 bundle->ofproto->need_revalidate = true;
1251 list_remove(&port->bundle_node);
1252 port->bundle = NULL;
1255 lacp_slave_unregister(bundle->lacp, port);
1258 bond_slave_unregister(bundle->bond, port);
1261 bundle_update(bundle);
1265 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1266 struct lacp_slave_settings *lacp,
1267 uint32_t bond_stable_id)
1269 struct ofport_dpif *port;
1271 port = get_ofp_port(bundle->ofproto, ofp_port);
1276 if (port->bundle != bundle) {
1277 bundle->ofproto->need_revalidate = true;
1279 bundle_del_port(port);
1282 port->bundle = bundle;
1283 list_push_back(&bundle->ports, &port->bundle_node);
1284 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1285 || !stp_forward_in_state(port->stp_state)) {
1286 bundle->floodable = false;
1290 port->bundle->ofproto->need_revalidate = true;
1291 lacp_slave_register(bundle->lacp, port, lacp);
1294 port->bond_stable_id = bond_stable_id;
1300 bundle_destroy(struct ofbundle *bundle)
1302 struct ofproto_dpif *ofproto;
1303 struct ofport_dpif *port, *next_port;
1310 ofproto = bundle->ofproto;
1311 for (i = 0; i < MAX_MIRRORS; i++) {
1312 struct ofmirror *m = ofproto->mirrors[i];
1314 if (m->out == bundle) {
1316 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1317 || hmapx_find_and_delete(&m->dsts, bundle)) {
1318 ofproto->need_revalidate = true;
1323 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1324 bundle_del_port(port);
1327 bundle_flush_macs(bundle);
1328 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1330 free(bundle->trunks);
1331 lacp_destroy(bundle->lacp);
1332 bond_destroy(bundle->bond);
1337 bundle_set(struct ofproto *ofproto_, void *aux,
1338 const struct ofproto_bundle_settings *s)
1340 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1341 bool need_flush = false;
1342 struct ofport_dpif *port;
1343 struct ofbundle *bundle;
1344 unsigned long *trunks;
1350 bundle_destroy(bundle_lookup(ofproto, aux));
1354 assert(s->n_slaves == 1 || s->bond != NULL);
1355 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1357 bundle = bundle_lookup(ofproto, aux);
1359 bundle = xmalloc(sizeof *bundle);
1361 bundle->ofproto = ofproto;
1362 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1363 hash_pointer(aux, 0));
1365 bundle->name = NULL;
1367 list_init(&bundle->ports);
1368 bundle->vlan_mode = PORT_VLAN_TRUNK;
1370 bundle->trunks = NULL;
1371 bundle->lacp = NULL;
1372 bundle->bond = NULL;
1374 bundle->floodable = true;
1376 bundle->src_mirrors = 0;
1377 bundle->dst_mirrors = 0;
1378 bundle->mirror_out = 0;
1381 if (!bundle->name || strcmp(s->name, bundle->name)) {
1383 bundle->name = xstrdup(s->name);
1388 if (!bundle->lacp) {
1389 ofproto->need_revalidate = true;
1390 bundle->lacp = lacp_create();
1392 lacp_configure(bundle->lacp, s->lacp);
1394 lacp_destroy(bundle->lacp);
1395 bundle->lacp = NULL;
1398 /* Update set of ports. */
1400 for (i = 0; i < s->n_slaves; i++) {
1401 if (!bundle_add_port(bundle, s->slaves[i],
1402 s->lacp ? &s->lacp_slaves[i] : NULL,
1403 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1407 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1408 struct ofport_dpif *next_port;
1410 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1411 for (i = 0; i < s->n_slaves; i++) {
1412 if (s->slaves[i] == port->up.ofp_port) {
1417 bundle_del_port(port);
1421 assert(list_size(&bundle->ports) <= s->n_slaves);
1423 if (list_is_empty(&bundle->ports)) {
1424 bundle_destroy(bundle);
1428 /* Set VLAN tagging mode */
1429 if (s->vlan_mode != bundle->vlan_mode) {
1430 bundle->vlan_mode = s->vlan_mode;
1435 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1436 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1438 if (vlan != bundle->vlan) {
1439 bundle->vlan = vlan;
1443 /* Get trunked VLANs. */
1444 switch (s->vlan_mode) {
1445 case PORT_VLAN_ACCESS:
1449 case PORT_VLAN_TRUNK:
1450 trunks = (unsigned long *) s->trunks;
1453 case PORT_VLAN_NATIVE_UNTAGGED:
1454 case PORT_VLAN_NATIVE_TAGGED:
1455 if (vlan != 0 && (!s->trunks
1456 || !bitmap_is_set(s->trunks, vlan)
1457 || bitmap_is_set(s->trunks, 0))) {
1458 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1460 trunks = bitmap_clone(s->trunks, 4096);
1462 trunks = bitmap_allocate1(4096);
1464 bitmap_set1(trunks, vlan);
1465 bitmap_set0(trunks, 0);
1467 trunks = (unsigned long *) s->trunks;
1474 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1475 free(bundle->trunks);
1476 if (trunks == s->trunks) {
1477 bundle->trunks = vlan_bitmap_clone(trunks);
1479 bundle->trunks = trunks;
1484 if (trunks != s->trunks) {
1489 if (!list_is_short(&bundle->ports)) {
1490 bundle->ofproto->has_bonded_bundles = true;
1492 if (bond_reconfigure(bundle->bond, s->bond)) {
1493 ofproto->need_revalidate = true;
1496 bundle->bond = bond_create(s->bond);
1497 ofproto->need_revalidate = true;
1500 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1501 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1505 bond_destroy(bundle->bond);
1506 bundle->bond = NULL;
1509 /* If we changed something that would affect MAC learning, un-learn
1510 * everything on this port and force flow revalidation. */
1512 bundle_flush_macs(bundle);
1519 bundle_remove(struct ofport *port_)
1521 struct ofport_dpif *port = ofport_dpif_cast(port_);
1522 struct ofbundle *bundle = port->bundle;
1525 bundle_del_port(port);
1526 if (list_is_empty(&bundle->ports)) {
1527 bundle_destroy(bundle);
1528 } else if (list_is_short(&bundle->ports)) {
1529 bond_destroy(bundle->bond);
1530 bundle->bond = NULL;
1536 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1538 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1539 struct ofport_dpif *port = port_;
1540 uint8_t ea[ETH_ADDR_LEN];
1543 error = netdev_get_etheraddr(port->up.netdev, ea);
1545 struct ofpbuf packet;
1548 ofpbuf_init(&packet, 0);
1549 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1551 memcpy(packet_pdu, pdu, pdu_size);
1553 error = netdev_send(port->up.netdev, &packet);
1555 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1556 "(%s)", port->bundle->name,
1557 netdev_get_name(port->up.netdev), strerror(error));
1559 ofpbuf_uninit(&packet);
1561 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1562 "%s (%s)", port->bundle->name,
1563 netdev_get_name(port->up.netdev), strerror(error));
1568 bundle_send_learning_packets(struct ofbundle *bundle)
1570 struct ofproto_dpif *ofproto = bundle->ofproto;
1571 int error, n_packets, n_errors;
1572 struct mac_entry *e;
1574 error = n_packets = n_errors = 0;
1575 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1576 if (e->port.p != bundle) {
1577 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1587 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1588 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1589 "packets, last error was: %s",
1590 bundle->name, n_errors, n_packets, strerror(error));
1592 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1593 bundle->name, n_packets);
1598 bundle_run(struct ofbundle *bundle)
1601 lacp_run(bundle->lacp, send_pdu_cb);
1604 struct ofport_dpif *port;
1606 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1607 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1610 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1611 lacp_negotiated(bundle->lacp));
1612 if (bond_should_send_learning_packets(bundle->bond)) {
1613 bundle_send_learning_packets(bundle);
1619 bundle_wait(struct ofbundle *bundle)
1622 lacp_wait(bundle->lacp);
1625 bond_wait(bundle->bond);
1632 mirror_scan(struct ofproto_dpif *ofproto)
1636 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1637 if (!ofproto->mirrors[idx]) {
1644 static struct ofmirror *
1645 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1649 for (i = 0; i < MAX_MIRRORS; i++) {
1650 struct ofmirror *mirror = ofproto->mirrors[i];
1651 if (mirror && mirror->aux == aux) {
1660 mirror_set(struct ofproto *ofproto_, void *aux,
1661 const struct ofproto_mirror_settings *s)
1663 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1664 mirror_mask_t mirror_bit;
1665 struct ofbundle *bundle;
1666 struct ofmirror *mirror;
1667 struct ofbundle *out;
1668 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1669 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1672 mirror = mirror_lookup(ofproto, aux);
1674 mirror_destroy(mirror);
1680 idx = mirror_scan(ofproto);
1682 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1684 ofproto->up.name, MAX_MIRRORS, s->name);
1688 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1689 mirror->ofproto = ofproto;
1692 mirror->out_vlan = -1;
1693 mirror->name = NULL;
1696 if (!mirror->name || strcmp(s->name, mirror->name)) {
1698 mirror->name = xstrdup(s->name);
1701 /* Get the new configuration. */
1702 if (s->out_bundle) {
1703 out = bundle_lookup(ofproto, s->out_bundle);
1705 mirror_destroy(mirror);
1711 out_vlan = s->out_vlan;
1713 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1714 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1716 /* If the configuration has not changed, do nothing. */
1717 if (hmapx_equals(&srcs, &mirror->srcs)
1718 && hmapx_equals(&dsts, &mirror->dsts)
1719 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1720 && mirror->out == out
1721 && mirror->out_vlan == out_vlan)
1723 hmapx_destroy(&srcs);
1724 hmapx_destroy(&dsts);
1728 hmapx_swap(&srcs, &mirror->srcs);
1729 hmapx_destroy(&srcs);
1731 hmapx_swap(&dsts, &mirror->dsts);
1732 hmapx_destroy(&dsts);
1734 free(mirror->vlans);
1735 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1738 mirror->out_vlan = out_vlan;
1740 /* Update bundles. */
1741 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1742 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1743 if (hmapx_contains(&mirror->srcs, bundle)) {
1744 bundle->src_mirrors |= mirror_bit;
1746 bundle->src_mirrors &= ~mirror_bit;
1749 if (hmapx_contains(&mirror->dsts, bundle)) {
1750 bundle->dst_mirrors |= mirror_bit;
1752 bundle->dst_mirrors &= ~mirror_bit;
1755 if (mirror->out == bundle) {
1756 bundle->mirror_out |= mirror_bit;
1758 bundle->mirror_out &= ~mirror_bit;
1762 ofproto->need_revalidate = true;
1763 mac_learning_flush(ofproto->ml);
1769 mirror_destroy(struct ofmirror *mirror)
1771 struct ofproto_dpif *ofproto;
1772 mirror_mask_t mirror_bit;
1773 struct ofbundle *bundle;
1779 ofproto = mirror->ofproto;
1780 ofproto->need_revalidate = true;
1781 mac_learning_flush(ofproto->ml);
1783 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1784 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1785 bundle->src_mirrors &= ~mirror_bit;
1786 bundle->dst_mirrors &= ~mirror_bit;
1787 bundle->mirror_out &= ~mirror_bit;
1790 hmapx_destroy(&mirror->srcs);
1791 hmapx_destroy(&mirror->dsts);
1792 free(mirror->vlans);
1794 ofproto->mirrors[mirror->idx] = NULL;
1800 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1802 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1803 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1804 ofproto->need_revalidate = true;
1805 mac_learning_flush(ofproto->ml);
1811 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
1813 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1814 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1815 return bundle && bundle->mirror_out != 0;
1819 forward_bpdu_changed(struct ofproto *ofproto_)
1821 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1822 /* Revalidate cached flows whenever forward_bpdu option changes. */
1823 ofproto->need_revalidate = true;
1828 static struct ofport_dpif *
1829 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1831 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1832 return ofport ? ofport_dpif_cast(ofport) : NULL;
1835 static struct ofport_dpif *
1836 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1838 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1842 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1843 struct dpif_port *dpif_port)
1845 ofproto_port->name = dpif_port->name;
1846 ofproto_port->type = dpif_port->type;
1847 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1851 port_run(struct ofport_dpif *ofport)
1853 bool enable = netdev_get_carrier(ofport->up.netdev);
1856 cfm_run(ofport->cfm);
1858 if (cfm_should_send_ccm(ofport->cfm)) {
1859 struct ofpbuf packet;
1861 ofpbuf_init(&packet, 0);
1862 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1863 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1864 ofport->odp_port, &packet);
1865 ofpbuf_uninit(&packet);
1868 enable = enable && !cfm_get_fault(ofport->cfm)
1869 && cfm_get_opup(ofport->cfm);
1872 if (ofport->bundle) {
1873 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1876 if (ofport->may_enable != enable) {
1877 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1879 if (ofproto->has_bundle_action) {
1880 ofproto->need_revalidate = true;
1884 ofport->may_enable = enable;
1888 port_wait(struct ofport_dpif *ofport)
1891 cfm_wait(ofport->cfm);
1896 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1897 struct ofproto_port *ofproto_port)
1899 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1900 struct dpif_port dpif_port;
1903 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1905 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1911 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1917 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1919 *ofp_portp = odp_port_to_ofp_port(odp_port);
1925 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1927 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1930 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1932 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1934 /* The caller is going to close ofport->up.netdev. If this is a
1935 * bonded port, then the bond is using that netdev, so remove it
1936 * from the bond. The client will need to reconfigure everything
1937 * after deleting ports, so then the slave will get re-added. */
1938 bundle_remove(&ofport->up);
1944 struct port_dump_state {
1945 struct dpif_port_dump dump;
1950 port_dump_start(const struct ofproto *ofproto_, void **statep)
1952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1953 struct port_dump_state *state;
1955 *statep = state = xmalloc(sizeof *state);
1956 dpif_port_dump_start(&state->dump, ofproto->dpif);
1957 state->done = false;
1962 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1963 struct ofproto_port *port)
1965 struct port_dump_state *state = state_;
1966 struct dpif_port dpif_port;
1968 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1969 ofproto_port_from_dpif_port(port, &dpif_port);
1972 int error = dpif_port_dump_done(&state->dump);
1974 return error ? error : EOF;
1979 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1981 struct port_dump_state *state = state_;
1984 dpif_port_dump_done(&state->dump);
1991 port_poll(const struct ofproto *ofproto_, char **devnamep)
1993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1994 return dpif_port_poll(ofproto->dpif, devnamep);
1998 port_poll_wait(const struct ofproto *ofproto_)
2000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2001 dpif_port_poll_wait(ofproto->dpif);
2005 port_is_lacp_current(const struct ofport *ofport_)
2007 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2008 return (ofport->bundle && ofport->bundle->lacp
2009 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2013 /* Upcall handling. */
2015 /* Flow miss batching.
2017 * Some dpifs implement operations faster when you hand them off in a batch.
2018 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2019 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2020 * more packets, plus possibly installing the flow in the dpif.
2022 * So far we only batch the operations that affect flow setup time the most.
2023 * It's possible to batch more than that, but the benefit might be minimal. */
2025 struct hmap_node hmap_node;
2027 const struct nlattr *key;
2029 struct list packets;
2032 struct flow_miss_op {
2033 union dpif_op dpif_op;
2034 struct facet *facet;
2037 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2038 * OpenFlow controller as necessary according to their individual
2041 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2042 * ownership is transferred to this function. */
2044 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2045 const struct flow *flow, bool clone)
2047 struct ofputil_packet_in pin;
2049 pin.packet = packet;
2050 pin.in_port = flow->in_port;
2051 pin.reason = OFPR_NO_MATCH;
2052 pin.buffer_id = 0; /* not yet known */
2053 pin.send_len = 0; /* not used for flow table misses */
2054 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2055 clone ? NULL : packet);
2058 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2059 * OpenFlow controller as necessary according to their individual
2062 * 'send_len' should be the number of bytes of 'packet' to send to the
2063 * controller, as specified in the action that caused the packet to be sent.
2065 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2066 * Otherwise, ownership is transferred to this function. */
2068 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2069 uint64_t userdata, const struct flow *flow, bool clone)
2071 struct ofputil_packet_in pin;
2072 struct user_action_cookie cookie;
2074 memcpy(&cookie, &userdata, sizeof(cookie));
2076 pin.packet = packet;
2077 pin.in_port = flow->in_port;
2078 pin.reason = OFPR_ACTION;
2079 pin.buffer_id = 0; /* not yet known */
2080 pin.send_len = cookie.data;
2081 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2082 clone ? NULL : packet);
2086 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2087 const struct ofpbuf *packet)
2089 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2095 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2097 cfm_process_heartbeat(ofport->cfm, packet);
2100 } else if (ofport->bundle && ofport->bundle->lacp
2101 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2103 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2106 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2108 stp_process_packet(ofport, packet);
2115 static struct flow_miss *
2116 flow_miss_create(struct hmap *todo, const struct flow *flow,
2117 const struct nlattr *key, size_t key_len)
2119 uint32_t hash = flow_hash(flow, 0);
2120 struct flow_miss *miss;
2122 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2123 if (flow_equal(&miss->flow, flow)) {
2128 miss = xmalloc(sizeof *miss);
2129 hmap_insert(todo, &miss->hmap_node, hash);
2132 miss->key_len = key_len;
2133 list_init(&miss->packets);
2138 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2139 struct flow_miss_op *ops, size_t *n_ops)
2141 const struct flow *flow = &miss->flow;
2142 struct ofpbuf *packet, *next_packet;
2143 struct facet *facet;
2145 facet = facet_lookup_valid(ofproto, flow);
2147 struct rule_dpif *rule;
2149 rule = rule_dpif_lookup(ofproto, flow, 0);
2151 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2152 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2154 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2155 COVERAGE_INC(ofproto_dpif_no_packet_in);
2156 /* XXX install 'drop' flow entry */
2160 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2164 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2166 list_remove(&packet->list_node);
2167 send_packet_in_miss(ofproto, packet, flow, false);
2173 facet = facet_create(rule, flow);
2176 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2177 list_remove(&packet->list_node);
2178 ofproto->n_matches++;
2180 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2182 * Extra-special case for fail-open mode.
2184 * We are in fail-open mode and the packet matched the fail-open
2185 * rule, but we are connected to a controller too. We should send
2186 * the packet up to the controller in the hope that it will try to
2187 * set up a flow and thereby allow us to exit fail-open.
2189 * See the top-level comment in fail-open.c for more information.
2191 send_packet_in_miss(ofproto, packet, flow, true);
2194 if (!facet->may_install) {
2195 facet_make_actions(ofproto, facet, packet);
2197 if (!execute_controller_action(ofproto, &facet->flow,
2198 facet->actions, facet->actions_len,
2200 struct flow_miss_op *op = &ops[(*n_ops)++];
2201 struct dpif_execute *execute = &op->dpif_op.execute;
2204 execute->type = DPIF_OP_EXECUTE;
2205 execute->key = miss->key;
2206 execute->key_len = miss->key_len;
2208 = (facet->may_install
2210 : xmemdup(facet->actions, facet->actions_len));
2211 execute->actions_len = facet->actions_len;
2212 execute->packet = packet;
2216 if (facet->may_install) {
2217 struct flow_miss_op *op = &ops[(*n_ops)++];
2218 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2221 put->type = DPIF_OP_FLOW_PUT;
2222 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2223 put->key = miss->key;
2224 put->key_len = miss->key_len;
2225 put->actions = facet->actions;
2226 put->actions_len = facet->actions_len;
2232 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2235 struct dpif_upcall *upcall;
2236 struct flow_miss *miss, *next_miss;
2237 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2238 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2247 /* Construct the to-do list.
2249 * This just amounts to extracting the flow from each packet and sticking
2250 * the packets that have the same flow in the same "flow_miss" structure so
2251 * that we can process them together. */
2253 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2254 struct flow_miss *miss;
2257 /* Obtain in_port and tun_id, at least, then set 'flow''s header
2259 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2260 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
2262 /* Handle 802.1ag, LACP, and STP specially. */
2263 if (process_special(ofproto, &flow, upcall->packet)) {
2264 ofpbuf_delete(upcall->packet);
2265 ofproto->n_matches++;
2269 /* Add other packets to a to-do list. */
2270 miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len);
2271 list_push_back(&miss->packets, &upcall->packet->list_node);
2274 /* Process each element in the to-do list, constructing the set of
2275 * operations to batch. */
2277 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2278 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2279 ofpbuf_list_delete(&miss->packets);
2280 hmap_remove(&todo, &miss->hmap_node);
2283 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2284 hmap_destroy(&todo);
2286 /* Execute batch. */
2287 for (i = 0; i < n_ops; i++) {
2288 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2290 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2292 /* Free memory and update facets. */
2293 for (i = 0; i < n_ops; i++) {
2294 struct flow_miss_op *op = &flow_miss_ops[i];
2295 struct dpif_execute *execute;
2296 struct dpif_flow_put *put;
2298 switch (op->dpif_op.type) {
2299 case DPIF_OP_EXECUTE:
2300 execute = &op->dpif_op.execute;
2301 if (op->facet->actions != execute->actions) {
2302 free((struct nlattr *) execute->actions);
2304 ofpbuf_delete((struct ofpbuf *) execute->packet);
2307 case DPIF_OP_FLOW_PUT:
2308 put = &op->dpif_op.flow_put;
2310 op->facet->installed = true;
2318 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2319 struct dpif_upcall *upcall)
2322 struct user_action_cookie cookie;
2324 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2326 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2327 if (ofproto->sflow) {
2328 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2329 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
2331 ofpbuf_delete(upcall->packet);
2333 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2334 COVERAGE_INC(ofproto_dpif_ctlr_action);
2335 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2336 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2339 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2344 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
2346 switch (upcall->type) {
2347 case DPIF_UC_ACTION:
2348 handle_userspace_upcall(ofproto, upcall);
2352 /* The caller handles these. */
2355 case DPIF_N_UC_TYPES:
2357 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
2362 /* Flow expiration. */
2364 static int facet_max_idle(const struct ofproto_dpif *);
2365 static void update_stats(struct ofproto_dpif *);
2366 static void rule_expire(struct rule_dpif *);
2367 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
2369 /* This function is called periodically by run(). Its job is to collect
2370 * updates for the flows that have been installed into the datapath, most
2371 * importantly when they last were used, and then use that information to
2372 * expire flows that have not been used recently.
2374 * Returns the number of milliseconds after which it should be called again. */
2376 expire(struct ofproto_dpif *ofproto)
2378 struct rule_dpif *rule, *next_rule;
2379 struct classifier *table;
2382 /* Update stats for each flow in the datapath. */
2383 update_stats(ofproto);
2385 /* Expire facets that have been idle too long. */
2386 dp_max_idle = facet_max_idle(ofproto);
2387 expire_facets(ofproto, dp_max_idle);
2389 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2390 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2391 struct cls_cursor cursor;
2393 cls_cursor_init(&cursor, table, NULL);
2394 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2399 /* All outstanding data in existing flows has been accounted, so it's a
2400 * good time to do bond rebalancing. */
2401 if (ofproto->has_bonded_bundles) {
2402 struct ofbundle *bundle;
2404 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2406 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2411 return MIN(dp_max_idle, 1000);
2414 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2416 * This function also pushes statistics updates to rules which each facet
2417 * resubmits into. Generally these statistics will be accurate. However, if a
2418 * facet changes the rule it resubmits into at some time in between
2419 * update_stats() runs, it is possible that statistics accrued to the
2420 * old rule will be incorrectly attributed to the new rule. This could be
2421 * avoided by calling update_stats() whenever rules are created or
2422 * deleted. However, the performance impact of making so many calls to the
2423 * datapath do not justify the benefit of having perfectly accurate statistics.
2426 update_stats(struct ofproto_dpif *p)
2428 const struct dpif_flow_stats *stats;
2429 struct dpif_flow_dump dump;
2430 const struct nlattr *key;
2433 dpif_flow_dump_start(&dump, p->dpif);
2434 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2435 struct facet *facet;
2438 if (odp_flow_key_to_flow(key, key_len, &flow)) {
2442 odp_flow_key_format(key, key_len, &s);
2443 VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
2449 facet = facet_find(p, &flow);
2451 if (facet && facet->installed) {
2453 if (stats->n_packets >= facet->dp_packet_count) {
2454 uint64_t extra = stats->n_packets - facet->dp_packet_count;
2455 facet->packet_count += extra;
2457 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2460 if (stats->n_bytes >= facet->dp_byte_count) {
2461 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
2463 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2466 facet->dp_packet_count = stats->n_packets;
2467 facet->dp_byte_count = stats->n_bytes;
2469 facet_update_time(p, facet, stats->used);
2470 facet_account(p, facet);
2471 facet_push_stats(facet);
2473 /* There's a flow in the datapath that we know nothing about.
2475 COVERAGE_INC(facet_unexpected);
2476 dpif_flow_del(p->dpif, key, key_len, NULL);
2479 dpif_flow_dump_done(&dump);
2482 /* Calculates and returns the number of milliseconds of idle time after which
2483 * facets should expire from the datapath and we should fold their statistics
2484 * into their parent rules in userspace. */
2486 facet_max_idle(const struct ofproto_dpif *ofproto)
2489 * Idle time histogram.
2491 * Most of the time a switch has a relatively small number of facets. When
2492 * this is the case we might as well keep statistics for all of them in
2493 * userspace and to cache them in the kernel datapath for performance as
2496 * As the number of facets increases, the memory required to maintain
2497 * statistics about them in userspace and in the kernel becomes
2498 * significant. However, with a large number of facets it is likely that
2499 * only a few of them are "heavy hitters" that consume a large amount of
2500 * bandwidth. At this point, only heavy hitters are worth caching in the
2501 * kernel and maintaining in userspaces; other facets we can discard.
2503 * The technique used to compute the idle time is to build a histogram with
2504 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
2505 * that is installed in the kernel gets dropped in the appropriate bucket.
2506 * After the histogram has been built, we compute the cutoff so that only
2507 * the most-recently-used 1% of facets (but at least
2508 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2509 * the most-recently-used bucket of facets is kept, so actually an
2510 * arbitrary number of facets can be kept in any given expiration run
2511 * (though the next run will delete most of those unless they receive
2514 * This requires a second pass through the facets, in addition to the pass
2515 * made by update_stats(), because the former function never looks
2516 * at uninstallable facets.
2518 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2519 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2520 int buckets[N_BUCKETS] = { 0 };
2521 int total, subtotal, bucket;
2522 struct facet *facet;
2526 total = hmap_count(&ofproto->facets);
2527 if (total <= ofproto->up.flow_eviction_threshold) {
2528 return N_BUCKETS * BUCKET_WIDTH;
2531 /* Build histogram. */
2533 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
2534 long long int idle = now - facet->used;
2535 int bucket = (idle <= 0 ? 0
2536 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2537 : (unsigned int) idle / BUCKET_WIDTH);
2541 /* Find the first bucket whose flows should be expired. */
2542 subtotal = bucket = 0;
2544 subtotal += buckets[bucket++];
2545 } while (bucket < N_BUCKETS &&
2546 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2548 if (VLOG_IS_DBG_ENABLED()) {
2552 ds_put_cstr(&s, "keep");
2553 for (i = 0; i < N_BUCKETS; i++) {
2555 ds_put_cstr(&s, ", drop");
2558 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2561 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2565 return bucket * BUCKET_WIDTH;
2569 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
2571 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
2572 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
2573 struct ofexpired expired;
2575 if (facet->installed) {
2576 struct dpif_flow_stats stats;
2578 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
2580 facet_update_stats(ofproto, facet, &stats);
2583 expired.flow = facet->flow;
2584 expired.packet_count = facet->packet_count;
2585 expired.byte_count = facet->byte_count;
2586 expired.used = facet->used;
2587 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2592 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
2594 long long int cutoff = time_msec() - dp_max_idle;
2595 struct facet *facet, *next_facet;
2597 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
2598 facet_active_timeout(ofproto, facet);
2599 if (facet->used < cutoff) {
2600 facet_remove(ofproto, facet);
2605 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2606 * then delete it entirely. */
2608 rule_expire(struct rule_dpif *rule)
2610 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2611 struct facet *facet, *next_facet;
2615 /* Has 'rule' expired? */
2617 if (rule->up.hard_timeout
2618 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
2619 reason = OFPRR_HARD_TIMEOUT;
2620 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2621 && now > rule->used + rule->up.idle_timeout * 1000) {
2622 reason = OFPRR_IDLE_TIMEOUT;
2627 COVERAGE_INC(ofproto_dpif_expired);
2629 /* Update stats. (This is a no-op if the rule expired due to an idle
2630 * timeout, because that only happens when the rule has no facets left.) */
2631 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2632 facet_remove(ofproto, facet);
2635 /* Get rid of the rule. */
2636 ofproto_rule_expire(&rule->up, reason);
2641 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
2643 * The caller must already have determined that no facet with an identical
2644 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2645 * the ofproto's classifier table.
2647 * The facet will initially have no ODP actions. The caller should fix that
2648 * by calling facet_make_actions(). */
2649 static struct facet *
2650 facet_create(struct rule_dpif *rule, const struct flow *flow)
2652 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2653 struct facet *facet;
2655 facet = xzalloc(sizeof *facet);
2656 facet->used = time_msec();
2657 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2658 list_push_back(&rule->facets, &facet->list_node);
2660 facet->flow = *flow;
2661 netflow_flow_init(&facet->nf_flow);
2662 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2668 facet_free(struct facet *facet)
2670 free(facet->actions);
2675 execute_controller_action(struct ofproto_dpif *ofproto,
2676 const struct flow *flow,
2677 const struct nlattr *odp_actions, size_t actions_len,
2678 struct ofpbuf *packet)
2681 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
2682 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
2683 /* As an optimization, avoid a round-trip from userspace to kernel to
2684 * userspace. This also avoids possibly filling up kernel packet
2685 * buffers along the way.
2687 * This optimization will not accidentally catch sFlow
2688 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
2689 * inside OVS_ACTION_ATTR_SAMPLE. */
2690 const struct nlattr *nla;
2692 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
2693 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
2701 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2702 * 'packet', which arrived on 'in_port'.
2704 * Takes ownership of 'packet'. */
2706 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2707 const struct nlattr *odp_actions, size_t actions_len,
2708 struct ofpbuf *packet)
2710 struct odputil_keybuf keybuf;
2714 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
2719 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2720 odp_flow_key_from_flow(&key, flow);
2722 error = dpif_execute(ofproto->dpif, key.data, key.size,
2723 odp_actions, actions_len, packet);
2725 ofpbuf_delete(packet);
2729 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2730 * statistics appropriately. 'packet' must have at least sizeof(struct
2731 * ofp_packet_in) bytes of headroom.
2733 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2734 * applying flow_extract() to 'packet' would yield the same flow as
2737 * 'facet' must have accurately composed datapath actions; that is, it must
2738 * not be in need of revalidation.
2740 * Takes ownership of 'packet'. */
2742 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2743 struct ofpbuf *packet)
2745 struct dpif_flow_stats stats;
2747 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2749 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2750 stats.used = time_msec();
2751 if (execute_odp_actions(ofproto, &facet->flow,
2752 facet->actions, facet->actions_len, packet)) {
2753 facet_update_stats(ofproto, facet, &stats);
2757 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2759 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2760 * rule's statistics, via facet_uninstall().
2762 * - Removes 'facet' from its rule and from ofproto->facets.
2765 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2767 facet_uninstall(ofproto, facet);
2768 facet_flush_stats(ofproto, facet);
2769 hmap_remove(&ofproto->facets, &facet->hmap_node);
2770 list_remove(&facet->list_node);
2774 /* Composes the datapath actions for 'facet' based on its rule's actions. */
2776 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2777 const struct ofpbuf *packet)
2779 const struct rule_dpif *rule = facet->rule;
2780 struct ofpbuf *odp_actions;
2781 struct action_xlate_ctx ctx;
2783 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2784 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2785 facet->tags = ctx.tags;
2786 facet->may_install = ctx.may_set_up_flow;
2787 facet->has_learn = ctx.has_learn;
2788 facet->has_normal = ctx.has_normal;
2789 facet->nf_flow.output_iface = ctx.nf_output_iface;
2791 if (facet->actions_len != odp_actions->size
2792 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2793 free(facet->actions);
2794 facet->actions_len = odp_actions->size;
2795 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2798 ofpbuf_delete(odp_actions);
2801 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2802 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2803 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2804 * since 'facet' was last updated.
2806 * Returns 0 if successful, otherwise a positive errno value.*/
2808 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2809 const struct nlattr *actions, size_t actions_len,
2810 struct dpif_flow_stats *stats)
2812 struct odputil_keybuf keybuf;
2813 enum dpif_flow_put_flags flags;
2817 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2819 flags |= DPIF_FP_ZERO_STATS;
2822 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2823 odp_flow_key_from_flow(&key, &facet->flow);
2825 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2826 actions, actions_len, stats);
2829 facet_reset_dp_stats(facet, stats);
2835 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2836 * 'zero_stats' is true, clears any existing statistics from the datapath for
2839 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2841 struct dpif_flow_stats stats;
2843 if (facet->may_install
2844 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2845 zero_stats ? &stats : NULL)) {
2846 facet->installed = true;
2851 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
2854 const struct nlattr *a;
2858 if (facet->byte_count <= facet->accounted_bytes) {
2861 n_bytes = facet->byte_count - facet->accounted_bytes;
2862 facet->accounted_bytes = facet->byte_count;
2864 /* Feed information from the active flows back into the learning table to
2865 * ensure that table is always in sync with what is actually flowing
2866 * through the datapath. */
2867 if (facet->has_learn || facet->has_normal) {
2868 struct action_xlate_ctx ctx;
2870 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2871 ctx.may_learn = true;
2872 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
2873 facet->rule->up.n_actions));
2876 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
2880 /* This loop feeds byte counters to bond_account() for rebalancing to use
2881 * as a basis. We also need to track the actual VLAN on which the packet
2882 * is going to be sent to ensure that it matches the one passed to
2883 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2885 vlan_tci = facet->flow.vlan_tci;
2886 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2887 struct ofport_dpif *port;
2889 switch (nl_attr_type(a)) {
2890 const struct nlattr *nested;
2891 case OVS_ACTION_ATTR_OUTPUT:
2892 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2893 if (port && port->bundle && port->bundle->bond) {
2894 bond_account(port->bundle->bond, &facet->flow,
2895 vlan_tci_to_vid(vlan_tci), n_bytes);
2899 case OVS_ACTION_ATTR_POP:
2900 if (nl_attr_get_u16(a) == OVS_KEY_ATTR_8021Q) {
2901 vlan_tci = htons(0);
2905 case OVS_ACTION_ATTR_PUSH:
2906 nested = nl_attr_get(a);
2907 if (nl_attr_type(nested) == OVS_KEY_ATTR_8021Q) {
2908 const struct ovs_key_8021q *q_key;
2910 q_key = nl_attr_get_unspec(nested, sizeof(*q_key));
2911 vlan_tci = q_key->q_tci;
2918 /* If 'rule' is installed in the datapath, uninstalls it. */
2920 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2922 if (facet->installed) {
2923 struct odputil_keybuf keybuf;
2924 struct dpif_flow_stats stats;
2928 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2929 odp_flow_key_from_flow(&key, &facet->flow);
2931 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2932 facet_reset_dp_stats(facet, &stats);
2934 facet_update_stats(p, facet, &stats);
2936 facet->installed = false;
2938 assert(facet->dp_packet_count == 0);
2939 assert(facet->dp_byte_count == 0);
2943 /* Returns true if the only action for 'facet' is to send to the controller.
2944 * (We don't report NetFlow expiration messages for such facets because they
2945 * are just part of the control logic for the network, not real traffic). */
2947 facet_is_controller_flow(struct facet *facet)
2950 && facet->rule->up.n_actions == 1
2951 && action_outputs_to_port(&facet->rule->up.actions[0],
2952 htons(OFPP_CONTROLLER)));
2955 /* Resets 'facet''s datapath statistics counters. This should be called when
2956 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2957 * it should contain the statistics returned by dpif when 'facet' was reset in
2958 * the datapath. 'stats' will be modified to only included statistics new
2959 * since 'facet' was last updated. */
2961 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2963 if (stats && facet->dp_packet_count <= stats->n_packets
2964 && facet->dp_byte_count <= stats->n_bytes) {
2965 stats->n_packets -= facet->dp_packet_count;
2966 stats->n_bytes -= facet->dp_byte_count;
2969 facet->dp_packet_count = 0;
2970 facet->dp_byte_count = 0;
2973 /* Folds all of 'facet''s statistics into its rule. Also updates the
2974 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2975 * 'facet''s statistics in the datapath should have been zeroed and folded into
2976 * its packet and byte counts before this function is called. */
2978 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2980 assert(!facet->dp_byte_count);
2981 assert(!facet->dp_packet_count);
2983 facet_push_stats(facet);
2984 facet_account(ofproto, facet);
2986 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2987 struct ofexpired expired;
2988 expired.flow = facet->flow;
2989 expired.packet_count = facet->packet_count;
2990 expired.byte_count = facet->byte_count;
2991 expired.used = facet->used;
2992 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2995 facet->rule->packet_count += facet->packet_count;
2996 facet->rule->byte_count += facet->byte_count;
2998 /* Reset counters to prevent double counting if 'facet' ever gets
3000 facet_reset_counters(facet);
3002 netflow_flow_clear(&facet->nf_flow);
3005 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3006 * Returns it if found, otherwise a null pointer.
3008 * The returned facet might need revalidation; use facet_lookup_valid()
3009 * instead if that is important. */
3010 static struct facet *
3011 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3013 struct facet *facet;
3015 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3017 if (flow_equal(flow, &facet->flow)) {
3025 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3026 * Returns it if found, otherwise a null pointer.
3028 * The returned facet is guaranteed to be valid. */
3029 static struct facet *
3030 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3032 struct facet *facet = facet_find(ofproto, flow);
3034 /* The facet we found might not be valid, since we could be in need of
3035 * revalidation. If it is not valid, don't return it. */
3037 && (ofproto->need_revalidate
3038 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3039 && !facet_revalidate(ofproto, facet)) {
3040 COVERAGE_INC(facet_invalidated);
3047 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3049 * - If the rule found is different from 'facet''s current rule, moves
3050 * 'facet' to the new rule and recompiles its actions.
3052 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3053 * where it is and recompiles its actions anyway.
3055 * - If there is none, destroys 'facet'.
3057 * Returns true if 'facet' still exists, false if it has been destroyed. */
3059 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3061 struct action_xlate_ctx ctx;
3062 struct ofpbuf *odp_actions;
3063 struct rule_dpif *new_rule;
3064 bool actions_changed;
3066 COVERAGE_INC(facet_revalidate);
3068 /* Determine the new rule. */
3069 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3071 /* No new rule, so delete the facet. */
3072 facet_remove(ofproto, facet);
3076 /* Calculate new datapath actions.
3078 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3079 * emit a NetFlow expiration and, if so, we need to have the old state
3080 * around to properly compose it. */
3081 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
3082 odp_actions = xlate_actions(&ctx,
3083 new_rule->up.actions, new_rule->up.n_actions);
3084 actions_changed = (facet->actions_len != odp_actions->size
3085 || memcmp(facet->actions, odp_actions->data,
3086 facet->actions_len));
3088 /* If the datapath actions changed or the installability changed,
3089 * then we need to talk to the datapath. */
3090 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
3091 if (ctx.may_set_up_flow) {
3092 struct dpif_flow_stats stats;
3094 facet_put__(ofproto, facet,
3095 odp_actions->data, odp_actions->size, &stats);
3096 facet_update_stats(ofproto, facet, &stats);
3098 facet_uninstall(ofproto, facet);
3101 /* The datapath flow is gone or has zeroed stats, so push stats out of
3102 * 'facet' into 'rule'. */
3103 facet_flush_stats(ofproto, facet);
3106 /* Update 'facet' now that we've taken care of all the old state. */
3107 facet->tags = ctx.tags;
3108 facet->nf_flow.output_iface = ctx.nf_output_iface;
3109 facet->may_install = ctx.may_set_up_flow;
3110 facet->has_learn = ctx.has_learn;
3111 facet->has_normal = ctx.has_normal;
3112 if (actions_changed) {
3113 free(facet->actions);
3114 facet->actions_len = odp_actions->size;
3115 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
3117 if (facet->rule != new_rule) {
3118 COVERAGE_INC(facet_changed_rule);
3119 list_remove(&facet->list_node);
3120 list_push_back(&new_rule->facets, &facet->list_node);
3121 facet->rule = new_rule;
3122 facet->used = new_rule->up.created;
3123 facet->rs_used = facet->used;
3126 ofpbuf_delete(odp_actions);
3131 /* Updates 'facet''s used time. Caller is responsible for calling
3132 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3134 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3137 if (used > facet->used) {
3139 if (used > facet->rule->used) {
3140 facet->rule->used = used;
3142 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3146 /* Folds the statistics from 'stats' into the counters in 'facet'.
3148 * Because of the meaning of a facet's counters, it only makes sense to do this
3149 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3150 * packet that was sent by hand or if it represents statistics that have been
3151 * cleared out of the datapath. */
3153 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
3154 const struct dpif_flow_stats *stats)
3156 if (stats->n_packets || stats->used > facet->used) {
3157 facet_update_time(ofproto, facet, stats->used);
3158 facet->packet_count += stats->n_packets;
3159 facet->byte_count += stats->n_bytes;
3160 facet_push_stats(facet);
3161 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3166 facet_reset_counters(struct facet *facet)
3168 facet->packet_count = 0;
3169 facet->byte_count = 0;
3170 facet->rs_packet_count = 0;
3171 facet->rs_byte_count = 0;
3172 facet->accounted_bytes = 0;
3176 facet_push_stats(struct facet *facet)
3178 uint64_t rs_packets, rs_bytes;
3180 assert(facet->packet_count >= facet->rs_packet_count);
3181 assert(facet->byte_count >= facet->rs_byte_count);
3182 assert(facet->used >= facet->rs_used);
3184 rs_packets = facet->packet_count - facet->rs_packet_count;
3185 rs_bytes = facet->byte_count - facet->rs_byte_count;
3187 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3188 facet->rs_packet_count = facet->packet_count;
3189 facet->rs_byte_count = facet->byte_count;
3190 facet->rs_used = facet->used;
3192 flow_push_stats(facet->rule, &facet->flow,
3193 rs_packets, rs_bytes, facet->used);
3197 struct ofproto_push {
3198 struct action_xlate_ctx ctx;
3205 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3207 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3210 rule->packet_count += push->packets;
3211 rule->byte_count += push->bytes;
3212 rule->used = MAX(push->used, rule->used);
3216 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3217 * 'rule''s actions. */
3219 flow_push_stats(const struct rule_dpif *rule,
3220 struct flow *flow, uint64_t packets, uint64_t bytes,
3223 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3224 struct ofproto_push push;
3226 push.packets = packets;
3230 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3231 push.ctx.resubmit_hook = push_resubmit;
3232 ofpbuf_delete(xlate_actions(&push.ctx,
3233 rule->up.actions, rule->up.n_actions));
3238 static struct rule_dpif *
3239 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3242 struct cls_rule *cls_rule;
3243 struct classifier *cls;
3245 if (table_id >= N_TABLES) {
3249 cls = &ofproto->up.tables[table_id];
3250 if (flow->tos_frag & FLOW_FRAG_ANY
3251 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3252 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3253 * are unavailable. */
3254 struct flow ofpc_normal_flow = *flow;
3255 ofpc_normal_flow.tp_src = htons(0);
3256 ofpc_normal_flow.tp_dst = htons(0);
3257 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3259 cls_rule = classifier_lookup(cls, flow);
3261 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3265 complete_operation(struct rule_dpif *rule)
3267 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3269 rule_invalidate(rule);
3271 struct dpif_completion *c = xmalloc(sizeof *c);
3272 c->op = rule->up.pending;
3273 list_push_back(&ofproto->completions, &c->list_node);
3275 ofoperation_complete(rule->up.pending, 0);
3279 static struct rule *
3282 struct rule_dpif *rule = xmalloc(sizeof *rule);
3287 rule_dealloc(struct rule *rule_)
3289 struct rule_dpif *rule = rule_dpif_cast(rule_);
3294 rule_construct(struct rule *rule_)
3296 struct rule_dpif *rule = rule_dpif_cast(rule_);
3297 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3298 struct rule_dpif *victim;
3302 error = validate_actions(rule->up.actions, rule->up.n_actions,
3303 &rule->up.cr.flow, ofproto->max_ports);
3308 rule->used = rule->up.created;
3309 rule->packet_count = 0;
3310 rule->byte_count = 0;
3312 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3313 if (victim && !list_is_empty(&victim->facets)) {
3314 struct facet *facet;
3316 rule->facets = victim->facets;
3317 list_moved(&rule->facets);
3318 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3319 /* XXX: We're only clearing our local counters here. It's possible
3320 * that quite a few packets are unaccounted for in the datapath
3321 * statistics. These will be accounted to the new rule instead of
3322 * cleared as required. This could be fixed by clearing out the
3323 * datapath statistics for this facet, but currently it doesn't
3325 facet_reset_counters(facet);
3329 /* Must avoid list_moved() in this case. */
3330 list_init(&rule->facets);
3333 table_id = rule->up.table_id;
3334 rule->tag = (victim ? victim->tag
3336 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3337 ofproto->tables[table_id].basis));
3339 complete_operation(rule);
3344 rule_destruct(struct rule *rule_)
3346 struct rule_dpif *rule = rule_dpif_cast(rule_);
3347 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3348 struct facet *facet, *next_facet;
3350 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3351 facet_revalidate(ofproto, facet);
3354 complete_operation(rule);
3358 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3360 struct rule_dpif *rule = rule_dpif_cast(rule_);
3361 struct facet *facet;
3363 /* Start from historical data for 'rule' itself that are no longer tracked
3364 * in facets. This counts, for example, facets that have expired. */
3365 *packets = rule->packet_count;
3366 *bytes = rule->byte_count;
3368 /* Add any statistics that are tracked by facets. This includes
3369 * statistical data recently updated by ofproto_update_stats() as well as
3370 * stats for packets that were executed "by hand" via dpif_execute(). */
3371 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3372 *packets += facet->packet_count;
3373 *bytes += facet->byte_count;
3378 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
3380 struct rule_dpif *rule = rule_dpif_cast(rule_);
3381 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3382 struct action_xlate_ctx ctx;
3383 struct ofpbuf *odp_actions;
3384 struct facet *facet;
3387 /* First look for a related facet. If we find one, account it to that. */
3388 facet = facet_lookup_valid(ofproto, flow);
3389 if (facet && facet->rule == rule) {
3390 if (!facet->may_install) {
3391 facet_make_actions(ofproto, facet, packet);
3393 facet_execute(ofproto, facet, packet);
3397 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
3398 * create a new facet for it and use that. */
3399 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
3400 facet = facet_create(rule, flow);
3401 facet_make_actions(ofproto, facet, packet);
3402 facet_execute(ofproto, facet, packet);
3403 facet_install(ofproto, facet, true);
3407 /* We can't account anything to a facet. If we were to try, then that
3408 * facet would have a non-matching rule, busting our invariants. */
3409 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3410 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3411 size = packet->size;
3412 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3413 odp_actions->size, packet)) {
3414 rule->used = time_msec();
3415 rule->packet_count++;
3416 rule->byte_count += size;
3417 flow_push_stats(rule, flow, 1, size, rule->used);
3419 ofpbuf_delete(odp_actions);
3425 rule_modify_actions(struct rule *rule_)
3427 struct rule_dpif *rule = rule_dpif_cast(rule_);
3428 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3431 error = validate_actions(rule->up.actions, rule->up.n_actions,
3432 &rule->up.cr.flow, ofproto->max_ports);
3434 ofoperation_complete(rule->up.pending, error);
3438 complete_operation(rule);
3441 /* Sends 'packet' out of port 'odp_port' within 'ofproto'.
3442 * Returns 0 if successful, otherwise a positive errno value. */
3444 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
3445 const struct ofpbuf *packet)
3447 struct ofpbuf key, odp_actions;
3448 struct odputil_keybuf keybuf;
3452 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
3453 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3454 odp_flow_key_from_flow(&key, &flow);
3456 ofpbuf_init(&odp_actions, 32);
3457 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
3459 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3460 error = dpif_execute(ofproto->dpif,
3462 odp_actions.data, odp_actions.size,
3464 ofpbuf_uninit(&odp_actions);
3467 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
3468 ofproto->up.name, odp_port, strerror(error));
3473 /* OpenFlow to datapath action translation. */
3475 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
3476 struct action_xlate_ctx *ctx);
3477 static void xlate_normal(struct action_xlate_ctx *);
3480 put_userspace_action(const struct ofproto_dpif *ofproto,
3481 struct ofpbuf *odp_actions,
3482 const struct flow *flow,
3483 const struct user_action_cookie *cookie)
3488 pid = dpif_port_get_pid(ofproto->dpif,
3489 ofp_port_to_odp_port(flow->in_port));
3491 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
3492 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
3493 nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
3494 cookie, sizeof *cookie);
3495 nl_msg_end_nested(odp_actions, offset);
3497 return odp_actions->size - NLA_ALIGN(sizeof *cookie);
3500 /* Compose SAMPLE action for sFlow. */
3502 compose_sflow_action(const struct ofproto_dpif *ofproto,
3503 struct ofpbuf *odp_actions,
3504 const struct flow *flow,
3507 uint32_t port_ifindex;
3508 uint32_t probability;
3509 struct user_action_cookie cookie;
3510 size_t sample_offset, actions_offset;
3511 int cookie_offset, n_output;
3513 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
3517 if (odp_port == OVSP_NONE) {
3521 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
3525 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
3527 /* Number of packets out of UINT_MAX to sample. */
3528 probability = dpif_sflow_get_probability(ofproto->sflow);
3529 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
3531 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
3533 cookie.type = USER_ACTION_COOKIE_SFLOW;
3534 cookie.data = port_ifindex;
3535 cookie.n_output = n_output;
3536 cookie.vlan_tci = 0;
3537 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
3539 nl_msg_end_nested(odp_actions, actions_offset);
3540 nl_msg_end_nested(odp_actions, sample_offset);
3541 return cookie_offset;
3544 /* SAMPLE action must be first action in any given list of actions.
3545 * At this point we do not have all information required to build it. So try to
3546 * build sample action as complete as possible. */
3548 add_sflow_action(struct action_xlate_ctx *ctx)
3550 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
3552 &ctx->flow, OVSP_NONE);
3553 ctx->sflow_odp_port = 0;
3554 ctx->sflow_n_outputs = 0;
3557 /* Fix SAMPLE action according to data collected while composing ODP actions.
3558 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
3559 * USERSPACE action's user-cookie which is required for sflow. */
3561 fix_sflow_action(struct action_xlate_ctx *ctx)
3563 const struct flow *base = &ctx->base_flow;
3564 struct user_action_cookie *cookie;
3566 if (!ctx->user_cookie_offset) {
3570 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
3572 assert(cookie != NULL);
3573 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
3575 if (ctx->sflow_n_outputs) {
3576 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
3577 ctx->sflow_odp_port);
3579 if (ctx->sflow_n_outputs >= 255) {
3580 cookie->n_output = 255;
3582 cookie->n_output = ctx->sflow_n_outputs;
3584 cookie->vlan_tci = base->vlan_tci;
3588 commit_action__(struct ofpbuf *odp_actions,
3589 enum ovs_action_attr act_type,
3590 enum ovs_key_attr key_type,
3591 const void *key, size_t key_size)
3593 size_t offset = nl_msg_start_nested(odp_actions, act_type);
3595 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
3596 nl_msg_end_nested(odp_actions, offset);
3600 commit_set_tun_id_action(const struct flow *flow, struct flow *base,
3601 struct ofpbuf *odp_actions)
3603 if (base->tun_id == flow->tun_id) {
3606 base->tun_id = flow->tun_id;
3608 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3609 OVS_KEY_ATTR_TUN_ID, &base->tun_id, sizeof(base->tun_id));
3613 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
3614 struct ofpbuf *odp_actions)
3616 struct ovs_key_ethernet eth_key;
3618 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
3619 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
3623 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
3624 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
3626 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
3627 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
3629 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3630 OVS_KEY_ATTR_ETHERNET, ð_key, sizeof(eth_key));
3634 commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci)
3636 struct flow *base = &ctx->base_flow;
3638 if (base->vlan_tci == new_tci) {
3642 if (base->vlan_tci & htons(VLAN_CFI)) {
3643 nl_msg_put_u16(ctx->odp_actions, OVS_ACTION_ATTR_POP,
3644 OVS_KEY_ATTR_8021Q);
3647 if (new_tci & htons(VLAN_CFI)) {
3648 struct ovs_key_8021q q_key;
3650 q_key.q_tpid = htons(ETH_TYPE_VLAN);
3651 q_key.q_tci = new_tci & ~htons(VLAN_CFI);
3653 commit_action__(ctx->odp_actions, OVS_ACTION_ATTR_PUSH,
3654 OVS_KEY_ATTR_8021Q, &q_key, sizeof(q_key));
3656 base->vlan_tci = new_tci;
3660 commit_set_nw_action(const struct flow *flow, struct flow *base,
3661 struct ofpbuf *odp_actions)
3663 int frag = base->tos_frag & FLOW_FRAG_MASK;
3664 struct ovs_key_ipv4 ipv4_key;
3666 if (base->dl_type != htons(ETH_TYPE_IP) ||
3667 !base->nw_src || !base->nw_dst) {
3671 if (base->nw_src == flow->nw_src &&
3672 base->nw_dst == flow->nw_dst &&
3673 base->tos_frag == flow->tos_frag) {
3678 memset(&ipv4_key, 0, sizeof(ipv4_key));
3679 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
3680 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
3681 ipv4_key.ipv4_proto = base->nw_proto;
3682 ipv4_key.ipv4_tos = flow->tos_frag & IP_DSCP_MASK;
3683 ipv4_key.ipv4_frag = (frag == 0 ? OVS_FRAG_TYPE_NONE
3684 : frag == FLOW_FRAG_ANY ? OVS_FRAG_TYPE_FIRST
3685 : OVS_FRAG_TYPE_LATER);
3687 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3688 OVS_KEY_ATTR_IPV4, &ipv4_key, sizeof(ipv4_key));
3692 commit_set_port_action(const struct flow *flow, struct flow *base,
3693 struct ofpbuf *odp_actions)
3695 if (!base->tp_src || !base->tp_dst) {
3699 if (base->tp_src == flow->tp_src &&
3700 base->tp_dst == flow->tp_dst) {
3704 if (flow->nw_proto == IPPROTO_TCP) {
3705 struct ovs_key_tcp port_key;
3707 port_key.tcp_src = base->tp_src = flow->tp_src;
3708 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
3710 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3711 OVS_KEY_ATTR_TCP, &port_key, sizeof(port_key));
3713 } else if (flow->nw_proto == IPPROTO_UDP) {
3714 struct ovs_key_udp port_key;
3716 port_key.udp_src = base->tp_src = flow->tp_src;
3717 port_key.udp_dst = base->tp_dst = flow->tp_dst;
3719 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3720 OVS_KEY_ATTR_UDP, &port_key, sizeof(port_key));
3725 commit_priority_action(struct action_xlate_ctx *ctx)
3727 if (ctx->base_priority == ctx->priority) {
3731 if (ctx->priority) {
3732 nl_msg_put_u32(ctx->odp_actions,
3733 OVS_ACTION_ATTR_SET_PRIORITY, ctx->priority);
3735 nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_PRIORITY);
3737 ctx->base_priority = ctx->priority;
3741 commit_odp_actions(struct action_xlate_ctx *ctx)
3743 const struct flow *flow = &ctx->flow;
3744 struct flow *base = &ctx->base_flow;
3745 struct ofpbuf *odp_actions = ctx->odp_actions;
3747 commit_set_tun_id_action(flow, base, odp_actions);
3748 commit_set_ether_addr_action(flow, base, odp_actions);
3749 commit_vlan_action(ctx, flow->vlan_tci);
3750 commit_set_nw_action(flow, base, odp_actions);
3751 commit_set_port_action(flow, base, odp_actions);
3752 commit_priority_action(ctx);
3756 compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port)
3758 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3759 ctx->sflow_odp_port = odp_port;
3760 ctx->sflow_n_outputs++;
3764 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3766 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3767 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
3770 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
3771 || !stp_forward_in_state(ofport->stp_state)) {
3772 /* Forwarding disabled on port. */
3777 * We don't have an ofport record for this port, but it doesn't hurt to
3778 * allow forwarding to it anyhow. Maybe such a port will appear later
3779 * and we're pre-populating the flow table.
3783 commit_odp_actions(ctx);
3784 compose_output_action(ctx, odp_port);
3785 ctx->nf_output_iface = ofp_port;
3789 xlate_table_action(struct action_xlate_ctx *ctx,
3790 uint16_t in_port, uint8_t table_id)
3792 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
3793 struct ofproto_dpif *ofproto = ctx->ofproto;
3794 struct rule_dpif *rule;
3795 uint16_t old_in_port;
3796 uint8_t old_table_id;
3798 old_table_id = ctx->table_id;
3799 ctx->table_id = table_id;
3801 /* Look up a flow with 'in_port' as the input port. */
3802 old_in_port = ctx->flow.in_port;
3803 ctx->flow.in_port = in_port;
3804 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
3807 if (table_id > 0 && table_id < N_TABLES) {
3808 struct table_dpif *table = &ofproto->tables[table_id];
3809 if (table->other_table) {
3812 : rule_calculate_tag(&ctx->flow,
3813 &table->other_table->wc,
3818 /* Restore the original input port. Otherwise OFPP_NORMAL and
3819 * OFPP_IN_PORT will have surprising behavior. */
3820 ctx->flow.in_port = old_in_port;
3822 if (ctx->resubmit_hook) {
3823 ctx->resubmit_hook(ctx, rule);
3828 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
3832 ctx->table_id = old_table_id;
3834 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3836 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
3837 MAX_RESUBMIT_RECURSION);
3842 xlate_resubmit_table(struct action_xlate_ctx *ctx,
3843 const struct nx_action_resubmit *nar)
3848 in_port = (nar->in_port == htons(OFPP_IN_PORT)
3850 : ntohs(nar->in_port));
3851 table_id = nar->table == 255 ? ctx->table_id : nar->table;
3853 xlate_table_action(ctx, in_port, table_id);
3857 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
3859 struct ofport_dpif *ofport;
3861 commit_odp_actions(ctx);
3862 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
3863 uint16_t ofp_port = ofport->up.ofp_port;
3864 if (ofp_port != ctx->flow.in_port
3865 && !(ofport->up.opp.config & mask)
3866 && stp_forward_in_state(ofport->stp_state)) {
3867 compose_output_action(ctx, ofport->odp_port);
3871 ctx->nf_output_iface = NF_OUT_FLOOD;
3875 compose_controller_action(struct action_xlate_ctx *ctx, int len)
3877 struct user_action_cookie cookie;
3879 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
3881 cookie.n_output = 0;
3882 cookie.vlan_tci = 0;
3883 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
3887 xlate_output_action__(struct action_xlate_ctx *ctx,
3888 uint16_t port, uint16_t max_len)
3890 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
3892 ctx->nf_output_iface = NF_OUT_DROP;
3896 add_output_action(ctx, ctx->flow.in_port);
3899 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
3905 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
3908 flood_packets(ctx, htonl(0));
3910 case OFPP_CONTROLLER:
3911 commit_odp_actions(ctx);
3912 compose_controller_action(ctx, max_len);
3915 add_output_action(ctx, OFPP_LOCAL);
3920 if (port != ctx->flow.in_port) {
3921 add_output_action(ctx, port);
3926 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3927 ctx->nf_output_iface = NF_OUT_FLOOD;
3928 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3929 ctx->nf_output_iface = prev_nf_output_iface;
3930 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3931 ctx->nf_output_iface != NF_OUT_FLOOD) {
3932 ctx->nf_output_iface = NF_OUT_MULTI;
3937 xlate_output_reg_action(struct action_xlate_ctx *ctx,
3938 const struct nx_action_output_reg *naor)
3942 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
3944 if (ofp_port <= UINT16_MAX) {
3945 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
3950 xlate_output_action(struct action_xlate_ctx *ctx,
3951 const struct ofp_action_output *oao)
3953 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3957 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3958 const struct ofp_action_enqueue *oae)
3960 uint16_t ofp_port, odp_port;
3961 uint32_t ctx_priority, priority;
3964 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3967 /* Fall back to ordinary output action. */
3968 xlate_output_action__(ctx, ntohs(oae->port), 0);
3972 /* Figure out datapath output port. */
3973 ofp_port = ntohs(oae->port);
3974 if (ofp_port == OFPP_IN_PORT) {
3975 ofp_port = ctx->flow.in_port;
3976 } else if (ofp_port == ctx->flow.in_port) {
3979 odp_port = ofp_port_to_odp_port(ofp_port);
3981 /* Add datapath actions. */
3982 ctx_priority = ctx->priority;
3983 ctx->priority = priority;
3984 add_output_action(ctx, odp_port);
3985 ctx->priority = ctx_priority;
3987 /* Update NetFlow output port. */
3988 if (ctx->nf_output_iface == NF_OUT_DROP) {
3989 ctx->nf_output_iface = odp_port;
3990 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3991 ctx->nf_output_iface = NF_OUT_MULTI;
3996 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3997 const struct nx_action_set_queue *nasq)
4002 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4005 /* Couldn't translate queue to a priority, so ignore. A warning
4006 * has already been logged. */
4010 ctx->priority = priority;
4013 struct xlate_reg_state {
4019 xlate_autopath(struct action_xlate_ctx *ctx,
4020 const struct nx_action_autopath *naa)
4022 uint16_t ofp_port = ntohl(naa->id);
4023 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4025 if (!port || !port->bundle) {
4026 ofp_port = OFPP_NONE;
4027 } else if (port->bundle->bond) {
4028 /* Autopath does not support VLAN hashing. */
4029 struct ofport_dpif *slave = bond_choose_output_slave(
4030 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4032 ofp_port = slave->up.ofp_port;
4035 autopath_execute(naa, &ctx->flow, ofp_port);
4039 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4041 struct ofproto_dpif *ofproto = ofproto_;
4042 struct ofport_dpif *port;
4052 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4055 port = get_ofp_port(ofproto, ofp_port);
4056 return port ? port->may_enable : false;
4061 xlate_learn_action(struct action_xlate_ctx *ctx,
4062 const struct nx_action_learn *learn)
4064 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4065 struct ofputil_flow_mod fm;
4068 learn_execute(learn, &ctx->flow, &fm);
4070 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4071 if (error && !VLOG_DROP_WARN(&rl)) {
4072 char *msg = ofputil_error_to_string(error);
4073 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4081 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4083 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4084 ? htonl(OFPPC_NO_RECV_STP)
4085 : htonl(OFPPC_NO_RECV))) {
4089 /* Only drop packets here if both forwarding and learning are
4090 * disabled. If just learning is enabled, we need to have
4091 * OFPP_NORMAL and the learning action have a look at the packet
4092 * before we can drop it. */
4093 if (!stp_forward_in_state(port->stp_state)
4094 && !stp_learn_in_state(port->stp_state)) {
4102 do_xlate_actions(const union ofp_action *in, size_t n_in,
4103 struct action_xlate_ctx *ctx)
4105 const struct ofport_dpif *port;
4106 const union ofp_action *ia;
4109 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4110 if (port && !may_receive(port, ctx)) {
4111 /* Drop this flow. */
4115 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4116 const struct ofp_action_dl_addr *oada;
4117 const struct nx_action_resubmit *nar;
4118 const struct nx_action_set_tunnel *nast;
4119 const struct nx_action_set_queue *nasq;
4120 const struct nx_action_multipath *nam;
4121 const struct nx_action_autopath *naa;
4122 const struct nx_action_bundle *nab;
4123 const struct nx_action_output_reg *naor;
4124 enum ofputil_action_code code;
4131 code = ofputil_decode_action_unsafe(ia);
4133 case OFPUTIL_OFPAT_OUTPUT:
4134 xlate_output_action(ctx, &ia->output);
4137 case OFPUTIL_OFPAT_SET_VLAN_VID:
4138 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4139 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4142 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4143 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4144 ctx->flow.vlan_tci |= htons(
4145 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4148 case OFPUTIL_OFPAT_STRIP_VLAN:
4149 ctx->flow.vlan_tci = htons(0);
4152 case OFPUTIL_OFPAT_SET_DL_SRC:
4153 oada = ((struct ofp_action_dl_addr *) ia);
4154 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4157 case OFPUTIL_OFPAT_SET_DL_DST:
4158 oada = ((struct ofp_action_dl_addr *) ia);
4159 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4162 case OFPUTIL_OFPAT_SET_NW_SRC:
4163 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4166 case OFPUTIL_OFPAT_SET_NW_DST:
4167 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4170 case OFPUTIL_OFPAT_SET_NW_TOS:
4171 ctx->flow.tos_frag &= ~IP_DSCP_MASK;
4172 ctx->flow.tos_frag |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4175 case OFPUTIL_OFPAT_SET_TP_SRC:
4176 ctx->flow.tp_src = ia->tp_port.tp_port;
4179 case OFPUTIL_OFPAT_SET_TP_DST:
4180 ctx->flow.tp_dst = ia->tp_port.tp_port;
4183 case OFPUTIL_OFPAT_ENQUEUE:
4184 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4187 case OFPUTIL_NXAST_RESUBMIT:
4188 nar = (const struct nx_action_resubmit *) ia;
4189 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4192 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4193 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4196 case OFPUTIL_NXAST_SET_TUNNEL:
4197 nast = (const struct nx_action_set_tunnel *) ia;
4198 tun_id = htonll(ntohl(nast->tun_id));
4199 ctx->flow.tun_id = tun_id;
4202 case OFPUTIL_NXAST_SET_QUEUE:
4203 nasq = (const struct nx_action_set_queue *) ia;
4204 xlate_set_queue_action(ctx, nasq);
4207 case OFPUTIL_NXAST_POP_QUEUE:
4211 case OFPUTIL_NXAST_REG_MOVE:
4212 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4216 case OFPUTIL_NXAST_REG_LOAD:
4217 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4221 case OFPUTIL_NXAST_NOTE:
4222 /* Nothing to do. */
4225 case OFPUTIL_NXAST_SET_TUNNEL64:
4226 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4227 ctx->flow.tun_id = tun_id;
4230 case OFPUTIL_NXAST_MULTIPATH:
4231 nam = (const struct nx_action_multipath *) ia;
4232 multipath_execute(nam, &ctx->flow);
4235 case OFPUTIL_NXAST_AUTOPATH:
4236 naa = (const struct nx_action_autopath *) ia;
4237 xlate_autopath(ctx, naa);
4240 case OFPUTIL_NXAST_BUNDLE:
4241 ctx->ofproto->has_bundle_action = true;
4242 nab = (const struct nx_action_bundle *) ia;
4243 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4248 case OFPUTIL_NXAST_BUNDLE_LOAD:
4249 ctx->ofproto->has_bundle_action = true;
4250 nab = (const struct nx_action_bundle *) ia;
4251 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4255 case OFPUTIL_NXAST_OUTPUT_REG:
4256 naor = (const struct nx_action_output_reg *) ia;
4257 xlate_output_reg_action(ctx, naor);
4260 case OFPUTIL_NXAST_LEARN:
4261 ctx->has_learn = true;
4262 if (ctx->may_learn) {
4263 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4267 case OFPUTIL_NXAST_EXIT:
4273 /* We've let OFPP_NORMAL and the learning action look at the packet,
4274 * so drop it now if forwarding is disabled. */
4275 if (port && !stp_forward_in_state(port->stp_state)) {
4276 ofpbuf_clear(ctx->odp_actions);
4277 add_sflow_action(ctx);
4282 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4283 struct ofproto_dpif *ofproto, const struct flow *flow,
4284 const struct ofpbuf *packet)
4286 ctx->ofproto = ofproto;
4288 ctx->packet = packet;
4289 ctx->may_learn = packet != NULL;
4290 ctx->resubmit_hook = NULL;
4293 static struct ofpbuf *
4294 xlate_actions(struct action_xlate_ctx *ctx,
4295 const union ofp_action *in, size_t n_in)
4297 COVERAGE_INC(ofproto_dpif_xlate);
4299 ctx->odp_actions = ofpbuf_new(512);
4300 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4302 ctx->may_set_up_flow = true;
4303 ctx->has_learn = false;
4304 ctx->has_normal = false;
4305 ctx->nf_output_iface = NF_OUT_DROP;
4308 ctx->base_priority = 0;
4309 ctx->base_flow = ctx->flow;
4310 ctx->base_flow.tun_id = 0;
4314 if (ctx->flow.tos_frag & FLOW_FRAG_ANY) {
4315 switch (ctx->ofproto->up.frag_handling) {
4316 case OFPC_FRAG_NORMAL:
4317 /* We must pretend that transport ports are unavailable. */
4318 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4319 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4322 case OFPC_FRAG_DROP:
4323 return ctx->odp_actions;
4325 case OFPC_FRAG_REASM:
4328 case OFPC_FRAG_NX_MATCH:
4329 /* Nothing to do. */
4334 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4335 ctx->may_set_up_flow = false;
4336 return ctx->odp_actions;
4338 add_sflow_action(ctx);
4339 do_xlate_actions(in, n_in, ctx);
4341 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4342 ctx->odp_actions->data,
4343 ctx->odp_actions->size)) {
4344 ctx->may_set_up_flow = false;
4346 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4348 compose_output_action(ctx, OVSP_LOCAL);
4351 fix_sflow_action(ctx);
4354 return ctx->odp_actions;
4357 /* OFPP_NORMAL implementation. */
4360 struct ofport_dpif *port;
4365 struct dst builtin[32];
4367 size_t n, allocated;
4370 static void dst_set_init(struct dst_set *);
4371 static void dst_set_add(struct dst_set *, const struct dst *);
4372 static void dst_set_free(struct dst_set *);
4374 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4376 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4377 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4378 * the bundle on which the packet was received, returns the VLAN to which the
4381 * Both 'vid' and the return value are in the range 0...4095. */
4383 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4385 switch (in_bundle->vlan_mode) {
4386 case PORT_VLAN_ACCESS:
4387 return in_bundle->vlan;
4390 case PORT_VLAN_TRUNK:
4393 case PORT_VLAN_NATIVE_UNTAGGED:
4394 case PORT_VLAN_NATIVE_TAGGED:
4395 return vid ? vid : in_bundle->vlan;
4402 /* Given 'vlan', the VLAN that a packet belongs to, and
4403 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4404 * that should be included in the 802.1Q header. (If the return value is 0,
4405 * then the 802.1Q header should only be included in the packet if there is a
4408 * Both 'vlan' and the return value are in the range 0...4095. */
4410 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4412 switch (out_bundle->vlan_mode) {
4413 case PORT_VLAN_ACCESS:
4416 case PORT_VLAN_TRUNK:
4417 case PORT_VLAN_NATIVE_TAGGED:
4420 case PORT_VLAN_NATIVE_UNTAGGED:
4421 return vlan == out_bundle->vlan ? 0 : vlan;
4429 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
4430 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
4434 vlan = input_vid_to_vlan(in_bundle, vlan_tci_to_vid(ctx->flow.vlan_tci));
4435 dst->vid = output_vlan_to_vid(out_bundle, vlan);
4437 dst->port = (!out_bundle->bond
4438 ? ofbundle_get_a_port(out_bundle)
4439 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4440 dst->vid, &ctx->tags));
4441 return dst->port != NULL;
4445 mirror_mask_ffs(mirror_mask_t mask)
4447 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4452 dst_set_init(struct dst_set *set)
4454 set->dsts = set->builtin;
4456 set->allocated = ARRAY_SIZE(set->builtin);
4460 dst_set_add(struct dst_set *set, const struct dst *dst)
4462 if (set->n >= set->allocated) {
4463 size_t new_allocated;
4464 struct dst *new_dsts;
4466 new_allocated = set->allocated * 2;
4467 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
4468 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
4472 set->dsts = new_dsts;
4473 set->allocated = new_allocated;
4475 set->dsts[set->n++] = *dst;
4479 dst_set_free(struct dst_set *set)
4481 if (set->dsts != set->builtin) {
4487 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
4490 for (i = 0; i < set->n; i++) {
4491 if (set->dsts[i].vid == test->vid
4492 && set->dsts[i].port == test->port) {
4500 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4502 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4503 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4507 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4509 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4512 /* Returns an arbitrary interface within 'bundle'. */
4513 static struct ofport_dpif *
4514 ofbundle_get_a_port(const struct ofbundle *bundle)
4516 return CONTAINER_OF(list_front(&bundle->ports),
4517 struct ofport_dpif, bundle_node);
4521 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
4522 const struct ofbundle *in_bundle,
4523 const struct ofbundle *out_bundle, struct dst_set *set)
4527 if (out_bundle == OFBUNDLE_FLOOD) {
4528 struct ofbundle *bundle;
4530 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
4531 if (bundle != in_bundle
4532 && ofbundle_includes_vlan(bundle, vlan)
4533 && bundle->floodable
4534 && !bundle->mirror_out
4535 && set_dst(ctx, &dst, in_bundle, bundle)) {
4536 dst_set_add(set, &dst);
4539 ctx->nf_output_iface = NF_OUT_FLOOD;
4540 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
4541 dst_set_add(set, &dst);
4542 ctx->nf_output_iface = dst.port->odp_port;
4547 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4549 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4552 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4553 * to a VLAN. In general most packets may be mirrored but we want to drop
4554 * protocols that may confuse switches. */
4556 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4558 /* If you change this function's behavior, please update corresponding
4559 * documentation in vswitch.xml at the same time. */
4560 if (dst[0] != 0x01) {
4561 /* All the currently banned MACs happen to start with 01 currently, so
4562 * this is a quick way to eliminate most of the good ones. */
4564 if (eth_addr_is_reserved(dst)) {
4565 /* Drop STP, IEEE pause frames, and other reserved protocols
4566 * (01-80-c2-00-00-0x). */
4570 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4572 if ((dst[3] & 0xfe) == 0xcc &&
4573 (dst[4] & 0xfe) == 0xcc &&
4574 (dst[5] & 0xfe) == 0xcc) {
4575 /* Drop the following protocols plus others following the same
4578 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4579 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4580 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4584 if (!(dst[3] | dst[4] | dst[5])) {
4585 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4594 compose_mirror_dsts(struct action_xlate_ctx *ctx,
4595 uint16_t vlan, const struct ofbundle *in_bundle,
4596 struct dst_set *set)
4598 struct ofproto_dpif *ofproto = ctx->ofproto;
4599 mirror_mask_t mirrors;
4603 mirrors = in_bundle->src_mirrors;
4604 for (i = 0; i < set->n; i++) {
4605 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
4612 flow_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4614 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
4615 if (vlan_is_mirrored(m, vlan)) {
4619 if (set_dst(ctx, &dst, in_bundle, m->out)
4620 && !dst_is_duplicate(set, &dst)) {
4621 dst_set_add(set, &dst);
4623 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
4624 struct ofbundle *bundle;
4626 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4627 if (ofbundle_includes_vlan(bundle, m->out_vlan)
4628 && set_dst(ctx, &dst, in_bundle, bundle))
4630 /* set_dst() got dst->vid from the input packet's VLAN,
4631 * not from m->out_vlan, so recompute it. */
4632 dst.vid = output_vlan_to_vid(bundle, m->out_vlan);
4634 if (dst_is_duplicate(set, &dst)) {
4638 if (bundle == in_bundle && dst.vid == flow_vid) {
4639 /* Don't send out input port on same VLAN. */
4642 dst_set_add(set, &dst);
4647 mirrors &= mirrors - 1;
4652 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
4653 const struct ofbundle *in_bundle,
4654 const struct ofbundle *out_bundle)
4656 uint16_t initial_vid, cur_vid;
4657 const struct dst *dst;
4661 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
4662 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
4668 /* Output all the packets we can without having to change the VLAN. */
4669 commit_odp_actions(ctx);
4670 initial_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4671 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
4672 if (dst->vid != initial_vid) {
4675 compose_output_action(ctx, dst->port->odp_port);
4678 /* Then output the rest. */
4679 cur_vid = initial_vid;
4680 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
4681 if (dst->vid == initial_vid) {
4684 if (dst->vid != cur_vid) {
4687 tci = htons(dst->vid);
4688 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4690 tci |= htons(VLAN_CFI);
4692 commit_vlan_action(ctx, tci);
4696 compose_output_action(ctx, dst->port->odp_port);
4702 /* Returns the effective vlan of a packet, taking into account both the
4703 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
4704 * the packet is untagged and -1 indicates it has an invalid header and
4705 * should be dropped. */
4707 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
4708 struct ofbundle *in_bundle, bool have_packet)
4710 int vlan = vlan_tci_to_vid(flow->vlan_tci);
4712 if (in_bundle->vlan_mode == PORT_VLAN_ACCESS) {
4713 /* Drop tagged packet on access port */
4715 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4716 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
4717 "packet received on port %s configured with "
4718 "implicit VLAN %"PRIu16,
4719 ofproto->up.name, vlan,
4720 in_bundle->name, in_bundle->vlan);
4723 } else if (ofbundle_includes_vlan(in_bundle, vlan)) {
4726 /* Drop packets from a VLAN not member of the trunk */
4728 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4729 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
4730 "packet received on port %s not configured for "
4732 ofproto->up.name, vlan, in_bundle->name, vlan);
4737 if (in_bundle->vlan_mode != PORT_VLAN_TRUNK) {
4738 return in_bundle->vlan;
4740 return ofbundle_includes_vlan(in_bundle, 0) ? 0 : -1;
4745 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
4746 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
4747 * indicate this; newer upstream kernels use gratuitous ARP requests. */
4749 is_gratuitous_arp(const struct flow *flow)
4751 return (flow->dl_type == htons(ETH_TYPE_ARP)
4752 && eth_addr_is_broadcast(flow->dl_dst)
4753 && (flow->nw_proto == ARP_OP_REPLY
4754 || (flow->nw_proto == ARP_OP_REQUEST
4755 && flow->nw_src == flow->nw_dst)));
4759 update_learning_table(struct ofproto_dpif *ofproto,
4760 const struct flow *flow, int vlan,
4761 struct ofbundle *in_bundle)
4763 struct mac_entry *mac;
4765 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
4769 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
4770 if (is_gratuitous_arp(flow)) {
4771 /* We don't want to learn from gratuitous ARP packets that are
4772 * reflected back over bond slaves so we lock the learning table. */
4773 if (!in_bundle->bond) {
4774 mac_entry_set_grat_arp_lock(mac);
4775 } else if (mac_entry_is_grat_arp_locked(mac)) {
4780 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
4781 /* The log messages here could actually be useful in debugging,
4782 * so keep the rate limit relatively high. */
4783 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4784 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
4785 "on port %s in VLAN %d",
4786 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
4787 in_bundle->name, vlan);
4789 mac->port.p = in_bundle;
4790 tag_set_add(&ofproto->revalidate_set,
4791 mac_learning_changed(ofproto->ml, mac));
4795 /* Determines whether packets in 'flow' within 'br' should be forwarded or
4796 * dropped. Returns true if they may be forwarded, false if they should be
4799 * If 'have_packet' is true, it indicates that the caller is processing a
4800 * received packet. If 'have_packet' is false, then the caller is just
4801 * revalidating an existing flow because configuration has changed. Either
4802 * way, 'have_packet' only affects logging (there is no point in logging errors
4803 * during revalidation).
4805 * Sets '*in_portp' to the input port. This will be a null pointer if
4806 * flow->in_port does not designate a known input port (in which case
4807 * is_admissible() returns false).
4809 * When returning true, sets '*vlanp' to the effective VLAN of the input
4810 * packet, as returned by flow_get_vlan().
4812 * May also add tags to '*tags', although the current implementation only does
4813 * so in one special case.
4816 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
4818 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
4820 struct ofport_dpif *in_port;
4821 struct ofbundle *in_bundle;
4824 /* Find the port and bundle for the received packet. */
4825 in_port = get_ofp_port(ofproto, flow->in_port);
4826 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
4827 if (!in_port || !in_bundle) {
4828 /* No interface? Something fishy... */
4830 /* Odd. A few possible reasons here:
4832 * - We deleted a port but there are still a few packets queued up
4835 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
4836 * we don't know about.
4838 * - Packet arrived on the local port but the local port is not
4841 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4843 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
4845 ofproto->up.name, flow->in_port);
4850 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
4855 /* Drop frames for reserved multicast addresses only if forward_bpdu
4856 * option is absent. */
4857 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
4861 /* Drop frames on bundles reserved for mirroring. */
4862 if (in_bundle->mirror_out) {
4864 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4865 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4866 "%s, which is reserved exclusively for mirroring",
4867 ofproto->up.name, in_bundle->name);
4872 if (in_bundle->bond) {
4873 struct mac_entry *mac;
4875 switch (bond_check_admissibility(in_bundle->bond, in_port,
4876 flow->dl_dst, tags)) {
4883 case BV_DROP_IF_MOVED:
4884 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
4885 if (mac && mac->port.p != in_bundle &&
4886 (!is_gratuitous_arp(flow)
4887 || mac_entry_is_grat_arp_locked(mac))) {
4898 xlate_normal(struct action_xlate_ctx *ctx)
4900 struct ofbundle *in_bundle;
4901 struct ofbundle *out_bundle;
4902 struct mac_entry *mac;
4905 ctx->has_normal = true;
4907 /* Check whether we should drop packets in this flow. */
4908 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
4909 &ctx->tags, &vlan, &in_bundle)) {
4914 /* Learn source MAC. */
4915 if (ctx->may_learn) {
4916 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
4919 /* Determine output bundle. */
4920 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
4923 out_bundle = mac->port.p;
4924 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
4925 /* If we are revalidating but don't have a learning entry then eject
4926 * the flow. Installing a flow that floods packets opens up a window
4927 * of time where we could learn from a packet reflected on a bond and
4928 * blackhole packets before the learning table is updated to reflect
4929 * the correct port. */
4930 ctx->may_set_up_flow = false;
4933 out_bundle = OFBUNDLE_FLOOD;
4936 /* Don't send packets out their input bundles. */
4937 if (in_bundle == out_bundle) {
4943 compose_actions(ctx, vlan, in_bundle, out_bundle);
4947 /* Optimized flow revalidation.
4949 * It's a difficult problem, in general, to tell which facets need to have
4950 * their actions recalculated whenever the OpenFlow flow table changes. We
4951 * don't try to solve that general problem: for most kinds of OpenFlow flow
4952 * table changes, we recalculate the actions for every facet. This is
4953 * relatively expensive, but it's good enough if the OpenFlow flow table
4954 * doesn't change very often.
4956 * However, we can expect one particular kind of OpenFlow flow table change to
4957 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
4958 * of CPU on revalidating every facet whenever MAC learning modifies the flow
4959 * table, we add a special case that applies to flow tables in which every rule
4960 * has the same form (that is, the same wildcards), except that the table is
4961 * also allowed to have a single "catch-all" flow that matches all packets. We
4962 * optimize this case by tagging all of the facets that resubmit into the table
4963 * and invalidating the same tag whenever a flow changes in that table. The
4964 * end result is that we revalidate just the facets that need it (and sometimes
4965 * a few more, but not all of the facets or even all of the facets that
4966 * resubmit to the table modified by MAC learning). */
4968 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
4969 * into an OpenFlow table with the given 'basis'. */
4971 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
4974 if (flow_wildcards_is_catchall(wc)) {
4977 struct flow tag_flow = *flow;
4978 flow_zero_wildcards(&tag_flow, wc);
4979 return tag_create_deterministic(flow_hash(&tag_flow, secret));
4983 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
4984 * taggability of that table.
4986 * This function must be called after *each* change to a flow table. If you
4987 * skip calling it on some changes then the pointer comparisons at the end can
4988 * be invalid if you get unlucky. For example, if a flow removal causes a
4989 * cls_table to be destroyed and then a flow insertion causes a cls_table with
4990 * different wildcards to be created with the same address, then this function
4991 * will incorrectly skip revalidation. */
4993 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
4995 struct table_dpif *table = &ofproto->tables[table_id];
4996 const struct classifier *cls = &ofproto->up.tables[table_id];
4997 struct cls_table *catchall, *other;
4998 struct cls_table *t;
5000 catchall = other = NULL;
5002 switch (hmap_count(&cls->tables)) {
5004 /* We could tag this OpenFlow table but it would make the logic a
5005 * little harder and it's a corner case that doesn't seem worth it
5011 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5012 if (cls_table_is_catchall(t)) {
5014 } else if (!other) {
5017 /* Indicate that we can't tag this by setting both tables to
5018 * NULL. (We know that 'catchall' is already NULL.) */
5025 /* Can't tag this table. */
5029 if (table->catchall_table != catchall || table->other_table != other) {
5030 table->catchall_table = catchall;
5031 table->other_table = other;
5032 ofproto->need_revalidate = true;
5036 /* Given 'rule' that has changed in some way (either it is a rule being
5037 * inserted, a rule being deleted, or a rule whose actions are being
5038 * modified), marks facets for revalidation to ensure that packets will be
5039 * forwarded correctly according to the new state of the flow table.
5041 * This function must be called after *each* change to a flow table. See
5042 * the comment on table_update_taggable() for more information. */
5044 rule_invalidate(const struct rule_dpif *rule)
5046 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5048 table_update_taggable(ofproto, rule->up.table_id);
5050 if (!ofproto->need_revalidate) {
5051 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5053 if (table->other_table && rule->tag) {
5054 tag_set_add(&ofproto->revalidate_set, rule->tag);
5056 ofproto->need_revalidate = true;
5062 set_frag_handling(struct ofproto *ofproto_,
5063 enum ofp_config_flags frag_handling)
5065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5067 if (frag_handling != OFPC_FRAG_REASM) {
5068 ofproto->need_revalidate = true;
5076 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5077 const struct flow *flow,
5078 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5080 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5083 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5084 ofproto->max_ports);
5086 struct odputil_keybuf keybuf;
5087 struct action_xlate_ctx ctx;
5088 struct ofpbuf *odp_actions;
5091 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5092 odp_flow_key_from_flow(&key, flow);
5094 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
5095 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5096 dpif_execute(ofproto->dpif, key.data, key.size,
5097 odp_actions->data, odp_actions->size, packet);
5098 ofpbuf_delete(odp_actions);
5104 get_netflow_ids(const struct ofproto *ofproto_,
5105 uint8_t *engine_type, uint8_t *engine_id)
5107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5109 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5112 static struct ofproto_dpif *
5113 ofproto_dpif_lookup(const char *name)
5115 struct ofproto *ofproto = ofproto_lookup(name);
5116 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
5117 ? ofproto_dpif_cast(ofproto)
5122 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5123 const char *args, void *aux OVS_UNUSED)
5125 struct ds ds = DS_EMPTY_INITIALIZER;
5126 const struct ofproto_dpif *ofproto;
5127 const struct mac_entry *e;
5129 ofproto = ofproto_dpif_lookup(args);
5131 unixctl_command_reply(conn, 501, "no such bridge");
5135 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5136 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5137 struct ofbundle *bundle = e->port.p;
5138 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5139 ofbundle_get_a_port(bundle)->odp_port,
5140 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5142 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5146 struct ofproto_trace {
5147 struct action_xlate_ctx ctx;
5153 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5154 const struct rule_dpif *rule)
5156 ds_put_char_multiple(result, '\t', level);
5158 ds_put_cstr(result, "No match\n");
5162 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5163 table_id, ntohll(rule->up.flow_cookie));
5164 cls_rule_format(&rule->up.cr, result);
5165 ds_put_char(result, '\n');
5167 ds_put_char_multiple(result, '\t', level);
5168 ds_put_cstr(result, "OpenFlow ");
5169 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5170 ds_put_char(result, '\n');
5174 trace_format_flow(struct ds *result, int level, const char *title,
5175 struct ofproto_trace *trace)
5177 ds_put_char_multiple(result, '\t', level);
5178 ds_put_format(result, "%s: ", title);
5179 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5180 ds_put_cstr(result, "unchanged");
5182 flow_format(result, &trace->ctx.flow);
5183 trace->flow = trace->ctx.flow;
5185 ds_put_char(result, '\n');
5189 trace_format_regs(struct ds *result, int level, const char *title,
5190 struct ofproto_trace *trace)
5194 ds_put_char_multiple(result, '\t', level);
5195 ds_put_format(result, "%s:", title);
5196 for (i = 0; i < FLOW_N_REGS; i++) {
5197 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5199 ds_put_char(result, '\n');
5203 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5205 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5206 struct ds *result = trace->result;
5208 ds_put_char(result, '\n');
5209 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5210 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5211 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5215 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5216 void *aux OVS_UNUSED)
5218 char *dpname, *arg1, *arg2, *arg3;
5219 char *args = xstrdup(args_);
5220 char *save_ptr = NULL;
5221 struct ofproto_dpif *ofproto;
5222 struct ofpbuf odp_key;
5223 struct ofpbuf *packet;
5224 struct rule_dpif *rule;
5230 ofpbuf_init(&odp_key, 0);
5233 dpname = strtok_r(args, " ", &save_ptr);
5234 arg1 = strtok_r(NULL, " ", &save_ptr);
5235 arg2 = strtok_r(NULL, " ", &save_ptr);
5236 arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5237 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5238 /* ofproto/trace dpname flow [-generate] */
5241 /* Convert string to datapath key. */
5242 ofpbuf_init(&odp_key, 0);
5243 error = odp_flow_key_from_string(arg1, &odp_key);
5245 unixctl_command_reply(conn, 501, "Bad flow syntax");
5249 /* Convert odp_key to flow. */
5250 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
5252 unixctl_command_reply(conn, 501, "Invalid flow");
5256 /* Generate a packet, if requested. */
5258 packet = ofpbuf_new(0);
5259 flow_compose(packet, &flow);
5261 } else if (dpname && arg1 && arg2 && arg3) {
5262 /* ofproto/trace dpname tun_id in_port packet */
5266 tun_id = htonll(strtoull(arg1, NULL, 0));
5267 in_port = ofp_port_to_odp_port(atoi(arg2));
5269 packet = ofpbuf_new(strlen(args) / 2);
5270 arg3 = ofpbuf_put_hex(packet, arg3, NULL);
5271 arg3 += strspn(arg3, " ");
5272 if (*arg3 != '\0') {
5273 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5276 if (packet->size < ETH_HEADER_LEN) {
5277 unixctl_command_reply(conn, 501,
5278 "Packet data too short for Ethernet");
5282 ds_put_cstr(&result, "Packet: ");
5283 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5284 ds_put_cstr(&result, s);
5287 flow_extract(packet, tun_id, in_port, &flow);
5289 unixctl_command_reply(conn, 501, "Bad command syntax");
5293 ofproto = ofproto_dpif_lookup(dpname);
5295 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5300 ds_put_cstr(&result, "Flow: ");
5301 flow_format(&result, &flow);
5302 ds_put_char(&result, '\n');
5304 rule = rule_dpif_lookup(ofproto, &flow, 0);
5305 trace_format_rule(&result, 0, 0, rule);
5307 struct ofproto_trace trace;
5308 struct ofpbuf *odp_actions;
5310 trace.result = &result;
5312 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
5313 trace.ctx.resubmit_hook = trace_resubmit;
5314 odp_actions = xlate_actions(&trace.ctx,
5315 rule->up.actions, rule->up.n_actions);
5317 ds_put_char(&result, '\n');
5318 trace_format_flow(&result, 0, "Final flow", &trace);
5319 ds_put_cstr(&result, "Datapath actions: ");
5320 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5321 ofpbuf_delete(odp_actions);
5323 if (!trace.ctx.may_set_up_flow) {
5325 ds_put_cstr(&result, "\nThis flow is not cachable.");
5327 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5328 "for complete actions, please supply a packet.");
5333 unixctl_command_reply(conn, 200, ds_cstr(&result));
5336 ds_destroy(&result);
5337 ofpbuf_delete(packet);
5338 ofpbuf_uninit(&odp_key);
5343 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5344 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5347 unixctl_command_reply(conn, 200, NULL);
5351 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5352 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5355 unixctl_command_reply(conn, 200, NULL);
5359 ofproto_dpif_unixctl_init(void)
5361 static bool registered;
5367 unixctl_command_register("ofproto/trace",
5368 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5369 ofproto_unixctl_trace, NULL);
5370 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5372 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5373 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5376 const struct ofproto_class ofproto_dpif_class = {
5403 port_is_lacp_current,
5404 NULL, /* rule_choose_table */
5411 rule_modify_actions,
5419 get_cfm_remote_mpids,
5423 get_stp_port_status,
5428 is_mirror_output_bundle,
5429 forward_bpdu_changed,