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 || !stp_forward_in_state(port->stp_state)) {
1235 bundle->floodable = false;
1242 bundle_del_port(struct ofport_dpif *port)
1244 struct ofbundle *bundle = port->bundle;
1246 bundle->ofproto->need_revalidate = true;
1248 list_remove(&port->bundle_node);
1249 port->bundle = NULL;
1252 lacp_slave_unregister(bundle->lacp, port);
1255 bond_slave_unregister(bundle->bond, port);
1258 bundle_update(bundle);
1262 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1263 struct lacp_slave_settings *lacp,
1264 uint32_t bond_stable_id)
1266 struct ofport_dpif *port;
1268 port = get_ofp_port(bundle->ofproto, ofp_port);
1273 if (port->bundle != bundle) {
1274 bundle->ofproto->need_revalidate = true;
1276 bundle_del_port(port);
1279 port->bundle = bundle;
1280 list_push_back(&bundle->ports, &port->bundle_node);
1281 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1282 || !stp_forward_in_state(port->stp_state)) {
1283 bundle->floodable = false;
1287 port->bundle->ofproto->need_revalidate = true;
1288 lacp_slave_register(bundle->lacp, port, lacp);
1291 port->bond_stable_id = bond_stable_id;
1297 bundle_destroy(struct ofbundle *bundle)
1299 struct ofproto_dpif *ofproto;
1300 struct ofport_dpif *port, *next_port;
1307 ofproto = bundle->ofproto;
1308 for (i = 0; i < MAX_MIRRORS; i++) {
1309 struct ofmirror *m = ofproto->mirrors[i];
1311 if (m->out == bundle) {
1313 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1314 || hmapx_find_and_delete(&m->dsts, bundle)) {
1315 ofproto->need_revalidate = true;
1320 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1321 bundle_del_port(port);
1324 bundle_flush_macs(bundle);
1325 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1327 free(bundle->trunks);
1328 lacp_destroy(bundle->lacp);
1329 bond_destroy(bundle->bond);
1334 bundle_set(struct ofproto *ofproto_, void *aux,
1335 const struct ofproto_bundle_settings *s)
1337 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1338 bool need_flush = false;
1339 struct ofport_dpif *port;
1340 struct ofbundle *bundle;
1341 unsigned long *trunks;
1347 bundle_destroy(bundle_lookup(ofproto, aux));
1351 assert(s->n_slaves == 1 || s->bond != NULL);
1352 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1354 bundle = bundle_lookup(ofproto, aux);
1356 bundle = xmalloc(sizeof *bundle);
1358 bundle->ofproto = ofproto;
1359 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1360 hash_pointer(aux, 0));
1362 bundle->name = NULL;
1364 list_init(&bundle->ports);
1365 bundle->vlan_mode = PORT_VLAN_TRUNK;
1367 bundle->trunks = NULL;
1368 bundle->use_priority_tags = s->use_priority_tags;
1369 bundle->lacp = NULL;
1370 bundle->bond = NULL;
1372 bundle->floodable = true;
1374 bundle->src_mirrors = 0;
1375 bundle->dst_mirrors = 0;
1376 bundle->mirror_out = 0;
1379 if (!bundle->name || strcmp(s->name, bundle->name)) {
1381 bundle->name = xstrdup(s->name);
1386 if (!bundle->lacp) {
1387 ofproto->need_revalidate = true;
1388 bundle->lacp = lacp_create();
1390 lacp_configure(bundle->lacp, s->lacp);
1392 lacp_destroy(bundle->lacp);
1393 bundle->lacp = NULL;
1396 /* Update set of ports. */
1398 for (i = 0; i < s->n_slaves; i++) {
1399 if (!bundle_add_port(bundle, s->slaves[i],
1400 s->lacp ? &s->lacp_slaves[i] : NULL,
1401 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1405 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1406 struct ofport_dpif *next_port;
1408 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1409 for (i = 0; i < s->n_slaves; i++) {
1410 if (s->slaves[i] == port->up.ofp_port) {
1415 bundle_del_port(port);
1419 assert(list_size(&bundle->ports) <= s->n_slaves);
1421 if (list_is_empty(&bundle->ports)) {
1422 bundle_destroy(bundle);
1426 /* Set VLAN tagging mode */
1427 if (s->vlan_mode != bundle->vlan_mode
1428 || s->use_priority_tags != bundle->use_priority_tags) {
1429 bundle->vlan_mode = s->vlan_mode;
1430 bundle->use_priority_tags = s->use_priority_tags;
1435 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1436 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1438 if (vlan != bundle->vlan) {
1439 bundle->vlan = vlan;
1443 /* Get trunked VLANs. */
1444 switch (s->vlan_mode) {
1445 case PORT_VLAN_ACCESS:
1449 case PORT_VLAN_TRUNK:
1450 trunks = (unsigned long *) s->trunks;
1453 case PORT_VLAN_NATIVE_UNTAGGED:
1454 case PORT_VLAN_NATIVE_TAGGED:
1455 if (vlan != 0 && (!s->trunks
1456 || !bitmap_is_set(s->trunks, vlan)
1457 || bitmap_is_set(s->trunks, 0))) {
1458 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1460 trunks = bitmap_clone(s->trunks, 4096);
1462 trunks = bitmap_allocate1(4096);
1464 bitmap_set1(trunks, vlan);
1465 bitmap_set0(trunks, 0);
1467 trunks = (unsigned long *) s->trunks;
1474 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1475 free(bundle->trunks);
1476 if (trunks == s->trunks) {
1477 bundle->trunks = vlan_bitmap_clone(trunks);
1479 bundle->trunks = trunks;
1484 if (trunks != s->trunks) {
1489 if (!list_is_short(&bundle->ports)) {
1490 bundle->ofproto->has_bonded_bundles = true;
1492 if (bond_reconfigure(bundle->bond, s->bond)) {
1493 ofproto->need_revalidate = true;
1496 bundle->bond = bond_create(s->bond);
1497 ofproto->need_revalidate = true;
1500 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1501 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1505 bond_destroy(bundle->bond);
1506 bundle->bond = NULL;
1509 /* If we changed something that would affect MAC learning, un-learn
1510 * everything on this port and force flow revalidation. */
1512 bundle_flush_macs(bundle);
1519 bundle_remove(struct ofport *port_)
1521 struct ofport_dpif *port = ofport_dpif_cast(port_);
1522 struct ofbundle *bundle = port->bundle;
1525 bundle_del_port(port);
1526 if (list_is_empty(&bundle->ports)) {
1527 bundle_destroy(bundle);
1528 } else if (list_is_short(&bundle->ports)) {
1529 bond_destroy(bundle->bond);
1530 bundle->bond = NULL;
1536 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1538 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1539 struct ofport_dpif *port = port_;
1540 uint8_t ea[ETH_ADDR_LEN];
1543 error = netdev_get_etheraddr(port->up.netdev, ea);
1545 struct ofpbuf packet;
1548 ofpbuf_init(&packet, 0);
1549 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1551 memcpy(packet_pdu, pdu, pdu_size);
1553 send_packet(ofproto_dpif_cast(port->up.ofproto), port->odp_port,
1555 ofpbuf_uninit(&packet);
1557 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1558 "%s (%s)", port->bundle->name,
1559 netdev_get_name(port->up.netdev), strerror(error));
1564 bundle_send_learning_packets(struct ofbundle *bundle)
1566 struct ofproto_dpif *ofproto = bundle->ofproto;
1567 int error, n_packets, n_errors;
1568 struct mac_entry *e;
1570 error = n_packets = n_errors = 0;
1571 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1572 if (e->port.p != bundle) {
1573 struct ofpbuf *learning_packet;
1574 struct ofport_dpif *port;
1577 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1580 ret = send_packet(ofproto_dpif_cast(port->up.ofproto),
1581 port->odp_port, learning_packet);
1582 ofpbuf_delete(learning_packet);
1592 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1593 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1594 "packets, last error was: %s",
1595 bundle->name, n_errors, n_packets, strerror(error));
1597 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1598 bundle->name, n_packets);
1603 bundle_run(struct ofbundle *bundle)
1606 lacp_run(bundle->lacp, send_pdu_cb);
1609 struct ofport_dpif *port;
1611 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1612 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1615 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1616 lacp_negotiated(bundle->lacp));
1617 if (bond_should_send_learning_packets(bundle->bond)) {
1618 bundle_send_learning_packets(bundle);
1624 bundle_wait(struct ofbundle *bundle)
1627 lacp_wait(bundle->lacp);
1630 bond_wait(bundle->bond);
1637 mirror_scan(struct ofproto_dpif *ofproto)
1641 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1642 if (!ofproto->mirrors[idx]) {
1649 static struct ofmirror *
1650 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1654 for (i = 0; i < MAX_MIRRORS; i++) {
1655 struct ofmirror *mirror = ofproto->mirrors[i];
1656 if (mirror && mirror->aux == aux) {
1664 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
1666 mirror_update_dups(struct ofproto_dpif *ofproto)
1670 for (i = 0; i < MAX_MIRRORS; i++) {
1671 struct ofmirror *m = ofproto->mirrors[i];
1674 m->dup_mirrors = MIRROR_MASK_C(1) << i;
1678 for (i = 0; i < MAX_MIRRORS; i++) {
1679 struct ofmirror *m1 = ofproto->mirrors[i];
1686 for (j = i + 1; j < MAX_MIRRORS; j++) {
1687 struct ofmirror *m2 = ofproto->mirrors[j];
1689 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
1690 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
1691 m2->dup_mirrors |= m1->dup_mirrors;
1698 mirror_set(struct ofproto *ofproto_, void *aux,
1699 const struct ofproto_mirror_settings *s)
1701 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1702 mirror_mask_t mirror_bit;
1703 struct ofbundle *bundle;
1704 struct ofmirror *mirror;
1705 struct ofbundle *out;
1706 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1707 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1710 mirror = mirror_lookup(ofproto, aux);
1712 mirror_destroy(mirror);
1718 idx = mirror_scan(ofproto);
1720 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1722 ofproto->up.name, MAX_MIRRORS, s->name);
1726 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1727 mirror->ofproto = ofproto;
1730 mirror->out_vlan = -1;
1731 mirror->name = NULL;
1734 if (!mirror->name || strcmp(s->name, mirror->name)) {
1736 mirror->name = xstrdup(s->name);
1739 /* Get the new configuration. */
1740 if (s->out_bundle) {
1741 out = bundle_lookup(ofproto, s->out_bundle);
1743 mirror_destroy(mirror);
1749 out_vlan = s->out_vlan;
1751 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1752 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1754 /* If the configuration has not changed, do nothing. */
1755 if (hmapx_equals(&srcs, &mirror->srcs)
1756 && hmapx_equals(&dsts, &mirror->dsts)
1757 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1758 && mirror->out == out
1759 && mirror->out_vlan == out_vlan)
1761 hmapx_destroy(&srcs);
1762 hmapx_destroy(&dsts);
1766 hmapx_swap(&srcs, &mirror->srcs);
1767 hmapx_destroy(&srcs);
1769 hmapx_swap(&dsts, &mirror->dsts);
1770 hmapx_destroy(&dsts);
1772 free(mirror->vlans);
1773 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1776 mirror->out_vlan = out_vlan;
1778 /* Update bundles. */
1779 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1780 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1781 if (hmapx_contains(&mirror->srcs, bundle)) {
1782 bundle->src_mirrors |= mirror_bit;
1784 bundle->src_mirrors &= ~mirror_bit;
1787 if (hmapx_contains(&mirror->dsts, bundle)) {
1788 bundle->dst_mirrors |= mirror_bit;
1790 bundle->dst_mirrors &= ~mirror_bit;
1793 if (mirror->out == bundle) {
1794 bundle->mirror_out |= mirror_bit;
1796 bundle->mirror_out &= ~mirror_bit;
1800 ofproto->need_revalidate = true;
1801 mac_learning_flush(ofproto->ml);
1802 mirror_update_dups(ofproto);
1808 mirror_destroy(struct ofmirror *mirror)
1810 struct ofproto_dpif *ofproto;
1811 mirror_mask_t mirror_bit;
1812 struct ofbundle *bundle;
1818 ofproto = mirror->ofproto;
1819 ofproto->need_revalidate = true;
1820 mac_learning_flush(ofproto->ml);
1822 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1823 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1824 bundle->src_mirrors &= ~mirror_bit;
1825 bundle->dst_mirrors &= ~mirror_bit;
1826 bundle->mirror_out &= ~mirror_bit;
1829 hmapx_destroy(&mirror->srcs);
1830 hmapx_destroy(&mirror->dsts);
1831 free(mirror->vlans);
1833 ofproto->mirrors[mirror->idx] = NULL;
1837 mirror_update_dups(ofproto);
1841 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1843 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1844 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1845 ofproto->need_revalidate = true;
1846 mac_learning_flush(ofproto->ml);
1852 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
1854 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1855 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1856 return bundle && bundle->mirror_out != 0;
1860 forward_bpdu_changed(struct ofproto *ofproto_)
1862 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1863 /* Revalidate cached flows whenever forward_bpdu option changes. */
1864 ofproto->need_revalidate = true;
1869 static struct ofport_dpif *
1870 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1872 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1873 return ofport ? ofport_dpif_cast(ofport) : NULL;
1876 static struct ofport_dpif *
1877 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1879 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1883 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1884 struct dpif_port *dpif_port)
1886 ofproto_port->name = dpif_port->name;
1887 ofproto_port->type = dpif_port->type;
1888 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1892 port_run(struct ofport_dpif *ofport)
1894 bool enable = netdev_get_carrier(ofport->up.netdev);
1897 cfm_run(ofport->cfm);
1899 if (cfm_should_send_ccm(ofport->cfm)) {
1900 struct ofpbuf packet;
1902 ofpbuf_init(&packet, 0);
1903 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1904 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1905 ofport->odp_port, &packet);
1906 ofpbuf_uninit(&packet);
1909 enable = enable && !cfm_get_fault(ofport->cfm)
1910 && cfm_get_opup(ofport->cfm);
1913 if (ofport->bundle) {
1914 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1917 if (ofport->may_enable != enable) {
1918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1920 if (ofproto->has_bundle_action) {
1921 ofproto->need_revalidate = true;
1925 ofport->may_enable = enable;
1929 port_wait(struct ofport_dpif *ofport)
1932 cfm_wait(ofport->cfm);
1937 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1938 struct ofproto_port *ofproto_port)
1940 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1941 struct dpif_port dpif_port;
1944 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1946 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1952 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1954 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1958 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1960 *ofp_portp = odp_port_to_ofp_port(odp_port);
1966 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1971 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1973 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1975 /* The caller is going to close ofport->up.netdev. If this is a
1976 * bonded port, then the bond is using that netdev, so remove it
1977 * from the bond. The client will need to reconfigure everything
1978 * after deleting ports, so then the slave will get re-added. */
1979 bundle_remove(&ofport->up);
1985 struct port_dump_state {
1986 struct dpif_port_dump dump;
1991 port_dump_start(const struct ofproto *ofproto_, void **statep)
1993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1994 struct port_dump_state *state;
1996 *statep = state = xmalloc(sizeof *state);
1997 dpif_port_dump_start(&state->dump, ofproto->dpif);
1998 state->done = false;
2003 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
2004 struct ofproto_port *port)
2006 struct port_dump_state *state = state_;
2007 struct dpif_port dpif_port;
2009 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
2010 ofproto_port_from_dpif_port(port, &dpif_port);
2013 int error = dpif_port_dump_done(&state->dump);
2015 return error ? error : EOF;
2020 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2022 struct port_dump_state *state = state_;
2025 dpif_port_dump_done(&state->dump);
2032 port_poll(const struct ofproto *ofproto_, char **devnamep)
2034 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2035 return dpif_port_poll(ofproto->dpif, devnamep);
2039 port_poll_wait(const struct ofproto *ofproto_)
2041 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2042 dpif_port_poll_wait(ofproto->dpif);
2046 port_is_lacp_current(const struct ofport *ofport_)
2048 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2049 return (ofport->bundle && ofport->bundle->lacp
2050 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2054 /* Upcall handling. */
2056 /* Flow miss batching.
2058 * Some dpifs implement operations faster when you hand them off in a batch.
2059 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2060 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2061 * more packets, plus possibly installing the flow in the dpif.
2063 * So far we only batch the operations that affect flow setup time the most.
2064 * It's possible to batch more than that, but the benefit might be minimal. */
2066 struct hmap_node hmap_node;
2068 const struct nlattr *key;
2070 struct list packets;
2073 struct flow_miss_op {
2074 union dpif_op dpif_op;
2075 struct facet *facet;
2078 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2079 * OpenFlow controller as necessary according to their individual
2082 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2083 * ownership is transferred to this function. */
2085 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2086 const struct flow *flow, bool clone)
2088 struct ofputil_packet_in pin;
2090 pin.packet = packet;
2091 pin.in_port = flow->in_port;
2092 pin.reason = OFPR_NO_MATCH;
2093 pin.buffer_id = 0; /* not yet known */
2094 pin.send_len = 0; /* not used for flow table misses */
2095 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2096 clone ? NULL : packet);
2099 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2100 * OpenFlow controller as necessary according to their individual
2103 * 'send_len' should be the number of bytes of 'packet' to send to the
2104 * controller, as specified in the action that caused the packet to be sent.
2106 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2107 * Otherwise, ownership is transferred to this function. */
2109 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2110 uint64_t userdata, const struct flow *flow, bool clone)
2112 struct ofputil_packet_in pin;
2113 struct user_action_cookie cookie;
2115 memcpy(&cookie, &userdata, sizeof(cookie));
2117 pin.packet = packet;
2118 pin.in_port = flow->in_port;
2119 pin.reason = OFPR_ACTION;
2120 pin.buffer_id = 0; /* not yet known */
2121 pin.send_len = cookie.data;
2122 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2123 clone ? NULL : packet);
2127 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2128 const struct ofpbuf *packet)
2130 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2136 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2138 cfm_process_heartbeat(ofport->cfm, packet);
2141 } else if (ofport->bundle && ofport->bundle->lacp
2142 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2144 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2147 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2149 stp_process_packet(ofport, packet);
2156 static struct flow_miss *
2157 flow_miss_create(struct hmap *todo, const struct flow *flow,
2158 const struct nlattr *key, size_t key_len)
2160 uint32_t hash = flow_hash(flow, 0);
2161 struct flow_miss *miss;
2163 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2164 if (flow_equal(&miss->flow, flow)) {
2169 miss = xmalloc(sizeof *miss);
2170 hmap_insert(todo, &miss->hmap_node, hash);
2173 miss->key_len = key_len;
2174 list_init(&miss->packets);
2179 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2180 struct flow_miss_op *ops, size_t *n_ops)
2182 const struct flow *flow = &miss->flow;
2183 struct ofpbuf *packet, *next_packet;
2184 struct facet *facet;
2186 facet = facet_lookup_valid(ofproto, flow);
2188 struct rule_dpif *rule;
2190 rule = rule_dpif_lookup(ofproto, flow, 0);
2192 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2193 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2195 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2196 COVERAGE_INC(ofproto_dpif_no_packet_in);
2197 /* XXX install 'drop' flow entry */
2201 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2205 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2207 list_remove(&packet->list_node);
2208 send_packet_in_miss(ofproto, packet, flow, false);
2214 facet = facet_create(rule, flow);
2217 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2218 list_remove(&packet->list_node);
2219 ofproto->n_matches++;
2221 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2223 * Extra-special case for fail-open mode.
2225 * We are in fail-open mode and the packet matched the fail-open
2226 * rule, but we are connected to a controller too. We should send
2227 * the packet up to the controller in the hope that it will try to
2228 * set up a flow and thereby allow us to exit fail-open.
2230 * See the top-level comment in fail-open.c for more information.
2232 send_packet_in_miss(ofproto, packet, flow, true);
2235 if (!facet->may_install) {
2236 facet_make_actions(ofproto, facet, packet);
2238 if (!execute_controller_action(ofproto, &facet->flow,
2239 facet->actions, facet->actions_len,
2241 struct flow_miss_op *op = &ops[(*n_ops)++];
2242 struct dpif_execute *execute = &op->dpif_op.execute;
2245 execute->type = DPIF_OP_EXECUTE;
2246 execute->key = miss->key;
2247 execute->key_len = miss->key_len;
2249 = (facet->may_install
2251 : xmemdup(facet->actions, facet->actions_len));
2252 execute->actions_len = facet->actions_len;
2253 execute->packet = packet;
2257 if (facet->may_install) {
2258 struct flow_miss_op *op = &ops[(*n_ops)++];
2259 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2262 put->type = DPIF_OP_FLOW_PUT;
2263 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2264 put->key = miss->key;
2265 put->key_len = miss->key_len;
2266 put->actions = facet->actions;
2267 put->actions_len = facet->actions_len;
2273 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2276 struct dpif_upcall *upcall;
2277 struct flow_miss *miss, *next_miss;
2278 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2279 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2288 /* Construct the to-do list.
2290 * This just amounts to extracting the flow from each packet and sticking
2291 * the packets that have the same flow in the same "flow_miss" structure so
2292 * that we can process them together. */
2294 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2295 struct flow_miss *miss;
2298 /* Obtain in_port and tun_id, at least, then set 'flow''s header
2300 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2301 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2302 flow.in_port, &flow);
2304 /* Handle 802.1ag, LACP, and STP specially. */
2305 if (process_special(ofproto, &flow, upcall->packet)) {
2306 ofpbuf_delete(upcall->packet);
2307 ofproto->n_matches++;
2311 /* Add other packets to a to-do list. */
2312 miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len);
2313 list_push_back(&miss->packets, &upcall->packet->list_node);
2316 /* Process each element in the to-do list, constructing the set of
2317 * operations to batch. */
2319 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2320 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2321 ofpbuf_list_delete(&miss->packets);
2322 hmap_remove(&todo, &miss->hmap_node);
2325 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2326 hmap_destroy(&todo);
2328 /* Execute batch. */
2329 for (i = 0; i < n_ops; i++) {
2330 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2332 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2334 /* Free memory and update facets. */
2335 for (i = 0; i < n_ops; i++) {
2336 struct flow_miss_op *op = &flow_miss_ops[i];
2337 struct dpif_execute *execute;
2338 struct dpif_flow_put *put;
2340 switch (op->dpif_op.type) {
2341 case DPIF_OP_EXECUTE:
2342 execute = &op->dpif_op.execute;
2343 if (op->facet->actions != execute->actions) {
2344 free((struct nlattr *) execute->actions);
2346 ofpbuf_delete((struct ofpbuf *) execute->packet);
2349 case DPIF_OP_FLOW_PUT:
2350 put = &op->dpif_op.flow_put;
2352 op->facet->installed = true;
2360 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2361 struct dpif_upcall *upcall)
2364 struct user_action_cookie cookie;
2366 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2368 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2369 if (ofproto->sflow) {
2370 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2371 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
2373 ofpbuf_delete(upcall->packet);
2375 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2376 COVERAGE_INC(ofproto_dpif_ctlr_action);
2377 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2378 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2381 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2386 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
2388 switch (upcall->type) {
2389 case DPIF_UC_ACTION:
2390 handle_userspace_upcall(ofproto, upcall);
2394 /* The caller handles these. */
2397 case DPIF_N_UC_TYPES:
2399 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
2404 /* Flow expiration. */
2406 static int facet_max_idle(const struct ofproto_dpif *);
2407 static void update_stats(struct ofproto_dpif *);
2408 static void rule_expire(struct rule_dpif *);
2409 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
2411 /* This function is called periodically by run(). Its job is to collect
2412 * updates for the flows that have been installed into the datapath, most
2413 * importantly when they last were used, and then use that information to
2414 * expire flows that have not been used recently.
2416 * Returns the number of milliseconds after which it should be called again. */
2418 expire(struct ofproto_dpif *ofproto)
2420 struct rule_dpif *rule, *next_rule;
2421 struct classifier *table;
2424 /* Update stats for each flow in the datapath. */
2425 update_stats(ofproto);
2427 /* Expire facets that have been idle too long. */
2428 dp_max_idle = facet_max_idle(ofproto);
2429 expire_facets(ofproto, dp_max_idle);
2431 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2432 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2433 struct cls_cursor cursor;
2435 cls_cursor_init(&cursor, table, NULL);
2436 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2441 /* All outstanding data in existing flows has been accounted, so it's a
2442 * good time to do bond rebalancing. */
2443 if (ofproto->has_bonded_bundles) {
2444 struct ofbundle *bundle;
2446 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2448 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2453 return MIN(dp_max_idle, 1000);
2456 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2458 * This function also pushes statistics updates to rules which each facet
2459 * resubmits into. Generally these statistics will be accurate. However, if a
2460 * facet changes the rule it resubmits into at some time in between
2461 * update_stats() runs, it is possible that statistics accrued to the
2462 * old rule will be incorrectly attributed to the new rule. This could be
2463 * avoided by calling update_stats() whenever rules are created or
2464 * deleted. However, the performance impact of making so many calls to the
2465 * datapath do not justify the benefit of having perfectly accurate statistics.
2468 update_stats(struct ofproto_dpif *p)
2470 const struct dpif_flow_stats *stats;
2471 struct dpif_flow_dump dump;
2472 const struct nlattr *key;
2475 dpif_flow_dump_start(&dump, p->dpif);
2476 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2477 struct facet *facet;
2480 if (odp_flow_key_to_flow(key, key_len, &flow)) {
2484 odp_flow_key_format(key, key_len, &s);
2485 VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
2491 facet = facet_find(p, &flow);
2493 if (facet && facet->installed) {
2495 if (stats->n_packets >= facet->dp_packet_count) {
2496 uint64_t extra = stats->n_packets - facet->dp_packet_count;
2497 facet->packet_count += extra;
2499 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2502 if (stats->n_bytes >= facet->dp_byte_count) {
2503 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
2505 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2508 facet->dp_packet_count = stats->n_packets;
2509 facet->dp_byte_count = stats->n_bytes;
2511 facet_update_time(p, facet, stats->used);
2512 facet_account(p, facet);
2513 facet_push_stats(facet);
2515 /* There's a flow in the datapath that we know nothing about.
2517 COVERAGE_INC(facet_unexpected);
2518 dpif_flow_del(p->dpif, key, key_len, NULL);
2521 dpif_flow_dump_done(&dump);
2524 /* Calculates and returns the number of milliseconds of idle time after which
2525 * facets should expire from the datapath and we should fold their statistics
2526 * into their parent rules in userspace. */
2528 facet_max_idle(const struct ofproto_dpif *ofproto)
2531 * Idle time histogram.
2533 * Most of the time a switch has a relatively small number of facets. When
2534 * this is the case we might as well keep statistics for all of them in
2535 * userspace and to cache them in the kernel datapath for performance as
2538 * As the number of facets increases, the memory required to maintain
2539 * statistics about them in userspace and in the kernel becomes
2540 * significant. However, with a large number of facets it is likely that
2541 * only a few of them are "heavy hitters" that consume a large amount of
2542 * bandwidth. At this point, only heavy hitters are worth caching in the
2543 * kernel and maintaining in userspaces; other facets we can discard.
2545 * The technique used to compute the idle time is to build a histogram with
2546 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
2547 * that is installed in the kernel gets dropped in the appropriate bucket.
2548 * After the histogram has been built, we compute the cutoff so that only
2549 * the most-recently-used 1% of facets (but at least
2550 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2551 * the most-recently-used bucket of facets is kept, so actually an
2552 * arbitrary number of facets can be kept in any given expiration run
2553 * (though the next run will delete most of those unless they receive
2556 * This requires a second pass through the facets, in addition to the pass
2557 * made by update_stats(), because the former function never looks
2558 * at uninstallable facets.
2560 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2561 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2562 int buckets[N_BUCKETS] = { 0 };
2563 int total, subtotal, bucket;
2564 struct facet *facet;
2568 total = hmap_count(&ofproto->facets);
2569 if (total <= ofproto->up.flow_eviction_threshold) {
2570 return N_BUCKETS * BUCKET_WIDTH;
2573 /* Build histogram. */
2575 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
2576 long long int idle = now - facet->used;
2577 int bucket = (idle <= 0 ? 0
2578 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2579 : (unsigned int) idle / BUCKET_WIDTH);
2583 /* Find the first bucket whose flows should be expired. */
2584 subtotal = bucket = 0;
2586 subtotal += buckets[bucket++];
2587 } while (bucket < N_BUCKETS &&
2588 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2590 if (VLOG_IS_DBG_ENABLED()) {
2594 ds_put_cstr(&s, "keep");
2595 for (i = 0; i < N_BUCKETS; i++) {
2597 ds_put_cstr(&s, ", drop");
2600 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2603 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2607 return bucket * BUCKET_WIDTH;
2611 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
2613 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
2614 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
2615 struct ofexpired expired;
2617 if (facet->installed) {
2618 struct dpif_flow_stats stats;
2620 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
2622 facet_update_stats(ofproto, facet, &stats);
2625 expired.flow = facet->flow;
2626 expired.packet_count = facet->packet_count;
2627 expired.byte_count = facet->byte_count;
2628 expired.used = facet->used;
2629 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2634 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
2636 long long int cutoff = time_msec() - dp_max_idle;
2637 struct facet *facet, *next_facet;
2639 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
2640 facet_active_timeout(ofproto, facet);
2641 if (facet->used < cutoff) {
2642 facet_remove(ofproto, facet);
2647 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2648 * then delete it entirely. */
2650 rule_expire(struct rule_dpif *rule)
2652 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2653 struct facet *facet, *next_facet;
2657 /* Has 'rule' expired? */
2659 if (rule->up.hard_timeout
2660 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
2661 reason = OFPRR_HARD_TIMEOUT;
2662 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2663 && now > rule->used + rule->up.idle_timeout * 1000) {
2664 reason = OFPRR_IDLE_TIMEOUT;
2669 COVERAGE_INC(ofproto_dpif_expired);
2671 /* Update stats. (This is a no-op if the rule expired due to an idle
2672 * timeout, because that only happens when the rule has no facets left.) */
2673 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2674 facet_remove(ofproto, facet);
2677 /* Get rid of the rule. */
2678 ofproto_rule_expire(&rule->up, reason);
2683 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
2685 * The caller must already have determined that no facet with an identical
2686 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2687 * the ofproto's classifier table.
2689 * The facet will initially have no ODP actions. The caller should fix that
2690 * by calling facet_make_actions(). */
2691 static struct facet *
2692 facet_create(struct rule_dpif *rule, const struct flow *flow)
2694 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2695 struct facet *facet;
2697 facet = xzalloc(sizeof *facet);
2698 facet->used = time_msec();
2699 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2700 list_push_back(&rule->facets, &facet->list_node);
2702 facet->flow = *flow;
2703 netflow_flow_init(&facet->nf_flow);
2704 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2710 facet_free(struct facet *facet)
2712 free(facet->actions);
2717 execute_controller_action(struct ofproto_dpif *ofproto,
2718 const struct flow *flow,
2719 const struct nlattr *odp_actions, size_t actions_len,
2720 struct ofpbuf *packet)
2723 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
2724 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
2725 /* As an optimization, avoid a round-trip from userspace to kernel to
2726 * userspace. This also avoids possibly filling up kernel packet
2727 * buffers along the way.
2729 * This optimization will not accidentally catch sFlow
2730 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
2731 * inside OVS_ACTION_ATTR_SAMPLE. */
2732 const struct nlattr *nla;
2734 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
2735 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
2743 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2744 * 'packet', which arrived on 'in_port'.
2746 * Takes ownership of 'packet'. */
2748 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2749 const struct nlattr *odp_actions, size_t actions_len,
2750 struct ofpbuf *packet)
2752 struct odputil_keybuf keybuf;
2756 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
2761 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2762 odp_flow_key_from_flow(&key, flow);
2764 error = dpif_execute(ofproto->dpif, key.data, key.size,
2765 odp_actions, actions_len, packet);
2767 ofpbuf_delete(packet);
2771 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2772 * statistics appropriately. 'packet' must have at least sizeof(struct
2773 * ofp_packet_in) bytes of headroom.
2775 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2776 * applying flow_extract() to 'packet' would yield the same flow as
2779 * 'facet' must have accurately composed datapath actions; that is, it must
2780 * not be in need of revalidation.
2782 * Takes ownership of 'packet'. */
2784 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2785 struct ofpbuf *packet)
2787 struct dpif_flow_stats stats;
2789 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2791 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2792 stats.used = time_msec();
2793 if (execute_odp_actions(ofproto, &facet->flow,
2794 facet->actions, facet->actions_len, packet)) {
2795 facet_update_stats(ofproto, facet, &stats);
2799 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2801 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2802 * rule's statistics, via facet_uninstall().
2804 * - Removes 'facet' from its rule and from ofproto->facets.
2807 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2809 facet_uninstall(ofproto, facet);
2810 facet_flush_stats(ofproto, facet);
2811 hmap_remove(&ofproto->facets, &facet->hmap_node);
2812 list_remove(&facet->list_node);
2816 /* Composes the datapath actions for 'facet' based on its rule's actions. */
2818 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2819 const struct ofpbuf *packet)
2821 const struct rule_dpif *rule = facet->rule;
2822 struct ofpbuf *odp_actions;
2823 struct action_xlate_ctx ctx;
2825 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2826 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2827 facet->tags = ctx.tags;
2828 facet->may_install = ctx.may_set_up_flow;
2829 facet->has_learn = ctx.has_learn;
2830 facet->has_normal = ctx.has_normal;
2831 facet->nf_flow.output_iface = ctx.nf_output_iface;
2833 if (facet->actions_len != odp_actions->size
2834 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2835 free(facet->actions);
2836 facet->actions_len = odp_actions->size;
2837 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2840 ofpbuf_delete(odp_actions);
2843 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2844 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2845 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2846 * since 'facet' was last updated.
2848 * Returns 0 if successful, otherwise a positive errno value.*/
2850 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2851 const struct nlattr *actions, size_t actions_len,
2852 struct dpif_flow_stats *stats)
2854 struct odputil_keybuf keybuf;
2855 enum dpif_flow_put_flags flags;
2859 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2861 flags |= DPIF_FP_ZERO_STATS;
2864 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2865 odp_flow_key_from_flow(&key, &facet->flow);
2867 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2868 actions, actions_len, stats);
2871 facet_reset_dp_stats(facet, stats);
2877 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2878 * 'zero_stats' is true, clears any existing statistics from the datapath for
2881 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2883 struct dpif_flow_stats stats;
2885 if (facet->may_install
2886 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2887 zero_stats ? &stats : NULL)) {
2888 facet->installed = true;
2893 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
2896 const struct nlattr *a;
2900 if (facet->byte_count <= facet->accounted_bytes) {
2903 n_bytes = facet->byte_count - facet->accounted_bytes;
2904 facet->accounted_bytes = facet->byte_count;
2906 /* Feed information from the active flows back into the learning table to
2907 * ensure that table is always in sync with what is actually flowing
2908 * through the datapath. */
2909 if (facet->has_learn || facet->has_normal) {
2910 struct action_xlate_ctx ctx;
2912 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2913 ctx.may_learn = true;
2914 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
2915 facet->rule->up.n_actions));
2918 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
2922 /* This loop feeds byte counters to bond_account() for rebalancing to use
2923 * as a basis. We also need to track the actual VLAN on which the packet
2924 * is going to be sent to ensure that it matches the one passed to
2925 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2927 vlan_tci = facet->flow.vlan_tci;
2928 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2929 const struct ovs_action_push_vlan *vlan;
2930 struct ofport_dpif *port;
2932 switch (nl_attr_type(a)) {
2933 case OVS_ACTION_ATTR_OUTPUT:
2934 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2935 if (port && port->bundle && port->bundle->bond) {
2936 bond_account(port->bundle->bond, &facet->flow,
2937 vlan_tci_to_vid(vlan_tci), n_bytes);
2941 case OVS_ACTION_ATTR_POP_VLAN:
2942 vlan_tci = htons(0);
2945 case OVS_ACTION_ATTR_PUSH_VLAN:
2946 vlan = nl_attr_get(a);
2947 vlan_tci = vlan->vlan_tci;
2953 /* If 'rule' is installed in the datapath, uninstalls it. */
2955 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2957 if (facet->installed) {
2958 struct odputil_keybuf keybuf;
2959 struct dpif_flow_stats stats;
2963 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2964 odp_flow_key_from_flow(&key, &facet->flow);
2966 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2967 facet_reset_dp_stats(facet, &stats);
2969 facet_update_stats(p, facet, &stats);
2971 facet->installed = false;
2973 assert(facet->dp_packet_count == 0);
2974 assert(facet->dp_byte_count == 0);
2978 /* Returns true if the only action for 'facet' is to send to the controller.
2979 * (We don't report NetFlow expiration messages for such facets because they
2980 * are just part of the control logic for the network, not real traffic). */
2982 facet_is_controller_flow(struct facet *facet)
2985 && facet->rule->up.n_actions == 1
2986 && action_outputs_to_port(&facet->rule->up.actions[0],
2987 htons(OFPP_CONTROLLER)));
2990 /* Resets 'facet''s datapath statistics counters. This should be called when
2991 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2992 * it should contain the statistics returned by dpif when 'facet' was reset in
2993 * the datapath. 'stats' will be modified to only included statistics new
2994 * since 'facet' was last updated. */
2996 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2998 if (stats && facet->dp_packet_count <= stats->n_packets
2999 && facet->dp_byte_count <= stats->n_bytes) {
3000 stats->n_packets -= facet->dp_packet_count;
3001 stats->n_bytes -= facet->dp_byte_count;
3004 facet->dp_packet_count = 0;
3005 facet->dp_byte_count = 0;
3008 /* Folds all of 'facet''s statistics into its rule. Also updates the
3009 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
3010 * 'facet''s statistics in the datapath should have been zeroed and folded into
3011 * its packet and byte counts before this function is called. */
3013 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3015 assert(!facet->dp_byte_count);
3016 assert(!facet->dp_packet_count);
3018 facet_push_stats(facet);
3019 facet_account(ofproto, facet);
3021 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3022 struct ofexpired expired;
3023 expired.flow = facet->flow;
3024 expired.packet_count = facet->packet_count;
3025 expired.byte_count = facet->byte_count;
3026 expired.used = facet->used;
3027 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3030 facet->rule->packet_count += facet->packet_count;
3031 facet->rule->byte_count += facet->byte_count;
3033 /* Reset counters to prevent double counting if 'facet' ever gets
3035 facet_reset_counters(facet);
3037 netflow_flow_clear(&facet->nf_flow);
3040 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3041 * Returns it if found, otherwise a null pointer.
3043 * The returned facet might need revalidation; use facet_lookup_valid()
3044 * instead if that is important. */
3045 static struct facet *
3046 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3048 struct facet *facet;
3050 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3052 if (flow_equal(flow, &facet->flow)) {
3060 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3061 * Returns it if found, otherwise a null pointer.
3063 * The returned facet is guaranteed to be valid. */
3064 static struct facet *
3065 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3067 struct facet *facet = facet_find(ofproto, flow);
3069 /* The facet we found might not be valid, since we could be in need of
3070 * revalidation. If it is not valid, don't return it. */
3072 && (ofproto->need_revalidate
3073 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3074 && !facet_revalidate(ofproto, facet)) {
3075 COVERAGE_INC(facet_invalidated);
3082 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3084 * - If the rule found is different from 'facet''s current rule, moves
3085 * 'facet' to the new rule and recompiles its actions.
3087 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3088 * where it is and recompiles its actions anyway.
3090 * - If there is none, destroys 'facet'.
3092 * Returns true if 'facet' still exists, false if it has been destroyed. */
3094 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3096 struct action_xlate_ctx ctx;
3097 struct ofpbuf *odp_actions;
3098 struct rule_dpif *new_rule;
3099 bool actions_changed;
3101 COVERAGE_INC(facet_revalidate);
3103 /* Determine the new rule. */
3104 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3106 /* No new rule, so delete the facet. */
3107 facet_remove(ofproto, facet);
3111 /* Calculate new datapath actions.
3113 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3114 * emit a NetFlow expiration and, if so, we need to have the old state
3115 * around to properly compose it. */
3116 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
3117 odp_actions = xlate_actions(&ctx,
3118 new_rule->up.actions, new_rule->up.n_actions);
3119 actions_changed = (facet->actions_len != odp_actions->size
3120 || memcmp(facet->actions, odp_actions->data,
3121 facet->actions_len));
3123 /* If the datapath actions changed or the installability changed,
3124 * then we need to talk to the datapath. */
3125 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
3126 if (ctx.may_set_up_flow) {
3127 struct dpif_flow_stats stats;
3129 facet_put__(ofproto, facet,
3130 odp_actions->data, odp_actions->size, &stats);
3131 facet_update_stats(ofproto, facet, &stats);
3133 facet_uninstall(ofproto, facet);
3136 /* The datapath flow is gone or has zeroed stats, so push stats out of
3137 * 'facet' into 'rule'. */
3138 facet_flush_stats(ofproto, facet);
3141 /* Update 'facet' now that we've taken care of all the old state. */
3142 facet->tags = ctx.tags;
3143 facet->nf_flow.output_iface = ctx.nf_output_iface;
3144 facet->may_install = ctx.may_set_up_flow;
3145 facet->has_learn = ctx.has_learn;
3146 facet->has_normal = ctx.has_normal;
3147 if (actions_changed) {
3148 free(facet->actions);
3149 facet->actions_len = odp_actions->size;
3150 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
3152 if (facet->rule != new_rule) {
3153 COVERAGE_INC(facet_changed_rule);
3154 list_remove(&facet->list_node);
3155 list_push_back(&new_rule->facets, &facet->list_node);
3156 facet->rule = new_rule;
3157 facet->used = new_rule->up.created;
3158 facet->rs_used = facet->used;
3161 ofpbuf_delete(odp_actions);
3166 /* Updates 'facet''s used time. Caller is responsible for calling
3167 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3169 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3172 if (used > facet->used) {
3174 if (used > facet->rule->used) {
3175 facet->rule->used = used;
3177 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3181 /* Folds the statistics from 'stats' into the counters in 'facet'.
3183 * Because of the meaning of a facet's counters, it only makes sense to do this
3184 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3185 * packet that was sent by hand or if it represents statistics that have been
3186 * cleared out of the datapath. */
3188 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
3189 const struct dpif_flow_stats *stats)
3191 if (stats->n_packets || stats->used > facet->used) {
3192 facet_update_time(ofproto, facet, stats->used);
3193 facet->packet_count += stats->n_packets;
3194 facet->byte_count += stats->n_bytes;
3195 facet_push_stats(facet);
3196 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3201 facet_reset_counters(struct facet *facet)
3203 facet->packet_count = 0;
3204 facet->byte_count = 0;
3205 facet->rs_packet_count = 0;
3206 facet->rs_byte_count = 0;
3207 facet->accounted_bytes = 0;
3211 facet_push_stats(struct facet *facet)
3213 uint64_t rs_packets, rs_bytes;
3215 assert(facet->packet_count >= facet->rs_packet_count);
3216 assert(facet->byte_count >= facet->rs_byte_count);
3217 assert(facet->used >= facet->rs_used);
3219 rs_packets = facet->packet_count - facet->rs_packet_count;
3220 rs_bytes = facet->byte_count - facet->rs_byte_count;
3222 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3223 facet->rs_packet_count = facet->packet_count;
3224 facet->rs_byte_count = facet->byte_count;
3225 facet->rs_used = facet->used;
3227 flow_push_stats(facet->rule, &facet->flow,
3228 rs_packets, rs_bytes, facet->used);
3232 struct ofproto_push {
3233 struct action_xlate_ctx ctx;
3240 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3242 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3245 rule->packet_count += push->packets;
3246 rule->byte_count += push->bytes;
3247 rule->used = MAX(push->used, rule->used);
3251 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3252 * 'rule''s actions. */
3254 flow_push_stats(const struct rule_dpif *rule,
3255 struct flow *flow, uint64_t packets, uint64_t bytes,
3258 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3259 struct ofproto_push push;
3261 push.packets = packets;
3265 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3266 push.ctx.resubmit_hook = push_resubmit;
3267 ofpbuf_delete(xlate_actions(&push.ctx,
3268 rule->up.actions, rule->up.n_actions));
3273 static struct rule_dpif *
3274 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3277 struct cls_rule *cls_rule;
3278 struct classifier *cls;
3280 if (table_id >= N_TABLES) {
3284 cls = &ofproto->up.tables[table_id];
3285 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3286 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3287 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3288 * are unavailable. */
3289 struct flow ofpc_normal_flow = *flow;
3290 ofpc_normal_flow.tp_src = htons(0);
3291 ofpc_normal_flow.tp_dst = htons(0);
3292 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3294 cls_rule = classifier_lookup(cls, flow);
3296 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3300 complete_operation(struct rule_dpif *rule)
3302 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3304 rule_invalidate(rule);
3306 struct dpif_completion *c = xmalloc(sizeof *c);
3307 c->op = rule->up.pending;
3308 list_push_back(&ofproto->completions, &c->list_node);
3310 ofoperation_complete(rule->up.pending, 0);
3314 static struct rule *
3317 struct rule_dpif *rule = xmalloc(sizeof *rule);
3322 rule_dealloc(struct rule *rule_)
3324 struct rule_dpif *rule = rule_dpif_cast(rule_);
3329 rule_construct(struct rule *rule_)
3331 struct rule_dpif *rule = rule_dpif_cast(rule_);
3332 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3333 struct rule_dpif *victim;
3337 error = validate_actions(rule->up.actions, rule->up.n_actions,
3338 &rule->up.cr.flow, ofproto->max_ports);
3343 rule->used = rule->up.created;
3344 rule->packet_count = 0;
3345 rule->byte_count = 0;
3347 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3348 if (victim && !list_is_empty(&victim->facets)) {
3349 struct facet *facet;
3351 rule->facets = victim->facets;
3352 list_moved(&rule->facets);
3353 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3354 /* XXX: We're only clearing our local counters here. It's possible
3355 * that quite a few packets are unaccounted for in the datapath
3356 * statistics. These will be accounted to the new rule instead of
3357 * cleared as required. This could be fixed by clearing out the
3358 * datapath statistics for this facet, but currently it doesn't
3360 facet_reset_counters(facet);
3364 /* Must avoid list_moved() in this case. */
3365 list_init(&rule->facets);
3368 table_id = rule->up.table_id;
3369 rule->tag = (victim ? victim->tag
3371 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3372 ofproto->tables[table_id].basis));
3374 complete_operation(rule);
3379 rule_destruct(struct rule *rule_)
3381 struct rule_dpif *rule = rule_dpif_cast(rule_);
3382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3383 struct facet *facet, *next_facet;
3385 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3386 facet_revalidate(ofproto, facet);
3389 complete_operation(rule);
3393 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3395 struct rule_dpif *rule = rule_dpif_cast(rule_);
3396 struct facet *facet;
3398 /* Start from historical data for 'rule' itself that are no longer tracked
3399 * in facets. This counts, for example, facets that have expired. */
3400 *packets = rule->packet_count;
3401 *bytes = rule->byte_count;
3403 /* Add any statistics that are tracked by facets. This includes
3404 * statistical data recently updated by ofproto_update_stats() as well as
3405 * stats for packets that were executed "by hand" via dpif_execute(). */
3406 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3407 *packets += facet->packet_count;
3408 *bytes += facet->byte_count;
3413 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
3415 struct rule_dpif *rule = rule_dpif_cast(rule_);
3416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3417 struct action_xlate_ctx ctx;
3418 struct ofpbuf *odp_actions;
3419 struct facet *facet;
3422 /* First look for a related facet. If we find one, account it to that. */
3423 facet = facet_lookup_valid(ofproto, flow);
3424 if (facet && facet->rule == rule) {
3425 if (!facet->may_install) {
3426 facet_make_actions(ofproto, facet, packet);
3428 facet_execute(ofproto, facet, packet);
3432 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
3433 * create a new facet for it and use that. */
3434 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
3435 facet = facet_create(rule, flow);
3436 facet_make_actions(ofproto, facet, packet);
3437 facet_execute(ofproto, facet, packet);
3438 facet_install(ofproto, facet, true);
3442 /* We can't account anything to a facet. If we were to try, then that
3443 * facet would have a non-matching rule, busting our invariants. */
3444 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3445 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3446 size = packet->size;
3447 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3448 odp_actions->size, packet)) {
3449 rule->used = time_msec();
3450 rule->packet_count++;
3451 rule->byte_count += size;
3452 flow_push_stats(rule, flow, 1, size, rule->used);
3454 ofpbuf_delete(odp_actions);
3460 rule_modify_actions(struct rule *rule_)
3462 struct rule_dpif *rule = rule_dpif_cast(rule_);
3463 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3466 error = validate_actions(rule->up.actions, rule->up.n_actions,
3467 &rule->up.cr.flow, ofproto->max_ports);
3469 ofoperation_complete(rule->up.pending, error);
3473 complete_operation(rule);
3476 /* Sends 'packet' out of port 'odp_port' within 'ofproto'.
3477 * Returns 0 if successful, otherwise a positive errno value. */
3479 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
3480 const struct ofpbuf *packet)
3482 struct ofpbuf key, odp_actions;
3483 struct odputil_keybuf keybuf;
3487 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
3488 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3489 odp_flow_key_from_flow(&key, &flow);
3491 ofpbuf_init(&odp_actions, 32);
3492 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
3494 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3495 error = dpif_execute(ofproto->dpif,
3497 odp_actions.data, odp_actions.size,
3499 ofpbuf_uninit(&odp_actions);
3502 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
3503 ofproto->up.name, odp_port, strerror(error));
3508 /* OpenFlow to datapath action translation. */
3510 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
3511 struct action_xlate_ctx *ctx);
3512 static void xlate_normal(struct action_xlate_ctx *);
3515 put_userspace_action(const struct ofproto_dpif *ofproto,
3516 struct ofpbuf *odp_actions,
3517 const struct flow *flow,
3518 const struct user_action_cookie *cookie)
3522 pid = dpif_port_get_pid(ofproto->dpif,
3523 ofp_port_to_odp_port(flow->in_port));
3525 return odp_put_userspace_action(pid, cookie, odp_actions);
3528 /* Compose SAMPLE action for sFlow. */
3530 compose_sflow_action(const struct ofproto_dpif *ofproto,
3531 struct ofpbuf *odp_actions,
3532 const struct flow *flow,
3535 uint32_t port_ifindex;
3536 uint32_t probability;
3537 struct user_action_cookie cookie;
3538 size_t sample_offset, actions_offset;
3539 int cookie_offset, n_output;
3541 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
3545 if (odp_port == OVSP_NONE) {
3549 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
3553 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
3555 /* Number of packets out of UINT_MAX to sample. */
3556 probability = dpif_sflow_get_probability(ofproto->sflow);
3557 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
3559 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
3561 cookie.type = USER_ACTION_COOKIE_SFLOW;
3562 cookie.data = port_ifindex;
3563 cookie.n_output = n_output;
3564 cookie.vlan_tci = 0;
3565 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
3567 nl_msg_end_nested(odp_actions, actions_offset);
3568 nl_msg_end_nested(odp_actions, sample_offset);
3569 return cookie_offset;
3572 /* SAMPLE action must be first action in any given list of actions.
3573 * At this point we do not have all information required to build it. So try to
3574 * build sample action as complete as possible. */
3576 add_sflow_action(struct action_xlate_ctx *ctx)
3578 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
3580 &ctx->flow, OVSP_NONE);
3581 ctx->sflow_odp_port = 0;
3582 ctx->sflow_n_outputs = 0;
3585 /* Fix SAMPLE action according to data collected while composing ODP actions.
3586 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
3587 * USERSPACE action's user-cookie which is required for sflow. */
3589 fix_sflow_action(struct action_xlate_ctx *ctx)
3591 const struct flow *base = &ctx->base_flow;
3592 struct user_action_cookie *cookie;
3594 if (!ctx->user_cookie_offset) {
3598 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
3600 assert(cookie != NULL);
3601 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
3603 if (ctx->sflow_n_outputs) {
3604 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
3605 ctx->sflow_odp_port);
3607 if (ctx->sflow_n_outputs >= 255) {
3608 cookie->n_output = 255;
3610 cookie->n_output = ctx->sflow_n_outputs;
3612 cookie->vlan_tci = base->vlan_tci;
3616 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
3617 const void *key, size_t key_size)
3619 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
3620 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
3621 nl_msg_end_nested(odp_actions, offset);
3625 commit_set_tun_id_action(const struct flow *flow, struct flow *base,
3626 struct ofpbuf *odp_actions)
3628 if (base->tun_id == flow->tun_id) {
3631 base->tun_id = flow->tun_id;
3633 commit_set_action(odp_actions, OVS_KEY_ATTR_TUN_ID,
3634 &base->tun_id, sizeof(base->tun_id));
3638 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
3639 struct ofpbuf *odp_actions)
3641 struct ovs_key_ethernet eth_key;
3643 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
3644 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
3648 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
3649 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
3651 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
3652 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
3654 commit_set_action(odp_actions, OVS_KEY_ATTR_ETHERNET,
3655 ð_key, sizeof(eth_key));
3659 commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci)
3661 struct flow *base = &ctx->base_flow;
3663 if (base->vlan_tci == new_tci) {
3667 if (base->vlan_tci & htons(VLAN_CFI)) {
3668 nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_VLAN);
3671 if (new_tci & htons(VLAN_CFI)) {
3672 struct ovs_action_push_vlan vlan;
3674 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
3675 vlan.vlan_tci = new_tci;
3676 nl_msg_put_unspec(ctx->odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
3677 &vlan, sizeof vlan);
3679 base->vlan_tci = new_tci;
3683 commit_set_nw_action(const struct flow *flow, struct flow *base,
3684 struct ofpbuf *odp_actions)
3686 struct ovs_key_ipv4 ipv4_key;
3688 if (base->dl_type != htons(ETH_TYPE_IP) ||
3689 !base->nw_src || !base->nw_dst) {
3693 if (base->nw_src == flow->nw_src &&
3694 base->nw_dst == flow->nw_dst &&
3695 base->nw_tos == flow->nw_tos &&
3696 base->nw_ttl == flow->nw_ttl &&
3697 base->nw_frag == flow->nw_frag) {
3701 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
3702 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
3703 ipv4_key.ipv4_proto = base->nw_proto;
3704 ipv4_key.ipv4_tos = flow->nw_tos;
3705 ipv4_key.ipv4_ttl = flow->nw_ttl;
3706 ipv4_key.ipv4_frag = (base->nw_frag == 0 ? OVS_FRAG_TYPE_NONE
3707 : base->nw_frag == FLOW_NW_FRAG_ANY
3708 ? OVS_FRAG_TYPE_FIRST : OVS_FRAG_TYPE_LATER);
3710 commit_set_action(odp_actions, OVS_KEY_ATTR_IPV4,
3711 &ipv4_key, sizeof(ipv4_key));
3715 commit_set_port_action(const struct flow *flow, struct flow *base,
3716 struct ofpbuf *odp_actions)
3718 if (!base->tp_src || !base->tp_dst) {
3722 if (base->tp_src == flow->tp_src &&
3723 base->tp_dst == flow->tp_dst) {
3727 if (flow->nw_proto == IPPROTO_TCP) {
3728 struct ovs_key_tcp port_key;
3730 port_key.tcp_src = base->tp_src = flow->tp_src;
3731 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
3733 commit_set_action(odp_actions, OVS_KEY_ATTR_TCP,
3734 &port_key, sizeof(port_key));
3736 } else if (flow->nw_proto == IPPROTO_UDP) {
3737 struct ovs_key_udp port_key;
3739 port_key.udp_src = base->tp_src = flow->tp_src;
3740 port_key.udp_dst = base->tp_dst = flow->tp_dst;
3742 commit_set_action(odp_actions, OVS_KEY_ATTR_UDP,
3743 &port_key, sizeof(port_key));
3748 commit_set_priority_action(const struct flow *flow, struct flow *base,
3749 struct ofpbuf *odp_actions)
3751 if (base->priority == flow->priority) {
3754 base->priority = flow->priority;
3756 commit_set_action(odp_actions, OVS_KEY_ATTR_PRIORITY,
3757 &base->priority, sizeof(base->priority));
3761 commit_odp_actions(struct action_xlate_ctx *ctx)
3763 const struct flow *flow = &ctx->flow;
3764 struct flow *base = &ctx->base_flow;
3765 struct ofpbuf *odp_actions = ctx->odp_actions;
3767 commit_set_tun_id_action(flow, base, odp_actions);
3768 commit_set_ether_addr_action(flow, base, odp_actions);
3769 commit_vlan_action(ctx, flow->vlan_tci);
3770 commit_set_nw_action(flow, base, odp_actions);
3771 commit_set_port_action(flow, base, odp_actions);
3772 commit_set_priority_action(flow, base, odp_actions);
3776 compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port)
3778 uint16_t ofp_port = odp_port_to_ofp_port(odp_port);
3779 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3781 if (ofport && ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
3785 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3786 ctx->sflow_odp_port = odp_port;
3787 ctx->sflow_n_outputs++;
3791 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3793 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3794 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
3796 if (ofport && !stp_forward_in_state(ofport->stp_state)) {
3797 /* Forwarding disabled on port. */
3801 /* We may not have an ofport record for this port, but it doesn't hurt to
3802 * allow forwarding to it anyhow. Maybe such a port will appear later and
3803 * we're pre-populating the flow table. */
3805 commit_odp_actions(ctx);
3806 compose_output_action(ctx, odp_port);
3807 ctx->nf_output_iface = ofp_port;
3811 xlate_table_action(struct action_xlate_ctx *ctx,
3812 uint16_t in_port, uint8_t table_id)
3814 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
3815 struct ofproto_dpif *ofproto = ctx->ofproto;
3816 struct rule_dpif *rule;
3817 uint16_t old_in_port;
3818 uint8_t old_table_id;
3820 old_table_id = ctx->table_id;
3821 ctx->table_id = table_id;
3823 /* Look up a flow with 'in_port' as the input port. */
3824 old_in_port = ctx->flow.in_port;
3825 ctx->flow.in_port = in_port;
3826 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
3829 if (table_id > 0 && table_id < N_TABLES) {
3830 struct table_dpif *table = &ofproto->tables[table_id];
3831 if (table->other_table) {
3834 : rule_calculate_tag(&ctx->flow,
3835 &table->other_table->wc,
3840 /* Restore the original input port. Otherwise OFPP_NORMAL and
3841 * OFPP_IN_PORT will have surprising behavior. */
3842 ctx->flow.in_port = old_in_port;
3844 if (ctx->resubmit_hook) {
3845 ctx->resubmit_hook(ctx, rule);
3850 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
3854 ctx->table_id = old_table_id;
3856 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3858 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
3859 MAX_RESUBMIT_RECURSION);
3864 xlate_resubmit_table(struct action_xlate_ctx *ctx,
3865 const struct nx_action_resubmit *nar)
3870 in_port = (nar->in_port == htons(OFPP_IN_PORT)
3872 : ntohs(nar->in_port));
3873 table_id = nar->table == 255 ? ctx->table_id : nar->table;
3875 xlate_table_action(ctx, in_port, table_id);
3879 flood_packets(struct action_xlate_ctx *ctx, bool all)
3881 struct ofport_dpif *ofport;
3883 commit_odp_actions(ctx);
3884 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
3885 uint16_t ofp_port = ofport->up.ofp_port;
3887 if (ofp_port == ctx->flow.in_port) {
3891 if (all || (!(ofport->up.opp.config & htonl(OFPPC_NO_FLOOD))
3892 && stp_forward_in_state(ofport->stp_state))) {
3893 compose_output_action(ctx, ofport->odp_port);
3897 ctx->nf_output_iface = NF_OUT_FLOOD;
3901 compose_controller_action(struct action_xlate_ctx *ctx, int len)
3903 struct user_action_cookie cookie;
3905 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
3907 cookie.n_output = 0;
3908 cookie.vlan_tci = 0;
3909 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
3913 xlate_output_action__(struct action_xlate_ctx *ctx,
3914 uint16_t port, uint16_t max_len)
3916 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
3918 ctx->nf_output_iface = NF_OUT_DROP;
3922 add_output_action(ctx, ctx->flow.in_port);
3925 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
3931 flood_packets(ctx, false);
3934 flood_packets(ctx, true);
3936 case OFPP_CONTROLLER:
3937 commit_odp_actions(ctx);
3938 compose_controller_action(ctx, max_len);
3941 add_output_action(ctx, OFPP_LOCAL);
3946 if (port != ctx->flow.in_port) {
3947 add_output_action(ctx, port);
3952 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3953 ctx->nf_output_iface = NF_OUT_FLOOD;
3954 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3955 ctx->nf_output_iface = prev_nf_output_iface;
3956 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3957 ctx->nf_output_iface != NF_OUT_FLOOD) {
3958 ctx->nf_output_iface = NF_OUT_MULTI;
3963 xlate_output_reg_action(struct action_xlate_ctx *ctx,
3964 const struct nx_action_output_reg *naor)
3968 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
3970 if (ofp_port <= UINT16_MAX) {
3971 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
3976 xlate_output_action(struct action_xlate_ctx *ctx,
3977 const struct ofp_action_output *oao)
3979 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3983 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3984 const struct ofp_action_enqueue *oae)
3987 uint32_t flow_priority, priority;
3990 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3993 /* Fall back to ordinary output action. */
3994 xlate_output_action__(ctx, ntohs(oae->port), 0);
3998 /* Figure out datapath output port. */
3999 ofp_port = ntohs(oae->port);
4000 if (ofp_port == OFPP_IN_PORT) {
4001 ofp_port = ctx->flow.in_port;
4002 } else if (ofp_port == ctx->flow.in_port) {
4006 /* Add datapath actions. */
4007 flow_priority = ctx->flow.priority;
4008 ctx->flow.priority = priority;
4009 add_output_action(ctx, ofp_port);
4010 ctx->flow.priority = flow_priority;
4012 /* Update NetFlow output port. */
4013 if (ctx->nf_output_iface == NF_OUT_DROP) {
4014 ctx->nf_output_iface = ofp_port;
4015 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4016 ctx->nf_output_iface = NF_OUT_MULTI;
4021 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4022 const struct nx_action_set_queue *nasq)
4027 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4030 /* Couldn't translate queue to a priority, so ignore. A warning
4031 * has already been logged. */
4035 ctx->flow.priority = priority;
4038 struct xlate_reg_state {
4044 xlate_autopath(struct action_xlate_ctx *ctx,
4045 const struct nx_action_autopath *naa)
4047 uint16_t ofp_port = ntohl(naa->id);
4048 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4050 if (!port || !port->bundle) {
4051 ofp_port = OFPP_NONE;
4052 } else if (port->bundle->bond) {
4053 /* Autopath does not support VLAN hashing. */
4054 struct ofport_dpif *slave = bond_choose_output_slave(
4055 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4057 ofp_port = slave->up.ofp_port;
4060 autopath_execute(naa, &ctx->flow, ofp_port);
4064 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4066 struct ofproto_dpif *ofproto = ofproto_;
4067 struct ofport_dpif *port;
4077 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4080 port = get_ofp_port(ofproto, ofp_port);
4081 return port ? port->may_enable : false;
4086 xlate_learn_action(struct action_xlate_ctx *ctx,
4087 const struct nx_action_learn *learn)
4089 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4090 struct ofputil_flow_mod fm;
4093 learn_execute(learn, &ctx->flow, &fm);
4095 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4096 if (error && !VLOG_DROP_WARN(&rl)) {
4097 char *msg = ofputil_error_to_string(error);
4098 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4106 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4108 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4109 ? htonl(OFPPC_NO_RECV_STP)
4110 : htonl(OFPPC_NO_RECV))) {
4114 /* Only drop packets here if both forwarding and learning are
4115 * disabled. If just learning is enabled, we need to have
4116 * OFPP_NORMAL and the learning action have a look at the packet
4117 * before we can drop it. */
4118 if (!stp_forward_in_state(port->stp_state)
4119 && !stp_learn_in_state(port->stp_state)) {
4127 do_xlate_actions(const union ofp_action *in, size_t n_in,
4128 struct action_xlate_ctx *ctx)
4130 const struct ofport_dpif *port;
4131 const union ofp_action *ia;
4134 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4135 if (port && !may_receive(port, ctx)) {
4136 /* Drop this flow. */
4140 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4141 const struct ofp_action_dl_addr *oada;
4142 const struct nx_action_resubmit *nar;
4143 const struct nx_action_set_tunnel *nast;
4144 const struct nx_action_set_queue *nasq;
4145 const struct nx_action_multipath *nam;
4146 const struct nx_action_autopath *naa;
4147 const struct nx_action_bundle *nab;
4148 const struct nx_action_output_reg *naor;
4149 enum ofputil_action_code code;
4156 code = ofputil_decode_action_unsafe(ia);
4158 case OFPUTIL_OFPAT_OUTPUT:
4159 xlate_output_action(ctx, &ia->output);
4162 case OFPUTIL_OFPAT_SET_VLAN_VID:
4163 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4164 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4167 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4168 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4169 ctx->flow.vlan_tci |= htons(
4170 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4173 case OFPUTIL_OFPAT_STRIP_VLAN:
4174 ctx->flow.vlan_tci = htons(0);
4177 case OFPUTIL_OFPAT_SET_DL_SRC:
4178 oada = ((struct ofp_action_dl_addr *) ia);
4179 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4182 case OFPUTIL_OFPAT_SET_DL_DST:
4183 oada = ((struct ofp_action_dl_addr *) ia);
4184 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4187 case OFPUTIL_OFPAT_SET_NW_SRC:
4188 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4191 case OFPUTIL_OFPAT_SET_NW_DST:
4192 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4195 case OFPUTIL_OFPAT_SET_NW_TOS:
4196 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4197 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4200 case OFPUTIL_OFPAT_SET_TP_SRC:
4201 ctx->flow.tp_src = ia->tp_port.tp_port;
4204 case OFPUTIL_OFPAT_SET_TP_DST:
4205 ctx->flow.tp_dst = ia->tp_port.tp_port;
4208 case OFPUTIL_OFPAT_ENQUEUE:
4209 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4212 case OFPUTIL_NXAST_RESUBMIT:
4213 nar = (const struct nx_action_resubmit *) ia;
4214 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4217 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4218 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4221 case OFPUTIL_NXAST_SET_TUNNEL:
4222 nast = (const struct nx_action_set_tunnel *) ia;
4223 tun_id = htonll(ntohl(nast->tun_id));
4224 ctx->flow.tun_id = tun_id;
4227 case OFPUTIL_NXAST_SET_QUEUE:
4228 nasq = (const struct nx_action_set_queue *) ia;
4229 xlate_set_queue_action(ctx, nasq);
4232 case OFPUTIL_NXAST_POP_QUEUE:
4233 ctx->flow.priority = ctx->original_priority;
4236 case OFPUTIL_NXAST_REG_MOVE:
4237 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4241 case OFPUTIL_NXAST_REG_LOAD:
4242 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4246 case OFPUTIL_NXAST_NOTE:
4247 /* Nothing to do. */
4250 case OFPUTIL_NXAST_SET_TUNNEL64:
4251 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4252 ctx->flow.tun_id = tun_id;
4255 case OFPUTIL_NXAST_MULTIPATH:
4256 nam = (const struct nx_action_multipath *) ia;
4257 multipath_execute(nam, &ctx->flow);
4260 case OFPUTIL_NXAST_AUTOPATH:
4261 naa = (const struct nx_action_autopath *) ia;
4262 xlate_autopath(ctx, naa);
4265 case OFPUTIL_NXAST_BUNDLE:
4266 ctx->ofproto->has_bundle_action = true;
4267 nab = (const struct nx_action_bundle *) ia;
4268 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4273 case OFPUTIL_NXAST_BUNDLE_LOAD:
4274 ctx->ofproto->has_bundle_action = true;
4275 nab = (const struct nx_action_bundle *) ia;
4276 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4280 case OFPUTIL_NXAST_OUTPUT_REG:
4281 naor = (const struct nx_action_output_reg *) ia;
4282 xlate_output_reg_action(ctx, naor);
4285 case OFPUTIL_NXAST_LEARN:
4286 ctx->has_learn = true;
4287 if (ctx->may_learn) {
4288 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4292 case OFPUTIL_NXAST_EXIT:
4298 /* We've let OFPP_NORMAL and the learning action look at the packet,
4299 * so drop it now if forwarding is disabled. */
4300 if (port && !stp_forward_in_state(port->stp_state)) {
4301 ofpbuf_clear(ctx->odp_actions);
4302 add_sflow_action(ctx);
4307 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4308 struct ofproto_dpif *ofproto, const struct flow *flow,
4309 const struct ofpbuf *packet)
4311 ctx->ofproto = ofproto;
4313 ctx->packet = packet;
4314 ctx->may_learn = packet != NULL;
4315 ctx->resubmit_hook = NULL;
4318 static struct ofpbuf *
4319 xlate_actions(struct action_xlate_ctx *ctx,
4320 const union ofp_action *in, size_t n_in)
4322 COVERAGE_INC(ofproto_dpif_xlate);
4324 ctx->odp_actions = ofpbuf_new(512);
4325 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4327 ctx->may_set_up_flow = true;
4328 ctx->has_learn = false;
4329 ctx->has_normal = false;
4330 ctx->nf_output_iface = NF_OUT_DROP;
4332 ctx->original_priority = ctx->flow.priority;
4333 ctx->base_flow = ctx->flow;
4334 ctx->base_flow.tun_id = 0;
4338 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4339 switch (ctx->ofproto->up.frag_handling) {
4340 case OFPC_FRAG_NORMAL:
4341 /* We must pretend that transport ports are unavailable. */
4342 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4343 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4346 case OFPC_FRAG_DROP:
4347 return ctx->odp_actions;
4349 case OFPC_FRAG_REASM:
4352 case OFPC_FRAG_NX_MATCH:
4353 /* Nothing to do. */
4358 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4359 ctx->may_set_up_flow = false;
4360 return ctx->odp_actions;
4362 add_sflow_action(ctx);
4363 do_xlate_actions(in, n_in, ctx);
4365 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4366 ctx->odp_actions->data,
4367 ctx->odp_actions->size)) {
4368 ctx->may_set_up_flow = false;
4370 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4372 compose_output_action(ctx, OVSP_LOCAL);
4375 fix_sflow_action(ctx);
4378 return ctx->odp_actions;
4381 /* OFPP_NORMAL implementation. */
4383 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4385 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4386 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4387 * the bundle on which the packet was received, returns the VLAN to which the
4390 * Both 'vid' and the return value are in the range 0...4095. */
4392 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4394 switch (in_bundle->vlan_mode) {
4395 case PORT_VLAN_ACCESS:
4396 return in_bundle->vlan;
4399 case PORT_VLAN_TRUNK:
4402 case PORT_VLAN_NATIVE_UNTAGGED:
4403 case PORT_VLAN_NATIVE_TAGGED:
4404 return vid ? vid : in_bundle->vlan;
4411 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
4412 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
4415 * 'vid' should be the VID obtained from the 802.1Q header that was received as
4416 * part of a packet (specify 0 if there was no 802.1Q header), in the range
4419 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
4421 switch (in_bundle->vlan_mode) {
4422 case PORT_VLAN_ACCESS:
4425 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4426 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
4427 "packet received on port %s configured as VLAN "
4428 "%"PRIu16" access port",
4429 in_bundle->ofproto->up.name, vid,
4430 in_bundle->name, in_bundle->vlan);
4436 case PORT_VLAN_NATIVE_UNTAGGED:
4437 case PORT_VLAN_NATIVE_TAGGED:
4439 /* Port must always carry its native VLAN. */
4443 case PORT_VLAN_TRUNK:
4444 if (!ofbundle_includes_vlan(in_bundle, vid)) {
4446 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4447 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
4448 "received on port %s not configured for trunking "
4450 in_bundle->ofproto->up.name, vid,
4451 in_bundle->name, vid);
4463 /* Given 'vlan', the VLAN that a packet belongs to, and
4464 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4465 * that should be included in the 802.1Q header. (If the return value is 0,
4466 * then the 802.1Q header should only be included in the packet if there is a
4469 * Both 'vlan' and the return value are in the range 0...4095. */
4471 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4473 switch (out_bundle->vlan_mode) {
4474 case PORT_VLAN_ACCESS:
4477 case PORT_VLAN_TRUNK:
4478 case PORT_VLAN_NATIVE_TAGGED:
4481 case PORT_VLAN_NATIVE_UNTAGGED:
4482 return vlan == out_bundle->vlan ? 0 : vlan;
4490 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
4493 struct ofport_dpif *port;
4497 vid = output_vlan_to_vid(out_bundle, vlan);
4498 if (!out_bundle->bond) {
4499 port = ofbundle_get_a_port(out_bundle);
4501 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4504 /* No slaves enabled, so drop packet. */
4510 if (tci || out_bundle->use_priority_tags) {
4511 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4513 tci |= htons(VLAN_CFI);
4516 commit_vlan_action(ctx, tci);
4518 compose_output_action(ctx, port->odp_port);
4519 ctx->nf_output_iface = port->up.ofp_port;
4523 mirror_mask_ffs(mirror_mask_t mask)
4525 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4530 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4532 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4533 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4537 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4539 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4542 /* Returns an arbitrary interface within 'bundle'. */
4543 static struct ofport_dpif *
4544 ofbundle_get_a_port(const struct ofbundle *bundle)
4546 return CONTAINER_OF(list_front(&bundle->ports),
4547 struct ofport_dpif, bundle_node);
4550 static mirror_mask_t
4551 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
4552 const struct ofbundle *in_bundle,
4553 const struct ofbundle *out_bundle)
4555 mirror_mask_t dst_mirrors = 0;
4557 if (out_bundle == OFBUNDLE_FLOOD) {
4558 struct ofbundle *bundle;
4560 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
4561 if (bundle != in_bundle
4562 && ofbundle_includes_vlan(bundle, vlan)
4563 && bundle->floodable
4564 && !bundle->mirror_out) {
4565 output_normal(ctx, bundle, vlan);
4566 dst_mirrors |= bundle->dst_mirrors;
4569 ctx->nf_output_iface = NF_OUT_FLOOD;
4570 } else if (out_bundle) {
4571 output_normal(ctx, out_bundle, vlan);
4572 dst_mirrors = out_bundle->dst_mirrors;
4579 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4581 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4584 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4585 * to a VLAN. In general most packets may be mirrored but we want to drop
4586 * protocols that may confuse switches. */
4588 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4590 /* If you change this function's behavior, please update corresponding
4591 * documentation in vswitch.xml at the same time. */
4592 if (dst[0] != 0x01) {
4593 /* All the currently banned MACs happen to start with 01 currently, so
4594 * this is a quick way to eliminate most of the good ones. */
4596 if (eth_addr_is_reserved(dst)) {
4597 /* Drop STP, IEEE pause frames, and other reserved protocols
4598 * (01-80-c2-00-00-0x). */
4602 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4604 if ((dst[3] & 0xfe) == 0xcc &&
4605 (dst[4] & 0xfe) == 0xcc &&
4606 (dst[5] & 0xfe) == 0xcc) {
4607 /* Drop the following protocols plus others following the same
4610 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4611 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4612 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4616 if (!(dst[3] | dst[4] | dst[5])) {
4617 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4626 output_mirrors(struct action_xlate_ctx *ctx,
4627 uint16_t vlan, const struct ofbundle *in_bundle,
4628 mirror_mask_t dst_mirrors)
4630 struct ofproto_dpif *ofproto = ctx->ofproto;
4631 mirror_mask_t mirrors;
4633 mirrors = in_bundle->src_mirrors | dst_mirrors;
4641 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
4643 if (!vlan_is_mirrored(m, vlan)) {
4644 mirrors &= mirrors - 1;
4648 mirrors &= ~m->dup_mirrors;
4650 output_normal(ctx, m->out, vlan);
4651 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)
4652 && vlan != m->out_vlan) {
4653 struct ofbundle *bundle;
4655 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4656 if (ofbundle_includes_vlan(bundle, m->out_vlan)
4657 && !bundle->mirror_out) {
4658 output_normal(ctx, bundle, m->out_vlan);
4665 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
4666 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
4667 * indicate this; newer upstream kernels use gratuitous ARP requests. */
4669 is_gratuitous_arp(const struct flow *flow)
4671 return (flow->dl_type == htons(ETH_TYPE_ARP)
4672 && eth_addr_is_broadcast(flow->dl_dst)
4673 && (flow->nw_proto == ARP_OP_REPLY
4674 || (flow->nw_proto == ARP_OP_REQUEST
4675 && flow->nw_src == flow->nw_dst)));
4679 update_learning_table(struct ofproto_dpif *ofproto,
4680 const struct flow *flow, int vlan,
4681 struct ofbundle *in_bundle)
4683 struct mac_entry *mac;
4685 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
4689 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
4690 if (is_gratuitous_arp(flow)) {
4691 /* We don't want to learn from gratuitous ARP packets that are
4692 * reflected back over bond slaves so we lock the learning table. */
4693 if (!in_bundle->bond) {
4694 mac_entry_set_grat_arp_lock(mac);
4695 } else if (mac_entry_is_grat_arp_locked(mac)) {
4700 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
4701 /* The log messages here could actually be useful in debugging,
4702 * so keep the rate limit relatively high. */
4703 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4704 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
4705 "on port %s in VLAN %d",
4706 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
4707 in_bundle->name, vlan);
4709 mac->port.p = in_bundle;
4710 tag_set_add(&ofproto->revalidate_set,
4711 mac_learning_changed(ofproto->ml, mac));
4715 static struct ofport_dpif *
4716 lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn)
4718 struct ofport_dpif *ofport;
4720 /* Find the port and bundle for the received packet. */
4721 ofport = get_ofp_port(ofproto, in_port);
4722 if (ofport && ofport->bundle) {
4726 /* Odd. A few possible reasons here:
4728 * - We deleted a port but there are still a few packets queued up
4731 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
4732 * we don't know about.
4734 * - The ofproto client didn't configure the port as part of a bundle.
4737 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4739 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
4740 "port %"PRIu16, ofproto->up.name, in_port);
4745 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
4746 * dropped. Returns true if they may be forwarded, false if they should be
4749 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
4750 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
4752 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
4753 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
4754 * checked by input_vid_is_valid().
4756 * May also add tags to '*tags', although the current implementation only does
4757 * so in one special case.
4760 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
4761 struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags)
4763 struct ofbundle *in_bundle = in_port->bundle;
4765 /* Drop frames for reserved multicast addresses
4766 * only if forward_bpdu option is absent. */
4767 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
4771 if (in_bundle->bond) {
4772 struct mac_entry *mac;
4774 switch (bond_check_admissibility(in_bundle->bond, in_port,
4775 flow->dl_dst, tags)) {
4782 case BV_DROP_IF_MOVED:
4783 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
4784 if (mac && mac->port.p != in_bundle &&
4785 (!is_gratuitous_arp(flow)
4786 || mac_entry_is_grat_arp_locked(mac))) {
4797 xlate_normal(struct action_xlate_ctx *ctx)
4799 mirror_mask_t dst_mirrors = 0;
4800 struct ofport_dpif *in_port;
4801 struct ofbundle *in_bundle;
4802 struct ofbundle *out_bundle;
4803 struct mac_entry *mac;
4807 ctx->has_normal = true;
4809 /* Obtain in_port from ctx->flow.in_port.
4811 * lookup_input_bundle() also ensures that in_port belongs to a bundle. */
4812 in_port = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
4813 ctx->packet != NULL);
4817 in_bundle = in_port->bundle;
4819 /* Drop malformed frames. */
4820 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
4821 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
4822 if (ctx->packet != NULL) {
4823 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4824 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
4825 "VLAN tag received on port %s",
4826 ctx->ofproto->up.name, in_bundle->name);
4831 /* Drop frames on bundles reserved for mirroring. */
4832 if (in_bundle->mirror_out) {
4833 if (ctx->packet != NULL) {
4834 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4835 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4836 "%s, which is reserved exclusively for mirroring",
4837 ctx->ofproto->up.name, in_bundle->name);
4843 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4844 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
4847 vlan = input_vid_to_vlan(in_bundle, vid);
4849 /* Check other admissibility requirements. */
4850 if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) {
4851 output_mirrors(ctx, vlan, in_bundle, 0);
4855 /* Learn source MAC. */
4856 if (ctx->may_learn) {
4857 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
4860 /* Determine output bundle. */
4861 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
4864 out_bundle = mac->port.p;
4865 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
4866 /* If we are revalidating but don't have a learning entry then eject
4867 * the flow. Installing a flow that floods packets opens up a window
4868 * of time where we could learn from a packet reflected on a bond and
4869 * blackhole packets before the learning table is updated to reflect
4870 * the correct port. */
4871 ctx->may_set_up_flow = false;
4874 out_bundle = OFBUNDLE_FLOOD;
4877 /* Don't send packets out their input bundles. */
4878 if (in_bundle != out_bundle) {
4879 dst_mirrors = compose_dsts(ctx, vlan, in_bundle, out_bundle);
4881 output_mirrors(ctx, vlan, in_bundle, dst_mirrors);
4884 /* Optimized flow revalidation.
4886 * It's a difficult problem, in general, to tell which facets need to have
4887 * their actions recalculated whenever the OpenFlow flow table changes. We
4888 * don't try to solve that general problem: for most kinds of OpenFlow flow
4889 * table changes, we recalculate the actions for every facet. This is
4890 * relatively expensive, but it's good enough if the OpenFlow flow table
4891 * doesn't change very often.
4893 * However, we can expect one particular kind of OpenFlow flow table change to
4894 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
4895 * of CPU on revalidating every facet whenever MAC learning modifies the flow
4896 * table, we add a special case that applies to flow tables in which every rule
4897 * has the same form (that is, the same wildcards), except that the table is
4898 * also allowed to have a single "catch-all" flow that matches all packets. We
4899 * optimize this case by tagging all of the facets that resubmit into the table
4900 * and invalidating the same tag whenever a flow changes in that table. The
4901 * end result is that we revalidate just the facets that need it (and sometimes
4902 * a few more, but not all of the facets or even all of the facets that
4903 * resubmit to the table modified by MAC learning). */
4905 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
4906 * into an OpenFlow table with the given 'basis'. */
4908 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
4911 if (flow_wildcards_is_catchall(wc)) {
4914 struct flow tag_flow = *flow;
4915 flow_zero_wildcards(&tag_flow, wc);
4916 return tag_create_deterministic(flow_hash(&tag_flow, secret));
4920 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
4921 * taggability of that table.
4923 * This function must be called after *each* change to a flow table. If you
4924 * skip calling it on some changes then the pointer comparisons at the end can
4925 * be invalid if you get unlucky. For example, if a flow removal causes a
4926 * cls_table to be destroyed and then a flow insertion causes a cls_table with
4927 * different wildcards to be created with the same address, then this function
4928 * will incorrectly skip revalidation. */
4930 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
4932 struct table_dpif *table = &ofproto->tables[table_id];
4933 const struct classifier *cls = &ofproto->up.tables[table_id];
4934 struct cls_table *catchall, *other;
4935 struct cls_table *t;
4937 catchall = other = NULL;
4939 switch (hmap_count(&cls->tables)) {
4941 /* We could tag this OpenFlow table but it would make the logic a
4942 * little harder and it's a corner case that doesn't seem worth it
4948 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
4949 if (cls_table_is_catchall(t)) {
4951 } else if (!other) {
4954 /* Indicate that we can't tag this by setting both tables to
4955 * NULL. (We know that 'catchall' is already NULL.) */
4962 /* Can't tag this table. */
4966 if (table->catchall_table != catchall || table->other_table != other) {
4967 table->catchall_table = catchall;
4968 table->other_table = other;
4969 ofproto->need_revalidate = true;
4973 /* Given 'rule' that has changed in some way (either it is a rule being
4974 * inserted, a rule being deleted, or a rule whose actions are being
4975 * modified), marks facets for revalidation to ensure that packets will be
4976 * forwarded correctly according to the new state of the flow table.
4978 * This function must be called after *each* change to a flow table. See
4979 * the comment on table_update_taggable() for more information. */
4981 rule_invalidate(const struct rule_dpif *rule)
4983 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4985 table_update_taggable(ofproto, rule->up.table_id);
4987 if (!ofproto->need_revalidate) {
4988 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
4990 if (table->other_table && rule->tag) {
4991 tag_set_add(&ofproto->revalidate_set, rule->tag);
4993 ofproto->need_revalidate = true;
4999 set_frag_handling(struct ofproto *ofproto_,
5000 enum ofp_config_flags frag_handling)
5002 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5004 if (frag_handling != OFPC_FRAG_REASM) {
5005 ofproto->need_revalidate = true;
5013 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5014 const struct flow *flow,
5015 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5017 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5020 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5021 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5024 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5025 ofproto->max_ports);
5027 struct odputil_keybuf keybuf;
5028 struct action_xlate_ctx ctx;
5029 struct ofpbuf *odp_actions;
5032 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5033 odp_flow_key_from_flow(&key, flow);
5035 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
5036 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5037 dpif_execute(ofproto->dpif, key.data, key.size,
5038 odp_actions->data, odp_actions->size, packet);
5039 ofpbuf_delete(odp_actions);
5045 get_netflow_ids(const struct ofproto *ofproto_,
5046 uint8_t *engine_type, uint8_t *engine_id)
5048 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5050 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5053 static struct ofproto_dpif *
5054 ofproto_dpif_lookup(const char *name)
5056 struct ofproto *ofproto = ofproto_lookup(name);
5057 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
5058 ? ofproto_dpif_cast(ofproto)
5063 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5064 const char *args, void *aux OVS_UNUSED)
5066 const struct ofproto_dpif *ofproto;
5068 ofproto = ofproto_dpif_lookup(args);
5070 unixctl_command_reply(conn, 501, "no such bridge");
5073 mac_learning_flush(ofproto->ml);
5075 unixctl_command_reply(conn, 200, "table successfully flushed");
5079 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5080 const char *args, void *aux OVS_UNUSED)
5082 struct ds ds = DS_EMPTY_INITIALIZER;
5083 const struct ofproto_dpif *ofproto;
5084 const struct mac_entry *e;
5086 ofproto = ofproto_dpif_lookup(args);
5088 unixctl_command_reply(conn, 501, "no such bridge");
5092 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5093 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5094 struct ofbundle *bundle = e->port.p;
5095 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5096 ofbundle_get_a_port(bundle)->odp_port,
5097 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5099 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5103 struct ofproto_trace {
5104 struct action_xlate_ctx ctx;
5110 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5111 const struct rule_dpif *rule)
5113 ds_put_char_multiple(result, '\t', level);
5115 ds_put_cstr(result, "No match\n");
5119 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5120 table_id, ntohll(rule->up.flow_cookie));
5121 cls_rule_format(&rule->up.cr, result);
5122 ds_put_char(result, '\n');
5124 ds_put_char_multiple(result, '\t', level);
5125 ds_put_cstr(result, "OpenFlow ");
5126 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5127 ds_put_char(result, '\n');
5131 trace_format_flow(struct ds *result, int level, const char *title,
5132 struct ofproto_trace *trace)
5134 ds_put_char_multiple(result, '\t', level);
5135 ds_put_format(result, "%s: ", title);
5136 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5137 ds_put_cstr(result, "unchanged");
5139 flow_format(result, &trace->ctx.flow);
5140 trace->flow = trace->ctx.flow;
5142 ds_put_char(result, '\n');
5146 trace_format_regs(struct ds *result, int level, const char *title,
5147 struct ofproto_trace *trace)
5151 ds_put_char_multiple(result, '\t', level);
5152 ds_put_format(result, "%s:", title);
5153 for (i = 0; i < FLOW_N_REGS; i++) {
5154 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5156 ds_put_char(result, '\n');
5160 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5162 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5163 struct ds *result = trace->result;
5165 ds_put_char(result, '\n');
5166 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5167 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5168 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5172 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5173 void *aux OVS_UNUSED)
5175 char *dpname, *arg1, *arg2, *arg3, *arg4;
5176 char *args = xstrdup(args_);
5177 char *save_ptr = NULL;
5178 struct ofproto_dpif *ofproto;
5179 struct ofpbuf odp_key;
5180 struct ofpbuf *packet;
5181 struct rule_dpif *rule;
5187 ofpbuf_init(&odp_key, 0);
5190 dpname = strtok_r(args, " ", &save_ptr);
5191 arg1 = strtok_r(NULL, " ", &save_ptr);
5192 arg2 = strtok_r(NULL, " ", &save_ptr);
5193 arg3 = strtok_r(NULL, " ", &save_ptr);
5194 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5195 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5196 /* ofproto/trace dpname flow [-generate] */
5199 /* Convert string to datapath key. */
5200 ofpbuf_init(&odp_key, 0);
5201 error = odp_flow_key_from_string(arg1, NULL, &odp_key);
5203 unixctl_command_reply(conn, 501, "Bad flow syntax");
5207 /* Convert odp_key to flow. */
5208 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
5210 unixctl_command_reply(conn, 501, "Invalid flow");
5214 /* Generate a packet, if requested. */
5216 packet = ofpbuf_new(0);
5217 flow_compose(packet, &flow);
5219 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5220 /* ofproto/trace dpname priority tun_id in_port packet */
5225 priority = atoi(arg1);
5226 tun_id = htonll(strtoull(arg2, NULL, 0));
5227 in_port = ofp_port_to_odp_port(atoi(arg3));
5229 packet = ofpbuf_new(strlen(args) / 2);
5230 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5231 arg4 += strspn(arg4, " ");
5232 if (*arg4 != '\0') {
5233 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5236 if (packet->size < ETH_HEADER_LEN) {
5237 unixctl_command_reply(conn, 501,
5238 "Packet data too short for Ethernet");
5242 ds_put_cstr(&result, "Packet: ");
5243 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5244 ds_put_cstr(&result, s);
5247 flow_extract(packet, priority, tun_id, in_port, &flow);
5249 unixctl_command_reply(conn, 501, "Bad command syntax");
5253 ofproto = ofproto_dpif_lookup(dpname);
5255 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5260 ds_put_cstr(&result, "Flow: ");
5261 flow_format(&result, &flow);
5262 ds_put_char(&result, '\n');
5264 rule = rule_dpif_lookup(ofproto, &flow, 0);
5265 trace_format_rule(&result, 0, 0, rule);
5267 struct ofproto_trace trace;
5268 struct ofpbuf *odp_actions;
5270 trace.result = &result;
5272 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
5273 trace.ctx.resubmit_hook = trace_resubmit;
5274 odp_actions = xlate_actions(&trace.ctx,
5275 rule->up.actions, rule->up.n_actions);
5277 ds_put_char(&result, '\n');
5278 trace_format_flow(&result, 0, "Final flow", &trace);
5279 ds_put_cstr(&result, "Datapath actions: ");
5280 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5281 ofpbuf_delete(odp_actions);
5283 if (!trace.ctx.may_set_up_flow) {
5285 ds_put_cstr(&result, "\nThis flow is not cachable.");
5287 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5288 "for complete actions, please supply a packet.");
5293 unixctl_command_reply(conn, 200, ds_cstr(&result));
5296 ds_destroy(&result);
5297 ofpbuf_delete(packet);
5298 ofpbuf_uninit(&odp_key);
5303 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5304 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5307 unixctl_command_reply(conn, 200, NULL);
5311 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5312 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5315 unixctl_command_reply(conn, 200, NULL);
5319 ofproto_dpif_unixctl_init(void)
5321 static bool registered;
5327 unixctl_command_register("ofproto/trace",
5328 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5329 ofproto_unixctl_trace, NULL);
5330 unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
5332 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5334 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5335 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5338 const struct ofproto_class ofproto_dpif_class = {
5365 port_is_lacp_current,
5366 NULL, /* rule_choose_table */
5373 rule_modify_actions,
5381 get_cfm_remote_mpids,
5385 get_stp_port_status,
5390 is_mirror_output_bundle,
5391 forward_bpdu_changed,