2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any facet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 #define MAX_MIRRORS 32
110 typedef uint32_t mirror_mask_t;
111 #define MIRROR_MASK_C(X) UINT32_C(X)
112 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
114 struct ofproto_dpif *ofproto; /* Owning ofproto. */
115 size_t idx; /* In ofproto's "mirrors" array. */
116 void *aux; /* Key supplied by ofproto's client. */
117 char *name; /* Identifier for log messages. */
119 /* Selection criteria. */
120 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
121 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
122 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
124 /* Output (mutually exclusive). */
125 struct ofbundle *out; /* Output port or NULL. */
126 int out_vlan; /* Output VLAN or -1. */
129 static void mirror_destroy(struct ofmirror *);
131 /* A group of one or more OpenFlow ports. */
132 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
140 struct list ports; /* Contains "struct ofport"s. */
141 enum port_vlan_mode vlan_mode; /* VLAN mode */
142 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
143 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
144 * NULL if all VLANs are trunked. */
145 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
146 struct bond *bond; /* Nonnull iff more than one port. */
149 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
151 /* Port mirroring info. */
152 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
153 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
154 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
157 static void bundle_remove(struct ofport *);
158 static void bundle_update(struct ofbundle *);
159 static void bundle_destroy(struct ofbundle *);
160 static void bundle_del_port(struct ofport_dpif *);
161 static void bundle_run(struct ofbundle *);
162 static void bundle_wait(struct ofbundle *);
164 static void stp_run(struct ofproto_dpif *ofproto);
165 static void stp_wait(struct ofproto_dpif *ofproto);
167 struct action_xlate_ctx {
168 /* action_xlate_ctx_init() initializes these members. */
171 struct ofproto_dpif *ofproto;
173 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
174 * this flow when actions change header fields. */
177 /* The packet corresponding to 'flow', or a null pointer if we are
178 * revalidating without a packet to refer to. */
179 const struct ofpbuf *packet;
181 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
182 * want to execute them if we are actually processing a packet, or if we
183 * are accounting for packets that the datapath has processed, but not if
184 * we are just revalidating. */
187 /* If nonnull, called just before executing a resubmit action.
189 * This is normally null so the client has to set it manually after
190 * calling action_xlate_ctx_init(). */
191 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
193 /* xlate_actions() initializes and uses these members. The client might want
194 * to look at them after it returns. */
196 struct ofpbuf *odp_actions; /* Datapath actions. */
197 tag_type tags; /* Tags associated with actions. */
198 bool may_set_up_flow; /* True ordinarily; false if the actions must
199 * be reassessed for every packet. */
200 bool has_learn; /* Actions include NXAST_LEARN? */
201 bool has_normal; /* Actions output to OFPP_NORMAL? */
202 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
204 /* xlate_actions() initializes and uses these members, but the client has no
205 * reason to look at them. */
207 int recurse; /* Recursion level, via xlate_table_action. */
208 struct flow base_flow; /* Flow at the last commit. */
209 uint32_t original_priority; /* Priority when packet arrived. */
210 uint8_t table_id; /* OpenFlow table ID where flow was found. */
211 uint32_t sflow_n_outputs; /* Number of output ports. */
212 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
213 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
214 bool exit; /* No further actions should be processed. */
217 static void action_xlate_ctx_init(struct action_xlate_ctx *,
218 struct ofproto_dpif *, const struct flow *,
219 const struct ofpbuf *);
220 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
221 const union ofp_action *in, size_t n_in);
223 /* An exact-match instantiation of an OpenFlow flow. */
225 long long int used; /* Time last used; time created if not used. */
229 * - Do include packets and bytes sent "by hand", e.g. with
232 * - Do include packets and bytes that were obtained from the datapath
233 * when its statistics were reset (e.g. dpif_flow_put() with
234 * DPIF_FP_ZERO_STATS).
236 uint64_t packet_count; /* Number of packets received. */
237 uint64_t byte_count; /* Number of bytes received. */
239 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
240 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
242 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
243 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
244 long long int rs_used; /* Used time pushed to resubmit children. */
246 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
248 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
249 struct list list_node; /* In owning rule's 'facets' list. */
250 struct rule_dpif *rule; /* Owning rule. */
251 struct flow flow; /* Exact-match flow. */
252 bool installed; /* Installed in datapath? */
253 bool may_install; /* True ordinarily; false if actions must
254 * be reassessed for every packet. */
255 bool has_learn; /* Actions include NXAST_LEARN? */
256 bool has_normal; /* Actions output to OFPP_NORMAL? */
257 size_t actions_len; /* Number of bytes in actions[]. */
258 struct nlattr *actions; /* Datapath actions. */
259 tag_type tags; /* Tags. */
260 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
263 static struct facet *facet_create(struct rule_dpif *, const struct flow *);
264 static void facet_remove(struct ofproto_dpif *, struct facet *);
265 static void facet_free(struct facet *);
267 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
268 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
269 const struct flow *);
270 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
272 static bool execute_controller_action(struct ofproto_dpif *,
274 const struct nlattr *odp_actions,
276 struct ofpbuf *packet);
277 static void facet_execute(struct ofproto_dpif *, struct facet *,
278 struct ofpbuf *packet);
280 static int facet_put__(struct ofproto_dpif *, struct facet *,
281 const struct nlattr *actions, size_t actions_len,
282 struct dpif_flow_stats *);
283 static void facet_install(struct ofproto_dpif *, struct facet *,
285 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
286 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
288 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
289 const struct ofpbuf *packet);
290 static void facet_update_time(struct ofproto_dpif *, struct facet *,
292 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
293 const struct dpif_flow_stats *);
294 static void facet_reset_counters(struct facet *);
295 static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
296 static void facet_push_stats(struct facet *);
297 static void facet_account(struct ofproto_dpif *, struct facet *);
299 static bool facet_is_controller_flow(struct facet *);
301 static void flow_push_stats(const struct rule_dpif *,
302 struct flow *, uint64_t packets, uint64_t bytes,
305 static uint32_t rule_calculate_tag(const struct flow *,
306 const struct flow_wildcards *,
308 static void rule_invalidate(const struct rule_dpif *);
314 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
315 struct list bundle_node; /* In struct ofbundle's "ports" list. */
316 struct cfm *cfm; /* Connectivity Fault Management, if any. */
317 tag_type tag; /* Tag associated with this port. */
318 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
319 bool may_enable; /* May be enabled in bonds. */
321 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
322 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
323 long long int stp_state_entered;
326 static struct ofport_dpif *
327 ofport_dpif_cast(const struct ofport *ofport)
329 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
330 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
333 static void port_run(struct ofport_dpif *);
334 static void port_wait(struct ofport_dpif *);
335 static int set_cfm(struct ofport *, const struct cfm_settings *);
337 struct dpif_completion {
338 struct list list_node;
339 struct ofoperation *op;
342 /* Extra information about a classifier table.
343 * Currently used just for optimized flow revalidation. */
345 /* If either of these is nonnull, then this table has a form that allows
346 * flows to be tagged to avoid revalidating most flows for the most common
347 * kinds of flow table changes. */
348 struct cls_table *catchall_table; /* Table that wildcards all fields. */
349 struct cls_table *other_table; /* Table with any other wildcard set. */
350 uint32_t basis; /* Keeps each table's tags separate. */
353 struct ofproto_dpif {
362 struct netflow *netflow;
363 struct dpif_sflow *sflow;
364 struct hmap bundles; /* Contains "struct ofbundle"s. */
365 struct mac_learning *ml;
366 struct ofmirror *mirrors[MAX_MIRRORS];
367 bool has_bonded_bundles;
370 struct timer next_expiration;
376 struct table_dpif tables[N_TABLES];
377 bool need_revalidate;
378 struct tag_set revalidate_set;
380 /* Support for debugging async flow mods. */
381 struct list completions;
383 bool has_bundle_action; /* True when the first bundle action appears. */
387 long long int stp_last_tick;
390 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
391 * for debugging the asynchronous flow_mod implementation.) */
394 static void ofproto_dpif_unixctl_init(void);
396 static struct ofproto_dpif *
397 ofproto_dpif_cast(const struct ofproto *ofproto)
399 assert(ofproto->ofproto_class == &ofproto_dpif_class);
400 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
403 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
405 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
408 /* Packet processing. */
409 static void update_learning_table(struct ofproto_dpif *,
410 const struct flow *, int vlan,
412 static bool is_admissible(struct ofproto_dpif *, const struct flow *,
413 bool have_packet, tag_type *, int *vlanp,
414 struct ofbundle **in_bundlep);
417 #define FLOW_MISS_MAX_BATCH 50
418 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
419 static void handle_miss_upcalls(struct ofproto_dpif *,
420 struct dpif_upcall *, size_t n);
422 /* Flow expiration. */
423 static int expire(struct ofproto_dpif *);
426 static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
427 const struct ofpbuf *packet);
429 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
430 const struct flow *, uint32_t odp_port);
431 /* Global variables. */
432 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
434 /* Factory functions. */
437 enumerate_types(struct sset *types)
439 dp_enumerate_types(types);
443 enumerate_names(const char *type, struct sset *names)
445 return dp_enumerate_names(type, names);
449 del(const char *type, const char *name)
454 error = dpif_open(name, type, &dpif);
456 error = dpif_delete(dpif);
462 /* Basic life-cycle. */
464 static struct ofproto *
467 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
472 dealloc(struct ofproto *ofproto_)
474 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
479 construct(struct ofproto *ofproto_, int *n_tablesp)
481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
482 const char *name = ofproto->up.name;
486 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
488 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
492 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
493 ofproto->n_matches = 0;
495 dpif_flow_flush(ofproto->dpif);
496 dpif_recv_purge(ofproto->dpif);
498 error = dpif_recv_set_mask(ofproto->dpif,
499 ((1u << DPIF_UC_MISS) |
500 (1u << DPIF_UC_ACTION)));
502 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
503 dpif_close(ofproto->dpif);
507 ofproto->netflow = NULL;
508 ofproto->sflow = NULL;
510 hmap_init(&ofproto->bundles);
511 ofproto->ml = mac_learning_create();
512 for (i = 0; i < MAX_MIRRORS; i++) {
513 ofproto->mirrors[i] = NULL;
515 ofproto->has_bonded_bundles = false;
517 timer_set_duration(&ofproto->next_expiration, 1000);
519 hmap_init(&ofproto->facets);
521 for (i = 0; i < N_TABLES; i++) {
522 struct table_dpif *table = &ofproto->tables[i];
524 table->catchall_table = NULL;
525 table->other_table = NULL;
526 table->basis = random_uint32();
528 ofproto->need_revalidate = false;
529 tag_set_init(&ofproto->revalidate_set);
531 list_init(&ofproto->completions);
533 ofproto_dpif_unixctl_init();
535 ofproto->has_bundle_action = false;
537 *n_tablesp = N_TABLES;
542 complete_operations(struct ofproto_dpif *ofproto)
544 struct dpif_completion *c, *next;
546 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
547 ofoperation_complete(c->op, 0);
548 list_remove(&c->list_node);
554 destruct(struct ofproto *ofproto_)
556 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
557 struct rule_dpif *rule, *next_rule;
558 struct classifier *table;
561 complete_operations(ofproto);
563 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
564 struct cls_cursor cursor;
566 cls_cursor_init(&cursor, table, NULL);
567 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
568 ofproto_rule_destroy(&rule->up);
572 for (i = 0; i < MAX_MIRRORS; i++) {
573 mirror_destroy(ofproto->mirrors[i]);
576 netflow_destroy(ofproto->netflow);
577 dpif_sflow_destroy(ofproto->sflow);
578 hmap_destroy(&ofproto->bundles);
579 mac_learning_destroy(ofproto->ml);
581 hmap_destroy(&ofproto->facets);
583 dpif_close(ofproto->dpif);
587 run(struct ofproto *ofproto_)
589 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
590 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
591 struct ofport_dpif *ofport;
592 struct ofbundle *bundle;
597 complete_operations(ofproto);
599 dpif_run(ofproto->dpif);
602 for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) {
603 struct dpif_upcall *upcall = &misses[n_misses];
606 error = dpif_recv(ofproto->dpif, upcall);
608 if (error == ENODEV && n_misses == 0) {
614 if (upcall->type == DPIF_UC_MISS) {
615 /* Handle it later. */
618 handle_upcall(ofproto, upcall);
622 handle_miss_upcalls(ofproto, misses, n_misses);
624 if (timer_expired(&ofproto->next_expiration)) {
625 int delay = expire(ofproto);
626 timer_set_duration(&ofproto->next_expiration, delay);
629 if (ofproto->netflow) {
630 netflow_run(ofproto->netflow);
632 if (ofproto->sflow) {
633 dpif_sflow_run(ofproto->sflow);
636 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
639 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
644 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
646 /* Now revalidate if there's anything to do. */
647 if (ofproto->need_revalidate
648 || !tag_set_is_empty(&ofproto->revalidate_set)) {
649 struct tag_set revalidate_set = ofproto->revalidate_set;
650 bool revalidate_all = ofproto->need_revalidate;
651 struct facet *facet, *next;
653 /* Clear the revalidation flags. */
654 tag_set_init(&ofproto->revalidate_set);
655 ofproto->need_revalidate = false;
657 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
659 || tag_set_intersects(&revalidate_set, facet->tags)) {
660 facet_revalidate(ofproto, facet);
669 wait(struct ofproto *ofproto_)
671 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
672 struct ofport_dpif *ofport;
673 struct ofbundle *bundle;
675 if (!clogged && !list_is_empty(&ofproto->completions)) {
676 poll_immediate_wake();
679 dpif_wait(ofproto->dpif);
680 dpif_recv_wait(ofproto->dpif);
681 if (ofproto->sflow) {
682 dpif_sflow_wait(ofproto->sflow);
684 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
685 poll_immediate_wake();
687 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
690 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
693 mac_learning_wait(ofproto->ml);
695 if (ofproto->need_revalidate) {
696 /* Shouldn't happen, but if it does just go around again. */
697 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
698 poll_immediate_wake();
700 timer_wait(&ofproto->next_expiration);
705 flush(struct ofproto *ofproto_)
707 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
708 struct facet *facet, *next_facet;
710 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
711 /* Mark the facet as not installed so that facet_remove() doesn't
712 * bother trying to uninstall it. There is no point in uninstalling it
713 * individually since we are about to blow away all the facets with
714 * dpif_flow_flush(). */
715 facet->installed = false;
716 facet->dp_packet_count = 0;
717 facet->dp_byte_count = 0;
718 facet_remove(ofproto, facet);
720 dpif_flow_flush(ofproto->dpif);
724 get_features(struct ofproto *ofproto_ OVS_UNUSED,
725 bool *arp_match_ip, uint32_t *actions)
727 *arp_match_ip = true;
728 *actions = ((1u << OFPAT_OUTPUT) |
729 (1u << OFPAT_SET_VLAN_VID) |
730 (1u << OFPAT_SET_VLAN_PCP) |
731 (1u << OFPAT_STRIP_VLAN) |
732 (1u << OFPAT_SET_DL_SRC) |
733 (1u << OFPAT_SET_DL_DST) |
734 (1u << OFPAT_SET_NW_SRC) |
735 (1u << OFPAT_SET_NW_DST) |
736 (1u << OFPAT_SET_NW_TOS) |
737 (1u << OFPAT_SET_TP_SRC) |
738 (1u << OFPAT_SET_TP_DST) |
739 (1u << OFPAT_ENQUEUE));
743 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
746 struct dpif_dp_stats s;
748 strcpy(ots->name, "classifier");
750 dpif_get_dp_stats(ofproto->dpif, &s);
751 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
752 put_32aligned_be64(&ots->matched_count,
753 htonll(s.n_hit + ofproto->n_matches));
757 set_netflow(struct ofproto *ofproto_,
758 const struct netflow_options *netflow_options)
760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
762 if (netflow_options) {
763 if (!ofproto->netflow) {
764 ofproto->netflow = netflow_create();
766 return netflow_set_options(ofproto->netflow, netflow_options);
768 netflow_destroy(ofproto->netflow);
769 ofproto->netflow = NULL;
774 static struct ofport *
777 struct ofport_dpif *port = xmalloc(sizeof *port);
782 port_dealloc(struct ofport *port_)
784 struct ofport_dpif *port = ofport_dpif_cast(port_);
789 port_construct(struct ofport *port_)
791 struct ofport_dpif *port = ofport_dpif_cast(port_);
792 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
794 ofproto->need_revalidate = true;
795 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
798 port->tag = tag_create_random();
799 port->may_enable = true;
800 port->stp_port = NULL;
801 port->stp_state = STP_DISABLED;
803 if (ofproto->sflow) {
804 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
805 netdev_get_name(port->up.netdev));
812 port_destruct(struct ofport *port_)
814 struct ofport_dpif *port = ofport_dpif_cast(port_);
815 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
817 ofproto->need_revalidate = true;
818 bundle_remove(port_);
819 set_cfm(port_, NULL);
820 if (ofproto->sflow) {
821 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
826 port_modified(struct ofport *port_)
828 struct ofport_dpif *port = ofport_dpif_cast(port_);
830 if (port->bundle && port->bundle->bond) {
831 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
836 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
838 struct ofport_dpif *port = ofport_dpif_cast(port_);
839 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
840 ovs_be32 changed = old_config ^ port->up.opp.config;
842 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
843 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
844 ofproto->need_revalidate = true;
846 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
847 bundle_update(port->bundle);
853 set_sflow(struct ofproto *ofproto_,
854 const struct ofproto_sflow_options *sflow_options)
856 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
857 struct dpif_sflow *ds = ofproto->sflow;
861 struct ofport_dpif *ofport;
863 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
864 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
865 dpif_sflow_add_port(ds, ofport->odp_port,
866 netdev_get_name(ofport->up.netdev));
868 ofproto->need_revalidate = true;
870 dpif_sflow_set_options(ds, sflow_options);
873 dpif_sflow_destroy(ds);
874 ofproto->need_revalidate = true;
875 ofproto->sflow = NULL;
882 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
884 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
891 struct ofproto_dpif *ofproto;
893 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
894 ofproto->need_revalidate = true;
895 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
898 if (cfm_configure(ofport->cfm, s)) {
904 cfm_destroy(ofport->cfm);
910 get_cfm_fault(const struct ofport *ofport_)
912 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
914 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
918 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
921 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
924 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
934 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
936 struct ofproto_dpif *ofproto = ofproto_;
937 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
938 struct ofport_dpif *ofport;
940 ofport = stp_port_get_aux(sp);
942 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
943 ofproto->up.name, port_num);
945 struct eth_header *eth = pkt->l2;
947 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
948 if (eth_addr_is_zero(eth->eth_src)) {
949 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
950 "with unknown MAC", ofproto->up.name, port_num);
952 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
953 ofport->odp_port, pkt);
959 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
961 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
963 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
965 /* Only revalidate flows if the configuration changed. */
966 if (!s != !ofproto->stp) {
967 ofproto->need_revalidate = true;
972 ofproto->stp = stp_create(ofproto_->name, s->system_id,
973 send_bpdu_cb, ofproto);
974 ofproto->stp_last_tick = time_msec();
977 stp_set_bridge_id(ofproto->stp, s->system_id);
978 stp_set_bridge_priority(ofproto->stp, s->priority);
979 stp_set_hello_time(ofproto->stp, s->hello_time);
980 stp_set_max_age(ofproto->stp, s->max_age);
981 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
983 stp_destroy(ofproto->stp);
991 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
993 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
997 s->bridge_id = stp_get_bridge_id(ofproto->stp);
998 s->designated_root = stp_get_designated_root(ofproto->stp);
999 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1008 update_stp_port_state(struct ofport_dpif *ofport)
1010 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1011 enum stp_state state;
1013 /* Figure out new state. */
1014 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1018 if (ofport->stp_state != state) {
1022 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1023 netdev_get_name(ofport->up.netdev),
1024 stp_state_name(ofport->stp_state),
1025 stp_state_name(state));
1026 if (stp_learn_in_state(ofport->stp_state)
1027 != stp_learn_in_state(state)) {
1028 /* xxx Learning action flows should also be flushed. */
1029 mac_learning_flush(ofproto->ml);
1031 fwd_change = stp_forward_in_state(ofport->stp_state)
1032 != stp_forward_in_state(state);
1034 ofproto->need_revalidate = true;
1035 ofport->stp_state = state;
1036 ofport->stp_state_entered = time_msec();
1038 if (fwd_change && ofport->bundle) {
1039 bundle_update(ofport->bundle);
1042 /* Update the STP state bits in the OpenFlow port description. */
1043 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1044 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1045 : state == STP_LEARNING ? OFPPS_STP_LEARN
1046 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1047 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1049 ofproto_port_set_state(&ofport->up, of_state);
1053 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1054 * caller is responsible for assigning STP port numbers and ensuring
1055 * there are no duplicates. */
1057 set_stp_port(struct ofport *ofport_,
1058 const struct ofproto_port_stp_settings *s)
1060 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1061 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1062 struct stp_port *sp = ofport->stp_port;
1064 if (!s || !s->enable) {
1066 ofport->stp_port = NULL;
1067 stp_port_disable(sp);
1068 update_stp_port_state(ofport);
1071 } else if (sp && stp_port_no(sp) != s->port_num
1072 && ofport == stp_port_get_aux(sp)) {
1073 /* The port-id changed, so disable the old one if it's not
1074 * already in use by another port. */
1075 stp_port_disable(sp);
1078 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1079 stp_port_enable(sp);
1081 stp_port_set_aux(sp, ofport);
1082 stp_port_set_priority(sp, s->priority);
1083 stp_port_set_path_cost(sp, s->path_cost);
1085 update_stp_port_state(ofport);
1091 get_stp_port_status(struct ofport *ofport_,
1092 struct ofproto_port_stp_status *s)
1094 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1095 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1096 struct stp_port *sp = ofport->stp_port;
1098 if (!ofproto->stp || !sp) {
1104 s->port_id = stp_port_get_id(sp);
1105 s->state = stp_port_get_state(sp);
1106 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1107 s->role = stp_port_get_role(sp);
1108 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1114 stp_run(struct ofproto_dpif *ofproto)
1117 long long int now = time_msec();
1118 long long int elapsed = now - ofproto->stp_last_tick;
1119 struct stp_port *sp;
1122 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1123 ofproto->stp_last_tick = now;
1125 while (stp_get_changed_port(ofproto->stp, &sp)) {
1126 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1129 update_stp_port_state(ofport);
1136 stp_wait(struct ofproto_dpif *ofproto)
1139 poll_timer_wait(1000);
1143 /* Returns true if STP should process 'flow'. */
1145 stp_should_process_flow(const struct flow *flow)
1147 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1151 stp_process_packet(const struct ofport_dpif *ofport,
1152 const struct ofpbuf *packet)
1154 struct ofpbuf payload = *packet;
1155 struct eth_header *eth = payload.data;
1156 struct stp_port *sp = ofport->stp_port;
1158 /* Sink packets on ports that have STP disabled when the bridge has
1160 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1164 /* Trim off padding on payload. */
1165 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1166 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1169 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1170 stp_received_bpdu(sp, payload.data, payload.size);
1176 /* Expires all MAC learning entries associated with 'port' and forces ofproto
1177 * to revalidate every flow. */
1179 bundle_flush_macs(struct ofbundle *bundle)
1181 struct ofproto_dpif *ofproto = bundle->ofproto;
1182 struct mac_learning *ml = ofproto->ml;
1183 struct mac_entry *mac, *next_mac;
1185 ofproto->need_revalidate = true;
1186 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1187 if (mac->port.p == bundle) {
1188 mac_learning_expire(ml, mac);
1193 static struct ofbundle *
1194 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1196 struct ofbundle *bundle;
1198 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1199 &ofproto->bundles) {
1200 if (bundle->aux == aux) {
1207 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1208 * ones that are found to 'bundles'. */
1210 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1211 void **auxes, size_t n_auxes,
1212 struct hmapx *bundles)
1216 hmapx_init(bundles);
1217 for (i = 0; i < n_auxes; i++) {
1218 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1220 hmapx_add(bundles, bundle);
1226 bundle_update(struct ofbundle *bundle)
1228 struct ofport_dpif *port;
1230 bundle->floodable = true;
1231 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1232 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1233 || !stp_forward_in_state(port->stp_state)) {
1234 bundle->floodable = false;
1241 bundle_del_port(struct ofport_dpif *port)
1243 struct ofbundle *bundle = port->bundle;
1245 bundle->ofproto->need_revalidate = true;
1247 list_remove(&port->bundle_node);
1248 port->bundle = NULL;
1251 lacp_slave_unregister(bundle->lacp, port);
1254 bond_slave_unregister(bundle->bond, port);
1257 bundle_update(bundle);
1261 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1262 struct lacp_slave_settings *lacp,
1263 uint32_t bond_stable_id)
1265 struct ofport_dpif *port;
1267 port = get_ofp_port(bundle->ofproto, ofp_port);
1272 if (port->bundle != bundle) {
1273 bundle->ofproto->need_revalidate = true;
1275 bundle_del_port(port);
1278 port->bundle = bundle;
1279 list_push_back(&bundle->ports, &port->bundle_node);
1280 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1281 || !stp_forward_in_state(port->stp_state)) {
1282 bundle->floodable = false;
1286 port->bundle->ofproto->need_revalidate = true;
1287 lacp_slave_register(bundle->lacp, port, lacp);
1290 port->bond_stable_id = bond_stable_id;
1296 bundle_destroy(struct ofbundle *bundle)
1298 struct ofproto_dpif *ofproto;
1299 struct ofport_dpif *port, *next_port;
1306 ofproto = bundle->ofproto;
1307 for (i = 0; i < MAX_MIRRORS; i++) {
1308 struct ofmirror *m = ofproto->mirrors[i];
1310 if (m->out == bundle) {
1312 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1313 || hmapx_find_and_delete(&m->dsts, bundle)) {
1314 ofproto->need_revalidate = true;
1319 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1320 bundle_del_port(port);
1323 bundle_flush_macs(bundle);
1324 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1326 free(bundle->trunks);
1327 lacp_destroy(bundle->lacp);
1328 bond_destroy(bundle->bond);
1333 bundle_set(struct ofproto *ofproto_, void *aux,
1334 const struct ofproto_bundle_settings *s)
1336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1337 bool need_flush = false;
1338 struct ofport_dpif *port;
1339 struct ofbundle *bundle;
1340 unsigned long *trunks;
1346 bundle_destroy(bundle_lookup(ofproto, aux));
1350 assert(s->n_slaves == 1 || s->bond != NULL);
1351 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1353 bundle = bundle_lookup(ofproto, aux);
1355 bundle = xmalloc(sizeof *bundle);
1357 bundle->ofproto = ofproto;
1358 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1359 hash_pointer(aux, 0));
1361 bundle->name = NULL;
1363 list_init(&bundle->ports);
1364 bundle->vlan_mode = PORT_VLAN_TRUNK;
1366 bundle->trunks = NULL;
1367 bundle->lacp = NULL;
1368 bundle->bond = NULL;
1370 bundle->floodable = true;
1372 bundle->src_mirrors = 0;
1373 bundle->dst_mirrors = 0;
1374 bundle->mirror_out = 0;
1377 if (!bundle->name || strcmp(s->name, bundle->name)) {
1379 bundle->name = xstrdup(s->name);
1384 if (!bundle->lacp) {
1385 ofproto->need_revalidate = true;
1386 bundle->lacp = lacp_create();
1388 lacp_configure(bundle->lacp, s->lacp);
1390 lacp_destroy(bundle->lacp);
1391 bundle->lacp = NULL;
1394 /* Update set of ports. */
1396 for (i = 0; i < s->n_slaves; i++) {
1397 if (!bundle_add_port(bundle, s->slaves[i],
1398 s->lacp ? &s->lacp_slaves[i] : NULL,
1399 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1403 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1404 struct ofport_dpif *next_port;
1406 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1407 for (i = 0; i < s->n_slaves; i++) {
1408 if (s->slaves[i] == port->up.ofp_port) {
1413 bundle_del_port(port);
1417 assert(list_size(&bundle->ports) <= s->n_slaves);
1419 if (list_is_empty(&bundle->ports)) {
1420 bundle_destroy(bundle);
1424 /* Set VLAN tagging mode */
1425 if (s->vlan_mode != bundle->vlan_mode) {
1426 bundle->vlan_mode = s->vlan_mode;
1431 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1432 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1434 if (vlan != bundle->vlan) {
1435 bundle->vlan = vlan;
1439 /* Get trunked VLANs. */
1440 switch (s->vlan_mode) {
1441 case PORT_VLAN_ACCESS:
1445 case PORT_VLAN_TRUNK:
1446 trunks = (unsigned long *) s->trunks;
1449 case PORT_VLAN_NATIVE_UNTAGGED:
1450 case PORT_VLAN_NATIVE_TAGGED:
1451 if (vlan != 0 && (!s->trunks
1452 || !bitmap_is_set(s->trunks, vlan)
1453 || bitmap_is_set(s->trunks, 0))) {
1454 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1456 trunks = bitmap_clone(s->trunks, 4096);
1458 trunks = bitmap_allocate1(4096);
1460 bitmap_set1(trunks, vlan);
1461 bitmap_set0(trunks, 0);
1463 trunks = (unsigned long *) s->trunks;
1470 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1471 free(bundle->trunks);
1472 if (trunks == s->trunks) {
1473 bundle->trunks = vlan_bitmap_clone(trunks);
1475 bundle->trunks = trunks;
1480 if (trunks != s->trunks) {
1485 if (!list_is_short(&bundle->ports)) {
1486 bundle->ofproto->has_bonded_bundles = true;
1488 if (bond_reconfigure(bundle->bond, s->bond)) {
1489 ofproto->need_revalidate = true;
1492 bundle->bond = bond_create(s->bond);
1493 ofproto->need_revalidate = true;
1496 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1497 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1501 bond_destroy(bundle->bond);
1502 bundle->bond = NULL;
1505 /* If we changed something that would affect MAC learning, un-learn
1506 * everything on this port and force flow revalidation. */
1508 bundle_flush_macs(bundle);
1515 bundle_remove(struct ofport *port_)
1517 struct ofport_dpif *port = ofport_dpif_cast(port_);
1518 struct ofbundle *bundle = port->bundle;
1521 bundle_del_port(port);
1522 if (list_is_empty(&bundle->ports)) {
1523 bundle_destroy(bundle);
1524 } else if (list_is_short(&bundle->ports)) {
1525 bond_destroy(bundle->bond);
1526 bundle->bond = NULL;
1532 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1534 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1535 struct ofport_dpif *port = port_;
1536 uint8_t ea[ETH_ADDR_LEN];
1539 error = netdev_get_etheraddr(port->up.netdev, ea);
1541 struct ofpbuf packet;
1544 ofpbuf_init(&packet, 0);
1545 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1547 memcpy(packet_pdu, pdu, pdu_size);
1549 send_packet(ofproto_dpif_cast(port->up.ofproto), port->odp_port,
1551 ofpbuf_uninit(&packet);
1553 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1554 "%s (%s)", port->bundle->name,
1555 netdev_get_name(port->up.netdev), strerror(error));
1560 bundle_send_learning_packets(struct ofbundle *bundle)
1562 struct ofproto_dpif *ofproto = bundle->ofproto;
1563 int error, n_packets, n_errors;
1564 struct mac_entry *e;
1566 error = n_packets = n_errors = 0;
1567 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1568 if (e->port.p != bundle) {
1569 struct ofpbuf *learning_packet;
1570 struct ofport_dpif *port;
1573 learning_packet = bond_compose_learning_packet(bundle->bond, e->mac,
1576 ret = send_packet(ofproto_dpif_cast(port->up.ofproto),
1577 port->odp_port, learning_packet);
1578 ofpbuf_delete(learning_packet);
1588 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1589 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1590 "packets, last error was: %s",
1591 bundle->name, n_errors, n_packets, strerror(error));
1593 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1594 bundle->name, n_packets);
1599 bundle_run(struct ofbundle *bundle)
1602 lacp_run(bundle->lacp, send_pdu_cb);
1605 struct ofport_dpif *port;
1607 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1608 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1611 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1612 lacp_negotiated(bundle->lacp));
1613 if (bond_should_send_learning_packets(bundle->bond)) {
1614 bundle_send_learning_packets(bundle);
1620 bundle_wait(struct ofbundle *bundle)
1623 lacp_wait(bundle->lacp);
1626 bond_wait(bundle->bond);
1633 mirror_scan(struct ofproto_dpif *ofproto)
1637 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1638 if (!ofproto->mirrors[idx]) {
1645 static struct ofmirror *
1646 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1650 for (i = 0; i < MAX_MIRRORS; i++) {
1651 struct ofmirror *mirror = ofproto->mirrors[i];
1652 if (mirror && mirror->aux == aux) {
1661 mirror_set(struct ofproto *ofproto_, void *aux,
1662 const struct ofproto_mirror_settings *s)
1664 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1665 mirror_mask_t mirror_bit;
1666 struct ofbundle *bundle;
1667 struct ofmirror *mirror;
1668 struct ofbundle *out;
1669 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1670 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1673 mirror = mirror_lookup(ofproto, aux);
1675 mirror_destroy(mirror);
1681 idx = mirror_scan(ofproto);
1683 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1685 ofproto->up.name, MAX_MIRRORS, s->name);
1689 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1690 mirror->ofproto = ofproto;
1693 mirror->out_vlan = -1;
1694 mirror->name = NULL;
1697 if (!mirror->name || strcmp(s->name, mirror->name)) {
1699 mirror->name = xstrdup(s->name);
1702 /* Get the new configuration. */
1703 if (s->out_bundle) {
1704 out = bundle_lookup(ofproto, s->out_bundle);
1706 mirror_destroy(mirror);
1712 out_vlan = s->out_vlan;
1714 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1715 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1717 /* If the configuration has not changed, do nothing. */
1718 if (hmapx_equals(&srcs, &mirror->srcs)
1719 && hmapx_equals(&dsts, &mirror->dsts)
1720 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1721 && mirror->out == out
1722 && mirror->out_vlan == out_vlan)
1724 hmapx_destroy(&srcs);
1725 hmapx_destroy(&dsts);
1729 hmapx_swap(&srcs, &mirror->srcs);
1730 hmapx_destroy(&srcs);
1732 hmapx_swap(&dsts, &mirror->dsts);
1733 hmapx_destroy(&dsts);
1735 free(mirror->vlans);
1736 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1739 mirror->out_vlan = out_vlan;
1741 /* Update bundles. */
1742 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1743 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1744 if (hmapx_contains(&mirror->srcs, bundle)) {
1745 bundle->src_mirrors |= mirror_bit;
1747 bundle->src_mirrors &= ~mirror_bit;
1750 if (hmapx_contains(&mirror->dsts, bundle)) {
1751 bundle->dst_mirrors |= mirror_bit;
1753 bundle->dst_mirrors &= ~mirror_bit;
1756 if (mirror->out == bundle) {
1757 bundle->mirror_out |= mirror_bit;
1759 bundle->mirror_out &= ~mirror_bit;
1763 ofproto->need_revalidate = true;
1764 mac_learning_flush(ofproto->ml);
1770 mirror_destroy(struct ofmirror *mirror)
1772 struct ofproto_dpif *ofproto;
1773 mirror_mask_t mirror_bit;
1774 struct ofbundle *bundle;
1780 ofproto = mirror->ofproto;
1781 ofproto->need_revalidate = true;
1782 mac_learning_flush(ofproto->ml);
1784 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1785 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1786 bundle->src_mirrors &= ~mirror_bit;
1787 bundle->dst_mirrors &= ~mirror_bit;
1788 bundle->mirror_out &= ~mirror_bit;
1791 hmapx_destroy(&mirror->srcs);
1792 hmapx_destroy(&mirror->dsts);
1793 free(mirror->vlans);
1795 ofproto->mirrors[mirror->idx] = NULL;
1801 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1803 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1804 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1805 ofproto->need_revalidate = true;
1806 mac_learning_flush(ofproto->ml);
1812 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
1814 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1815 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1816 return bundle && bundle->mirror_out != 0;
1820 forward_bpdu_changed(struct ofproto *ofproto_)
1822 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1823 /* Revalidate cached flows whenever forward_bpdu option changes. */
1824 ofproto->need_revalidate = true;
1829 static struct ofport_dpif *
1830 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1832 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1833 return ofport ? ofport_dpif_cast(ofport) : NULL;
1836 static struct ofport_dpif *
1837 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1839 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1843 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1844 struct dpif_port *dpif_port)
1846 ofproto_port->name = dpif_port->name;
1847 ofproto_port->type = dpif_port->type;
1848 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1852 port_run(struct ofport_dpif *ofport)
1854 bool enable = netdev_get_carrier(ofport->up.netdev);
1857 cfm_run(ofport->cfm);
1859 if (cfm_should_send_ccm(ofport->cfm)) {
1860 struct ofpbuf packet;
1862 ofpbuf_init(&packet, 0);
1863 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1864 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1865 ofport->odp_port, &packet);
1866 ofpbuf_uninit(&packet);
1869 enable = enable && !cfm_get_fault(ofport->cfm)
1870 && cfm_get_opup(ofport->cfm);
1873 if (ofport->bundle) {
1874 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1877 if (ofport->may_enable != enable) {
1878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1880 if (ofproto->has_bundle_action) {
1881 ofproto->need_revalidate = true;
1885 ofport->may_enable = enable;
1889 port_wait(struct ofport_dpif *ofport)
1892 cfm_wait(ofport->cfm);
1897 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1898 struct ofproto_port *ofproto_port)
1900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1901 struct dpif_port dpif_port;
1904 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1906 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1912 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1918 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1920 *ofp_portp = odp_port_to_ofp_port(odp_port);
1926 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1928 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1931 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1933 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1935 /* The caller is going to close ofport->up.netdev. If this is a
1936 * bonded port, then the bond is using that netdev, so remove it
1937 * from the bond. The client will need to reconfigure everything
1938 * after deleting ports, so then the slave will get re-added. */
1939 bundle_remove(&ofport->up);
1945 struct port_dump_state {
1946 struct dpif_port_dump dump;
1951 port_dump_start(const struct ofproto *ofproto_, void **statep)
1953 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1954 struct port_dump_state *state;
1956 *statep = state = xmalloc(sizeof *state);
1957 dpif_port_dump_start(&state->dump, ofproto->dpif);
1958 state->done = false;
1963 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1964 struct ofproto_port *port)
1966 struct port_dump_state *state = state_;
1967 struct dpif_port dpif_port;
1969 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1970 ofproto_port_from_dpif_port(port, &dpif_port);
1973 int error = dpif_port_dump_done(&state->dump);
1975 return error ? error : EOF;
1980 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1982 struct port_dump_state *state = state_;
1985 dpif_port_dump_done(&state->dump);
1992 port_poll(const struct ofproto *ofproto_, char **devnamep)
1994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1995 return dpif_port_poll(ofproto->dpif, devnamep);
1999 port_poll_wait(const struct ofproto *ofproto_)
2001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2002 dpif_port_poll_wait(ofproto->dpif);
2006 port_is_lacp_current(const struct ofport *ofport_)
2008 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009 return (ofport->bundle && ofport->bundle->lacp
2010 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2014 /* Upcall handling. */
2016 /* Flow miss batching.
2018 * Some dpifs implement operations faster when you hand them off in a batch.
2019 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2020 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2021 * more packets, plus possibly installing the flow in the dpif.
2023 * So far we only batch the operations that affect flow setup time the most.
2024 * It's possible to batch more than that, but the benefit might be minimal. */
2026 struct hmap_node hmap_node;
2028 const struct nlattr *key;
2030 struct list packets;
2033 struct flow_miss_op {
2034 union dpif_op dpif_op;
2035 struct facet *facet;
2038 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2039 * OpenFlow controller as necessary according to their individual
2042 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2043 * ownership is transferred to this function. */
2045 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2046 const struct flow *flow, bool clone)
2048 struct ofputil_packet_in pin;
2050 pin.packet = packet;
2051 pin.in_port = flow->in_port;
2052 pin.reason = OFPR_NO_MATCH;
2053 pin.buffer_id = 0; /* not yet known */
2054 pin.send_len = 0; /* not used for flow table misses */
2055 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2056 clone ? NULL : packet);
2059 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2060 * OpenFlow controller as necessary according to their individual
2063 * 'send_len' should be the number of bytes of 'packet' to send to the
2064 * controller, as specified in the action that caused the packet to be sent.
2066 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2067 * Otherwise, ownership is transferred to this function. */
2069 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2070 uint64_t userdata, const struct flow *flow, bool clone)
2072 struct ofputil_packet_in pin;
2073 struct user_action_cookie cookie;
2075 memcpy(&cookie, &userdata, sizeof(cookie));
2077 pin.packet = packet;
2078 pin.in_port = flow->in_port;
2079 pin.reason = OFPR_ACTION;
2080 pin.buffer_id = 0; /* not yet known */
2081 pin.send_len = cookie.data;
2082 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2083 clone ? NULL : packet);
2087 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2088 const struct ofpbuf *packet)
2090 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2096 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2098 cfm_process_heartbeat(ofport->cfm, packet);
2101 } else if (ofport->bundle && ofport->bundle->lacp
2102 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2104 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2107 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2109 stp_process_packet(ofport, packet);
2116 static struct flow_miss *
2117 flow_miss_create(struct hmap *todo, const struct flow *flow,
2118 const struct nlattr *key, size_t key_len)
2120 uint32_t hash = flow_hash(flow, 0);
2121 struct flow_miss *miss;
2123 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2124 if (flow_equal(&miss->flow, flow)) {
2129 miss = xmalloc(sizeof *miss);
2130 hmap_insert(todo, &miss->hmap_node, hash);
2133 miss->key_len = key_len;
2134 list_init(&miss->packets);
2139 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2140 struct flow_miss_op *ops, size_t *n_ops)
2142 const struct flow *flow = &miss->flow;
2143 struct ofpbuf *packet, *next_packet;
2144 struct facet *facet;
2146 facet = facet_lookup_valid(ofproto, flow);
2148 struct rule_dpif *rule;
2150 rule = rule_dpif_lookup(ofproto, flow, 0);
2152 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2153 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2155 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2156 COVERAGE_INC(ofproto_dpif_no_packet_in);
2157 /* XXX install 'drop' flow entry */
2161 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2165 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2167 list_remove(&packet->list_node);
2168 send_packet_in_miss(ofproto, packet, flow, false);
2174 facet = facet_create(rule, flow);
2177 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2178 list_remove(&packet->list_node);
2179 ofproto->n_matches++;
2181 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2183 * Extra-special case for fail-open mode.
2185 * We are in fail-open mode and the packet matched the fail-open
2186 * rule, but we are connected to a controller too. We should send
2187 * the packet up to the controller in the hope that it will try to
2188 * set up a flow and thereby allow us to exit fail-open.
2190 * See the top-level comment in fail-open.c for more information.
2192 send_packet_in_miss(ofproto, packet, flow, true);
2195 if (!facet->may_install) {
2196 facet_make_actions(ofproto, facet, packet);
2198 if (!execute_controller_action(ofproto, &facet->flow,
2199 facet->actions, facet->actions_len,
2201 struct flow_miss_op *op = &ops[(*n_ops)++];
2202 struct dpif_execute *execute = &op->dpif_op.execute;
2205 execute->type = DPIF_OP_EXECUTE;
2206 execute->key = miss->key;
2207 execute->key_len = miss->key_len;
2209 = (facet->may_install
2211 : xmemdup(facet->actions, facet->actions_len));
2212 execute->actions_len = facet->actions_len;
2213 execute->packet = packet;
2217 if (facet->may_install) {
2218 struct flow_miss_op *op = &ops[(*n_ops)++];
2219 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2222 put->type = DPIF_OP_FLOW_PUT;
2223 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2224 put->key = miss->key;
2225 put->key_len = miss->key_len;
2226 put->actions = facet->actions;
2227 put->actions_len = facet->actions_len;
2233 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2236 struct dpif_upcall *upcall;
2237 struct flow_miss *miss, *next_miss;
2238 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2239 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2248 /* Construct the to-do list.
2250 * This just amounts to extracting the flow from each packet and sticking
2251 * the packets that have the same flow in the same "flow_miss" structure so
2252 * that we can process them together. */
2254 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2255 struct flow_miss *miss;
2258 /* Obtain in_port and tun_id, at least, then set 'flow''s header
2260 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2261 flow_extract(upcall->packet, flow.priority, flow.tun_id,
2262 flow.in_port, &flow);
2264 /* Handle 802.1ag, LACP, and STP specially. */
2265 if (process_special(ofproto, &flow, upcall->packet)) {
2266 ofpbuf_delete(upcall->packet);
2267 ofproto->n_matches++;
2271 /* Add other packets to a to-do list. */
2272 miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len);
2273 list_push_back(&miss->packets, &upcall->packet->list_node);
2276 /* Process each element in the to-do list, constructing the set of
2277 * operations to batch. */
2279 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2280 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2281 ofpbuf_list_delete(&miss->packets);
2282 hmap_remove(&todo, &miss->hmap_node);
2285 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2286 hmap_destroy(&todo);
2288 /* Execute batch. */
2289 for (i = 0; i < n_ops; i++) {
2290 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2292 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2294 /* Free memory and update facets. */
2295 for (i = 0; i < n_ops; i++) {
2296 struct flow_miss_op *op = &flow_miss_ops[i];
2297 struct dpif_execute *execute;
2298 struct dpif_flow_put *put;
2300 switch (op->dpif_op.type) {
2301 case DPIF_OP_EXECUTE:
2302 execute = &op->dpif_op.execute;
2303 if (op->facet->actions != execute->actions) {
2304 free((struct nlattr *) execute->actions);
2306 ofpbuf_delete((struct ofpbuf *) execute->packet);
2309 case DPIF_OP_FLOW_PUT:
2310 put = &op->dpif_op.flow_put;
2312 op->facet->installed = true;
2320 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2321 struct dpif_upcall *upcall)
2324 struct user_action_cookie cookie;
2326 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2328 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2329 if (ofproto->sflow) {
2330 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2331 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
2333 ofpbuf_delete(upcall->packet);
2335 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2336 COVERAGE_INC(ofproto_dpif_ctlr_action);
2337 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2338 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2341 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2346 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
2348 switch (upcall->type) {
2349 case DPIF_UC_ACTION:
2350 handle_userspace_upcall(ofproto, upcall);
2354 /* The caller handles these. */
2357 case DPIF_N_UC_TYPES:
2359 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
2364 /* Flow expiration. */
2366 static int facet_max_idle(const struct ofproto_dpif *);
2367 static void update_stats(struct ofproto_dpif *);
2368 static void rule_expire(struct rule_dpif *);
2369 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
2371 /* This function is called periodically by run(). Its job is to collect
2372 * updates for the flows that have been installed into the datapath, most
2373 * importantly when they last were used, and then use that information to
2374 * expire flows that have not been used recently.
2376 * Returns the number of milliseconds after which it should be called again. */
2378 expire(struct ofproto_dpif *ofproto)
2380 struct rule_dpif *rule, *next_rule;
2381 struct classifier *table;
2384 /* Update stats for each flow in the datapath. */
2385 update_stats(ofproto);
2387 /* Expire facets that have been idle too long. */
2388 dp_max_idle = facet_max_idle(ofproto);
2389 expire_facets(ofproto, dp_max_idle);
2391 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2392 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2393 struct cls_cursor cursor;
2395 cls_cursor_init(&cursor, table, NULL);
2396 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2401 /* All outstanding data in existing flows has been accounted, so it's a
2402 * good time to do bond rebalancing. */
2403 if (ofproto->has_bonded_bundles) {
2404 struct ofbundle *bundle;
2406 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2408 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2413 return MIN(dp_max_idle, 1000);
2416 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2418 * This function also pushes statistics updates to rules which each facet
2419 * resubmits into. Generally these statistics will be accurate. However, if a
2420 * facet changes the rule it resubmits into at some time in between
2421 * update_stats() runs, it is possible that statistics accrued to the
2422 * old rule will be incorrectly attributed to the new rule. This could be
2423 * avoided by calling update_stats() whenever rules are created or
2424 * deleted. However, the performance impact of making so many calls to the
2425 * datapath do not justify the benefit of having perfectly accurate statistics.
2428 update_stats(struct ofproto_dpif *p)
2430 const struct dpif_flow_stats *stats;
2431 struct dpif_flow_dump dump;
2432 const struct nlattr *key;
2435 dpif_flow_dump_start(&dump, p->dpif);
2436 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2437 struct facet *facet;
2440 if (odp_flow_key_to_flow(key, key_len, &flow)) {
2444 odp_flow_key_format(key, key_len, &s);
2445 VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
2451 facet = facet_find(p, &flow);
2453 if (facet && facet->installed) {
2455 if (stats->n_packets >= facet->dp_packet_count) {
2456 uint64_t extra = stats->n_packets - facet->dp_packet_count;
2457 facet->packet_count += extra;
2459 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2462 if (stats->n_bytes >= facet->dp_byte_count) {
2463 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
2465 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2468 facet->dp_packet_count = stats->n_packets;
2469 facet->dp_byte_count = stats->n_bytes;
2471 facet_update_time(p, facet, stats->used);
2472 facet_account(p, facet);
2473 facet_push_stats(facet);
2475 /* There's a flow in the datapath that we know nothing about.
2477 COVERAGE_INC(facet_unexpected);
2478 dpif_flow_del(p->dpif, key, key_len, NULL);
2481 dpif_flow_dump_done(&dump);
2484 /* Calculates and returns the number of milliseconds of idle time after which
2485 * facets should expire from the datapath and we should fold their statistics
2486 * into their parent rules in userspace. */
2488 facet_max_idle(const struct ofproto_dpif *ofproto)
2491 * Idle time histogram.
2493 * Most of the time a switch has a relatively small number of facets. When
2494 * this is the case we might as well keep statistics for all of them in
2495 * userspace and to cache them in the kernel datapath for performance as
2498 * As the number of facets increases, the memory required to maintain
2499 * statistics about them in userspace and in the kernel becomes
2500 * significant. However, with a large number of facets it is likely that
2501 * only a few of them are "heavy hitters" that consume a large amount of
2502 * bandwidth. At this point, only heavy hitters are worth caching in the
2503 * kernel and maintaining in userspaces; other facets we can discard.
2505 * The technique used to compute the idle time is to build a histogram with
2506 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
2507 * that is installed in the kernel gets dropped in the appropriate bucket.
2508 * After the histogram has been built, we compute the cutoff so that only
2509 * the most-recently-used 1% of facets (but at least
2510 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2511 * the most-recently-used bucket of facets is kept, so actually an
2512 * arbitrary number of facets can be kept in any given expiration run
2513 * (though the next run will delete most of those unless they receive
2516 * This requires a second pass through the facets, in addition to the pass
2517 * made by update_stats(), because the former function never looks
2518 * at uninstallable facets.
2520 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2521 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2522 int buckets[N_BUCKETS] = { 0 };
2523 int total, subtotal, bucket;
2524 struct facet *facet;
2528 total = hmap_count(&ofproto->facets);
2529 if (total <= ofproto->up.flow_eviction_threshold) {
2530 return N_BUCKETS * BUCKET_WIDTH;
2533 /* Build histogram. */
2535 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
2536 long long int idle = now - facet->used;
2537 int bucket = (idle <= 0 ? 0
2538 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2539 : (unsigned int) idle / BUCKET_WIDTH);
2543 /* Find the first bucket whose flows should be expired. */
2544 subtotal = bucket = 0;
2546 subtotal += buckets[bucket++];
2547 } while (bucket < N_BUCKETS &&
2548 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2550 if (VLOG_IS_DBG_ENABLED()) {
2554 ds_put_cstr(&s, "keep");
2555 for (i = 0; i < N_BUCKETS; i++) {
2557 ds_put_cstr(&s, ", drop");
2560 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2563 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2567 return bucket * BUCKET_WIDTH;
2571 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
2573 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
2574 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
2575 struct ofexpired expired;
2577 if (facet->installed) {
2578 struct dpif_flow_stats stats;
2580 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
2582 facet_update_stats(ofproto, facet, &stats);
2585 expired.flow = facet->flow;
2586 expired.packet_count = facet->packet_count;
2587 expired.byte_count = facet->byte_count;
2588 expired.used = facet->used;
2589 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2594 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
2596 long long int cutoff = time_msec() - dp_max_idle;
2597 struct facet *facet, *next_facet;
2599 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
2600 facet_active_timeout(ofproto, facet);
2601 if (facet->used < cutoff) {
2602 facet_remove(ofproto, facet);
2607 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2608 * then delete it entirely. */
2610 rule_expire(struct rule_dpif *rule)
2612 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2613 struct facet *facet, *next_facet;
2617 /* Has 'rule' expired? */
2619 if (rule->up.hard_timeout
2620 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
2621 reason = OFPRR_HARD_TIMEOUT;
2622 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2623 && now > rule->used + rule->up.idle_timeout * 1000) {
2624 reason = OFPRR_IDLE_TIMEOUT;
2629 COVERAGE_INC(ofproto_dpif_expired);
2631 /* Update stats. (This is a no-op if the rule expired due to an idle
2632 * timeout, because that only happens when the rule has no facets left.) */
2633 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2634 facet_remove(ofproto, facet);
2637 /* Get rid of the rule. */
2638 ofproto_rule_expire(&rule->up, reason);
2643 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
2645 * The caller must already have determined that no facet with an identical
2646 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2647 * the ofproto's classifier table.
2649 * The facet will initially have no ODP actions. The caller should fix that
2650 * by calling facet_make_actions(). */
2651 static struct facet *
2652 facet_create(struct rule_dpif *rule, const struct flow *flow)
2654 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2655 struct facet *facet;
2657 facet = xzalloc(sizeof *facet);
2658 facet->used = time_msec();
2659 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2660 list_push_back(&rule->facets, &facet->list_node);
2662 facet->flow = *flow;
2663 netflow_flow_init(&facet->nf_flow);
2664 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2670 facet_free(struct facet *facet)
2672 free(facet->actions);
2677 execute_controller_action(struct ofproto_dpif *ofproto,
2678 const struct flow *flow,
2679 const struct nlattr *odp_actions, size_t actions_len,
2680 struct ofpbuf *packet)
2683 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
2684 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
2685 /* As an optimization, avoid a round-trip from userspace to kernel to
2686 * userspace. This also avoids possibly filling up kernel packet
2687 * buffers along the way.
2689 * This optimization will not accidentally catch sFlow
2690 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
2691 * inside OVS_ACTION_ATTR_SAMPLE. */
2692 const struct nlattr *nla;
2694 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
2695 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
2703 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2704 * 'packet', which arrived on 'in_port'.
2706 * Takes ownership of 'packet'. */
2708 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2709 const struct nlattr *odp_actions, size_t actions_len,
2710 struct ofpbuf *packet)
2712 struct odputil_keybuf keybuf;
2716 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
2721 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2722 odp_flow_key_from_flow(&key, flow);
2724 error = dpif_execute(ofproto->dpif, key.data, key.size,
2725 odp_actions, actions_len, packet);
2727 ofpbuf_delete(packet);
2731 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2732 * statistics appropriately. 'packet' must have at least sizeof(struct
2733 * ofp_packet_in) bytes of headroom.
2735 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2736 * applying flow_extract() to 'packet' would yield the same flow as
2739 * 'facet' must have accurately composed datapath actions; that is, it must
2740 * not be in need of revalidation.
2742 * Takes ownership of 'packet'. */
2744 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2745 struct ofpbuf *packet)
2747 struct dpif_flow_stats stats;
2749 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2751 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2752 stats.used = time_msec();
2753 if (execute_odp_actions(ofproto, &facet->flow,
2754 facet->actions, facet->actions_len, packet)) {
2755 facet_update_stats(ofproto, facet, &stats);
2759 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2761 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2762 * rule's statistics, via facet_uninstall().
2764 * - Removes 'facet' from its rule and from ofproto->facets.
2767 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2769 facet_uninstall(ofproto, facet);
2770 facet_flush_stats(ofproto, facet);
2771 hmap_remove(&ofproto->facets, &facet->hmap_node);
2772 list_remove(&facet->list_node);
2776 /* Composes the datapath actions for 'facet' based on its rule's actions. */
2778 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2779 const struct ofpbuf *packet)
2781 const struct rule_dpif *rule = facet->rule;
2782 struct ofpbuf *odp_actions;
2783 struct action_xlate_ctx ctx;
2785 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2786 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2787 facet->tags = ctx.tags;
2788 facet->may_install = ctx.may_set_up_flow;
2789 facet->has_learn = ctx.has_learn;
2790 facet->has_normal = ctx.has_normal;
2791 facet->nf_flow.output_iface = ctx.nf_output_iface;
2793 if (facet->actions_len != odp_actions->size
2794 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2795 free(facet->actions);
2796 facet->actions_len = odp_actions->size;
2797 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2800 ofpbuf_delete(odp_actions);
2803 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2804 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2805 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2806 * since 'facet' was last updated.
2808 * Returns 0 if successful, otherwise a positive errno value.*/
2810 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2811 const struct nlattr *actions, size_t actions_len,
2812 struct dpif_flow_stats *stats)
2814 struct odputil_keybuf keybuf;
2815 enum dpif_flow_put_flags flags;
2819 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2821 flags |= DPIF_FP_ZERO_STATS;
2824 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2825 odp_flow_key_from_flow(&key, &facet->flow);
2827 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2828 actions, actions_len, stats);
2831 facet_reset_dp_stats(facet, stats);
2837 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2838 * 'zero_stats' is true, clears any existing statistics from the datapath for
2841 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2843 struct dpif_flow_stats stats;
2845 if (facet->may_install
2846 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2847 zero_stats ? &stats : NULL)) {
2848 facet->installed = true;
2853 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
2856 const struct nlattr *a;
2860 if (facet->byte_count <= facet->accounted_bytes) {
2863 n_bytes = facet->byte_count - facet->accounted_bytes;
2864 facet->accounted_bytes = facet->byte_count;
2866 /* Feed information from the active flows back into the learning table to
2867 * ensure that table is always in sync with what is actually flowing
2868 * through the datapath. */
2869 if (facet->has_learn || facet->has_normal) {
2870 struct action_xlate_ctx ctx;
2872 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2873 ctx.may_learn = true;
2874 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
2875 facet->rule->up.n_actions));
2878 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
2882 /* This loop feeds byte counters to bond_account() for rebalancing to use
2883 * as a basis. We also need to track the actual VLAN on which the packet
2884 * is going to be sent to ensure that it matches the one passed to
2885 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2887 vlan_tci = facet->flow.vlan_tci;
2888 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2889 const struct ovs_action_push_vlan *vlan;
2890 struct ofport_dpif *port;
2892 switch (nl_attr_type(a)) {
2893 case OVS_ACTION_ATTR_OUTPUT:
2894 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2895 if (port && port->bundle && port->bundle->bond) {
2896 bond_account(port->bundle->bond, &facet->flow,
2897 vlan_tci_to_vid(vlan_tci), n_bytes);
2901 case OVS_ACTION_ATTR_POP_VLAN:
2902 vlan_tci = htons(0);
2905 case OVS_ACTION_ATTR_PUSH_VLAN:
2906 vlan = nl_attr_get(a);
2907 vlan_tci = vlan->vlan_tci;
2913 /* If 'rule' is installed in the datapath, uninstalls it. */
2915 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2917 if (facet->installed) {
2918 struct odputil_keybuf keybuf;
2919 struct dpif_flow_stats stats;
2923 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2924 odp_flow_key_from_flow(&key, &facet->flow);
2926 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2927 facet_reset_dp_stats(facet, &stats);
2929 facet_update_stats(p, facet, &stats);
2931 facet->installed = false;
2933 assert(facet->dp_packet_count == 0);
2934 assert(facet->dp_byte_count == 0);
2938 /* Returns true if the only action for 'facet' is to send to the controller.
2939 * (We don't report NetFlow expiration messages for such facets because they
2940 * are just part of the control logic for the network, not real traffic). */
2942 facet_is_controller_flow(struct facet *facet)
2945 && facet->rule->up.n_actions == 1
2946 && action_outputs_to_port(&facet->rule->up.actions[0],
2947 htons(OFPP_CONTROLLER)));
2950 /* Resets 'facet''s datapath statistics counters. This should be called when
2951 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2952 * it should contain the statistics returned by dpif when 'facet' was reset in
2953 * the datapath. 'stats' will be modified to only included statistics new
2954 * since 'facet' was last updated. */
2956 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2958 if (stats && facet->dp_packet_count <= stats->n_packets
2959 && facet->dp_byte_count <= stats->n_bytes) {
2960 stats->n_packets -= facet->dp_packet_count;
2961 stats->n_bytes -= facet->dp_byte_count;
2964 facet->dp_packet_count = 0;
2965 facet->dp_byte_count = 0;
2968 /* Folds all of 'facet''s statistics into its rule. Also updates the
2969 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2970 * 'facet''s statistics in the datapath should have been zeroed and folded into
2971 * its packet and byte counts before this function is called. */
2973 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2975 assert(!facet->dp_byte_count);
2976 assert(!facet->dp_packet_count);
2978 facet_push_stats(facet);
2979 facet_account(ofproto, facet);
2981 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2982 struct ofexpired expired;
2983 expired.flow = facet->flow;
2984 expired.packet_count = facet->packet_count;
2985 expired.byte_count = facet->byte_count;
2986 expired.used = facet->used;
2987 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2990 facet->rule->packet_count += facet->packet_count;
2991 facet->rule->byte_count += facet->byte_count;
2993 /* Reset counters to prevent double counting if 'facet' ever gets
2995 facet_reset_counters(facet);
2997 netflow_flow_clear(&facet->nf_flow);
3000 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3001 * Returns it if found, otherwise a null pointer.
3003 * The returned facet might need revalidation; use facet_lookup_valid()
3004 * instead if that is important. */
3005 static struct facet *
3006 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3008 struct facet *facet;
3010 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3012 if (flow_equal(flow, &facet->flow)) {
3020 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3021 * Returns it if found, otherwise a null pointer.
3023 * The returned facet is guaranteed to be valid. */
3024 static struct facet *
3025 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3027 struct facet *facet = facet_find(ofproto, flow);
3029 /* The facet we found might not be valid, since we could be in need of
3030 * revalidation. If it is not valid, don't return it. */
3032 && (ofproto->need_revalidate
3033 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3034 && !facet_revalidate(ofproto, facet)) {
3035 COVERAGE_INC(facet_invalidated);
3042 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3044 * - If the rule found is different from 'facet''s current rule, moves
3045 * 'facet' to the new rule and recompiles its actions.
3047 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3048 * where it is and recompiles its actions anyway.
3050 * - If there is none, destroys 'facet'.
3052 * Returns true if 'facet' still exists, false if it has been destroyed. */
3054 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3056 struct action_xlate_ctx ctx;
3057 struct ofpbuf *odp_actions;
3058 struct rule_dpif *new_rule;
3059 bool actions_changed;
3061 COVERAGE_INC(facet_revalidate);
3063 /* Determine the new rule. */
3064 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3066 /* No new rule, so delete the facet. */
3067 facet_remove(ofproto, facet);
3071 /* Calculate new datapath actions.
3073 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3074 * emit a NetFlow expiration and, if so, we need to have the old state
3075 * around to properly compose it. */
3076 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
3077 odp_actions = xlate_actions(&ctx,
3078 new_rule->up.actions, new_rule->up.n_actions);
3079 actions_changed = (facet->actions_len != odp_actions->size
3080 || memcmp(facet->actions, odp_actions->data,
3081 facet->actions_len));
3083 /* If the datapath actions changed or the installability changed,
3084 * then we need to talk to the datapath. */
3085 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
3086 if (ctx.may_set_up_flow) {
3087 struct dpif_flow_stats stats;
3089 facet_put__(ofproto, facet,
3090 odp_actions->data, odp_actions->size, &stats);
3091 facet_update_stats(ofproto, facet, &stats);
3093 facet_uninstall(ofproto, facet);
3096 /* The datapath flow is gone or has zeroed stats, so push stats out of
3097 * 'facet' into 'rule'. */
3098 facet_flush_stats(ofproto, facet);
3101 /* Update 'facet' now that we've taken care of all the old state. */
3102 facet->tags = ctx.tags;
3103 facet->nf_flow.output_iface = ctx.nf_output_iface;
3104 facet->may_install = ctx.may_set_up_flow;
3105 facet->has_learn = ctx.has_learn;
3106 facet->has_normal = ctx.has_normal;
3107 if (actions_changed) {
3108 free(facet->actions);
3109 facet->actions_len = odp_actions->size;
3110 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
3112 if (facet->rule != new_rule) {
3113 COVERAGE_INC(facet_changed_rule);
3114 list_remove(&facet->list_node);
3115 list_push_back(&new_rule->facets, &facet->list_node);
3116 facet->rule = new_rule;
3117 facet->used = new_rule->up.created;
3118 facet->rs_used = facet->used;
3121 ofpbuf_delete(odp_actions);
3126 /* Updates 'facet''s used time. Caller is responsible for calling
3127 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3129 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3132 if (used > facet->used) {
3134 if (used > facet->rule->used) {
3135 facet->rule->used = used;
3137 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3141 /* Folds the statistics from 'stats' into the counters in 'facet'.
3143 * Because of the meaning of a facet's counters, it only makes sense to do this
3144 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3145 * packet that was sent by hand or if it represents statistics that have been
3146 * cleared out of the datapath. */
3148 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
3149 const struct dpif_flow_stats *stats)
3151 if (stats->n_packets || stats->used > facet->used) {
3152 facet_update_time(ofproto, facet, stats->used);
3153 facet->packet_count += stats->n_packets;
3154 facet->byte_count += stats->n_bytes;
3155 facet_push_stats(facet);
3156 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3161 facet_reset_counters(struct facet *facet)
3163 facet->packet_count = 0;
3164 facet->byte_count = 0;
3165 facet->rs_packet_count = 0;
3166 facet->rs_byte_count = 0;
3167 facet->accounted_bytes = 0;
3171 facet_push_stats(struct facet *facet)
3173 uint64_t rs_packets, rs_bytes;
3175 assert(facet->packet_count >= facet->rs_packet_count);
3176 assert(facet->byte_count >= facet->rs_byte_count);
3177 assert(facet->used >= facet->rs_used);
3179 rs_packets = facet->packet_count - facet->rs_packet_count;
3180 rs_bytes = facet->byte_count - facet->rs_byte_count;
3182 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3183 facet->rs_packet_count = facet->packet_count;
3184 facet->rs_byte_count = facet->byte_count;
3185 facet->rs_used = facet->used;
3187 flow_push_stats(facet->rule, &facet->flow,
3188 rs_packets, rs_bytes, facet->used);
3192 struct ofproto_push {
3193 struct action_xlate_ctx ctx;
3200 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3202 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3205 rule->packet_count += push->packets;
3206 rule->byte_count += push->bytes;
3207 rule->used = MAX(push->used, rule->used);
3211 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3212 * 'rule''s actions. */
3214 flow_push_stats(const struct rule_dpif *rule,
3215 struct flow *flow, uint64_t packets, uint64_t bytes,
3218 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3219 struct ofproto_push push;
3221 push.packets = packets;
3225 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3226 push.ctx.resubmit_hook = push_resubmit;
3227 ofpbuf_delete(xlate_actions(&push.ctx,
3228 rule->up.actions, rule->up.n_actions));
3233 static struct rule_dpif *
3234 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3237 struct cls_rule *cls_rule;
3238 struct classifier *cls;
3240 if (table_id >= N_TABLES) {
3244 cls = &ofproto->up.tables[table_id];
3245 if (flow->nw_frag & FLOW_NW_FRAG_ANY
3246 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3247 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3248 * are unavailable. */
3249 struct flow ofpc_normal_flow = *flow;
3250 ofpc_normal_flow.tp_src = htons(0);
3251 ofpc_normal_flow.tp_dst = htons(0);
3252 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3254 cls_rule = classifier_lookup(cls, flow);
3256 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3260 complete_operation(struct rule_dpif *rule)
3262 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3264 rule_invalidate(rule);
3266 struct dpif_completion *c = xmalloc(sizeof *c);
3267 c->op = rule->up.pending;
3268 list_push_back(&ofproto->completions, &c->list_node);
3270 ofoperation_complete(rule->up.pending, 0);
3274 static struct rule *
3277 struct rule_dpif *rule = xmalloc(sizeof *rule);
3282 rule_dealloc(struct rule *rule_)
3284 struct rule_dpif *rule = rule_dpif_cast(rule_);
3289 rule_construct(struct rule *rule_)
3291 struct rule_dpif *rule = rule_dpif_cast(rule_);
3292 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3293 struct rule_dpif *victim;
3297 error = validate_actions(rule->up.actions, rule->up.n_actions,
3298 &rule->up.cr.flow, ofproto->max_ports);
3303 rule->used = rule->up.created;
3304 rule->packet_count = 0;
3305 rule->byte_count = 0;
3307 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3308 if (victim && !list_is_empty(&victim->facets)) {
3309 struct facet *facet;
3311 rule->facets = victim->facets;
3312 list_moved(&rule->facets);
3313 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3314 /* XXX: We're only clearing our local counters here. It's possible
3315 * that quite a few packets are unaccounted for in the datapath
3316 * statistics. These will be accounted to the new rule instead of
3317 * cleared as required. This could be fixed by clearing out the
3318 * datapath statistics for this facet, but currently it doesn't
3320 facet_reset_counters(facet);
3324 /* Must avoid list_moved() in this case. */
3325 list_init(&rule->facets);
3328 table_id = rule->up.table_id;
3329 rule->tag = (victim ? victim->tag
3331 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3332 ofproto->tables[table_id].basis));
3334 complete_operation(rule);
3339 rule_destruct(struct rule *rule_)
3341 struct rule_dpif *rule = rule_dpif_cast(rule_);
3342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3343 struct facet *facet, *next_facet;
3345 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3346 facet_revalidate(ofproto, facet);
3349 complete_operation(rule);
3353 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3355 struct rule_dpif *rule = rule_dpif_cast(rule_);
3356 struct facet *facet;
3358 /* Start from historical data for 'rule' itself that are no longer tracked
3359 * in facets. This counts, for example, facets that have expired. */
3360 *packets = rule->packet_count;
3361 *bytes = rule->byte_count;
3363 /* Add any statistics that are tracked by facets. This includes
3364 * statistical data recently updated by ofproto_update_stats() as well as
3365 * stats for packets that were executed "by hand" via dpif_execute(). */
3366 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3367 *packets += facet->packet_count;
3368 *bytes += facet->byte_count;
3373 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
3375 struct rule_dpif *rule = rule_dpif_cast(rule_);
3376 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3377 struct action_xlate_ctx ctx;
3378 struct ofpbuf *odp_actions;
3379 struct facet *facet;
3382 /* First look for a related facet. If we find one, account it to that. */
3383 facet = facet_lookup_valid(ofproto, flow);
3384 if (facet && facet->rule == rule) {
3385 if (!facet->may_install) {
3386 facet_make_actions(ofproto, facet, packet);
3388 facet_execute(ofproto, facet, packet);
3392 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
3393 * create a new facet for it and use that. */
3394 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
3395 facet = facet_create(rule, flow);
3396 facet_make_actions(ofproto, facet, packet);
3397 facet_execute(ofproto, facet, packet);
3398 facet_install(ofproto, facet, true);
3402 /* We can't account anything to a facet. If we were to try, then that
3403 * facet would have a non-matching rule, busting our invariants. */
3404 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3405 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3406 size = packet->size;
3407 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3408 odp_actions->size, packet)) {
3409 rule->used = time_msec();
3410 rule->packet_count++;
3411 rule->byte_count += size;
3412 flow_push_stats(rule, flow, 1, size, rule->used);
3414 ofpbuf_delete(odp_actions);
3420 rule_modify_actions(struct rule *rule_)
3422 struct rule_dpif *rule = rule_dpif_cast(rule_);
3423 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3426 error = validate_actions(rule->up.actions, rule->up.n_actions,
3427 &rule->up.cr.flow, ofproto->max_ports);
3429 ofoperation_complete(rule->up.pending, error);
3433 complete_operation(rule);
3436 /* Sends 'packet' out of port 'odp_port' within 'ofproto'.
3437 * Returns 0 if successful, otherwise a positive errno value. */
3439 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
3440 const struct ofpbuf *packet)
3442 struct ofpbuf key, odp_actions;
3443 struct odputil_keybuf keybuf;
3447 flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow);
3448 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3449 odp_flow_key_from_flow(&key, &flow);
3451 ofpbuf_init(&odp_actions, 32);
3452 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
3454 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3455 error = dpif_execute(ofproto->dpif,
3457 odp_actions.data, odp_actions.size,
3459 ofpbuf_uninit(&odp_actions);
3462 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
3463 ofproto->up.name, odp_port, strerror(error));
3468 /* OpenFlow to datapath action translation. */
3470 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
3471 struct action_xlate_ctx *ctx);
3472 static void xlate_normal(struct action_xlate_ctx *);
3475 put_userspace_action(const struct ofproto_dpif *ofproto,
3476 struct ofpbuf *odp_actions,
3477 const struct flow *flow,
3478 const struct user_action_cookie *cookie)
3483 pid = dpif_port_get_pid(ofproto->dpif,
3484 ofp_port_to_odp_port(flow->in_port));
3486 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
3487 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
3488 nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
3489 cookie, sizeof *cookie);
3490 nl_msg_end_nested(odp_actions, offset);
3492 return odp_actions->size - NLA_ALIGN(sizeof *cookie);
3495 /* Compose SAMPLE action for sFlow. */
3497 compose_sflow_action(const struct ofproto_dpif *ofproto,
3498 struct ofpbuf *odp_actions,
3499 const struct flow *flow,
3502 uint32_t port_ifindex;
3503 uint32_t probability;
3504 struct user_action_cookie cookie;
3505 size_t sample_offset, actions_offset;
3506 int cookie_offset, n_output;
3508 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
3512 if (odp_port == OVSP_NONE) {
3516 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
3520 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
3522 /* Number of packets out of UINT_MAX to sample. */
3523 probability = dpif_sflow_get_probability(ofproto->sflow);
3524 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
3526 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
3528 cookie.type = USER_ACTION_COOKIE_SFLOW;
3529 cookie.data = port_ifindex;
3530 cookie.n_output = n_output;
3531 cookie.vlan_tci = 0;
3532 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
3534 nl_msg_end_nested(odp_actions, actions_offset);
3535 nl_msg_end_nested(odp_actions, sample_offset);
3536 return cookie_offset;
3539 /* SAMPLE action must be first action in any given list of actions.
3540 * At this point we do not have all information required to build it. So try to
3541 * build sample action as complete as possible. */
3543 add_sflow_action(struct action_xlate_ctx *ctx)
3545 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
3547 &ctx->flow, OVSP_NONE);
3548 ctx->sflow_odp_port = 0;
3549 ctx->sflow_n_outputs = 0;
3552 /* Fix SAMPLE action according to data collected while composing ODP actions.
3553 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
3554 * USERSPACE action's user-cookie which is required for sflow. */
3556 fix_sflow_action(struct action_xlate_ctx *ctx)
3558 const struct flow *base = &ctx->base_flow;
3559 struct user_action_cookie *cookie;
3561 if (!ctx->user_cookie_offset) {
3565 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
3567 assert(cookie != NULL);
3568 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
3570 if (ctx->sflow_n_outputs) {
3571 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
3572 ctx->sflow_odp_port);
3574 if (ctx->sflow_n_outputs >= 255) {
3575 cookie->n_output = 255;
3577 cookie->n_output = ctx->sflow_n_outputs;
3579 cookie->vlan_tci = base->vlan_tci;
3583 commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type,
3584 const void *key, size_t key_size)
3586 size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET);
3587 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
3588 nl_msg_end_nested(odp_actions, offset);
3592 commit_set_tun_id_action(const struct flow *flow, struct flow *base,
3593 struct ofpbuf *odp_actions)
3595 if (base->tun_id == flow->tun_id) {
3598 base->tun_id = flow->tun_id;
3600 commit_set_action(odp_actions, OVS_KEY_ATTR_TUN_ID,
3601 &base->tun_id, sizeof(base->tun_id));
3605 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
3606 struct ofpbuf *odp_actions)
3608 struct ovs_key_ethernet eth_key;
3610 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
3611 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
3615 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
3616 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
3618 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
3619 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
3621 commit_set_action(odp_actions, OVS_KEY_ATTR_ETHERNET,
3622 ð_key, sizeof(eth_key));
3626 commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci)
3628 struct flow *base = &ctx->base_flow;
3630 if (base->vlan_tci == new_tci) {
3634 if (base->vlan_tci & htons(VLAN_CFI)) {
3635 nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_VLAN);
3638 if (new_tci & htons(VLAN_CFI)) {
3639 struct ovs_action_push_vlan vlan;
3641 vlan.vlan_tpid = htons(ETH_TYPE_VLAN);
3642 vlan.vlan_tci = new_tci;
3643 nl_msg_put_unspec(ctx->odp_actions, OVS_ACTION_ATTR_PUSH_VLAN,
3644 &vlan, sizeof vlan);
3646 base->vlan_tci = new_tci;
3650 commit_set_nw_action(const struct flow *flow, struct flow *base,
3651 struct ofpbuf *odp_actions)
3653 struct ovs_key_ipv4 ipv4_key;
3655 if (base->dl_type != htons(ETH_TYPE_IP) ||
3656 !base->nw_src || !base->nw_dst) {
3660 if (base->nw_src == flow->nw_src &&
3661 base->nw_dst == flow->nw_dst &&
3662 base->nw_tos == flow->nw_tos &&
3663 base->nw_ttl == flow->nw_ttl &&
3664 base->nw_frag == flow->nw_frag) {
3668 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
3669 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
3670 ipv4_key.ipv4_proto = base->nw_proto;
3671 ipv4_key.ipv4_tos = flow->nw_tos;
3672 ipv4_key.ipv4_ttl = flow->nw_ttl;
3673 ipv4_key.ipv4_frag = (base->nw_frag == 0 ? OVS_FRAG_TYPE_NONE
3674 : base->nw_frag == FLOW_NW_FRAG_ANY
3675 ? OVS_FRAG_TYPE_FIRST : OVS_FRAG_TYPE_LATER);
3677 commit_set_action(odp_actions, OVS_KEY_ATTR_IPV4,
3678 &ipv4_key, sizeof(ipv4_key));
3682 commit_set_port_action(const struct flow *flow, struct flow *base,
3683 struct ofpbuf *odp_actions)
3685 if (!base->tp_src || !base->tp_dst) {
3689 if (base->tp_src == flow->tp_src &&
3690 base->tp_dst == flow->tp_dst) {
3694 if (flow->nw_proto == IPPROTO_TCP) {
3695 struct ovs_key_tcp port_key;
3697 port_key.tcp_src = base->tp_src = flow->tp_src;
3698 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
3700 commit_set_action(odp_actions, OVS_KEY_ATTR_TCP,
3701 &port_key, sizeof(port_key));
3703 } else if (flow->nw_proto == IPPROTO_UDP) {
3704 struct ovs_key_udp port_key;
3706 port_key.udp_src = base->tp_src = flow->tp_src;
3707 port_key.udp_dst = base->tp_dst = flow->tp_dst;
3709 commit_set_action(odp_actions, OVS_KEY_ATTR_UDP,
3710 &port_key, sizeof(port_key));
3715 commit_set_priority_action(const struct flow *flow, struct flow *base,
3716 struct ofpbuf *odp_actions)
3718 if (base->priority == flow->priority) {
3721 base->priority = flow->priority;
3723 commit_set_action(odp_actions, OVS_KEY_ATTR_PRIORITY,
3724 &base->priority, sizeof(base->priority));
3728 commit_odp_actions(struct action_xlate_ctx *ctx)
3730 const struct flow *flow = &ctx->flow;
3731 struct flow *base = &ctx->base_flow;
3732 struct ofpbuf *odp_actions = ctx->odp_actions;
3734 commit_set_tun_id_action(flow, base, odp_actions);
3735 commit_set_ether_addr_action(flow, base, odp_actions);
3736 commit_vlan_action(ctx, flow->vlan_tci);
3737 commit_set_nw_action(flow, base, odp_actions);
3738 commit_set_port_action(flow, base, odp_actions);
3739 commit_set_priority_action(flow, base, odp_actions);
3743 compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port)
3745 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3746 ctx->sflow_odp_port = odp_port;
3747 ctx->sflow_n_outputs++;
3751 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3753 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3754 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
3757 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
3758 || !stp_forward_in_state(ofport->stp_state)) {
3759 /* Forwarding disabled on port. */
3764 * We don't have an ofport record for this port, but it doesn't hurt to
3765 * allow forwarding to it anyhow. Maybe such a port will appear later
3766 * and we're pre-populating the flow table.
3770 commit_odp_actions(ctx);
3771 compose_output_action(ctx, odp_port);
3772 ctx->nf_output_iface = ofp_port;
3776 xlate_table_action(struct action_xlate_ctx *ctx,
3777 uint16_t in_port, uint8_t table_id)
3779 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
3780 struct ofproto_dpif *ofproto = ctx->ofproto;
3781 struct rule_dpif *rule;
3782 uint16_t old_in_port;
3783 uint8_t old_table_id;
3785 old_table_id = ctx->table_id;
3786 ctx->table_id = table_id;
3788 /* Look up a flow with 'in_port' as the input port. */
3789 old_in_port = ctx->flow.in_port;
3790 ctx->flow.in_port = in_port;
3791 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
3794 if (table_id > 0 && table_id < N_TABLES) {
3795 struct table_dpif *table = &ofproto->tables[table_id];
3796 if (table->other_table) {
3799 : rule_calculate_tag(&ctx->flow,
3800 &table->other_table->wc,
3805 /* Restore the original input port. Otherwise OFPP_NORMAL and
3806 * OFPP_IN_PORT will have surprising behavior. */
3807 ctx->flow.in_port = old_in_port;
3809 if (ctx->resubmit_hook) {
3810 ctx->resubmit_hook(ctx, rule);
3815 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
3819 ctx->table_id = old_table_id;
3821 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3823 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
3824 MAX_RESUBMIT_RECURSION);
3829 xlate_resubmit_table(struct action_xlate_ctx *ctx,
3830 const struct nx_action_resubmit *nar)
3835 in_port = (nar->in_port == htons(OFPP_IN_PORT)
3837 : ntohs(nar->in_port));
3838 table_id = nar->table == 255 ? ctx->table_id : nar->table;
3840 xlate_table_action(ctx, in_port, table_id);
3844 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
3846 struct ofport_dpif *ofport;
3848 commit_odp_actions(ctx);
3849 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
3850 uint16_t ofp_port = ofport->up.ofp_port;
3851 if (ofp_port != ctx->flow.in_port
3852 && !(ofport->up.opp.config & mask)
3853 && stp_forward_in_state(ofport->stp_state)) {
3854 compose_output_action(ctx, ofport->odp_port);
3858 ctx->nf_output_iface = NF_OUT_FLOOD;
3862 compose_controller_action(struct action_xlate_ctx *ctx, int len)
3864 struct user_action_cookie cookie;
3866 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
3868 cookie.n_output = 0;
3869 cookie.vlan_tci = 0;
3870 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
3874 xlate_output_action__(struct action_xlate_ctx *ctx,
3875 uint16_t port, uint16_t max_len)
3877 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
3879 ctx->nf_output_iface = NF_OUT_DROP;
3883 add_output_action(ctx, ctx->flow.in_port);
3886 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
3892 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
3895 flood_packets(ctx, htonl(0));
3897 case OFPP_CONTROLLER:
3898 commit_odp_actions(ctx);
3899 compose_controller_action(ctx, max_len);
3902 add_output_action(ctx, OFPP_LOCAL);
3907 if (port != ctx->flow.in_port) {
3908 add_output_action(ctx, port);
3913 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3914 ctx->nf_output_iface = NF_OUT_FLOOD;
3915 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3916 ctx->nf_output_iface = prev_nf_output_iface;
3917 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3918 ctx->nf_output_iface != NF_OUT_FLOOD) {
3919 ctx->nf_output_iface = NF_OUT_MULTI;
3924 xlate_output_reg_action(struct action_xlate_ctx *ctx,
3925 const struct nx_action_output_reg *naor)
3929 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
3931 if (ofp_port <= UINT16_MAX) {
3932 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
3937 xlate_output_action(struct action_xlate_ctx *ctx,
3938 const struct ofp_action_output *oao)
3940 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3944 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3945 const struct ofp_action_enqueue *oae)
3947 uint16_t ofp_port, odp_port;
3948 uint32_t flow_priority, priority;
3951 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3954 /* Fall back to ordinary output action. */
3955 xlate_output_action__(ctx, ntohs(oae->port), 0);
3959 /* Figure out datapath output port. */
3960 ofp_port = ntohs(oae->port);
3961 if (ofp_port == OFPP_IN_PORT) {
3962 ofp_port = ctx->flow.in_port;
3963 } else if (ofp_port == ctx->flow.in_port) {
3966 odp_port = ofp_port_to_odp_port(ofp_port);
3968 /* Add datapath actions. */
3969 flow_priority = ctx->flow.priority;
3970 ctx->flow.priority = priority;
3971 add_output_action(ctx, odp_port);
3972 ctx->flow.priority = flow_priority;
3974 /* Update NetFlow output port. */
3975 if (ctx->nf_output_iface == NF_OUT_DROP) {
3976 ctx->nf_output_iface = odp_port;
3977 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3978 ctx->nf_output_iface = NF_OUT_MULTI;
3983 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3984 const struct nx_action_set_queue *nasq)
3989 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3992 /* Couldn't translate queue to a priority, so ignore. A warning
3993 * has already been logged. */
3997 ctx->flow.priority = priority;
4000 struct xlate_reg_state {
4006 xlate_autopath(struct action_xlate_ctx *ctx,
4007 const struct nx_action_autopath *naa)
4009 uint16_t ofp_port = ntohl(naa->id);
4010 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4012 if (!port || !port->bundle) {
4013 ofp_port = OFPP_NONE;
4014 } else if (port->bundle->bond) {
4015 /* Autopath does not support VLAN hashing. */
4016 struct ofport_dpif *slave = bond_choose_output_slave(
4017 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4019 ofp_port = slave->up.ofp_port;
4022 autopath_execute(naa, &ctx->flow, ofp_port);
4026 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4028 struct ofproto_dpif *ofproto = ofproto_;
4029 struct ofport_dpif *port;
4039 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4042 port = get_ofp_port(ofproto, ofp_port);
4043 return port ? port->may_enable : false;
4048 xlate_learn_action(struct action_xlate_ctx *ctx,
4049 const struct nx_action_learn *learn)
4051 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4052 struct ofputil_flow_mod fm;
4055 learn_execute(learn, &ctx->flow, &fm);
4057 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4058 if (error && !VLOG_DROP_WARN(&rl)) {
4059 char *msg = ofputil_error_to_string(error);
4060 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4068 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4070 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4071 ? htonl(OFPPC_NO_RECV_STP)
4072 : htonl(OFPPC_NO_RECV))) {
4076 /* Only drop packets here if both forwarding and learning are
4077 * disabled. If just learning is enabled, we need to have
4078 * OFPP_NORMAL and the learning action have a look at the packet
4079 * before we can drop it. */
4080 if (!stp_forward_in_state(port->stp_state)
4081 && !stp_learn_in_state(port->stp_state)) {
4089 do_xlate_actions(const union ofp_action *in, size_t n_in,
4090 struct action_xlate_ctx *ctx)
4092 const struct ofport_dpif *port;
4093 const union ofp_action *ia;
4096 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4097 if (port && !may_receive(port, ctx)) {
4098 /* Drop this flow. */
4102 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4103 const struct ofp_action_dl_addr *oada;
4104 const struct nx_action_resubmit *nar;
4105 const struct nx_action_set_tunnel *nast;
4106 const struct nx_action_set_queue *nasq;
4107 const struct nx_action_multipath *nam;
4108 const struct nx_action_autopath *naa;
4109 const struct nx_action_bundle *nab;
4110 const struct nx_action_output_reg *naor;
4111 enum ofputil_action_code code;
4118 code = ofputil_decode_action_unsafe(ia);
4120 case OFPUTIL_OFPAT_OUTPUT:
4121 xlate_output_action(ctx, &ia->output);
4124 case OFPUTIL_OFPAT_SET_VLAN_VID:
4125 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4126 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4129 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4130 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4131 ctx->flow.vlan_tci |= htons(
4132 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4135 case OFPUTIL_OFPAT_STRIP_VLAN:
4136 ctx->flow.vlan_tci = htons(0);
4139 case OFPUTIL_OFPAT_SET_DL_SRC:
4140 oada = ((struct ofp_action_dl_addr *) ia);
4141 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4144 case OFPUTIL_OFPAT_SET_DL_DST:
4145 oada = ((struct ofp_action_dl_addr *) ia);
4146 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4149 case OFPUTIL_OFPAT_SET_NW_SRC:
4150 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4153 case OFPUTIL_OFPAT_SET_NW_DST:
4154 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4157 case OFPUTIL_OFPAT_SET_NW_TOS:
4158 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
4159 ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4162 case OFPUTIL_OFPAT_SET_TP_SRC:
4163 ctx->flow.tp_src = ia->tp_port.tp_port;
4166 case OFPUTIL_OFPAT_SET_TP_DST:
4167 ctx->flow.tp_dst = ia->tp_port.tp_port;
4170 case OFPUTIL_OFPAT_ENQUEUE:
4171 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4174 case OFPUTIL_NXAST_RESUBMIT:
4175 nar = (const struct nx_action_resubmit *) ia;
4176 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4179 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4180 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4183 case OFPUTIL_NXAST_SET_TUNNEL:
4184 nast = (const struct nx_action_set_tunnel *) ia;
4185 tun_id = htonll(ntohl(nast->tun_id));
4186 ctx->flow.tun_id = tun_id;
4189 case OFPUTIL_NXAST_SET_QUEUE:
4190 nasq = (const struct nx_action_set_queue *) ia;
4191 xlate_set_queue_action(ctx, nasq);
4194 case OFPUTIL_NXAST_POP_QUEUE:
4195 ctx->flow.priority = ctx->original_priority;
4198 case OFPUTIL_NXAST_REG_MOVE:
4199 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4203 case OFPUTIL_NXAST_REG_LOAD:
4204 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4208 case OFPUTIL_NXAST_NOTE:
4209 /* Nothing to do. */
4212 case OFPUTIL_NXAST_SET_TUNNEL64:
4213 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4214 ctx->flow.tun_id = tun_id;
4217 case OFPUTIL_NXAST_MULTIPATH:
4218 nam = (const struct nx_action_multipath *) ia;
4219 multipath_execute(nam, &ctx->flow);
4222 case OFPUTIL_NXAST_AUTOPATH:
4223 naa = (const struct nx_action_autopath *) ia;
4224 xlate_autopath(ctx, naa);
4227 case OFPUTIL_NXAST_BUNDLE:
4228 ctx->ofproto->has_bundle_action = true;
4229 nab = (const struct nx_action_bundle *) ia;
4230 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4235 case OFPUTIL_NXAST_BUNDLE_LOAD:
4236 ctx->ofproto->has_bundle_action = true;
4237 nab = (const struct nx_action_bundle *) ia;
4238 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4242 case OFPUTIL_NXAST_OUTPUT_REG:
4243 naor = (const struct nx_action_output_reg *) ia;
4244 xlate_output_reg_action(ctx, naor);
4247 case OFPUTIL_NXAST_LEARN:
4248 ctx->has_learn = true;
4249 if (ctx->may_learn) {
4250 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4254 case OFPUTIL_NXAST_EXIT:
4260 /* We've let OFPP_NORMAL and the learning action look at the packet,
4261 * so drop it now if forwarding is disabled. */
4262 if (port && !stp_forward_in_state(port->stp_state)) {
4263 ofpbuf_clear(ctx->odp_actions);
4264 add_sflow_action(ctx);
4269 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4270 struct ofproto_dpif *ofproto, const struct flow *flow,
4271 const struct ofpbuf *packet)
4273 ctx->ofproto = ofproto;
4275 ctx->packet = packet;
4276 ctx->may_learn = packet != NULL;
4277 ctx->resubmit_hook = NULL;
4280 static struct ofpbuf *
4281 xlate_actions(struct action_xlate_ctx *ctx,
4282 const union ofp_action *in, size_t n_in)
4284 COVERAGE_INC(ofproto_dpif_xlate);
4286 ctx->odp_actions = ofpbuf_new(512);
4287 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4289 ctx->may_set_up_flow = true;
4290 ctx->has_learn = false;
4291 ctx->has_normal = false;
4292 ctx->nf_output_iface = NF_OUT_DROP;
4294 ctx->original_priority = ctx->flow.priority;
4295 ctx->base_flow = ctx->flow;
4296 ctx->base_flow.tun_id = 0;
4300 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
4301 switch (ctx->ofproto->up.frag_handling) {
4302 case OFPC_FRAG_NORMAL:
4303 /* We must pretend that transport ports are unavailable. */
4304 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4305 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4308 case OFPC_FRAG_DROP:
4309 return ctx->odp_actions;
4311 case OFPC_FRAG_REASM:
4314 case OFPC_FRAG_NX_MATCH:
4315 /* Nothing to do. */
4320 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4321 ctx->may_set_up_flow = false;
4322 return ctx->odp_actions;
4324 add_sflow_action(ctx);
4325 do_xlate_actions(in, n_in, ctx);
4327 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4328 ctx->odp_actions->data,
4329 ctx->odp_actions->size)) {
4330 ctx->may_set_up_flow = false;
4332 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4334 compose_output_action(ctx, OVSP_LOCAL);
4337 fix_sflow_action(ctx);
4340 return ctx->odp_actions;
4343 /* OFPP_NORMAL implementation. */
4346 struct ofport_dpif *port;
4351 struct dst builtin[32];
4353 size_t n, allocated;
4356 static void dst_set_init(struct dst_set *);
4357 static void dst_set_add(struct dst_set *, const struct dst *);
4358 static void dst_set_free(struct dst_set *);
4360 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4362 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4363 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4364 * the bundle on which the packet was received, returns the VLAN to which the
4367 * Both 'vid' and the return value are in the range 0...4095. */
4369 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4371 switch (in_bundle->vlan_mode) {
4372 case PORT_VLAN_ACCESS:
4373 return in_bundle->vlan;
4376 case PORT_VLAN_TRUNK:
4379 case PORT_VLAN_NATIVE_UNTAGGED:
4380 case PORT_VLAN_NATIVE_TAGGED:
4381 return vid ? vid : in_bundle->vlan;
4388 /* Given 'vlan', the VLAN that a packet belongs to, and
4389 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4390 * that should be included in the 802.1Q header. (If the return value is 0,
4391 * then the 802.1Q header should only be included in the packet if there is a
4394 * Both 'vlan' and the return value are in the range 0...4095. */
4396 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4398 switch (out_bundle->vlan_mode) {
4399 case PORT_VLAN_ACCESS:
4402 case PORT_VLAN_TRUNK:
4403 case PORT_VLAN_NATIVE_TAGGED:
4406 case PORT_VLAN_NATIVE_UNTAGGED:
4407 return vlan == out_bundle->vlan ? 0 : vlan;
4415 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
4416 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
4420 vlan = input_vid_to_vlan(in_bundle, vlan_tci_to_vid(ctx->flow.vlan_tci));
4421 dst->vid = output_vlan_to_vid(out_bundle, vlan);
4423 dst->port = (!out_bundle->bond
4424 ? ofbundle_get_a_port(out_bundle)
4425 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4426 dst->vid, &ctx->tags));
4427 return dst->port != NULL;
4431 mirror_mask_ffs(mirror_mask_t mask)
4433 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4438 dst_set_init(struct dst_set *set)
4440 set->dsts = set->builtin;
4442 set->allocated = ARRAY_SIZE(set->builtin);
4446 dst_set_add(struct dst_set *set, const struct dst *dst)
4448 if (set->n >= set->allocated) {
4449 size_t new_allocated;
4450 struct dst *new_dsts;
4452 new_allocated = set->allocated * 2;
4453 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
4454 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
4458 set->dsts = new_dsts;
4459 set->allocated = new_allocated;
4461 set->dsts[set->n++] = *dst;
4465 dst_set_free(struct dst_set *set)
4467 if (set->dsts != set->builtin) {
4473 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
4476 for (i = 0; i < set->n; i++) {
4477 if (set->dsts[i].vid == test->vid
4478 && set->dsts[i].port == test->port) {
4486 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4488 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4489 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4493 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4495 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4498 /* Returns an arbitrary interface within 'bundle'. */
4499 static struct ofport_dpif *
4500 ofbundle_get_a_port(const struct ofbundle *bundle)
4502 return CONTAINER_OF(list_front(&bundle->ports),
4503 struct ofport_dpif, bundle_node);
4507 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
4508 const struct ofbundle *in_bundle,
4509 const struct ofbundle *out_bundle, struct dst_set *set)
4513 if (out_bundle == OFBUNDLE_FLOOD) {
4514 struct ofbundle *bundle;
4516 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
4517 if (bundle != in_bundle
4518 && ofbundle_includes_vlan(bundle, vlan)
4519 && bundle->floodable
4520 && !bundle->mirror_out
4521 && set_dst(ctx, &dst, in_bundle, bundle)) {
4522 dst_set_add(set, &dst);
4525 ctx->nf_output_iface = NF_OUT_FLOOD;
4526 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
4527 dst_set_add(set, &dst);
4528 ctx->nf_output_iface = dst.port->odp_port;
4533 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4535 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4538 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4539 * to a VLAN. In general most packets may be mirrored but we want to drop
4540 * protocols that may confuse switches. */
4542 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4544 /* If you change this function's behavior, please update corresponding
4545 * documentation in vswitch.xml at the same time. */
4546 if (dst[0] != 0x01) {
4547 /* All the currently banned MACs happen to start with 01 currently, so
4548 * this is a quick way to eliminate most of the good ones. */
4550 if (eth_addr_is_reserved(dst)) {
4551 /* Drop STP, IEEE pause frames, and other reserved protocols
4552 * (01-80-c2-00-00-0x). */
4556 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4558 if ((dst[3] & 0xfe) == 0xcc &&
4559 (dst[4] & 0xfe) == 0xcc &&
4560 (dst[5] & 0xfe) == 0xcc) {
4561 /* Drop the following protocols plus others following the same
4564 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4565 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4566 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4570 if (!(dst[3] | dst[4] | dst[5])) {
4571 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4580 compose_mirror_dsts(struct action_xlate_ctx *ctx,
4581 uint16_t vlan, const struct ofbundle *in_bundle,
4582 struct dst_set *set)
4584 struct ofproto_dpif *ofproto = ctx->ofproto;
4585 mirror_mask_t mirrors;
4589 mirrors = in_bundle->src_mirrors;
4590 for (i = 0; i < set->n; i++) {
4591 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
4598 flow_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4600 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
4601 if (vlan_is_mirrored(m, vlan)) {
4605 if (set_dst(ctx, &dst, in_bundle, m->out)
4606 && !dst_is_duplicate(set, &dst)) {
4607 dst_set_add(set, &dst);
4609 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
4610 struct ofbundle *bundle;
4612 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4613 if (ofbundle_includes_vlan(bundle, m->out_vlan)
4614 && set_dst(ctx, &dst, in_bundle, bundle))
4616 /* set_dst() got dst->vid from the input packet's VLAN,
4617 * not from m->out_vlan, so recompute it. */
4618 dst.vid = output_vlan_to_vid(bundle, m->out_vlan);
4620 if (dst_is_duplicate(set, &dst)) {
4624 if (bundle == in_bundle && dst.vid == flow_vid) {
4625 /* Don't send out input port on same VLAN. */
4628 dst_set_add(set, &dst);
4633 mirrors &= mirrors - 1;
4638 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
4639 const struct ofbundle *in_bundle,
4640 const struct ofbundle *out_bundle)
4642 uint16_t initial_vid, cur_vid;
4643 const struct dst *dst;
4647 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
4648 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
4654 /* Output all the packets we can without having to change the VLAN. */
4655 commit_odp_actions(ctx);
4656 initial_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4657 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
4658 if (dst->vid != initial_vid) {
4661 compose_output_action(ctx, dst->port->odp_port);
4664 /* Then output the rest. */
4665 cur_vid = initial_vid;
4666 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
4667 if (dst->vid == initial_vid) {
4670 if (dst->vid != cur_vid) {
4673 tci = htons(dst->vid);
4674 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4676 tci |= htons(VLAN_CFI);
4678 commit_vlan_action(ctx, tci);
4682 compose_output_action(ctx, dst->port->odp_port);
4688 /* Returns the effective vlan of a packet, taking into account both the
4689 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
4690 * the packet is untagged and -1 indicates it has an invalid header and
4691 * should be dropped. */
4693 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
4694 struct ofbundle *in_bundle, bool have_packet)
4696 int vlan = vlan_tci_to_vid(flow->vlan_tci);
4698 if (in_bundle->vlan_mode == PORT_VLAN_ACCESS) {
4699 /* Drop tagged packet on access port */
4701 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4702 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
4703 "packet received on port %s configured with "
4704 "implicit VLAN %"PRIu16,
4705 ofproto->up.name, vlan,
4706 in_bundle->name, in_bundle->vlan);
4709 } else if (ofbundle_includes_vlan(in_bundle, vlan)) {
4712 /* Drop packets from a VLAN not member of the trunk */
4714 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4715 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
4716 "packet received on port %s not configured for "
4718 ofproto->up.name, vlan, in_bundle->name, vlan);
4723 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
4724 !(flow->vlan_tci & htons(VLAN_CFI))) {
4725 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4726 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
4727 "VLAN tag received on port %s",
4728 ofproto->up.name, in_bundle->name);
4731 if (in_bundle->vlan_mode != PORT_VLAN_TRUNK) {
4732 return in_bundle->vlan;
4734 return ofbundle_includes_vlan(in_bundle, 0) ? 0 : -1;
4739 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
4740 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
4741 * indicate this; newer upstream kernels use gratuitous ARP requests. */
4743 is_gratuitous_arp(const struct flow *flow)
4745 return (flow->dl_type == htons(ETH_TYPE_ARP)
4746 && eth_addr_is_broadcast(flow->dl_dst)
4747 && (flow->nw_proto == ARP_OP_REPLY
4748 || (flow->nw_proto == ARP_OP_REQUEST
4749 && flow->nw_src == flow->nw_dst)));
4753 update_learning_table(struct ofproto_dpif *ofproto,
4754 const struct flow *flow, int vlan,
4755 struct ofbundle *in_bundle)
4757 struct mac_entry *mac;
4759 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
4763 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
4764 if (is_gratuitous_arp(flow)) {
4765 /* We don't want to learn from gratuitous ARP packets that are
4766 * reflected back over bond slaves so we lock the learning table. */
4767 if (!in_bundle->bond) {
4768 mac_entry_set_grat_arp_lock(mac);
4769 } else if (mac_entry_is_grat_arp_locked(mac)) {
4774 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
4775 /* The log messages here could actually be useful in debugging,
4776 * so keep the rate limit relatively high. */
4777 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4778 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
4779 "on port %s in VLAN %d",
4780 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
4781 in_bundle->name, vlan);
4783 mac->port.p = in_bundle;
4784 tag_set_add(&ofproto->revalidate_set,
4785 mac_learning_changed(ofproto->ml, mac));
4789 /* Determines whether packets in 'flow' within 'br' should be forwarded or
4790 * dropped. Returns true if they may be forwarded, false if they should be
4793 * If 'have_packet' is true, it indicates that the caller is processing a
4794 * received packet. If 'have_packet' is false, then the caller is just
4795 * revalidating an existing flow because configuration has changed. Either
4796 * way, 'have_packet' only affects logging (there is no point in logging errors
4797 * during revalidation).
4799 * Sets '*in_portp' to the input port. This will be a null pointer if
4800 * flow->in_port does not designate a known input port (in which case
4801 * is_admissible() returns false).
4803 * When returning true, sets '*vlanp' to the effective VLAN of the input
4804 * packet, as returned by flow_get_vlan().
4806 * May also add tags to '*tags', although the current implementation only does
4807 * so in one special case.
4810 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
4812 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
4814 struct ofport_dpif *in_port;
4815 struct ofbundle *in_bundle;
4818 /* Find the port and bundle for the received packet. */
4819 in_port = get_ofp_port(ofproto, flow->in_port);
4820 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
4821 if (!in_port || !in_bundle) {
4822 /* No interface? Something fishy... */
4824 /* Odd. A few possible reasons here:
4826 * - We deleted a port but there are still a few packets queued up
4829 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
4830 * we don't know about.
4832 * - Packet arrived on the local port but the local port is not
4835 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4837 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
4839 ofproto->up.name, flow->in_port);
4844 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
4849 /* Drop frames for reserved multicast addresses only if forward_bpdu
4850 * option is absent. */
4851 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
4855 /* Drop frames on bundles reserved for mirroring. */
4856 if (in_bundle->mirror_out) {
4858 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4859 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4860 "%s, which is reserved exclusively for mirroring",
4861 ofproto->up.name, in_bundle->name);
4866 if (in_bundle->bond) {
4867 struct mac_entry *mac;
4869 switch (bond_check_admissibility(in_bundle->bond, in_port,
4870 flow->dl_dst, tags)) {
4877 case BV_DROP_IF_MOVED:
4878 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
4879 if (mac && mac->port.p != in_bundle &&
4880 (!is_gratuitous_arp(flow)
4881 || mac_entry_is_grat_arp_locked(mac))) {
4892 xlate_normal(struct action_xlate_ctx *ctx)
4894 struct ofbundle *in_bundle;
4895 struct ofbundle *out_bundle;
4896 struct mac_entry *mac;
4899 ctx->has_normal = true;
4901 /* Check whether we should drop packets in this flow. */
4902 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
4903 &ctx->tags, &vlan, &in_bundle)) {
4908 /* Learn source MAC. */
4909 if (ctx->may_learn) {
4910 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
4913 /* Determine output bundle. */
4914 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
4917 out_bundle = mac->port.p;
4918 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
4919 /* If we are revalidating but don't have a learning entry then eject
4920 * the flow. Installing a flow that floods packets opens up a window
4921 * of time where we could learn from a packet reflected on a bond and
4922 * blackhole packets before the learning table is updated to reflect
4923 * the correct port. */
4924 ctx->may_set_up_flow = false;
4927 out_bundle = OFBUNDLE_FLOOD;
4930 /* Don't send packets out their input bundles. */
4931 if (in_bundle == out_bundle) {
4937 compose_actions(ctx, vlan, in_bundle, out_bundle);
4941 /* Optimized flow revalidation.
4943 * It's a difficult problem, in general, to tell which facets need to have
4944 * their actions recalculated whenever the OpenFlow flow table changes. We
4945 * don't try to solve that general problem: for most kinds of OpenFlow flow
4946 * table changes, we recalculate the actions for every facet. This is
4947 * relatively expensive, but it's good enough if the OpenFlow flow table
4948 * doesn't change very often.
4950 * However, we can expect one particular kind of OpenFlow flow table change to
4951 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
4952 * of CPU on revalidating every facet whenever MAC learning modifies the flow
4953 * table, we add a special case that applies to flow tables in which every rule
4954 * has the same form (that is, the same wildcards), except that the table is
4955 * also allowed to have a single "catch-all" flow that matches all packets. We
4956 * optimize this case by tagging all of the facets that resubmit into the table
4957 * and invalidating the same tag whenever a flow changes in that table. The
4958 * end result is that we revalidate just the facets that need it (and sometimes
4959 * a few more, but not all of the facets or even all of the facets that
4960 * resubmit to the table modified by MAC learning). */
4962 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
4963 * into an OpenFlow table with the given 'basis'. */
4965 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
4968 if (flow_wildcards_is_catchall(wc)) {
4971 struct flow tag_flow = *flow;
4972 flow_zero_wildcards(&tag_flow, wc);
4973 return tag_create_deterministic(flow_hash(&tag_flow, secret));
4977 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
4978 * taggability of that table.
4980 * This function must be called after *each* change to a flow table. If you
4981 * skip calling it on some changes then the pointer comparisons at the end can
4982 * be invalid if you get unlucky. For example, if a flow removal causes a
4983 * cls_table to be destroyed and then a flow insertion causes a cls_table with
4984 * different wildcards to be created with the same address, then this function
4985 * will incorrectly skip revalidation. */
4987 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
4989 struct table_dpif *table = &ofproto->tables[table_id];
4990 const struct classifier *cls = &ofproto->up.tables[table_id];
4991 struct cls_table *catchall, *other;
4992 struct cls_table *t;
4994 catchall = other = NULL;
4996 switch (hmap_count(&cls->tables)) {
4998 /* We could tag this OpenFlow table but it would make the logic a
4999 * little harder and it's a corner case that doesn't seem worth it
5005 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5006 if (cls_table_is_catchall(t)) {
5008 } else if (!other) {
5011 /* Indicate that we can't tag this by setting both tables to
5012 * NULL. (We know that 'catchall' is already NULL.) */
5019 /* Can't tag this table. */
5023 if (table->catchall_table != catchall || table->other_table != other) {
5024 table->catchall_table = catchall;
5025 table->other_table = other;
5026 ofproto->need_revalidate = true;
5030 /* Given 'rule' that has changed in some way (either it is a rule being
5031 * inserted, a rule being deleted, or a rule whose actions are being
5032 * modified), marks facets for revalidation to ensure that packets will be
5033 * forwarded correctly according to the new state of the flow table.
5035 * This function must be called after *each* change to a flow table. See
5036 * the comment on table_update_taggable() for more information. */
5038 rule_invalidate(const struct rule_dpif *rule)
5040 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5042 table_update_taggable(ofproto, rule->up.table_id);
5044 if (!ofproto->need_revalidate) {
5045 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5047 if (table->other_table && rule->tag) {
5048 tag_set_add(&ofproto->revalidate_set, rule->tag);
5050 ofproto->need_revalidate = true;
5056 set_frag_handling(struct ofproto *ofproto_,
5057 enum ofp_config_flags frag_handling)
5059 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5061 if (frag_handling != OFPC_FRAG_REASM) {
5062 ofproto->need_revalidate = true;
5070 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5071 const struct flow *flow,
5072 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5074 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5077 if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) {
5078 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
5081 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5082 ofproto->max_ports);
5084 struct odputil_keybuf keybuf;
5085 struct action_xlate_ctx ctx;
5086 struct ofpbuf *odp_actions;
5089 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5090 odp_flow_key_from_flow(&key, flow);
5092 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
5093 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5094 dpif_execute(ofproto->dpif, key.data, key.size,
5095 odp_actions->data, odp_actions->size, packet);
5096 ofpbuf_delete(odp_actions);
5102 get_netflow_ids(const struct ofproto *ofproto_,
5103 uint8_t *engine_type, uint8_t *engine_id)
5105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5107 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5110 static struct ofproto_dpif *
5111 ofproto_dpif_lookup(const char *name)
5113 struct ofproto *ofproto = ofproto_lookup(name);
5114 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
5115 ? ofproto_dpif_cast(ofproto)
5120 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn,
5121 const char *args, void *aux OVS_UNUSED)
5123 const struct ofproto_dpif *ofproto;
5125 ofproto = ofproto_dpif_lookup(args);
5127 unixctl_command_reply(conn, 501, "no such bridge");
5130 mac_learning_flush(ofproto->ml);
5132 unixctl_command_reply(conn, 200, "table successfully flushed");
5136 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5137 const char *args, void *aux OVS_UNUSED)
5139 struct ds ds = DS_EMPTY_INITIALIZER;
5140 const struct ofproto_dpif *ofproto;
5141 const struct mac_entry *e;
5143 ofproto = ofproto_dpif_lookup(args);
5145 unixctl_command_reply(conn, 501, "no such bridge");
5149 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5150 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5151 struct ofbundle *bundle = e->port.p;
5152 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5153 ofbundle_get_a_port(bundle)->odp_port,
5154 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5156 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5160 struct ofproto_trace {
5161 struct action_xlate_ctx ctx;
5167 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5168 const struct rule_dpif *rule)
5170 ds_put_char_multiple(result, '\t', level);
5172 ds_put_cstr(result, "No match\n");
5176 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5177 table_id, ntohll(rule->up.flow_cookie));
5178 cls_rule_format(&rule->up.cr, result);
5179 ds_put_char(result, '\n');
5181 ds_put_char_multiple(result, '\t', level);
5182 ds_put_cstr(result, "OpenFlow ");
5183 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5184 ds_put_char(result, '\n');
5188 trace_format_flow(struct ds *result, int level, const char *title,
5189 struct ofproto_trace *trace)
5191 ds_put_char_multiple(result, '\t', level);
5192 ds_put_format(result, "%s: ", title);
5193 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5194 ds_put_cstr(result, "unchanged");
5196 flow_format(result, &trace->ctx.flow);
5197 trace->flow = trace->ctx.flow;
5199 ds_put_char(result, '\n');
5203 trace_format_regs(struct ds *result, int level, const char *title,
5204 struct ofproto_trace *trace)
5208 ds_put_char_multiple(result, '\t', level);
5209 ds_put_format(result, "%s:", title);
5210 for (i = 0; i < FLOW_N_REGS; i++) {
5211 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5213 ds_put_char(result, '\n');
5217 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5219 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5220 struct ds *result = trace->result;
5222 ds_put_char(result, '\n');
5223 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5224 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5225 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5229 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5230 void *aux OVS_UNUSED)
5232 char *dpname, *arg1, *arg2, *arg3, *arg4;
5233 char *args = xstrdup(args_);
5234 char *save_ptr = NULL;
5235 struct ofproto_dpif *ofproto;
5236 struct ofpbuf odp_key;
5237 struct ofpbuf *packet;
5238 struct rule_dpif *rule;
5244 ofpbuf_init(&odp_key, 0);
5247 dpname = strtok_r(args, " ", &save_ptr);
5248 arg1 = strtok_r(NULL, " ", &save_ptr);
5249 arg2 = strtok_r(NULL, " ", &save_ptr);
5250 arg3 = strtok_r(NULL, " ", &save_ptr);
5251 arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5252 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5253 /* ofproto/trace dpname flow [-generate] */
5256 /* Convert string to datapath key. */
5257 ofpbuf_init(&odp_key, 0);
5258 error = odp_flow_key_from_string(arg1, &odp_key);
5260 unixctl_command_reply(conn, 501, "Bad flow syntax");
5264 /* Convert odp_key to flow. */
5265 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
5267 unixctl_command_reply(conn, 501, "Invalid flow");
5271 /* Generate a packet, if requested. */
5273 packet = ofpbuf_new(0);
5274 flow_compose(packet, &flow);
5276 } else if (dpname && arg1 && arg2 && arg3 && arg4) {
5277 /* ofproto/trace dpname priority tun_id in_port packet */
5282 priority = atoi(arg1);
5283 tun_id = htonll(strtoull(arg2, NULL, 0));
5284 in_port = ofp_port_to_odp_port(atoi(arg3));
5286 packet = ofpbuf_new(strlen(args) / 2);
5287 arg4 = ofpbuf_put_hex(packet, arg4, NULL);
5288 arg4 += strspn(arg4, " ");
5289 if (*arg4 != '\0') {
5290 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5293 if (packet->size < ETH_HEADER_LEN) {
5294 unixctl_command_reply(conn, 501,
5295 "Packet data too short for Ethernet");
5299 ds_put_cstr(&result, "Packet: ");
5300 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5301 ds_put_cstr(&result, s);
5304 flow_extract(packet, priority, tun_id, in_port, &flow);
5306 unixctl_command_reply(conn, 501, "Bad command syntax");
5310 ofproto = ofproto_dpif_lookup(dpname);
5312 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5317 ds_put_cstr(&result, "Flow: ");
5318 flow_format(&result, &flow);
5319 ds_put_char(&result, '\n');
5321 rule = rule_dpif_lookup(ofproto, &flow, 0);
5322 trace_format_rule(&result, 0, 0, rule);
5324 struct ofproto_trace trace;
5325 struct ofpbuf *odp_actions;
5327 trace.result = &result;
5329 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
5330 trace.ctx.resubmit_hook = trace_resubmit;
5331 odp_actions = xlate_actions(&trace.ctx,
5332 rule->up.actions, rule->up.n_actions);
5334 ds_put_char(&result, '\n');
5335 trace_format_flow(&result, 0, "Final flow", &trace);
5336 ds_put_cstr(&result, "Datapath actions: ");
5337 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5338 ofpbuf_delete(odp_actions);
5340 if (!trace.ctx.may_set_up_flow) {
5342 ds_put_cstr(&result, "\nThis flow is not cachable.");
5344 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5345 "for complete actions, please supply a packet.");
5350 unixctl_command_reply(conn, 200, ds_cstr(&result));
5353 ds_destroy(&result);
5354 ofpbuf_delete(packet);
5355 ofpbuf_uninit(&odp_key);
5360 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5361 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5364 unixctl_command_reply(conn, 200, NULL);
5368 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5369 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5372 unixctl_command_reply(conn, 200, NULL);
5376 ofproto_dpif_unixctl_init(void)
5378 static bool registered;
5384 unixctl_command_register("ofproto/trace",
5385 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5386 ofproto_unixctl_trace, NULL);
5387 unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush,
5389 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5391 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5392 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5395 const struct ofproto_class ofproto_dpif_class = {
5422 port_is_lacp_current,
5423 NULL, /* rule_choose_table */
5430 rule_modify_actions,
5438 get_cfm_remote_mpids,
5442 get_stp_port_status,
5447 is_mirror_output_bundle,
5448 forward_bpdu_changed,