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 (exactly one of out == NULL and out_vlan == -1 is true). */
125 struct ofbundle *out; /* Output port or NULL. */
126 int out_vlan; /* Output VLAN or -1. */
127 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
130 static void mirror_destroy(struct ofmirror *);
132 /* A group of one or more OpenFlow ports. */
133 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
135 struct ofproto_dpif *ofproto; /* Owning ofproto. */
136 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
137 void *aux; /* Key supplied by ofproto's client. */
138 char *name; /* Identifier for log messages. */
141 struct list ports; /* Contains "struct ofport"s. */
142 enum port_vlan_mode vlan_mode; /* VLAN mode */
143 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
144 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
145 * NULL if all VLANs are trunked. */
146 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
147 struct bond *bond; /* Nonnull iff more than one port. */
148 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
151 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
153 /* Port mirroring info. */
154 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
155 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
156 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
159 static void bundle_remove(struct ofport *);
160 static void bundle_update(struct ofbundle *);
161 static void bundle_destroy(struct ofbundle *);
162 static void bundle_del_port(struct ofport_dpif *);
163 static void bundle_run(struct ofbundle *);
164 static void bundle_wait(struct ofbundle *);
166 static void stp_run(struct ofproto_dpif *ofproto);
167 static void stp_wait(struct ofproto_dpif *ofproto);
169 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
171 struct action_xlate_ctx {
172 /* action_xlate_ctx_init() initializes these members. */
175 struct ofproto_dpif *ofproto;
177 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
178 * this flow when actions change header fields. */
181 /* The packet corresponding to 'flow', or a null pointer if we are
182 * revalidating without a packet to refer to. */
183 const struct ofpbuf *packet;
185 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
186 * want to execute them if we are actually processing a packet, or if we
187 * are accounting for packets that the datapath has processed, but not if
188 * we are just revalidating. */
191 /* If nonnull, called just before executing a resubmit action.
193 * This is normally null so the client has to set it manually after
194 * calling action_xlate_ctx_init(). */
195 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
197 /* xlate_actions() initializes and uses these members. The client might want
198 * to look at them after it returns. */
200 struct ofpbuf *odp_actions; /* Datapath actions. */
201 tag_type tags; /* Tags associated with actions. */
202 bool may_set_up_flow; /* True ordinarily; false if the actions must
203 * be reassessed for every packet. */
204 bool has_learn; /* Actions include NXAST_LEARN? */
205 bool has_normal; /* Actions output to OFPP_NORMAL? */
206 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
208 /* xlate_actions() initializes and uses these members, but the client has no
209 * reason to look at them. */
211 int recurse; /* Recursion level, via xlate_table_action. */
212 struct flow base_flow; /* Flow at the last commit. */
213 uint32_t original_priority; /* Priority when packet arrived. */
214 uint8_t table_id; /* OpenFlow table ID where flow was found. */
215 uint32_t sflow_n_outputs; /* Number of output ports. */
216 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
217 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
218 bool exit; /* No further actions should be processed. */
221 static void action_xlate_ctx_init(struct action_xlate_ctx *,
222 struct ofproto_dpif *, const struct flow *,
223 const struct ofpbuf *);
224 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
225 const union ofp_action *in, size_t n_in);
227 /* An exact-match instantiation of an OpenFlow flow. */
229 long long int used; /* Time last used; time created if not used. */
233 * - Do include packets and bytes sent "by hand", e.g. with
236 * - Do include packets and bytes that were obtained from the datapath
237 * when its statistics were reset (e.g. dpif_flow_put() with
238 * DPIF_FP_ZERO_STATS).
240 uint64_t packet_count; /* Number of packets received. */
241 uint64_t byte_count; /* Number of bytes received. */
243 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
244 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
246 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
247 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
248 long long int rs_used; /* Used time pushed to resubmit children. */
250 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
252 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
253 struct list list_node; /* In owning rule's 'facets' list. */
254 struct rule_dpif *rule; /* Owning rule. */
255 struct flow flow; /* Exact-match flow. */
256 bool installed; /* Installed in datapath? */
257 bool may_install; /* True ordinarily; false if actions must
258 * be reassessed for every packet. */
259 bool has_learn; /* Actions include NXAST_LEARN? */
260 bool has_normal; /* Actions output to OFPP_NORMAL? */
261 size_t actions_len; /* Number of bytes in actions[]. */
262 struct nlattr *actions; /* Datapath actions. */
263 tag_type tags; /* Tags. */
264 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
267 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
268 static void facet_remove(struct ofproto_dpif *, struct facet *);
269 static void facet_free(struct facet *);
271 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
272 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
273 const struct flow *);
274 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
276 static bool execute_controller_action(struct ofproto_dpif *,
278 const struct nlattr *odp_actions,
280 struct ofpbuf *packet);
281 static void facet_execute(struct ofproto_dpif *, struct facet *,
282 struct ofpbuf *packet);
284 static int facet_put__(struct ofproto_dpif *, struct facet *,
285 const struct nlattr *actions, size_t actions_len,
286 struct dpif_flow_stats *);
287 static void facet_install(struct ofproto_dpif *, struct facet *,
289 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
290 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
292 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
293 const struct ofpbuf *packet);
294 static void facet_update_time(struct ofproto_dpif *, struct facet *,
296 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
297 const struct dpif_flow_stats *);
298 static void facet_reset_counters(struct facet *);
299 static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
300 static void facet_push_stats(struct facet *);
301 static void facet_account(struct ofproto_dpif *, struct facet *);
303 static bool facet_is_controller_flow(struct facet *);
305 static void flow_push_stats(const struct rule_dpif *,
306 struct flow *, uint64_t packets, uint64_t bytes,
309 static uint32_t rule_calculate_tag(const struct flow *,
310 const struct flow_wildcards *,
312 static void rule_invalidate(const struct rule_dpif *);
318 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
319 struct list bundle_node; /* In struct ofbundle's "ports" list. */
320 struct cfm *cfm; /* Connectivity Fault Management, if any. */
321 tag_type tag; /* Tag associated with this port. */
322 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
323 bool may_enable; /* May be enabled in bonds. */
325 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
326 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
327 long long int stp_state_entered;
330 static struct ofport_dpif *
331 ofport_dpif_cast(const struct ofport *ofport)
333 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
334 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
337 static void port_run(struct ofport_dpif *);
338 static void port_wait(struct ofport_dpif *);
339 static int set_cfm(struct ofport *, const struct cfm_settings *);
341 struct dpif_completion {
342 struct list list_node;
343 struct ofoperation *op;
346 /* Extra information about a classifier table.
347 * Currently used just for optimized flow revalidation. */
349 /* If either of these is nonnull, then this table has a form that allows
350 * flows to be tagged to avoid revalidating most flows for the most common
351 * kinds of flow table changes. */
352 struct cls_table *catchall_table; /* Table that wildcards all fields. */
353 struct cls_table *other_table; /* Table with any other wildcard set. */
354 uint32_t basis; /* Keeps each table's tags separate. */
357 struct ofproto_dpif {
366 struct netflow *netflow;
367 struct dpif_sflow *sflow;
368 struct hmap bundles; /* Contains "struct ofbundle"s. */
369 struct mac_learning *ml;
370 struct ofmirror *mirrors[MAX_MIRRORS];
371 bool has_bonded_bundles;
374 struct timer next_expiration;
380 struct table_dpif tables[N_TABLES];
381 bool need_revalidate;
382 struct tag_set revalidate_set;
384 /* Support for debugging async flow mods. */
385 struct list completions;
387 bool has_bundle_action; /* True when the first bundle action appears. */
391 long long int stp_last_tick;
394 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
395 * for debugging the asynchronous flow_mod implementation.) */
398 static void ofproto_dpif_unixctl_init(void);
400 static struct ofproto_dpif *
401 ofproto_dpif_cast(const struct ofproto *ofproto)
403 assert(ofproto->ofproto_class == &ofproto_dpif_class);
404 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
407 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
409 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
412 /* Packet processing. */
413 static void update_learning_table(struct ofproto_dpif *,
414 const struct flow *, int vlan,
417 #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 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
954 ofport->odp_port, pkt);
960 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
962 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
964 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
966 /* Only revalidate flows if the configuration changed. */
967 if (!s != !ofproto->stp) {
968 ofproto->need_revalidate = true;
973 ofproto->stp = stp_create(ofproto_->name, s->system_id,
974 send_bpdu_cb, ofproto);
975 ofproto->stp_last_tick = time_msec();
978 stp_set_bridge_id(ofproto->stp, s->system_id);
979 stp_set_bridge_priority(ofproto->stp, s->priority);
980 stp_set_hello_time(ofproto->stp, s->hello_time);
981 stp_set_max_age(ofproto->stp, s->max_age);
982 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
984 stp_destroy(ofproto->stp);
992 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
998 s->bridge_id = stp_get_bridge_id(ofproto->stp);
999 s->designated_root = stp_get_designated_root(ofproto->stp);
1000 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1009 update_stp_port_state(struct ofport_dpif *ofport)
1011 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1012 enum stp_state state;
1014 /* Figure out new state. */
1015 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1019 if (ofport->stp_state != state) {
1023 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1024 netdev_get_name(ofport->up.netdev),
1025 stp_state_name(ofport->stp_state),
1026 stp_state_name(state));
1027 if (stp_learn_in_state(ofport->stp_state)
1028 != stp_learn_in_state(state)) {
1029 /* xxx Learning action flows should also be flushed. */
1030 mac_learning_flush(ofproto->ml);
1032 fwd_change = stp_forward_in_state(ofport->stp_state)
1033 != stp_forward_in_state(state);
1035 ofproto->need_revalidate = true;
1036 ofport->stp_state = state;
1037 ofport->stp_state_entered = time_msec();
1039 if (fwd_change && ofport->bundle) {
1040 bundle_update(ofport->bundle);
1043 /* Update the STP state bits in the OpenFlow port description. */
1044 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1045 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1046 : state == STP_LEARNING ? OFPPS_STP_LEARN
1047 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1048 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1050 ofproto_port_set_state(&ofport->up, of_state);
1054 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1055 * caller is responsible for assigning STP port numbers and ensuring
1056 * there are no duplicates. */
1058 set_stp_port(struct ofport *ofport_,
1059 const struct ofproto_port_stp_settings *s)
1061 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1062 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1063 struct stp_port *sp = ofport->stp_port;
1065 if (!s || !s->enable) {
1067 ofport->stp_port = NULL;
1068 stp_port_disable(sp);
1069 update_stp_port_state(ofport);
1072 } else if (sp && stp_port_no(sp) != s->port_num
1073 && ofport == stp_port_get_aux(sp)) {
1074 /* The port-id changed, so disable the old one if it's not
1075 * already in use by another port. */
1076 stp_port_disable(sp);
1079 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1080 stp_port_enable(sp);
1082 stp_port_set_aux(sp, ofport);
1083 stp_port_set_priority(sp, s->priority);
1084 stp_port_set_path_cost(sp, s->path_cost);
1086 update_stp_port_state(ofport);
1092 get_stp_port_status(struct ofport *ofport_,
1093 struct ofproto_port_stp_status *s)
1095 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1097 struct stp_port *sp = ofport->stp_port;
1099 if (!ofproto->stp || !sp) {
1105 s->port_id = stp_port_get_id(sp);
1106 s->state = stp_port_get_state(sp);
1107 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1108 s->role = stp_port_get_role(sp);
1109 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1115 stp_run(struct ofproto_dpif *ofproto)
1118 long long int now = time_msec();
1119 long long int elapsed = now - ofproto->stp_last_tick;
1120 struct stp_port *sp;
1123 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1124 ofproto->stp_last_tick = now;
1126 while (stp_get_changed_port(ofproto->stp, &sp)) {
1127 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1130 update_stp_port_state(ofport);
1137 stp_wait(struct ofproto_dpif *ofproto)
1140 poll_timer_wait(1000);
1144 /* Returns true if STP should process 'flow'. */
1146 stp_should_process_flow(const struct flow *flow)
1148 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1152 stp_process_packet(const struct ofport_dpif *ofport,
1153 const struct ofpbuf *packet)
1155 struct ofpbuf payload = *packet;
1156 struct eth_header *eth = payload.data;
1157 struct stp_port *sp = ofport->stp_port;
1159 /* Sink packets on ports that have STP disabled when the bridge has
1161 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1165 /* Trim off padding on payload. */
1166 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1167 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1170 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1171 stp_received_bpdu(sp, payload.data, payload.size);
1177 /* Expires all MAC learning entries associated with 'port' and forces ofproto
1178 * to revalidate every flow. */
1180 bundle_flush_macs(struct ofbundle *bundle)
1182 struct ofproto_dpif *ofproto = bundle->ofproto;
1183 struct mac_learning *ml = ofproto->ml;
1184 struct mac_entry *mac, *next_mac;
1186 ofproto->need_revalidate = true;
1187 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1188 if (mac->port.p == bundle) {
1189 mac_learning_expire(ml, mac);
1194 static struct ofbundle *
1195 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1197 struct ofbundle *bundle;
1199 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1200 &ofproto->bundles) {
1201 if (bundle->aux == aux) {
1208 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1209 * ones that are found to 'bundles'. */
1211 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1212 void **auxes, size_t n_auxes,
1213 struct hmapx *bundles)
1217 hmapx_init(bundles);
1218 for (i = 0; i < n_auxes; i++) {
1219 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1221 hmapx_add(bundles, bundle);
1227 bundle_update(struct ofbundle *bundle)
1229 struct ofport_dpif *port;
1231 bundle->floodable = true;
1232 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1233 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1234 bundle->floodable = false;
1241 bundle_del_port(struct ofport_dpif *port)
1243 struct ofbundle *bundle = port->bundle;
1245 bundle->ofproto->need_revalidate = true;
1247 list_remove(&port->bundle_node);
1248 port->bundle = NULL;
1251 lacp_slave_unregister(bundle->lacp, port);
1254 bond_slave_unregister(bundle->bond, port);
1257 bundle_update(bundle);
1261 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1262 struct lacp_slave_settings *lacp,
1263 uint32_t bond_stable_id)
1265 struct ofport_dpif *port;
1267 port = get_ofp_port(bundle->ofproto, ofp_port);
1272 if (port->bundle != bundle) {
1273 bundle->ofproto->need_revalidate = true;
1275 bundle_del_port(port);
1278 port->bundle = bundle;
1279 list_push_back(&bundle->ports, &port->bundle_node);
1280 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
1281 bundle->floodable = false;
1285 port->bundle->ofproto->need_revalidate = true;
1286 lacp_slave_register(bundle->lacp, port, lacp);
1289 port->bond_stable_id = bond_stable_id;
1295 bundle_destroy(struct ofbundle *bundle)
1297 struct ofproto_dpif *ofproto;
1298 struct ofport_dpif *port, *next_port;
1305 ofproto = bundle->ofproto;
1306 for (i = 0; i < MAX_MIRRORS; i++) {
1307 struct ofmirror *m = ofproto->mirrors[i];
1309 if (m->out == bundle) {
1311 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1312 || hmapx_find_and_delete(&m->dsts, bundle)) {
1313 ofproto->need_revalidate = true;
1318 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1319 bundle_del_port(port);
1322 bundle_flush_macs(bundle);
1323 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1325 free(bundle->trunks);
1326 lacp_destroy(bundle->lacp);
1327 bond_destroy(bundle->bond);
1332 bundle_set(struct ofproto *ofproto_, void *aux,
1333 const struct ofproto_bundle_settings *s)
1335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1336 bool need_flush = false;
1337 struct ofport_dpif *port;
1338 struct ofbundle *bundle;
1339 unsigned long *trunks;
1345 bundle_destroy(bundle_lookup(ofproto, aux));
1349 assert(s->n_slaves == 1 || s->bond != NULL);
1350 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1352 bundle = bundle_lookup(ofproto, aux);
1354 bundle = xmalloc(sizeof *bundle);
1356 bundle->ofproto = ofproto;
1357 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1358 hash_pointer(aux, 0));
1360 bundle->name = NULL;
1362 list_init(&bundle->ports);
1363 bundle->vlan_mode = PORT_VLAN_TRUNK;
1365 bundle->trunks = NULL;
1366 bundle->use_priority_tags = s->use_priority_tags;
1367 bundle->lacp = NULL;
1368 bundle->bond = NULL;
1370 bundle->floodable = true;
1372 bundle->src_mirrors = 0;
1373 bundle->dst_mirrors = 0;
1374 bundle->mirror_out = 0;
1377 if (!bundle->name || strcmp(s->name, bundle->name)) {
1379 bundle->name = xstrdup(s->name);
1384 if (!bundle->lacp) {
1385 ofproto->need_revalidate = true;
1386 bundle->lacp = lacp_create();
1388 lacp_configure(bundle->lacp, s->lacp);
1390 lacp_destroy(bundle->lacp);
1391 bundle->lacp = NULL;
1394 /* Update set of ports. */
1396 for (i = 0; i < s->n_slaves; i++) {
1397 if (!bundle_add_port(bundle, s->slaves[i],
1398 s->lacp ? &s->lacp_slaves[i] : NULL,
1399 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1403 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1404 struct ofport_dpif *next_port;
1406 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1407 for (i = 0; i < s->n_slaves; i++) {
1408 if (s->slaves[i] == port->up.ofp_port) {
1413 bundle_del_port(port);
1417 assert(list_size(&bundle->ports) <= s->n_slaves);
1419 if (list_is_empty(&bundle->ports)) {
1420 bundle_destroy(bundle);
1424 /* Set VLAN tagging mode */
1425 if (s->vlan_mode != bundle->vlan_mode
1426 || s->use_priority_tags != bundle->use_priority_tags) {
1427 bundle->vlan_mode = s->vlan_mode;
1428 bundle->use_priority_tags = s->use_priority_tags;
1433 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1434 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1436 if (vlan != bundle->vlan) {
1437 bundle->vlan = vlan;
1441 /* Get trunked VLANs. */
1442 switch (s->vlan_mode) {
1443 case PORT_VLAN_ACCESS:
1447 case PORT_VLAN_TRUNK:
1448 trunks = (unsigned long *) s->trunks;
1451 case PORT_VLAN_NATIVE_UNTAGGED:
1452 case PORT_VLAN_NATIVE_TAGGED:
1453 if (vlan != 0 && (!s->trunks
1454 || !bitmap_is_set(s->trunks, vlan)
1455 || bitmap_is_set(s->trunks, 0))) {
1456 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1458 trunks = bitmap_clone(s->trunks, 4096);
1460 trunks = bitmap_allocate1(4096);
1462 bitmap_set1(trunks, vlan);
1463 bitmap_set0(trunks, 0);
1465 trunks = (unsigned long *) s->trunks;
1472 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1473 free(bundle->trunks);
1474 if (trunks == s->trunks) {
1475 bundle->trunks = vlan_bitmap_clone(trunks);
1477 bundle->trunks = trunks;
1482 if (trunks != s->trunks) {
1487 if (!list_is_short(&bundle->ports)) {
1488 bundle->ofproto->has_bonded_bundles = true;
1490 if (bond_reconfigure(bundle->bond, s->bond)) {
1491 ofproto->need_revalidate = true;
1494 bundle->bond = bond_create(s->bond);
1495 ofproto->need_revalidate = true;
1498 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1499 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1503 bond_destroy(bundle->bond);
1504 bundle->bond = NULL;
1507 /* If we changed something that would affect MAC learning, un-learn
1508 * everything on this port and force flow revalidation. */
1510 bundle_flush_macs(bundle);
1517 bundle_remove(struct ofport *port_)
1519 struct ofport_dpif *port = ofport_dpif_cast(port_);
1520 struct ofbundle *bundle = port->bundle;
1523 bundle_del_port(port);
1524 if (list_is_empty(&bundle->ports)) {
1525 bundle_destroy(bundle);
1526 } else if (list_is_short(&bundle->ports)) {
1527 bond_destroy(bundle->bond);
1528 bundle->bond = NULL;
1534 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1536 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1537 struct ofport_dpif *port = port_;
1538 uint8_t ea[ETH_ADDR_LEN];
1541 error = netdev_get_etheraddr(port->up.netdev, ea);
1543 struct ofpbuf packet;
1546 ofpbuf_init(&packet, 0);
1547 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1549 memcpy(packet_pdu, pdu, pdu_size);
1551 send_packet(ofproto_dpif_cast(port->up.ofproto), port->odp_port,
1553 ofpbuf_uninit(&packet);
1555 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1556 "%s (%s)", port->bundle->name,
1557 netdev_get_name(port->up.netdev), strerror(error));
1562 bundle_send_learning_packets(struct ofbundle *bundle)
1564 struct ofproto_dpif *ofproto = bundle->ofproto;
1565 int error, n_packets, n_errors;
1566 struct mac_entry *e;
1568 error = n_packets = n_errors = 0;
1569 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1570 if (e->port.p != bundle) {
1571 struct ofpbuf *learning_packet;
1572 struct ofport_dpif *port;
1575 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1578 ret = send_packet(ofproto_dpif_cast(port->up.ofproto),
1579 port->odp_port, learning_packet);
1580 ofpbuf_delete(learning_packet);
1590 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1591 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1592 "packets, last error was: %s",
1593 bundle->name, n_errors, n_packets, strerror(error));
1595 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1596 bundle->name, n_packets);
1601 bundle_run(struct ofbundle *bundle)
1604 lacp_run(bundle->lacp, send_pdu_cb);
1607 struct ofport_dpif *port;
1609 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1610 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1613 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1614 lacp_negotiated(bundle->lacp));
1615 if (bond_should_send_learning_packets(bundle->bond)) {
1616 bundle_send_learning_packets(bundle);
1622 bundle_wait(struct ofbundle *bundle)
1625 lacp_wait(bundle->lacp);
1628 bond_wait(bundle->bond);
1635 mirror_scan(struct ofproto_dpif *ofproto)
1639 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1640 if (!ofproto->mirrors[idx]) {
1647 static struct ofmirror *
1648 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1652 for (i = 0; i < MAX_MIRRORS; i++) {
1653 struct ofmirror *mirror = ofproto->mirrors[i];
1654 if (mirror && mirror->aux == aux) {
1662 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1664 mirror_update_dups(struct ofproto_dpif *ofproto)
1668 for (i = 0; i < MAX_MIRRORS; i++) {
1669 struct ofmirror *m = ofproto->mirrors[i];
1672 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1676 for (i = 0; i < MAX_MIRRORS; i++) {
1677 struct ofmirror *m1 = ofproto->mirrors[i];
1684 for (j = i + 1; j < MAX_MIRRORS; j++) {
1685 struct ofmirror *m2 = ofproto->mirrors[j];
1687 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1688 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1689 m2->dup_mirrors |= m1->dup_mirrors;
1696 mirror_set(struct ofproto *ofproto_, void *aux,
1697 const struct ofproto_mirror_settings *s)
1699 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1700 mirror_mask_t mirror_bit;
1701 struct ofbundle *bundle;
1702 struct ofmirror *mirror;
1703 struct ofbundle *out;
1704 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1705 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1708 mirror = mirror_lookup(ofproto, aux);
1710 mirror_destroy(mirror);
1716 idx = mirror_scan(ofproto);
1718 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1720 ofproto->up.name, MAX_MIRRORS, s->name);
1724 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1725 mirror->ofproto = ofproto;
1728 mirror->out_vlan = -1;
1729 mirror->name = NULL;
1732 if (!mirror->name || strcmp(s->name, mirror->name)) {
1734 mirror->name = xstrdup(s->name);
1737 /* Get the new configuration. */
1738 if (s->out_bundle) {
1739 out = bundle_lookup(ofproto, s->out_bundle);
1741 mirror_destroy(mirror);
1747 out_vlan = s->out_vlan;
1749 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1750 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1752 /* If the configuration has not changed, do nothing. */
1753 if (hmapx_equals(&srcs, &mirror->srcs)
1754 && hmapx_equals(&dsts, &mirror->dsts)
1755 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1756 && mirror->out == out
1757 && mirror->out_vlan == out_vlan)
1759 hmapx_destroy(&srcs);
1760 hmapx_destroy(&dsts);
1764 hmapx_swap(&srcs, &mirror->srcs);
1765 hmapx_destroy(&srcs);
1767 hmapx_swap(&dsts, &mirror->dsts);
1768 hmapx_destroy(&dsts);
1770 free(mirror->vlans);
1771 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1774 mirror->out_vlan = out_vlan;
1776 /* Update bundles. */
1777 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1778 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1779 if (hmapx_contains(&mirror->srcs, bundle)) {
1780 bundle->src_mirrors |= mirror_bit;
1782 bundle->src_mirrors &= ~mirror_bit;
1785 if (hmapx_contains(&mirror->dsts, bundle)) {
1786 bundle->dst_mirrors |= mirror_bit;
1788 bundle->dst_mirrors &= ~mirror_bit;
1791 if (mirror->out == bundle) {
1792 bundle->mirror_out |= mirror_bit;
1794 bundle->mirror_out &= ~mirror_bit;
1798 ofproto->need_revalidate = true;
1799 mac_learning_flush(ofproto->ml);
1800 mirror_update_dups(ofproto);
1806 mirror_destroy(struct ofmirror *mirror)
1808 struct ofproto_dpif *ofproto;
1809 mirror_mask_t mirror_bit;
1810 struct ofbundle *bundle;
1816 ofproto = mirror->ofproto;
1817 ofproto->need_revalidate = true;
1818 mac_learning_flush(ofproto->ml);
1820 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1821 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1822 bundle->src_mirrors &= ~mirror_bit;
1823 bundle->dst_mirrors &= ~mirror_bit;
1824 bundle->mirror_out &= ~mirror_bit;
1827 hmapx_destroy(&mirror->srcs);
1828 hmapx_destroy(&mirror->dsts);
1829 free(mirror->vlans);
1831 ofproto->mirrors[mirror->idx] = NULL;
1835 mirror_update_dups(ofproto);
1839 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1842 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1843 ofproto->need_revalidate = true;
1844 mac_learning_flush(ofproto->ml);
1850 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
1852 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1853 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1854 return bundle && bundle->mirror_out != 0;
1858 forward_bpdu_changed(struct ofproto *ofproto_)
1860 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1861 /* Revalidate cached flows whenever forward_bpdu option changes. */
1862 ofproto->need_revalidate = true;
1867 static struct ofport_dpif *
1868 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1870 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1871 return ofport ? ofport_dpif_cast(ofport) : NULL;
1874 static struct ofport_dpif *
1875 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1877 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1881 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1882 struct dpif_port *dpif_port)
1884 ofproto_port->name = dpif_port->name;
1885 ofproto_port->type = dpif_port->type;
1886 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1890 port_run(struct ofport_dpif *ofport)
1892 bool enable = netdev_get_carrier(ofport->up.netdev);
1895 cfm_run(ofport->cfm);
1897 if (cfm_should_send_ccm(ofport->cfm)) {
1898 struct ofpbuf packet;
1900 ofpbuf_init(&packet, 0);
1901 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1902 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1903 ofport->odp_port, &packet);
1904 ofpbuf_uninit(&packet);
1907 enable = enable && !cfm_get_fault(ofport->cfm)
1908 && cfm_get_opup(ofport->cfm);
1911 if (ofport->bundle) {
1912 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1915 if (ofport->may_enable != enable) {
1916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1918 if (ofproto->has_bundle_action) {
1919 ofproto->need_revalidate = true;
1923 ofport->may_enable = enable;
1927 port_wait(struct ofport_dpif *ofport)
1930 cfm_wait(ofport->cfm);
1935 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1936 struct ofproto_port *ofproto_port)
1938 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1939 struct dpif_port dpif_port;
1942 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1944 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1950 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1952 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1956 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1958 *ofp_portp = odp_port_to_ofp_port(odp_port);
1964 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1966 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1969 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1971 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1973 /* The caller is going to close ofport->up.netdev. If this is a
1974 * bonded port, then the bond is using that netdev, so remove it
1975 * from the bond. The client will need to reconfigure everything
1976 * after deleting ports, so then the slave will get re-added. */
1977 bundle_remove(&ofport->up);
1983 struct port_dump_state {
1984 struct dpif_port_dump dump;
1989 port_dump_start(const struct ofproto *ofproto_, void **statep)
1991 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1992 struct port_dump_state *state;
1994 *statep = state = xmalloc(sizeof *state);
1995 dpif_port_dump_start(&state->dump, ofproto->dpif);
1996 state->done = false;
2001 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2002 struct ofproto_port *port)
2004 struct port_dump_state *state = state_;
2005 struct dpif_port dpif_port;
2007 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2008 ofproto_port_from_dpif_port(port, &dpif_port);
2011 int error = dpif_port_dump_done(&state->dump);
2013 return error ? error : EOF;
2018 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2020 struct port_dump_state *state = state_;
2023 dpif_port_dump_done(&state->dump);
2030 port_poll(const struct ofproto *ofproto_, char **devnamep)
2032 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2033 return dpif_port_poll(ofproto->dpif, devnamep);
2037 port_poll_wait(const struct ofproto *ofproto_)
2039 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2040 dpif_port_poll_wait(ofproto->dpif);
2044 port_is_lacp_current(const struct ofport *ofport_)
2046 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2047 return (ofport->bundle && ofport->bundle->lacp
2048 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2052 /* Upcall handling. */
2054 /* Flow miss batching.
2056 * Some dpifs implement operations faster when you hand them off in a batch.
2057 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2058 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2059 * more packets, plus possibly installing the flow in the dpif.
2061 * So far we only batch the operations that affect flow setup time the most.
2062 * It's possible to batch more than that, but the benefit might be minimal. */
2064 struct hmap_node hmap_node;
2066 const struct nlattr *key;
2068 struct list packets;
2071 struct flow_miss_op {
2072 union dpif_op dpif_op;
2073 struct facet *facet;
2076 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2077 * OpenFlow controller as necessary according to their individual
2080 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2081 * ownership is transferred to this function. */
2083 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2084 const struct flow *flow, bool clone)
2086 struct ofputil_packet_in pin;
2088 pin.packet = packet;
2089 pin.in_port = flow->in_port;
2090 pin.reason = OFPR_NO_MATCH;
2091 pin.buffer_id = 0; /* not yet known */
2092 pin.send_len = 0; /* not used for flow table misses */
2093 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2094 clone ? NULL : packet);
2097 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2098 * OpenFlow controller as necessary according to their individual
2101 * 'send_len' should be the number of bytes of 'packet' to send to the
2102 * controller, as specified in the action that caused the packet to be sent.
2104 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2105 * Otherwise, ownership is transferred to this function. */
2107 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2108 uint64_t userdata, const struct flow *flow, bool clone)
2110 struct ofputil_packet_in pin;
2111 struct user_action_cookie cookie;
2113 memcpy(&cookie, &userdata, sizeof(cookie));
2115 pin.packet = packet;
2116 pin.in_port = flow->in_port;
2117 pin.reason = OFPR_ACTION;
2118 pin.buffer_id = 0; /* not yet known */
2119 pin.send_len = cookie.data;
2120 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2121 clone ? NULL : packet);
2125 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2126 const struct ofpbuf *packet)
2128 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2134 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2136 cfm_process_heartbeat(ofport->cfm, packet);
2139 } else if (ofport->bundle && ofport->bundle->lacp
2140 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2142 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2145 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2147 stp_process_packet(ofport, packet);
2154 static struct flow_miss *
2155 flow_miss_create(struct hmap *todo, const struct flow *flow,
2156 const struct nlattr *key, size_t key_len)
2158 uint32_t hash = flow_hash(flow, 0);
2159 struct flow_miss *miss;
2161 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2162 if (flow_equal(&miss->flow, flow)) {
2167 miss = xmalloc(sizeof *miss);
2168 hmap_insert(todo, &miss->hmap_node, hash);
2171 miss->key_len = key_len;
2172 list_init(&miss->packets);
2177 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2178 struct flow_miss_op *ops, size_t *n_ops)
2180 const struct flow *flow = &miss->flow;
2181 struct ofpbuf *packet, *next_packet;
2182 struct facet *facet;
2184 facet = facet_lookup_valid(ofproto, flow);
2186 struct rule_dpif *rule;
2188 rule = rule_dpif_lookup(ofproto, flow, 0);
2190 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2191 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2193 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2194 COVERAGE_INC(ofproto_dpif_no_packet_in);
2195 /* XXX install 'drop' flow entry */
2199 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2203 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2205 list_remove(&packet->list_node);
2206 send_packet_in_miss(ofproto, packet, flow, false);
2212 facet = facet_create(rule, flow);
2215 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2216 list_remove(&packet->list_node);
2217 ofproto->n_matches++;
2219 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2221 * Extra-special case for fail-open mode.
2223 * We are in fail-open mode and the packet matched the fail-open
2224 * rule, but we are connected to a controller too. We should send
2225 * the packet up to the controller in the hope that it will try to
2226 * set up a flow and thereby allow us to exit fail-open.
2228 * See the top-level comment in fail-open.c for more information.
2230 send_packet_in_miss(ofproto, packet, flow, true);
2233 if (!facet->may_install) {
2234 facet_make_actions(ofproto, facet, packet);
2236 if (!execute_controller_action(ofproto, &facet->flow,
2237 facet->actions, facet->actions_len,
2239 struct flow_miss_op *op = &ops[(*n_ops)++];
2240 struct dpif_execute *execute = &op->dpif_op.execute;
2243 execute->type = DPIF_OP_EXECUTE;
2244 execute->key = miss->key;
2245 execute->key_len = miss->key_len;
2247 = (facet->may_install
2249 : xmemdup(facet->actions, facet->actions_len));
2250 execute->actions_len = facet->actions_len;
2251 execute->packet = packet;
2255 if (facet->may_install) {
2256 struct flow_miss_op *op = &ops[(*n_ops)++];
2257 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2260 put->type = DPIF_OP_FLOW_PUT;
2261 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2262 put->key = miss->key;
2263 put->key_len = miss->key_len;
2264 put->actions = facet->actions;
2265 put->actions_len = facet->actions_len;
2271 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2274 struct dpif_upcall *upcall;
2275 struct flow_miss *miss, *next_miss;
2276 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2277 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2286 /* Construct the to-do list.
2288 * This just amounts to extracting the flow from each packet and sticking
2289 * the packets that have the same flow in the same "flow_miss" structure so
2290 * that we can process them together. */
2292 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2293 struct flow_miss *miss;
2296 /* Obtain in_port and tun_id, at least, then set 'flow''s header
2298 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2299 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2300 flow.in_port, &flow);
2302 /* Handle 802.1ag, LACP, and STP specially. */
2303 if (process_special(ofproto, &flow, upcall->packet)) {
2304 ofpbuf_delete(upcall->packet);
2305 ofproto->n_matches++;
2309 /* Add other packets to a to-do list. */
2310 miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len);
2311 list_push_back(&miss->packets, &upcall->packet->list_node);
2314 /* Process each element in the to-do list, constructing the set of
2315 * operations to batch. */
2317 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2318 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2319 ofpbuf_list_delete(&miss->packets);
2320 hmap_remove(&todo, &miss->hmap_node);
2323 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2324 hmap_destroy(&todo);
2326 /* Execute batch. */
2327 for (i = 0; i < n_ops; i++) {
2328 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2330 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2332 /* Free memory and update facets. */
2333 for (i = 0; i < n_ops; i++) {
2334 struct flow_miss_op *op = &flow_miss_ops[i];
2335 struct dpif_execute *execute;
2336 struct dpif_flow_put *put;
2338 switch (op->dpif_op.type) {
2339 case DPIF_OP_EXECUTE:
2340 execute = &op->dpif_op.execute;
2341 if (op->facet->actions != execute->actions) {
2342 free((struct nlattr *) execute->actions);
2344 ofpbuf_delete((struct ofpbuf *) execute->packet);
2347 case DPIF_OP_FLOW_PUT:
2348 put = &op->dpif_op.flow_put;
2350 op->facet->installed = true;
2358 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2359 struct dpif_upcall *upcall)
2362 struct user_action_cookie cookie;
2364 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2366 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2367 if (ofproto->sflow) {
2368 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2369 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
2371 ofpbuf_delete(upcall->packet);
2373 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2374 COVERAGE_INC(ofproto_dpif_ctlr_action);
2375 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2376 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2379 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2384 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
2386 switch (upcall->type) {
2387 case DPIF_UC_ACTION:
2388 handle_userspace_upcall(ofproto, upcall);
2392 /* The caller handles these. */
2395 case DPIF_N_UC_TYPES:
2397 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
2402 /* Flow expiration. */
2404 static int facet_max_idle(const struct ofproto_dpif *);
2405 static void update_stats(struct ofproto_dpif *);
2406 static void rule_expire(struct rule_dpif *);
2407 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
2409 /* This function is called periodically by run(). Its job is to collect
2410 * updates for the flows that have been installed into the datapath, most
2411 * importantly when they last were used, and then use that information to
2412 * expire flows that have not been used recently.
2414 * Returns the number of milliseconds after which it should be called again. */
2416 expire(struct ofproto_dpif *ofproto)
2418 struct rule_dpif *rule, *next_rule;
2419 struct classifier *table;
2422 /* Update stats for each flow in the datapath. */
2423 update_stats(ofproto);
2425 /* Expire facets that have been idle too long. */
2426 dp_max_idle = facet_max_idle(ofproto);
2427 expire_facets(ofproto, dp_max_idle);
2429 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2430 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2431 struct cls_cursor cursor;
2433 cls_cursor_init(&cursor, table, NULL);
2434 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2439 /* All outstanding data in existing flows has been accounted, so it's a
2440 * good time to do bond rebalancing. */
2441 if (ofproto->has_bonded_bundles) {
2442 struct ofbundle *bundle;
2444 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2446 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2451 return MIN(dp_max_idle, 1000);
2454 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2456 * This function also pushes statistics updates to rules which each facet
2457 * resubmits into. Generally these statistics will be accurate. However, if a
2458 * facet changes the rule it resubmits into at some time in between
2459 * update_stats() runs, it is possible that statistics accrued to the
2460 * old rule will be incorrectly attributed to the new rule. This could be
2461 * avoided by calling update_stats() whenever rules are created or
2462 * deleted. However, the performance impact of making so many calls to the
2463 * datapath do not justify the benefit of having perfectly accurate statistics.
2466 update_stats(struct ofproto_dpif *p)
2468 const struct dpif_flow_stats *stats;
2469 struct dpif_flow_dump dump;
2470 const struct nlattr *key;
2473 dpif_flow_dump_start(&dump, p->dpif);
2474 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2475 struct facet *facet;
2478 if (odp_flow_key_to_flow(key, key_len, &flow)) {
2482 odp_flow_key_format(key, key_len, &s);
2483 VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
2489 facet = facet_find(p, &flow);
2491 if (facet && facet->installed) {
2493 if (stats->n_packets >= facet->dp_packet_count) {
2494 uint64_t extra = stats->n_packets - facet->dp_packet_count;
2495 facet->packet_count += extra;
2497 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2500 if (stats->n_bytes >= facet->dp_byte_count) {
2501 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
2503 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2506 facet->dp_packet_count = stats->n_packets;
2507 facet->dp_byte_count = stats->n_bytes;
2509 facet_update_time(p, facet, stats->used);
2510 facet_account(p, facet);
2511 facet_push_stats(facet);
2513 /* There's a flow in the datapath that we know nothing about.
2515 COVERAGE_INC(facet_unexpected);
2516 dpif_flow_del(p->dpif, key, key_len, NULL);
2519 dpif_flow_dump_done(&dump);
2522 /* Calculates and returns the number of milliseconds of idle time after which
2523 * facets should expire from the datapath and we should fold their statistics
2524 * into their parent rules in userspace. */
2526 facet_max_idle(const struct ofproto_dpif *ofproto)
2529 * Idle time histogram.
2531 * Most of the time a switch has a relatively small number of facets. When
2532 * this is the case we might as well keep statistics for all of them in
2533 * userspace and to cache them in the kernel datapath for performance as
2536 * As the number of facets increases, the memory required to maintain
2537 * statistics about them in userspace and in the kernel becomes
2538 * significant. However, with a large number of facets it is likely that
2539 * only a few of them are "heavy hitters" that consume a large amount of
2540 * bandwidth. At this point, only heavy hitters are worth caching in the
2541 * kernel and maintaining in userspaces; other facets we can discard.
2543 * The technique used to compute the idle time is to build a histogram with
2544 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
2545 * that is installed in the kernel gets dropped in the appropriate bucket.
2546 * After the histogram has been built, we compute the cutoff so that only
2547 * the most-recently-used 1% of facets (but at least
2548 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2549 * the most-recently-used bucket of facets is kept, so actually an
2550 * arbitrary number of facets can be kept in any given expiration run
2551 * (though the next run will delete most of those unless they receive
2554 * This requires a second pass through the facets, in addition to the pass
2555 * made by update_stats(), because the former function never looks
2556 * at uninstallable facets.
2558 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2559 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2560 int buckets[N_BUCKETS] = { 0 };
2561 int total, subtotal, bucket;
2562 struct facet *facet;
2566 total = hmap_count(&ofproto->facets);
2567 if (total <= ofproto->up.flow_eviction_threshold) {
2568 return N_BUCKETS * BUCKET_WIDTH;
2571 /* Build histogram. */
2573 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
2574 long long int idle = now - facet->used;
2575 int bucket = (idle <= 0 ? 0
2576 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2577 : (unsigned int) idle / BUCKET_WIDTH);
2581 /* Find the first bucket whose flows should be expired. */
2582 subtotal = bucket = 0;
2584 subtotal += buckets[bucket++];
2585 } while (bucket < N_BUCKETS &&
2586 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2588 if (VLOG_IS_DBG_ENABLED()) {
2592 ds_put_cstr(&s, "keep");
2593 for (i = 0; i < N_BUCKETS; i++) {
2595 ds_put_cstr(&s, ", drop");
2598 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2601 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2605 return bucket * BUCKET_WIDTH;
2609 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
2611 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
2612 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
2613 struct ofexpired expired;
2615 if (facet->installed) {
2616 struct dpif_flow_stats stats;
2618 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
2620 facet_update_stats(ofproto, facet, &stats);
2623 expired.flow = facet->flow;
2624 expired.packet_count = facet->packet_count;
2625 expired.byte_count = facet->byte_count;
2626 expired.used = facet->used;
2627 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2632 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
2634 long long int cutoff = time_msec() - dp_max_idle;
2635 struct facet *facet, *next_facet;
2637 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
2638 facet_active_timeout(ofproto, facet);
2639 if (facet->used < cutoff) {
2640 facet_remove(ofproto, facet);
2645 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2646 * then delete it entirely. */
2648 rule_expire(struct rule_dpif *rule)
2650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2651 struct facet *facet, *next_facet;
2655 /* Has 'rule' expired? */
2657 if (rule->up.hard_timeout
2658 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
2659 reason = OFPRR_HARD_TIMEOUT;
2660 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2661 && now > rule->used + rule->up.idle_timeout * 1000) {
2662 reason = OFPRR_IDLE_TIMEOUT;
2667 COVERAGE_INC(ofproto_dpif_expired);
2669 /* Update stats. (This is a no-op if the rule expired due to an idle
2670 * timeout, because that only happens when the rule has no facets left.) */
2671 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2672 facet_remove(ofproto, facet);
2675 /* Get rid of the rule. */
2676 ofproto_rule_expire(&rule->up, reason);
2681 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
2683 * The caller must already have determined that no facet with an identical
2684 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2685 * the ofproto's classifier table.
2687 * The facet will initially have no ODP actions. The caller should fix that
2688 * by calling facet_make_actions(). */
2689 static struct facet *
2690 facet_create(struct rule_dpif *rule, const struct flow *flow)
2692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2693 struct facet *facet;
2695 facet = xzalloc(sizeof *facet);
2696 facet->used = time_msec();
2697 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2698 list_push_back(&rule->facets, &facet->list_node);
2700 facet->flow = *flow;
2701 netflow_flow_init(&facet->nf_flow);
2702 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2708 facet_free(struct facet *facet)
2710 free(facet->actions);
2715 execute_controller_action(struct ofproto_dpif *ofproto,
2716 const struct flow *flow,
2717 const struct nlattr *odp_actions, size_t actions_len,
2718 struct ofpbuf *packet)
2721 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
2722 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
2723 /* As an optimization, avoid a round-trip from userspace to kernel to
2724 * userspace. This also avoids possibly filling up kernel packet
2725 * buffers along the way.
2727 * This optimization will not accidentally catch sFlow
2728 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
2729 * inside OVS_ACTION_ATTR_SAMPLE. */
2730 const struct nlattr *nla;
2732 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
2733 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
2741 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2742 * 'packet', which arrived on 'in_port'.
2744 * Takes ownership of 'packet'. */
2746 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2747 const struct nlattr *odp_actions, size_t actions_len,
2748 struct ofpbuf *packet)
2750 struct odputil_keybuf keybuf;
2754 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
2759 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2760 odp_flow_key_from_flow(&key, flow);
2762 error = dpif_execute(ofproto->dpif, key.data, key.size,
2763 odp_actions, actions_len, packet);
2765 ofpbuf_delete(packet);
2769 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2770 * statistics appropriately. 'packet' must have at least sizeof(struct
2771 * ofp_packet_in) bytes of headroom.
2773 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2774 * applying flow_extract() to 'packet' would yield the same flow as
2777 * 'facet' must have accurately composed datapath actions; that is, it must
2778 * not be in need of revalidation.
2780 * Takes ownership of 'packet'. */
2782 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2783 struct ofpbuf *packet)
2785 struct dpif_flow_stats stats;
2787 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2789 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2790 stats.used = time_msec();
2791 if (execute_odp_actions(ofproto, &facet->flow,
2792 facet->actions, facet->actions_len, packet)) {
2793 facet_update_stats(ofproto, facet, &stats);
2797 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2799 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2800 * rule's statistics, via facet_uninstall().
2802 * - Removes 'facet' from its rule and from ofproto->facets.
2805 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2807 facet_uninstall(ofproto, facet);
2808 facet_flush_stats(ofproto, facet);
2809 hmap_remove(&ofproto->facets, &facet->hmap_node);
2810 list_remove(&facet->list_node);
2814 /* Composes the datapath actions for 'facet' based on its rule's actions. */
2816 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2817 const struct ofpbuf *packet)
2819 const struct rule_dpif *rule = facet->rule;
2820 struct ofpbuf *odp_actions;
2821 struct action_xlate_ctx ctx;
2823 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2824 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2825 facet->tags = ctx.tags;
2826 facet->may_install = ctx.may_set_up_flow;
2827 facet->has_learn = ctx.has_learn;
2828 facet->has_normal = ctx.has_normal;
2829 facet->nf_flow.output_iface = ctx.nf_output_iface;
2831 if (facet->actions_len != odp_actions->size
2832 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2833 free(facet->actions);
2834 facet->actions_len = odp_actions->size;
2835 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2838 ofpbuf_delete(odp_actions);
2841 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2842 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2843 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2844 * since 'facet' was last updated.
2846 * Returns 0 if successful, otherwise a positive errno value.*/
2848 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2849 const struct nlattr *actions, size_t actions_len,
2850 struct dpif_flow_stats *stats)
2852 struct odputil_keybuf keybuf;
2853 enum dpif_flow_put_flags flags;
2857 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2859 flags |= DPIF_FP_ZERO_STATS;
2862 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2863 odp_flow_key_from_flow(&key, &facet->flow);
2865 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2866 actions, actions_len, stats);
2869 facet_reset_dp_stats(facet, stats);
2875 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2876 * 'zero_stats' is true, clears any existing statistics from the datapath for
2879 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2881 struct dpif_flow_stats stats;
2883 if (facet->may_install
2884 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2885 zero_stats ? &stats : NULL)) {
2886 facet->installed = true;
2891 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
2894 const struct nlattr *a;
2898 if (facet->byte_count <= facet->accounted_bytes) {
2901 n_bytes = facet->byte_count - facet->accounted_bytes;
2902 facet->accounted_bytes = facet->byte_count;
2904 /* Feed information from the active flows back into the learning table to
2905 * ensure that table is always in sync with what is actually flowing
2906 * through the datapath. */
2907 if (facet->has_learn || facet->has_normal) {
2908 struct action_xlate_ctx ctx;
2910 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2911 ctx.may_learn = true;
2912 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
2913 facet->rule->up.n_actions));
2916 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
2920 /* This loop feeds byte counters to bond_account() for rebalancing to use
2921 * as a basis. We also need to track the actual VLAN on which the packet
2922 * is going to be sent to ensure that it matches the one passed to
2923 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2925 vlan_tci = facet->flow.vlan_tci;
2926 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2927 const struct ovs_action_push_vlan *vlan;
2928 struct ofport_dpif *port;
2930 switch (nl_attr_type(a)) {
2931 case OVS_ACTION_ATTR_OUTPUT:
2932 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2933 if (port && port->bundle && port->bundle->bond) {
2934 bond_account(port->bundle->bond, &facet->flow,
2935 vlan_tci_to_vid(vlan_tci), n_bytes);
2939 case OVS_ACTION_ATTR_POP_VLAN:
2940 vlan_tci = htons(0);
2943 case OVS_ACTION_ATTR_PUSH_VLAN:
2944 vlan = nl_attr_get(a);
2945 vlan_tci = vlan->vlan_tci;
2951 /* If 'rule' is installed in the datapath, uninstalls it. */
2953 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2955 if (facet->installed) {
2956 struct odputil_keybuf keybuf;
2957 struct dpif_flow_stats stats;
2961 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2962 odp_flow_key_from_flow(&key, &facet->flow);
2964 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2965 facet_reset_dp_stats(facet, &stats);
2967 facet_update_stats(p, facet, &stats);
2969 facet->installed = false;
2971 assert(facet->dp_packet_count == 0);
2972 assert(facet->dp_byte_count == 0);
2976 /* Returns true if the only action for 'facet' is to send to the controller.
2977 * (We don't report NetFlow expiration messages for such facets because they
2978 * are just part of the control logic for the network, not real traffic). */
2980 facet_is_controller_flow(struct facet *facet)
2983 && facet->rule->up.n_actions == 1
2984 && action_outputs_to_port(&facet->rule->up.actions[0],
2985 htons(OFPP_CONTROLLER)));
2988 /* Resets 'facet''s datapath statistics counters. This should be called when
2989 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2990 * it should contain the statistics returned by dpif when 'facet' was reset in
2991 * the datapath. 'stats' will be modified to only included statistics new
2992 * since 'facet' was last updated. */
2994 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2996 if (stats && facet->dp_packet_count <= stats->n_packets
2997 && facet->dp_byte_count <= stats->n_bytes) {
2998 stats->n_packets -= facet->dp_packet_count;
2999 stats->n_bytes -= facet->dp_byte_count;
3002 facet->dp_packet_count = 0;
3003 facet->dp_byte_count = 0;
3006 /* Folds all of 'facet''s statistics into its rule. Also updates the
3007 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3008 * 'facet''s statistics in the datapath should have been zeroed and folded into
3009 * its packet and byte counts before this function is called. */
3011 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3013 assert(!facet->dp_byte_count);
3014 assert(!facet->dp_packet_count);
3016 facet_push_stats(facet);
3017 facet_account(ofproto, facet);
3019 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3020 struct ofexpired expired;
3021 expired.flow = facet->flow;
3022 expired.packet_count = facet->packet_count;
3023 expired.byte_count = facet->byte_count;
3024 expired.used = facet->used;
3025 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3028 facet->rule->packet_count += facet->packet_count;
3029 facet->rule->byte_count += facet->byte_count;
3031 /* Reset counters to prevent double counting if 'facet' ever gets
3033 facet_reset_counters(facet);
3035 netflow_flow_clear(&facet->nf_flow);
3038 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3039 * Returns it if found, otherwise a null pointer.
3041 * The returned facet might need revalidation; use facet_lookup_valid()
3042 * instead if that is important. */
3043 static struct facet *
3044 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3046 struct facet *facet;
3048 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3050 if (flow_equal(flow, &facet->flow)) {
3058 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3059 * Returns it if found, otherwise a null pointer.
3061 * The returned facet is guaranteed to be valid. */
3062 static struct facet *
3063 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3065 struct facet *facet = facet_find(ofproto, flow);
3067 /* The facet we found might not be valid, since we could be in need of
3068 * revalidation. If it is not valid, don't return it. */
3070 && (ofproto->need_revalidate
3071 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3072 && !facet_revalidate(ofproto, facet)) {
3073 COVERAGE_INC(facet_invalidated);
3080 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3082 * - If the rule found is different from 'facet''s current rule, moves
3083 * 'facet' to the new rule and recompiles its actions.
3085 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3086 * where it is and recompiles its actions anyway.
3088 * - If there is none, destroys 'facet'.
3090 * Returns true if 'facet' still exists, false if it has been destroyed. */
3092 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3094 struct action_xlate_ctx ctx;
3095 struct ofpbuf *odp_actions;
3096 struct rule_dpif *new_rule;
3097 bool actions_changed;
3099 COVERAGE_INC(facet_revalidate);
3101 /* Determine the new rule. */
3102 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3104 /* No new rule, so delete the facet. */
3105 facet_remove(ofproto, facet);
3109 /* Calculate new datapath actions.
3111 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3112 * emit a NetFlow expiration and, if so, we need to have the old state
3113 * around to properly compose it. */
3114 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
3115 odp_actions = xlate_actions(&ctx,
3116 new_rule->up.actions, new_rule->up.n_actions);
3117 actions_changed = (facet->actions_len != odp_actions->size
3118 || memcmp(facet->actions, odp_actions->data,
3119 facet->actions_len));
3121 /* If the datapath actions changed or the installability changed,
3122 * then we need to talk to the datapath. */
3123 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
3124 if (ctx.may_set_up_flow) {
3125 struct dpif_flow_stats stats;
3127 facet_put__(ofproto, facet,
3128 odp_actions->data, odp_actions->size, &stats);
3129 facet_update_stats(ofproto, facet, &stats);
3131 facet_uninstall(ofproto, facet);
3134 /* The datapath flow is gone or has zeroed stats, so push stats out of
3135 * 'facet' into 'rule'. */
3136 facet_flush_stats(ofproto, facet);
3139 /* Update 'facet' now that we've taken care of all the old state. */
3140 facet->tags = ctx.tags;
3141 facet->nf_flow.output_iface = ctx.nf_output_iface;
3142 facet->may_install = ctx.may_set_up_flow;
3143 facet->has_learn = ctx.has_learn;
3144 facet->has_normal = ctx.has_normal;
3145 if (actions_changed) {
3146 free(facet->actions);
3147 facet->actions_len = odp_actions->size;
3148 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
3150 if (facet->rule != new_rule) {
3151 COVERAGE_INC(facet_changed_rule);
3152 list_remove(&facet->list_node);
3153 list_push_back(&new_rule->facets, &facet->list_node);
3154 facet->rule = new_rule;
3155 facet->used = new_rule->up.created;
3156 facet->rs_used = facet->used;
3159 ofpbuf_delete(odp_actions);
3164 /* Updates 'facet''s used time. Caller is responsible for calling
3165 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3167 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3170 if (used > facet->used) {
3172 if (used > facet->rule->used) {
3173 facet->rule->used = used;
3175 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3179 /* Folds the statistics from 'stats' into the counters in 'facet'.
3181 * Because of the meaning of a facet's counters, it only makes sense to do this
3182 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3183 * packet that was sent by hand or if it represents statistics that have been
3184 * cleared out of the datapath. */
3186 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
3187 const struct dpif_flow_stats *stats)
3189 if (stats->n_packets || stats->used > facet->used) {
3190 facet_update_time(ofproto, facet, stats->used);
3191 facet->packet_count += stats->n_packets;
3192 facet->byte_count += stats->n_bytes;
3193 facet_push_stats(facet);
3194 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3199 facet_reset_counters(struct facet *facet)
3201 facet->packet_count = 0;
3202 facet->byte_count = 0;
3203 facet->rs_packet_count = 0;
3204 facet->rs_byte_count = 0;
3205 facet->accounted_bytes = 0;
3209 facet_push_stats(struct facet *facet)
3211 uint64_t rs_packets, rs_bytes;
3213 assert(facet->packet_count >= facet->rs_packet_count);
3214 assert(facet->byte_count >= facet->rs_byte_count);
3215 assert(facet->used >= facet->rs_used);
3217 rs_packets = facet->packet_count - facet->rs_packet_count;
3218 rs_bytes = facet->byte_count - facet->rs_byte_count;
3220 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3221 facet->rs_packet_count = facet->packet_count;
3222 facet->rs_byte_count = facet->byte_count;
3223 facet->rs_used = facet->used;
3225 flow_push_stats(facet->rule, &facet->flow,
3226 rs_packets, rs_bytes, facet->used);
3230 struct ofproto_push {
3231 struct action_xlate_ctx ctx;
3238 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3240 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3243 rule->packet_count += push->packets;
3244 rule->byte_count += push->bytes;
3245 rule->used = MAX(push->used, rule->used);
3249 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3250 * 'rule''s actions. */
3252 flow_push_stats(const struct rule_dpif *rule,
3253 struct flow *flow, uint64_t packets, uint64_t bytes,
3256 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3257 struct ofproto_push push;
3259 push.packets = packets;
3263 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3264 push.ctx.resubmit_hook = push_resubmit;
3265 ofpbuf_delete(xlate_actions(&push.ctx,
3266 rule->up.actions, rule->up.n_actions));
3271 static struct rule_dpif *
3272 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3275 struct cls_rule *cls_rule;
3276 struct classifier *cls;
3278 if (table_id >= N_TABLES) {
3282 cls = &ofproto->up.tables[table_id];
3283 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3284 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3285 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3286 * are unavailable. */
3287 struct flow ofpc_normal_flow = *flow;
3288 ofpc_normal_flow.tp_src = htons(0);
3289 ofpc_normal_flow.tp_dst = htons(0);
3290 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3292 cls_rule = classifier_lookup(cls, flow);
3294 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3298 complete_operation(struct rule_dpif *rule)
3300 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3302 rule_invalidate(rule);
3304 struct dpif_completion *c = xmalloc(sizeof *c);
3305 c->op = rule->up.pending;
3306 list_push_back(&ofproto->completions, &c->list_node);
3308 ofoperation_complete(rule->up.pending, 0);
3312 static struct rule *
3315 struct rule_dpif *rule = xmalloc(sizeof *rule);
3320 rule_dealloc(struct rule *rule_)
3322 struct rule_dpif *rule = rule_dpif_cast(rule_);
3327 rule_construct(struct rule *rule_)
3329 struct rule_dpif *rule = rule_dpif_cast(rule_);
3330 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3331 struct rule_dpif *victim;
3335 error = validate_actions(rule->up.actions, rule->up.n_actions,
3336 &rule->up.cr.flow, ofproto->max_ports);
3341 rule->used = rule->up.created;
3342 rule->packet_count = 0;
3343 rule->byte_count = 0;
3345 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3346 if (victim && !list_is_empty(&victim->facets)) {
3347 struct facet *facet;
3349 rule->facets = victim->facets;
3350 list_moved(&rule->facets);
3351 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3352 /* XXX: We're only clearing our local counters here. It's possible
3353 * that quite a few packets are unaccounted for in the datapath
3354 * statistics. These will be accounted to the new rule instead of
3355 * cleared as required. This could be fixed by clearing out the
3356 * datapath statistics for this facet, but currently it doesn't
3358 facet_reset_counters(facet);
3362 /* Must avoid list_moved() in this case. */
3363 list_init(&rule->facets);
3366 table_id = rule->up.table_id;
3367 rule->tag = (victim ? victim->tag
3369 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3370 ofproto->tables[table_id].basis));
3372 complete_operation(rule);
3377 rule_destruct(struct rule *rule_)
3379 struct rule_dpif *rule = rule_dpif_cast(rule_);
3380 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3381 struct facet *facet, *next_facet;
3383 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3384 facet_revalidate(ofproto, facet);
3387 complete_operation(rule);
3391 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3393 struct rule_dpif *rule = rule_dpif_cast(rule_);
3394 struct facet *facet;
3396 /* Start from historical data for 'rule' itself that are no longer tracked
3397 * in facets. This counts, for example, facets that have expired. */
3398 *packets = rule->packet_count;
3399 *bytes = rule->byte_count;
3401 /* Add any statistics that are tracked by facets. This includes
3402 * statistical data recently updated by ofproto_update_stats() as well as
3403 * stats for packets that were executed "by hand" via dpif_execute(). */
3404 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3405 *packets += facet->packet_count;
3406 *bytes += facet->byte_count;
3411 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
3413 struct rule_dpif *rule = rule_dpif_cast(rule_);
3414 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3415 struct action_xlate_ctx ctx;
3416 struct ofpbuf *odp_actions;
3417 struct facet *facet;
3420 /* First look for a related facet. If we find one, account it to that. */
3421 facet = facet_lookup_valid(ofproto, flow);
3422 if (facet && facet->rule == rule) {
3423 if (!facet->may_install) {
3424 facet_make_actions(ofproto, facet, packet);
3426 facet_execute(ofproto, facet, packet);
3430 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
3431 * create a new facet for it and use that. */
3432 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
3433 facet = facet_create(rule, flow);
3434 facet_make_actions(ofproto, facet, packet);
3435 facet_execute(ofproto, facet, packet);
3436 facet_install(ofproto, facet, true);
3440 /* We can't account anything to a facet. If we were to try, then that
3441 * facet would have a non-matching rule, busting our invariants. */
3442 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3443 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3444 size = packet->size;
3445 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3446 odp_actions->size, packet)) {
3447 rule->used = time_msec();
3448 rule->packet_count++;
3449 rule->byte_count += size;
3450 flow_push_stats(rule, flow, 1, size, rule->used);
3452 ofpbuf_delete(odp_actions);
3458 rule_modify_actions(struct rule *rule_)
3460 struct rule_dpif *rule = rule_dpif_cast(rule_);
3461 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3464 error = validate_actions(rule->up.actions, rule->up.n_actions,
3465 &rule->up.cr.flow, ofproto->max_ports);
3467 ofoperation_complete(rule->up.pending, error);
3471 complete_operation(rule);
3474 /* Sends 'packet' out of port 'odp_port' within 'ofproto'.
3475 * Returns 0 if successful, otherwise a positive errno value. */
3477 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
3478 const struct ofpbuf *packet)
3480 struct ofpbuf key, odp_actions;
3481 struct odputil_keybuf keybuf;
3485 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
3486 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3487 odp_flow_key_from_flow(&key, &flow);
3489 ofpbuf_init(&odp_actions, 32);
3490 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
3492 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3493 error = dpif_execute(ofproto->dpif,
3495 odp_actions.data, odp_actions.size,
3497 ofpbuf_uninit(&odp_actions);
3500 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
3501 ofproto->up.name, odp_port, strerror(error));
3506 /* OpenFlow to datapath action translation. */
3508 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
3509 struct action_xlate_ctx *ctx);
3510 static void xlate_normal(struct action_xlate_ctx *);
3513 put_userspace_action(const struct ofproto_dpif *ofproto,
3514 struct ofpbuf *odp_actions,
3515 const struct flow *flow,
3516 const struct user_action_cookie *cookie)
3520 pid = dpif_port_get_pid(ofproto->dpif,
3521 ofp_port_to_odp_port(flow->in_port));
3523 return odp_put_userspace_action(pid, cookie, odp_actions);
3526 /* Compose SAMPLE action for sFlow. */
3528 compose_sflow_action(const struct ofproto_dpif *ofproto,
3529 struct ofpbuf *odp_actions,
3530 const struct flow *flow,
3533 uint32_t port_ifindex;
3534 uint32_t probability;
3535 struct user_action_cookie cookie;
3536 size_t sample_offset, actions_offset;
3537 int cookie_offset, n_output;
3539 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
3543 if (odp_port == OVSP_NONE) {
3547 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
3551 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
3553 /* Number of packets out of UINT_MAX to sample. */
3554 probability = dpif_sflow_get_probability(ofproto->sflow);
3555 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
3557 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
3559 cookie.type = USER_ACTION_COOKIE_SFLOW;
3560 cookie.data = port_ifindex;
3561 cookie.n_output = n_output;
3562 cookie.vlan_tci = 0;
3563 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
3565 nl_msg_end_nested(odp_actions, actions_offset);
3566 nl_msg_end_nested(odp_actions, sample_offset);
3567 return cookie_offset;
3570 /* SAMPLE action must be first action in any given list of actions.
3571 * At this point we do not have all information required to build it. So try to
3572 * build sample action as complete as possible. */
3574 add_sflow_action(struct action_xlate_ctx *ctx)
3576 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
3578 &ctx->flow, OVSP_NONE);
3579 ctx->sflow_odp_port = 0;
3580 ctx->sflow_n_outputs = 0;
3583 /* Fix SAMPLE action according to data collected while composing ODP actions.
3584 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
3585 * USERSPACE action's user-cookie which is required for sflow. */
3587 fix_sflow_action(struct action_xlate_ctx *ctx)
3589 const struct flow *base = &ctx->base_flow;
3590 struct user_action_cookie *cookie;
3592 if (!ctx->user_cookie_offset) {
3596 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
3598 assert(cookie != NULL);
3599 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
3601 if (ctx->sflow_n_outputs) {
3602 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
3603 ctx->sflow_odp_port);
3605 if (ctx->sflow_n_outputs >= 255) {
3606 cookie->n_output = 255;
3608 cookie->n_output = ctx->sflow_n_outputs;
3610 cookie->vlan_tci = base->vlan_tci;
3614 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
3615 const void *key, size_t key_size)
3617 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
3618 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
3619 nl_msg_end_nested(odp_actions, offset);
3623 commit_set_tun_id_action(const struct flow *flow, struct flow *base,
3624 struct ofpbuf *odp_actions)
3626 if (base->tun_id == flow->tun_id) {
3629 base->tun_id = flow->tun_id;
3631 commit_set_action(odp_actions, OVS_KEY_ATTR_TUN_ID,
3632 &base->tun_id, sizeof(base->tun_id));
3636 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
3637 struct ofpbuf *odp_actions)
3639 struct ovs_key_ethernet eth_key;
3641 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
3642 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
3646 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
3647 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
3649 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
3650 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
3652 commit_set_action(odp_actions, OVS_KEY_ATTR_ETHERNET,
3653 ð_key, sizeof(eth_key));
3657 commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci)
3659 struct flow *base = &ctx->base_flow;
3661 if (base->vlan_tci == new_tci) {
3665 if (base->vlan_tci & htons(VLAN_CFI)) {
3666 nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_VLAN);
3669 if (new_tci & htons(VLAN_CFI)) {
3670 struct ovs_action_push_vlan vlan;
3672 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
3673 vlan.vlan_tci = new_tci;
3674 nl_msg_put_unspec(ctx->odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
3675 &vlan, sizeof vlan);
3677 base->vlan_tci = new_tci;
3681 commit_set_nw_action(const struct flow *flow, struct flow *base,
3682 struct ofpbuf *odp_actions)
3684 struct ovs_key_ipv4 ipv4_key;
3686 if (base->dl_type != htons(ETH_TYPE_IP) ||
3687 !base->nw_src || !base->nw_dst) {
3691 if (base->nw_src == flow->nw_src &&
3692 base->nw_dst == flow->nw_dst &&
3693 base->nw_tos == flow->nw_tos &&
3694 base->nw_ttl == flow->nw_ttl &&
3695 base->nw_frag == flow->nw_frag) {
3699 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
3700 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
3701 ipv4_key.ipv4_proto = base->nw_proto;
3702 ipv4_key.ipv4_tos = flow->nw_tos;
3703 ipv4_key.ipv4_ttl = flow->nw_ttl;
3704 ipv4_key.ipv4_frag = (base->nw_frag == 0 ? OVS_FRAG_TYPE_NONE
3705 : base->nw_frag == FLOW_NW_FRAG_ANY
3706 ? OVS_FRAG_TYPE_FIRST : OVS_FRAG_TYPE_LATER);
3708 commit_set_action(odp_actions, OVS_KEY_ATTR_IPV4,
3709 &ipv4_key, sizeof(ipv4_key));
3713 commit_set_port_action(const struct flow *flow, struct flow *base,
3714 struct ofpbuf *odp_actions)
3716 if (!base->tp_src || !base->tp_dst) {
3720 if (base->tp_src == flow->tp_src &&
3721 base->tp_dst == flow->tp_dst) {
3725 if (flow->nw_proto == IPPROTO_TCP) {
3726 struct ovs_key_tcp port_key;
3728 port_key.tcp_src = base->tp_src = flow->tp_src;
3729 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
3731 commit_set_action(odp_actions, OVS_KEY_ATTR_TCP,
3732 &port_key, sizeof(port_key));
3734 } else if (flow->nw_proto == IPPROTO_UDP) {
3735 struct ovs_key_udp port_key;
3737 port_key.udp_src = base->tp_src = flow->tp_src;
3738 port_key.udp_dst = base->tp_dst = flow->tp_dst;
3740 commit_set_action(odp_actions, OVS_KEY_ATTR_UDP,
3741 &port_key, sizeof(port_key));
3746 commit_set_priority_action(const struct flow *flow, struct flow *base,
3747 struct ofpbuf *odp_actions)
3749 if (base->priority == flow->priority) {
3752 base->priority = flow->priority;
3754 commit_set_action(odp_actions, OVS_KEY_ATTR_PRIORITY,
3755 &base->priority, sizeof(base->priority));
3759 commit_odp_actions(struct action_xlate_ctx *ctx)
3761 const struct flow *flow = &ctx->flow;
3762 struct flow *base = &ctx->base_flow;
3763 struct ofpbuf *odp_actions = ctx->odp_actions;
3765 commit_set_tun_id_action(flow, base, odp_actions);
3766 commit_set_ether_addr_action(flow, base, odp_actions);
3767 commit_vlan_action(ctx, flow->vlan_tci);
3768 commit_set_nw_action(flow, base, odp_actions);
3769 commit_set_port_action(flow, base, odp_actions);
3770 commit_set_priority_action(flow, base, odp_actions);
3774 force_compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3776 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3777 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
3779 if (ofport && ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
3783 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3784 ctx->sflow_odp_port = odp_port;
3785 ctx->sflow_n_outputs++;
3789 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3791 struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3793 if (ofport && !stp_forward_in_state(ofport->stp_state)) {
3794 /* Forwarding disabled on port. */
3798 /* We may not have an ofport record for this port, but it doesn't hurt to
3799 * allow forwarding to it anyhow. Maybe such a port will appear later and
3800 * we're pre-populating the flow table. */
3801 force_compose_output_action(ctx, ofp_port);
3805 commit_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3807 commit_odp_actions(ctx);
3808 compose_output_action(ctx, ofp_port);
3809 ctx->nf_output_iface = ofp_port;
3813 xlate_table_action(struct action_xlate_ctx *ctx,
3814 uint16_t in_port, uint8_t table_id)
3816 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
3817 struct ofproto_dpif *ofproto = ctx->ofproto;
3818 struct rule_dpif *rule;
3819 uint16_t old_in_port;
3820 uint8_t old_table_id;
3822 old_table_id = ctx->table_id;
3823 ctx->table_id = table_id;
3825 /* Look up a flow with 'in_port' as the input port. */
3826 old_in_port = ctx->flow.in_port;
3827 ctx->flow.in_port = in_port;
3828 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
3831 if (table_id > 0 && table_id < N_TABLES) {
3832 struct table_dpif *table = &ofproto->tables[table_id];
3833 if (table->other_table) {
3836 : rule_calculate_tag(&ctx->flow,
3837 &table->other_table->wc,
3842 /* Restore the original input port. Otherwise OFPP_NORMAL and
3843 * OFPP_IN_PORT will have surprising behavior. */
3844 ctx->flow.in_port = old_in_port;
3846 if (ctx->resubmit_hook) {
3847 ctx->resubmit_hook(ctx, rule);
3852 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
3856 ctx->table_id = old_table_id;
3858 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3860 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
3861 MAX_RESUBMIT_RECURSION);
3866 xlate_resubmit_table(struct action_xlate_ctx *ctx,
3867 const struct nx_action_resubmit *nar)
3872 in_port = (nar->in_port == htons(OFPP_IN_PORT)
3874 : ntohs(nar->in_port));
3875 table_id = nar->table == 255 ? ctx->table_id : nar->table;
3877 xlate_table_action(ctx, in_port, table_id);
3881 flood_packets(struct action_xlate_ctx *ctx, bool all)
3883 struct ofport_dpif *ofport;
3885 commit_odp_actions(ctx);
3886 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
3887 uint16_t ofp_port = ofport->up.ofp_port;
3889 if (ofp_port == ctx->flow.in_port) {
3894 force_compose_output_action(ctx, ofp_port);
3895 } else if (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))) {
3896 compose_output_action(ctx, ofp_port);
3900 ctx->nf_output_iface = NF_OUT_FLOOD;
3904 compose_controller_action(struct action_xlate_ctx *ctx, int len)
3906 struct user_action_cookie cookie;
3908 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
3910 cookie.n_output = 0;
3911 cookie.vlan_tci = 0;
3912 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
3916 xlate_output_action__(struct action_xlate_ctx *ctx,
3917 uint16_t port, uint16_t max_len)
3919 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
3921 ctx->nf_output_iface = NF_OUT_DROP;
3925 commit_output_action(ctx, ctx->flow.in_port);
3928 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
3934 flood_packets(ctx, false);
3937 flood_packets(ctx, true);
3939 case OFPP_CONTROLLER:
3940 commit_odp_actions(ctx);
3941 compose_controller_action(ctx, max_len);
3944 commit_output_action(ctx, OFPP_LOCAL);
3949 if (port != ctx->flow.in_port) {
3950 commit_output_action(ctx, port);
3955 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3956 ctx->nf_output_iface = NF_OUT_FLOOD;
3957 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3958 ctx->nf_output_iface = prev_nf_output_iface;
3959 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3960 ctx->nf_output_iface != NF_OUT_FLOOD) {
3961 ctx->nf_output_iface = NF_OUT_MULTI;
3966 xlate_output_reg_action(struct action_xlate_ctx *ctx,
3967 const struct nx_action_output_reg *naor)
3971 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
3973 if (ofp_port <= UINT16_MAX) {
3974 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
3979 xlate_output_action(struct action_xlate_ctx *ctx,
3980 const struct ofp_action_output *oao)
3982 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3986 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3987 const struct ofp_action_enqueue *oae)
3990 uint32_t flow_priority, priority;
3993 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3996 /* Fall back to ordinary output action. */
3997 xlate_output_action__(ctx, ntohs(oae->port), 0);
4001 /* Figure out datapath output port. */
4002 ofp_port = ntohs(oae->port);
4003 if (ofp_port == OFPP_IN_PORT) {
4004 ofp_port = ctx->flow.in_port;
4005 } else if (ofp_port == ctx->flow.in_port) {
4009 /* Add datapath actions. */
4010 flow_priority = ctx->flow.priority;
4011 ctx->flow.priority = priority;
4012 commit_output_action(ctx, ofp_port);
4013 ctx->flow.priority = flow_priority;
4015 /* Update NetFlow output port. */
4016 if (ctx->nf_output_iface == NF_OUT_DROP) {
4017 ctx->nf_output_iface = ofp_port;
4018 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4019 ctx->nf_output_iface = NF_OUT_MULTI;
4024 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4025 const struct nx_action_set_queue *nasq)
4030 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4033 /* Couldn't translate queue to a priority, so ignore. A warning
4034 * has already been logged. */
4038 ctx->flow.priority = priority;
4041 struct xlate_reg_state {
4047 xlate_autopath(struct action_xlate_ctx *ctx,
4048 const struct nx_action_autopath *naa)
4050 uint16_t ofp_port = ntohl(naa->id);
4051 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4053 if (!port || !port->bundle) {
4054 ofp_port = OFPP_NONE;
4055 } else if (port->bundle->bond) {
4056 /* Autopath does not support VLAN hashing. */
4057 struct ofport_dpif *slave = bond_choose_output_slave(
4058 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4060 ofp_port = slave->up.ofp_port;
4063 autopath_execute(naa, &ctx->flow, ofp_port);
4067 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4069 struct ofproto_dpif *ofproto = ofproto_;
4070 struct ofport_dpif *port;
4080 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4083 port = get_ofp_port(ofproto, ofp_port);
4084 return port ? port->may_enable : false;
4089 xlate_learn_action(struct action_xlate_ctx *ctx,
4090 const struct nx_action_learn *learn)
4092 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4093 struct ofputil_flow_mod fm;
4096 learn_execute(learn, &ctx->flow, &fm);
4098 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4099 if (error && !VLOG_DROP_WARN(&rl)) {
4100 char *msg = ofputil_error_to_string(error);
4101 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4109 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4111 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4112 ? htonl(OFPPC_NO_RECV_STP)
4113 : htonl(OFPPC_NO_RECV))) {
4117 /* Only drop packets here if both forwarding and learning are
4118 * disabled. If just learning is enabled, we need to have
4119 * OFPP_NORMAL and the learning action have a look at the packet
4120 * before we can drop it. */
4121 if (!stp_forward_in_state(port->stp_state)
4122 && !stp_learn_in_state(port->stp_state)) {
4130 do_xlate_actions(const union ofp_action *in, size_t n_in,
4131 struct action_xlate_ctx *ctx)
4133 const struct ofport_dpif *port;
4134 const union ofp_action *ia;
4137 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4138 if (port && !may_receive(port, ctx)) {
4139 /* Drop this flow. */
4143 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4144 const struct ofp_action_dl_addr *oada;
4145 const struct nx_action_resubmit *nar;
4146 const struct nx_action_set_tunnel *nast;
4147 const struct nx_action_set_queue *nasq;
4148 const struct nx_action_multipath *nam;
4149 const struct nx_action_autopath *naa;
4150 const struct nx_action_bundle *nab;
4151 const struct nx_action_output_reg *naor;
4152 enum ofputil_action_code code;
4159 code = ofputil_decode_action_unsafe(ia);
4161 case OFPUTIL_OFPAT_OUTPUT:
4162 xlate_output_action(ctx, &ia->output);
4165 case OFPUTIL_OFPAT_SET_VLAN_VID:
4166 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4167 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4170 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4171 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4172 ctx->flow.vlan_tci |= htons(
4173 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4176 case OFPUTIL_OFPAT_STRIP_VLAN:
4177 ctx->flow.vlan_tci = htons(0);
4180 case OFPUTIL_OFPAT_SET_DL_SRC:
4181 oada = ((struct ofp_action_dl_addr *) ia);
4182 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4185 case OFPUTIL_OFPAT_SET_DL_DST:
4186 oada = ((struct ofp_action_dl_addr *) ia);
4187 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4190 case OFPUTIL_OFPAT_SET_NW_SRC:
4191 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4194 case OFPUTIL_OFPAT_SET_NW_DST:
4195 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4198 case OFPUTIL_OFPAT_SET_NW_TOS:
4199 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4200 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4203 case OFPUTIL_OFPAT_SET_TP_SRC:
4204 ctx->flow.tp_src = ia->tp_port.tp_port;
4207 case OFPUTIL_OFPAT_SET_TP_DST:
4208 ctx->flow.tp_dst = ia->tp_port.tp_port;
4211 case OFPUTIL_OFPAT_ENQUEUE:
4212 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4215 case OFPUTIL_NXAST_RESUBMIT:
4216 nar = (const struct nx_action_resubmit *) ia;
4217 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4220 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4221 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4224 case OFPUTIL_NXAST_SET_TUNNEL:
4225 nast = (const struct nx_action_set_tunnel *) ia;
4226 tun_id = htonll(ntohl(nast->tun_id));
4227 ctx->flow.tun_id = tun_id;
4230 case OFPUTIL_NXAST_SET_QUEUE:
4231 nasq = (const struct nx_action_set_queue *) ia;
4232 xlate_set_queue_action(ctx, nasq);
4235 case OFPUTIL_NXAST_POP_QUEUE:
4236 ctx->flow.priority = ctx->original_priority;
4239 case OFPUTIL_NXAST_REG_MOVE:
4240 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4244 case OFPUTIL_NXAST_REG_LOAD:
4245 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4249 case OFPUTIL_NXAST_NOTE:
4250 /* Nothing to do. */
4253 case OFPUTIL_NXAST_SET_TUNNEL64:
4254 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4255 ctx->flow.tun_id = tun_id;
4258 case OFPUTIL_NXAST_MULTIPATH:
4259 nam = (const struct nx_action_multipath *) ia;
4260 multipath_execute(nam, &ctx->flow);
4263 case OFPUTIL_NXAST_AUTOPATH:
4264 naa = (const struct nx_action_autopath *) ia;
4265 xlate_autopath(ctx, naa);
4268 case OFPUTIL_NXAST_BUNDLE:
4269 ctx->ofproto->has_bundle_action = true;
4270 nab = (const struct nx_action_bundle *) ia;
4271 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4276 case OFPUTIL_NXAST_BUNDLE_LOAD:
4277 ctx->ofproto->has_bundle_action = true;
4278 nab = (const struct nx_action_bundle *) ia;
4279 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4283 case OFPUTIL_NXAST_OUTPUT_REG:
4284 naor = (const struct nx_action_output_reg *) ia;
4285 xlate_output_reg_action(ctx, naor);
4288 case OFPUTIL_NXAST_LEARN:
4289 ctx->has_learn = true;
4290 if (ctx->may_learn) {
4291 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4295 case OFPUTIL_NXAST_EXIT:
4301 /* We've let OFPP_NORMAL and the learning action look at the packet,
4302 * so drop it now if forwarding is disabled. */
4303 if (port && !stp_forward_in_state(port->stp_state)) {
4304 ofpbuf_clear(ctx->odp_actions);
4305 add_sflow_action(ctx);
4310 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4311 struct ofproto_dpif *ofproto, const struct flow *flow,
4312 const struct ofpbuf *packet)
4314 ctx->ofproto = ofproto;
4316 ctx->packet = packet;
4317 ctx->may_learn = packet != NULL;
4318 ctx->resubmit_hook = NULL;
4321 static struct ofpbuf *
4322 xlate_actions(struct action_xlate_ctx *ctx,
4323 const union ofp_action *in, size_t n_in)
4325 COVERAGE_INC(ofproto_dpif_xlate);
4327 ctx->odp_actions = ofpbuf_new(512);
4328 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4330 ctx->may_set_up_flow = true;
4331 ctx->has_learn = false;
4332 ctx->has_normal = false;
4333 ctx->nf_output_iface = NF_OUT_DROP;
4335 ctx->original_priority = ctx->flow.priority;
4336 ctx->base_flow = ctx->flow;
4337 ctx->base_flow.tun_id = 0;
4341 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4342 switch (ctx->ofproto->up.frag_handling) {
4343 case OFPC_FRAG_NORMAL:
4344 /* We must pretend that transport ports are unavailable. */
4345 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4346 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4349 case OFPC_FRAG_DROP:
4350 return ctx->odp_actions;
4352 case OFPC_FRAG_REASM:
4355 case OFPC_FRAG_NX_MATCH:
4356 /* Nothing to do. */
4361 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4362 ctx->may_set_up_flow = false;
4363 return ctx->odp_actions;
4365 add_sflow_action(ctx);
4366 do_xlate_actions(in, n_in, ctx);
4368 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4369 ctx->odp_actions->data,
4370 ctx->odp_actions->size)) {
4371 ctx->may_set_up_flow = false;
4373 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4375 compose_output_action(ctx, OFPP_LOCAL);
4378 fix_sflow_action(ctx);
4381 return ctx->odp_actions;
4384 /* OFPP_NORMAL implementation. */
4386 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4388 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4389 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4390 * the bundle on which the packet was received, returns the VLAN to which the
4393 * Both 'vid' and the return value are in the range 0...4095. */
4395 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4397 switch (in_bundle->vlan_mode) {
4398 case PORT_VLAN_ACCESS:
4399 return in_bundle->vlan;
4402 case PORT_VLAN_TRUNK:
4405 case PORT_VLAN_NATIVE_UNTAGGED:
4406 case PORT_VLAN_NATIVE_TAGGED:
4407 return vid ? vid : in_bundle->vlan;
4414 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4415 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4418 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4419 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4422 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4424 switch (in_bundle->vlan_mode) {
4425 case PORT_VLAN_ACCESS:
4428 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4429 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4430 "packet received on port %s configured as VLAN "
4431 "%"PRIu16" access port",
4432 in_bundle->ofproto->up.name, vid,
4433 in_bundle->name, in_bundle->vlan);
4439 case PORT_VLAN_NATIVE_UNTAGGED:
4440 case PORT_VLAN_NATIVE_TAGGED:
4442 /* Port must always carry its native VLAN. */
4446 case PORT_VLAN_TRUNK:
4447 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4449 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4450 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4451 "received on port %s not configured for trunking "
4453 in_bundle->ofproto->up.name, vid,
4454 in_bundle->name, vid);
4466 /* Given 'vlan', the VLAN that a packet belongs to, and
4467 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4468 * that should be included in the 802.1Q header. (If the return value is 0,
4469 * then the 802.1Q header should only be included in the packet if there is a
4472 * Both 'vlan' and the return value are in the range 0...4095. */
4474 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4476 switch (out_bundle->vlan_mode) {
4477 case PORT_VLAN_ACCESS:
4480 case PORT_VLAN_TRUNK:
4481 case PORT_VLAN_NATIVE_TAGGED:
4484 case PORT_VLAN_NATIVE_UNTAGGED:
4485 return vlan == out_bundle->vlan ? 0 : vlan;
4493 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4496 struct ofport_dpif *port;
4500 vid = output_vlan_to_vid(out_bundle, vlan);
4501 if (!out_bundle->bond) {
4502 port = ofbundle_get_a_port(out_bundle);
4504 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4507 /* No slaves enabled, so drop packet. */
4513 if (tci || out_bundle->use_priority_tags) {
4514 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4516 tci |= htons(VLAN_CFI);
4519 commit_vlan_action(ctx, tci);
4521 compose_output_action(ctx, port->up.ofp_port);
4522 ctx->nf_output_iface = port->up.ofp_port;
4526 mirror_mask_ffs(mirror_mask_t mask)
4528 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4533 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4535 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4536 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4540 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4542 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4545 /* Returns an arbitrary interface within 'bundle'. */
4546 static struct ofport_dpif *
4547 ofbundle_get_a_port(const struct ofbundle *bundle)
4549 return CONTAINER_OF(list_front(&bundle->ports),
4550 struct ofport_dpif, bundle_node);
4553 static mirror_mask_t
4554 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
4555 const struct ofbundle *in_bundle,
4556 const struct ofbundle *out_bundle)
4558 mirror_mask_t dst_mirrors = 0;
4560 if (out_bundle == OFBUNDLE_FLOOD) {
4561 struct ofbundle *bundle;
4563 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
4564 if (bundle != in_bundle
4565 && ofbundle_includes_vlan(bundle, vlan)
4566 && bundle->floodable
4567 && !bundle->mirror_out) {
4568 output_normal(ctx, bundle, vlan);
4569 dst_mirrors |= bundle->dst_mirrors;
4572 ctx->nf_output_iface = NF_OUT_FLOOD;
4573 } else if (out_bundle) {
4574 output_normal(ctx, out_bundle, vlan);
4575 dst_mirrors = out_bundle->dst_mirrors;
4582 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4584 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4587 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4588 * to a VLAN. In general most packets may be mirrored but we want to drop
4589 * protocols that may confuse switches. */
4591 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4593 /* If you change this function's behavior, please update corresponding
4594 * documentation in vswitch.xml at the same time. */
4595 if (dst[0] != 0x01) {
4596 /* All the currently banned MACs happen to start with 01 currently, so
4597 * this is a quick way to eliminate most of the good ones. */
4599 if (eth_addr_is_reserved(dst)) {
4600 /* Drop STP, IEEE pause frames, and other reserved protocols
4601 * (01-80-c2-00-00-0x). */
4605 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4607 if ((dst[3] & 0xfe) == 0xcc &&
4608 (dst[4] & 0xfe) == 0xcc &&
4609 (dst[5] & 0xfe) == 0xcc) {
4610 /* Drop the following protocols plus others following the same
4613 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4614 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4615 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4619 if (!(dst[3] | dst[4] | dst[5])) {
4620 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4629 output_mirrors(struct action_xlate_ctx *ctx,
4630 uint16_t vlan, const struct ofbundle *in_bundle,
4631 mirror_mask_t dst_mirrors)
4633 struct ofproto_dpif *ofproto = ctx->ofproto;
4634 mirror_mask_t mirrors;
4636 mirrors = in_bundle->src_mirrors | dst_mirrors;
4644 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
4646 if (!vlan_is_mirrored(m, vlan)) {
4647 mirrors &= mirrors - 1;
4651 mirrors &= ~m->dup_mirrors;
4653 output_normal(ctx, m->out, vlan);
4654 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)
4655 && vlan != m->out_vlan) {
4656 struct ofbundle *bundle;
4658 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4659 if (ofbundle_includes_vlan(bundle, m->out_vlan)
4660 && !bundle->mirror_out) {
4661 output_normal(ctx, bundle, m->out_vlan);
4668 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
4669 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
4670 * indicate this; newer upstream kernels use gratuitous ARP requests. */
4672 is_gratuitous_arp(const struct flow *flow)
4674 return (flow->dl_type == htons(ETH_TYPE_ARP)
4675 && eth_addr_is_broadcast(flow->dl_dst)
4676 && (flow->nw_proto == ARP_OP_REPLY
4677 || (flow->nw_proto == ARP_OP_REQUEST
4678 && flow->nw_src == flow->nw_dst)));
4682 update_learning_table(struct ofproto_dpif *ofproto,
4683 const struct flow *flow, int vlan,
4684 struct ofbundle *in_bundle)
4686 struct mac_entry *mac;
4688 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
4692 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
4693 if (is_gratuitous_arp(flow)) {
4694 /* We don't want to learn from gratuitous ARP packets that are
4695 * reflected back over bond slaves so we lock the learning table. */
4696 if (!in_bundle->bond) {
4697 mac_entry_set_grat_arp_lock(mac);
4698 } else if (mac_entry_is_grat_arp_locked(mac)) {
4703 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
4704 /* The log messages here could actually be useful in debugging,
4705 * so keep the rate limit relatively high. */
4706 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4707 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
4708 "on port %s in VLAN %d",
4709 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
4710 in_bundle->name, vlan);
4712 mac->port.p = in_bundle;
4713 tag_set_add(&ofproto->revalidate_set,
4714 mac_learning_changed(ofproto->ml, mac));
4718 static struct ofport_dpif *
4719 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
4721 struct ofport_dpif *ofport;
4723 /* Find the port and bundle for the received packet. */
4724 ofport = get_ofp_port(ofproto, in_port);
4725 if (ofport && ofport->bundle) {
4729 /* Odd. A few possible reasons here:
4731 * - We deleted a port but there are still a few packets queued up
4734 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
4735 * we don't know about.
4737 * - The ofproto client didn't configure the port as part of a bundle.
4740 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4742 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
4743 "port %"PRIu16, ofproto->up.name, in_port);
4748 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
4749 * dropped. Returns true if they may be forwarded, false if they should be
4752 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
4753 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
4755 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
4756 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
4757 * checked by input_vid_is_valid().
4759 * May also add tags to '*tags', although the current implementation only does
4760 * so in one special case.
4763 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
4764 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
4766 struct ofbundle *in_bundle = in_port->bundle;
4768 /* Drop frames for reserved multicast addresses
4769 * only if forward_bpdu option is absent. */
4770 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
4774 if (in_bundle->bond) {
4775 struct mac_entry *mac;
4777 switch (bond_check_admissibility(in_bundle->bond, in_port,
4778 flow->dl_dst, tags)) {
4785 case BV_DROP_IF_MOVED:
4786 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
4787 if (mac && mac->port.p != in_bundle &&
4788 (!is_gratuitous_arp(flow)
4789 || mac_entry_is_grat_arp_locked(mac))) {
4800 xlate_normal(struct action_xlate_ctx *ctx)
4802 mirror_mask_t dst_mirrors = 0;
4803 struct ofport_dpif *in_port;
4804 struct ofbundle *in_bundle;
4805 struct ofbundle *out_bundle;
4806 struct mac_entry *mac;
4810 ctx->has_normal = true;
4812 /* Obtain in_port from ctx->flow.in_port.
4814 * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
4815 in_port = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
4816 ctx->packet != NULL);
4820 in_bundle = in_port->bundle;
4822 /* Drop malformed frames. */
4823 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
4824 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
4825 if (ctx->packet != NULL) {
4826 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4827 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
4828 "VLAN tag received on port %s",
4829 ctx->ofproto->up.name, in_bundle->name);
4834 /* Drop frames on bundles reserved for mirroring. */
4835 if (in_bundle->mirror_out) {
4836 if (ctx->packet != NULL) {
4837 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4838 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4839 "%s, which is reserved exclusively for mirroring",
4840 ctx->ofproto->up.name, in_bundle->name);
4846 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4847 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
4850 vlan = input_vid_to_vlan(in_bundle, vid);
4852 /* Check other admissibility requirements. */
4853 if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
4854 output_mirrors(ctx, vlan, in_bundle, 0);
4858 /* Learn source MAC. */
4859 if (ctx->may_learn) {
4860 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
4863 /* Determine output bundle. */
4864 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
4867 out_bundle = mac->port.p;
4868 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
4869 /* If we are revalidating but don't have a learning entry then eject
4870 * the flow. Installing a flow that floods packets opens up a window
4871 * of time where we could learn from a packet reflected on a bond and
4872 * blackhole packets before the learning table is updated to reflect
4873 * the correct port. */
4874 ctx->may_set_up_flow = false;
4877 out_bundle = OFBUNDLE_FLOOD;
4880 /* Don't send packets out their input bundles. */
4881 if (in_bundle != out_bundle) {
4882 dst_mirrors = compose_dsts(ctx, vlan, in_bundle, out_bundle);
4884 output_mirrors(ctx, vlan, in_bundle, dst_mirrors);
4887 /* Optimized flow revalidation.
4889 * It's a difficult problem, in general, to tell which facets need to have
4890 * their actions recalculated whenever the OpenFlow flow table changes. We
4891 * don't try to solve that general problem: for most kinds of OpenFlow flow
4892 * table changes, we recalculate the actions for every facet. This is
4893 * relatively expensive, but it's good enough if the OpenFlow flow table
4894 * doesn't change very often.
4896 * However, we can expect one particular kind of OpenFlow flow table change to
4897 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
4898 * of CPU on revalidating every facet whenever MAC learning modifies the flow
4899 * table, we add a special case that applies to flow tables in which every rule
4900 * has the same form (that is, the same wildcards), except that the table is
4901 * also allowed to have a single "catch-all" flow that matches all packets. We
4902 * optimize this case by tagging all of the facets that resubmit into the table
4903 * and invalidating the same tag whenever a flow changes in that table. The
4904 * end result is that we revalidate just the facets that need it (and sometimes
4905 * a few more, but not all of the facets or even all of the facets that
4906 * resubmit to the table modified by MAC learning). */
4908 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
4909 * into an OpenFlow table with the given 'basis'. */
4911 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
4914 if (flow_wildcards_is_catchall(wc)) {
4917 struct flow tag_flow = *flow;
4918 flow_zero_wildcards(&tag_flow, wc);
4919 return tag_create_deterministic(flow_hash(&tag_flow, secret));
4923 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
4924 * taggability of that table.
4926 * This function must be called after *each* change to a flow table. If you
4927 * skip calling it on some changes then the pointer comparisons at the end can
4928 * be invalid if you get unlucky. For example, if a flow removal causes a
4929 * cls_table to be destroyed and then a flow insertion causes a cls_table with
4930 * different wildcards to be created with the same address, then this function
4931 * will incorrectly skip revalidation. */
4933 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
4935 struct table_dpif *table = &ofproto->tables[table_id];
4936 const struct classifier *cls = &ofproto->up.tables[table_id];
4937 struct cls_table *catchall, *other;
4938 struct cls_table *t;
4940 catchall = other = NULL;
4942 switch (hmap_count(&cls->tables)) {
4944 /* We could tag this OpenFlow table but it would make the logic a
4945 * little harder and it's a corner case that doesn't seem worth it
4951 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
4952 if (cls_table_is_catchall(t)) {
4954 } else if (!other) {
4957 /* Indicate that we can't tag this by setting both tables to
4958 * NULL. (We know that 'catchall' is already NULL.) */
4965 /* Can't tag this table. */
4969 if (table->catchall_table != catchall || table->other_table != other) {
4970 table->catchall_table = catchall;
4971 table->other_table = other;
4972 ofproto->need_revalidate = true;
4976 /* Given 'rule' that has changed in some way (either it is a rule being
4977 * inserted, a rule being deleted, or a rule whose actions are being
4978 * modified), marks facets for revalidation to ensure that packets will be
4979 * forwarded correctly according to the new state of the flow table.
4981 * This function must be called after *each* change to a flow table. See
4982 * the comment on table_update_taggable() for more information. */
4984 rule_invalidate(const struct rule_dpif *rule)
4986 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4988 table_update_taggable(ofproto, rule->up.table_id);
4990 if (!ofproto->need_revalidate) {
4991 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
4993 if (table->other_table && rule->tag) {
4994 tag_set_add(&ofproto->revalidate_set, rule->tag);
4996 ofproto->need_revalidate = true;
5002 set_frag_handling(struct ofproto *ofproto_,
5003 enum ofp_config_flags frag_handling)
5005 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5007 if (frag_handling != OFPC_FRAG_REASM) {
5008 ofproto->need_revalidate = true;
5016 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5017 const struct flow *flow,
5018 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5020 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5023 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5024 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5027 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5028 ofproto->max_ports);
5030 struct odputil_keybuf keybuf;
5031 struct action_xlate_ctx ctx;
5032 struct ofpbuf *odp_actions;
5035 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5036 odp_flow_key_from_flow(&key, flow);
5038 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
5039 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5040 dpif_execute(ofproto->dpif, key.data, key.size,
5041 odp_actions->data, odp_actions->size, packet);
5042 ofpbuf_delete(odp_actions);
5048 get_netflow_ids(const struct ofproto *ofproto_,
5049 uint8_t *engine_type, uint8_t *engine_id)
5051 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5053 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5056 static struct ofproto_dpif *
5057 ofproto_dpif_lookup(const char *name)
5059 struct ofproto *ofproto = ofproto_lookup(name);
5060 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
5061 ? ofproto_dpif_cast(ofproto)
5066 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5067 const char *args, void *aux OVS_UNUSED)
5069 const struct ofproto_dpif *ofproto;
5071 ofproto = ofproto_dpif_lookup(args);
5073 unixctl_command_reply(conn, 501, "no such bridge");
5076 mac_learning_flush(ofproto->ml);
5078 unixctl_command_reply(conn, 200, "table successfully flushed");
5082 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5083 const char *args, void *aux OVS_UNUSED)
5085 struct ds ds = DS_EMPTY_INITIALIZER;
5086 const struct ofproto_dpif *ofproto;
5087 const struct mac_entry *e;
5089 ofproto = ofproto_dpif_lookup(args);
5091 unixctl_command_reply(conn, 501, "no such bridge");
5095 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5096 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5097 struct ofbundle *bundle = e->port.p;
5098 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5099 ofbundle_get_a_port(bundle)->odp_port,
5100 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5102 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5106 struct ofproto_trace {
5107 struct action_xlate_ctx ctx;
5113 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5114 const struct rule_dpif *rule)
5116 ds_put_char_multiple(result, '\t', level);
5118 ds_put_cstr(result, "No match\n");
5122 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5123 table_id, ntohll(rule->up.flow_cookie));
5124 cls_rule_format(&rule->up.cr, result);
5125 ds_put_char(result, '\n');
5127 ds_put_char_multiple(result, '\t', level);
5128 ds_put_cstr(result, "OpenFlow ");
5129 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5130 ds_put_char(result, '\n');
5134 trace_format_flow(struct ds *result, int level, const char *title,
5135 struct ofproto_trace *trace)
5137 ds_put_char_multiple(result, '\t', level);
5138 ds_put_format(result, "%s: ", title);
5139 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5140 ds_put_cstr(result, "unchanged");
5142 flow_format(result, &trace->ctx.flow);
5143 trace->flow = trace->ctx.flow;
5145 ds_put_char(result, '\n');
5149 trace_format_regs(struct ds *result, int level, const char *title,
5150 struct ofproto_trace *trace)
5154 ds_put_char_multiple(result, '\t', level);
5155 ds_put_format(result, "%s:", title);
5156 for (i = 0; i < FLOW_N_REGS; i++) {
5157 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5159 ds_put_char(result, '\n');
5163 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5165 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5166 struct ds *result = trace->result;
5168 ds_put_char(result, '\n');
5169 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5170 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5171 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5175 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5176 void *aux OVS_UNUSED)
5178 char *dpname, *arg1, *arg2, *arg3, *arg4;
5179 char *args = xstrdup(args_);
5180 char *save_ptr = NULL;
5181 struct ofproto_dpif *ofproto;
5182 struct ofpbuf odp_key;
5183 struct ofpbuf *packet;
5184 struct rule_dpif *rule;
5190 ofpbuf_init(&odp_key, 0);
5193 dpname = strtok_r(args, " ", &save_ptr);
5194 arg1 = strtok_r(NULL, " ", &save_ptr);
5195 arg2 = strtok_r(NULL, " ", &save_ptr);
5196 arg3 = strtok_r(NULL, " ", &save_ptr);
5197 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5198 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5199 /* ofproto/trace dpname flow [-generate] */
5202 /* Convert string to datapath key. */
5203 ofpbuf_init(&odp_key, 0);
5204 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5206 unixctl_command_reply(conn, 501, "Bad flow syntax");
5210 /* Convert odp_key to flow. */
5211 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
5213 unixctl_command_reply(conn, 501, "Invalid flow");
5217 /* Generate a packet, if requested. */
5219 packet = ofpbuf_new(0);
5220 flow_compose(packet, &flow);
5222 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5223 /* ofproto/trace dpname priority tun_id in_port packet */
5228 priority = atoi(arg1);
5229 tun_id = htonll(strtoull(arg2, NULL, 0));
5230 in_port = ofp_port_to_odp_port(atoi(arg3));
5232 packet = ofpbuf_new(strlen(args) / 2);
5233 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5234 arg4 += strspn(arg4, " ");
5235 if (*arg4 != '\0') {
5236 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5239 if (packet->size < ETH_HEADER_LEN) {
5240 unixctl_command_reply(conn, 501,
5241 "Packet data too short for Ethernet");
5245 ds_put_cstr(&result, "Packet: ");
5246 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5247 ds_put_cstr(&result, s);
5250 flow_extract(packet, priority, tun_id, in_port, &flow);
5252 unixctl_command_reply(conn, 501, "Bad command syntax");
5256 ofproto = ofproto_dpif_lookup(dpname);
5258 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5263 ds_put_cstr(&result, "Flow: ");
5264 flow_format(&result, &flow);
5265 ds_put_char(&result, '\n');
5267 rule = rule_dpif_lookup(ofproto, &flow, 0);
5268 trace_format_rule(&result, 0, 0, rule);
5270 struct ofproto_trace trace;
5271 struct ofpbuf *odp_actions;
5273 trace.result = &result;
5275 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
5276 trace.ctx.resubmit_hook = trace_resubmit;
5277 odp_actions = xlate_actions(&trace.ctx,
5278 rule->up.actions, rule->up.n_actions);
5280 ds_put_char(&result, '\n');
5281 trace_format_flow(&result, 0, "Final flow", &trace);
5282 ds_put_cstr(&result, "Datapath actions: ");
5283 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5284 ofpbuf_delete(odp_actions);
5286 if (!trace.ctx.may_set_up_flow) {
5288 ds_put_cstr(&result, "\nThis flow is not cachable.");
5290 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5291 "for complete actions, please supply a packet.");
5296 unixctl_command_reply(conn, 200, ds_cstr(&result));
5299 ds_destroy(&result);
5300 ofpbuf_delete(packet);
5301 ofpbuf_uninit(&odp_key);
5306 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5307 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5310 unixctl_command_reply(conn, 200, NULL);
5314 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5315 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5318 unixctl_command_reply(conn, 200, NULL);
5322 ofproto_dpif_unixctl_init(void)
5324 static bool registered;
5330 unixctl_command_register("ofproto/trace",
5331 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5332 ofproto_unixctl_trace, NULL);
5333 unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
5335 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5337 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5338 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5341 const struct ofproto_class ofproto_dpif_class = {
5368 port_is_lacp_current,
5369 NULL, /* rule_choose_table */
5376 rule_modify_actions,
5384 get_cfm_remote_mpids,
5388 get_stp_port_status,
5393 is_mirror_output_bundle,
5394 forward_bpdu_changed,