2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "multipath.h"
44 #include "ofp-print.h"
45 #include "ofproto-dpif-sflow.h"
46 #include "poll-loop.h"
48 #include "unaligned.h"
50 #include "vlan-bitmap.h"
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
55 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
56 COVERAGE_DEFINE(ofproto_dpif_expired);
57 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
58 COVERAGE_DEFINE(ofproto_dpif_xlate);
59 COVERAGE_DEFINE(facet_changed_rule);
60 COVERAGE_DEFINE(facet_invalidated);
61 COVERAGE_DEFINE(facet_revalidate);
62 COVERAGE_DEFINE(facet_unexpected);
64 /* Maximum depth of flow table recursion (due to resubmit actions) in a
65 * flow translation. */
66 #define MAX_RESUBMIT_RECURSION 32
68 /* Number of implemented OpenFlow tables. */
69 enum { N_TABLES = 255 };
70 BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255);
78 long long int used; /* Time last used; time created if not used. */
82 * - Do include packets and bytes from facets that have been deleted or
83 * whose own statistics have been folded into the rule.
85 * - Do include packets and bytes sent "by hand" that were accounted to
86 * the rule without any facet being involved (this is a rare corner
87 * case in rule_execute()).
89 * - Do not include packet or bytes that can be obtained from any facet's
90 * packet_count or byte_count member or that can be obtained from the
91 * datapath by, e.g., dpif_flow_get() for any facet.
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
96 tag_type tag; /* Caches rule_calculate_tag() result. */
98 struct list facets; /* List of "struct facet"s. */
101 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
103 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
106 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
107 const struct flow *, uint8_t table);
109 #define MAX_MIRRORS 32
110 typedef uint32_t mirror_mask_t;
111 #define MIRROR_MASK_C(X) UINT32_C(X)
112 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
114 struct ofproto_dpif *ofproto; /* Owning ofproto. */
115 size_t idx; /* In ofproto's "mirrors" array. */
116 void *aux; /* Key supplied by ofproto's client. */
117 char *name; /* Identifier for log messages. */
119 /* Selection criteria. */
120 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
121 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
122 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
124 /* Output (mutually exclusive). */
125 struct ofbundle *out; /* Output port or NULL. */
126 int out_vlan; /* Output VLAN or -1. */
129 static void mirror_destroy(struct ofmirror *);
131 /* A group of one or more OpenFlow ports. */
132 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
140 struct list ports; /* Contains "struct ofport"s. */
141 enum port_vlan_mode vlan_mode; /* VLAN mode */
142 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
143 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
144 * NULL if all VLANs are trunked. */
145 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
146 struct bond *bond; /* Nonnull iff more than one port. */
149 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
151 /* Port mirroring info. */
152 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
153 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
154 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
157 static void bundle_remove(struct ofport *);
158 static void bundle_update(struct ofbundle *);
159 static void bundle_destroy(struct ofbundle *);
160 static void bundle_del_port(struct ofport_dpif *);
161 static void bundle_run(struct ofbundle *);
162 static void bundle_wait(struct ofbundle *);
164 static void stp_run(struct ofproto_dpif *ofproto);
165 static void stp_wait(struct ofproto_dpif *ofproto);
167 struct action_xlate_ctx {
168 /* action_xlate_ctx_init() initializes these members. */
171 struct ofproto_dpif *ofproto;
173 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
174 * this flow when actions change header fields. */
177 /* The packet corresponding to 'flow', or a null pointer if we are
178 * revalidating without a packet to refer to. */
179 const struct ofpbuf *packet;
181 /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We
182 * want to execute them if we are actually processing a packet, or if we
183 * are accounting for packets that the datapath has processed, but not if
184 * we are just revalidating. */
187 /* If nonnull, called just before executing a resubmit action.
189 * This is normally null so the client has to set it manually after
190 * calling action_xlate_ctx_init(). */
191 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
193 /* xlate_actions() initializes and uses these members. The client might want
194 * to look at them after it returns. */
196 struct ofpbuf *odp_actions; /* Datapath actions. */
197 tag_type tags; /* Tags associated with actions. */
198 bool may_set_up_flow; /* True ordinarily; false if the actions must
199 * be reassessed for every packet. */
200 bool has_learn; /* Actions include NXAST_LEARN? */
201 bool has_normal; /* Actions output to OFPP_NORMAL? */
202 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
204 /* xlate_actions() initializes and uses these members, but the client has no
205 * reason to look at them. */
207 int recurse; /* Recursion level, via xlate_table_action. */
208 uint32_t priority; /* Current flow priority. 0 if none. */
209 struct flow base_flow; /* Flow at the last commit. */
210 uint32_t base_priority; /* Priority at the last commit. */
211 uint8_t table_id; /* OpenFlow table ID where flow was found. */
212 uint32_t sflow_n_outputs; /* Number of output ports. */
213 uint16_t sflow_odp_port; /* Output port for composing sFlow action. */
214 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
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 int error = netdev_send(ofport->up.netdev, pkt);
954 VLOG_WARN_RL(&rl, "%s: sending BPDU on port %s failed (%s)",
956 netdev_get_name(ofport->up.netdev),
964 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
966 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
968 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
970 /* Only revalidate flows if the configuration changed. */
971 if (!s != !ofproto->stp) {
972 ofproto->need_revalidate = true;
977 ofproto->stp = stp_create(ofproto_->name, s->system_id,
978 send_bpdu_cb, ofproto);
979 ofproto->stp_last_tick = time_msec();
982 stp_set_bridge_id(ofproto->stp, s->system_id);
983 stp_set_bridge_priority(ofproto->stp, s->priority);
984 stp_set_hello_time(ofproto->stp, s->hello_time);
985 stp_set_max_age(ofproto->stp, s->max_age);
986 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
988 stp_destroy(ofproto->stp);
996 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
998 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1002 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1003 s->designated_root = stp_get_designated_root(ofproto->stp);
1004 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1013 update_stp_port_state(struct ofport_dpif *ofport)
1015 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1016 enum stp_state state;
1018 /* Figure out new state. */
1019 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1023 if (ofport->stp_state != state) {
1027 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1028 netdev_get_name(ofport->up.netdev),
1029 stp_state_name(ofport->stp_state),
1030 stp_state_name(state));
1031 if (stp_learn_in_state(ofport->stp_state)
1032 != stp_learn_in_state(state)) {
1033 /* xxx Learning action flows should also be flushed. */
1034 mac_learning_flush(ofproto->ml);
1036 fwd_change = stp_forward_in_state(ofport->stp_state)
1037 != stp_forward_in_state(state);
1039 ofproto->need_revalidate = true;
1040 ofport->stp_state = state;
1041 ofport->stp_state_entered = time_msec();
1044 bundle_update(ofport->bundle);
1047 /* Update the STP state bits in the OpenFlow port description. */
1048 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1049 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1050 : state == STP_LEARNING ? OFPPS_STP_LEARN
1051 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1052 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1054 ofproto_port_set_state(&ofport->up, of_state);
1058 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1059 * caller is responsible for assigning STP port numbers and ensuring
1060 * there are no duplicates. */
1062 set_stp_port(struct ofport *ofport_,
1063 const struct ofproto_port_stp_settings *s)
1065 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1066 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1067 struct stp_port *sp = ofport->stp_port;
1069 if (!s || !s->enable) {
1071 ofport->stp_port = NULL;
1072 stp_port_disable(sp);
1075 } else if (sp && stp_port_no(sp) != s->port_num
1076 && ofport == stp_port_get_aux(sp)) {
1077 /* The port-id changed, so disable the old one if it's not
1078 * already in use by another port. */
1079 stp_port_disable(sp);
1082 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1083 stp_port_enable(sp);
1085 stp_port_set_aux(sp, ofport);
1086 stp_port_set_priority(sp, s->priority);
1087 stp_port_set_path_cost(sp, s->path_cost);
1089 update_stp_port_state(ofport);
1095 get_stp_port_status(struct ofport *ofport_,
1096 struct ofproto_port_stp_status *s)
1098 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1100 struct stp_port *sp = ofport->stp_port;
1102 if (!ofproto->stp || !sp) {
1108 s->port_id = stp_port_get_id(sp);
1109 s->state = stp_port_get_state(sp);
1110 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1111 s->role = stp_port_get_role(sp);
1117 stp_run(struct ofproto_dpif *ofproto)
1120 long long int now = time_msec();
1121 long long int elapsed = now - ofproto->stp_last_tick;
1122 struct stp_port *sp;
1125 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1126 ofproto->stp_last_tick = now;
1128 while (stp_get_changed_port(ofproto->stp, &sp)) {
1129 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1132 update_stp_port_state(ofport);
1139 stp_wait(struct ofproto_dpif *ofproto)
1142 poll_timer_wait(1000);
1146 /* Returns true if STP should process 'flow'. */
1148 stp_should_process_flow(const struct flow *flow)
1150 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1154 stp_process_packet(const struct ofport_dpif *ofport,
1155 const struct ofpbuf *packet)
1157 struct ofpbuf payload = *packet;
1158 struct eth_header *eth = payload.data;
1159 struct stp_port *sp = ofport->stp_port;
1161 /* Sink packets on ports that have STP disabled when the bridge has
1163 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1167 /* Trim off padding on payload. */
1168 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1169 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1172 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1173 stp_received_bpdu(sp, payload.data, payload.size);
1179 /* Expires all MAC learning entries associated with 'port' and forces ofproto
1180 * to revalidate every flow. */
1182 bundle_flush_macs(struct ofbundle *bundle)
1184 struct ofproto_dpif *ofproto = bundle->ofproto;
1185 struct mac_learning *ml = ofproto->ml;
1186 struct mac_entry *mac, *next_mac;
1188 ofproto->need_revalidate = true;
1189 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1190 if (mac->port.p == bundle) {
1191 mac_learning_expire(ml, mac);
1196 static struct ofbundle *
1197 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1199 struct ofbundle *bundle;
1201 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1202 &ofproto->bundles) {
1203 if (bundle->aux == aux) {
1210 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1211 * ones that are found to 'bundles'. */
1213 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1214 void **auxes, size_t n_auxes,
1215 struct hmapx *bundles)
1219 hmapx_init(bundles);
1220 for (i = 0; i < n_auxes; i++) {
1221 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1223 hmapx_add(bundles, bundle);
1229 bundle_update(struct ofbundle *bundle)
1231 struct ofport_dpif *port;
1233 bundle->floodable = true;
1234 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1235 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1236 || !stp_forward_in_state(port->stp_state)) {
1237 bundle->floodable = false;
1244 bundle_del_port(struct ofport_dpif *port)
1246 struct ofbundle *bundle = port->bundle;
1248 bundle->ofproto->need_revalidate = true;
1250 list_remove(&port->bundle_node);
1251 port->bundle = NULL;
1254 lacp_slave_unregister(bundle->lacp, port);
1257 bond_slave_unregister(bundle->bond, port);
1260 bundle_update(bundle);
1264 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1265 struct lacp_slave_settings *lacp,
1266 uint32_t bond_stable_id)
1268 struct ofport_dpif *port;
1270 port = get_ofp_port(bundle->ofproto, ofp_port);
1275 if (port->bundle != bundle) {
1276 bundle->ofproto->need_revalidate = true;
1278 bundle_del_port(port);
1281 port->bundle = bundle;
1282 list_push_back(&bundle->ports, &port->bundle_node);
1283 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1284 || !stp_forward_in_state(port->stp_state)) {
1285 bundle->floodable = false;
1289 port->bundle->ofproto->need_revalidate = true;
1290 lacp_slave_register(bundle->lacp, port, lacp);
1293 port->bond_stable_id = bond_stable_id;
1299 bundle_destroy(struct ofbundle *bundle)
1301 struct ofproto_dpif *ofproto;
1302 struct ofport_dpif *port, *next_port;
1309 ofproto = bundle->ofproto;
1310 for (i = 0; i < MAX_MIRRORS; i++) {
1311 struct ofmirror *m = ofproto->mirrors[i];
1313 if (m->out == bundle) {
1315 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1316 || hmapx_find_and_delete(&m->dsts, bundle)) {
1317 ofproto->need_revalidate = true;
1322 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1323 bundle_del_port(port);
1326 bundle_flush_macs(bundle);
1327 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1329 free(bundle->trunks);
1330 lacp_destroy(bundle->lacp);
1331 bond_destroy(bundle->bond);
1336 bundle_set(struct ofproto *ofproto_, void *aux,
1337 const struct ofproto_bundle_settings *s)
1339 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1340 bool need_flush = false;
1341 struct ofport_dpif *port;
1342 struct ofbundle *bundle;
1343 unsigned long *trunks;
1349 bundle_destroy(bundle_lookup(ofproto, aux));
1353 assert(s->n_slaves == 1 || s->bond != NULL);
1354 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1356 bundle = bundle_lookup(ofproto, aux);
1358 bundle = xmalloc(sizeof *bundle);
1360 bundle->ofproto = ofproto;
1361 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1362 hash_pointer(aux, 0));
1364 bundle->name = NULL;
1366 list_init(&bundle->ports);
1367 bundle->vlan_mode = PORT_VLAN_TRUNK;
1369 bundle->trunks = NULL;
1370 bundle->lacp = NULL;
1371 bundle->bond = NULL;
1373 bundle->floodable = true;
1375 bundle->src_mirrors = 0;
1376 bundle->dst_mirrors = 0;
1377 bundle->mirror_out = 0;
1380 if (!bundle->name || strcmp(s->name, bundle->name)) {
1382 bundle->name = xstrdup(s->name);
1387 if (!bundle->lacp) {
1388 ofproto->need_revalidate = true;
1389 bundle->lacp = lacp_create();
1391 lacp_configure(bundle->lacp, s->lacp);
1393 lacp_destroy(bundle->lacp);
1394 bundle->lacp = NULL;
1397 /* Update set of ports. */
1399 for (i = 0; i < s->n_slaves; i++) {
1400 if (!bundle_add_port(bundle, s->slaves[i],
1401 s->lacp ? &s->lacp_slaves[i] : NULL,
1402 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1406 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1407 struct ofport_dpif *next_port;
1409 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1410 for (i = 0; i < s->n_slaves; i++) {
1411 if (s->slaves[i] == port->up.ofp_port) {
1416 bundle_del_port(port);
1420 assert(list_size(&bundle->ports) <= s->n_slaves);
1422 if (list_is_empty(&bundle->ports)) {
1423 bundle_destroy(bundle);
1427 /* Set VLAN tagging mode */
1428 if (s->vlan_mode != bundle->vlan_mode) {
1429 bundle->vlan_mode = s->vlan_mode;
1434 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1435 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1437 if (vlan != bundle->vlan) {
1438 bundle->vlan = vlan;
1442 /* Get trunked VLANs. */
1443 switch (s->vlan_mode) {
1444 case PORT_VLAN_ACCESS:
1448 case PORT_VLAN_TRUNK:
1449 trunks = (unsigned long *) s->trunks;
1452 case PORT_VLAN_NATIVE_UNTAGGED:
1453 case PORT_VLAN_NATIVE_TAGGED:
1454 if (vlan != 0 && (!s->trunks
1455 || !bitmap_is_set(s->trunks, vlan)
1456 || bitmap_is_set(s->trunks, 0))) {
1457 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1459 trunks = bitmap_clone(s->trunks, 4096);
1461 trunks = bitmap_allocate1(4096);
1463 bitmap_set1(trunks, vlan);
1464 bitmap_set0(trunks, 0);
1466 trunks = (unsigned long *) s->trunks;
1473 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1474 free(bundle->trunks);
1475 if (trunks == s->trunks) {
1476 bundle->trunks = vlan_bitmap_clone(trunks);
1478 bundle->trunks = trunks;
1483 if (trunks != s->trunks) {
1488 if (!list_is_short(&bundle->ports)) {
1489 bundle->ofproto->has_bonded_bundles = true;
1491 if (bond_reconfigure(bundle->bond, s->bond)) {
1492 ofproto->need_revalidate = true;
1495 bundle->bond = bond_create(s->bond);
1496 ofproto->need_revalidate = true;
1499 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1500 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1504 bond_destroy(bundle->bond);
1505 bundle->bond = NULL;
1508 /* If we changed something that would affect MAC learning, un-learn
1509 * everything on this port and force flow revalidation. */
1511 bundle_flush_macs(bundle);
1518 bundle_remove(struct ofport *port_)
1520 struct ofport_dpif *port = ofport_dpif_cast(port_);
1521 struct ofbundle *bundle = port->bundle;
1524 bundle_del_port(port);
1525 if (list_is_empty(&bundle->ports)) {
1526 bundle_destroy(bundle);
1527 } else if (list_is_short(&bundle->ports)) {
1528 bond_destroy(bundle->bond);
1529 bundle->bond = NULL;
1535 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1537 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1538 struct ofport_dpif *port = port_;
1539 uint8_t ea[ETH_ADDR_LEN];
1542 error = netdev_get_etheraddr(port->up.netdev, ea);
1544 struct ofpbuf packet;
1547 ofpbuf_init(&packet, 0);
1548 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1550 memcpy(packet_pdu, pdu, pdu_size);
1552 error = netdev_send(port->up.netdev, &packet);
1554 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1555 "(%s)", port->bundle->name,
1556 netdev_get_name(port->up.netdev), strerror(error));
1558 ofpbuf_uninit(&packet);
1560 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1561 "%s (%s)", port->bundle->name,
1562 netdev_get_name(port->up.netdev), strerror(error));
1567 bundle_send_learning_packets(struct ofbundle *bundle)
1569 struct ofproto_dpif *ofproto = bundle->ofproto;
1570 int error, n_packets, n_errors;
1571 struct mac_entry *e;
1573 error = n_packets = n_errors = 0;
1574 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1575 if (e->port.p != bundle) {
1576 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1586 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1587 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1588 "packets, last error was: %s",
1589 bundle->name, n_errors, n_packets, strerror(error));
1591 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1592 bundle->name, n_packets);
1597 bundle_run(struct ofbundle *bundle)
1600 lacp_run(bundle->lacp, send_pdu_cb);
1603 struct ofport_dpif *port;
1605 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1606 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1609 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1610 lacp_negotiated(bundle->lacp));
1611 if (bond_should_send_learning_packets(bundle->bond)) {
1612 bundle_send_learning_packets(bundle);
1618 bundle_wait(struct ofbundle *bundle)
1621 lacp_wait(bundle->lacp);
1624 bond_wait(bundle->bond);
1631 mirror_scan(struct ofproto_dpif *ofproto)
1635 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1636 if (!ofproto->mirrors[idx]) {
1643 static struct ofmirror *
1644 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1648 for (i = 0; i < MAX_MIRRORS; i++) {
1649 struct ofmirror *mirror = ofproto->mirrors[i];
1650 if (mirror && mirror->aux == aux) {
1659 mirror_set(struct ofproto *ofproto_, void *aux,
1660 const struct ofproto_mirror_settings *s)
1662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1663 mirror_mask_t mirror_bit;
1664 struct ofbundle *bundle;
1665 struct ofmirror *mirror;
1666 struct ofbundle *out;
1667 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1668 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1671 mirror = mirror_lookup(ofproto, aux);
1673 mirror_destroy(mirror);
1679 idx = mirror_scan(ofproto);
1681 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1683 ofproto->up.name, MAX_MIRRORS, s->name);
1687 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1688 mirror->ofproto = ofproto;
1691 mirror->out_vlan = -1;
1692 mirror->name = NULL;
1695 if (!mirror->name || strcmp(s->name, mirror->name)) {
1697 mirror->name = xstrdup(s->name);
1700 /* Get the new configuration. */
1701 if (s->out_bundle) {
1702 out = bundle_lookup(ofproto, s->out_bundle);
1704 mirror_destroy(mirror);
1710 out_vlan = s->out_vlan;
1712 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1713 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1715 /* If the configuration has not changed, do nothing. */
1716 if (hmapx_equals(&srcs, &mirror->srcs)
1717 && hmapx_equals(&dsts, &mirror->dsts)
1718 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1719 && mirror->out == out
1720 && mirror->out_vlan == out_vlan)
1722 hmapx_destroy(&srcs);
1723 hmapx_destroy(&dsts);
1727 hmapx_swap(&srcs, &mirror->srcs);
1728 hmapx_destroy(&srcs);
1730 hmapx_swap(&dsts, &mirror->dsts);
1731 hmapx_destroy(&dsts);
1733 free(mirror->vlans);
1734 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1737 mirror->out_vlan = out_vlan;
1739 /* Update bundles. */
1740 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1741 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1742 if (hmapx_contains(&mirror->srcs, bundle)) {
1743 bundle->src_mirrors |= mirror_bit;
1745 bundle->src_mirrors &= ~mirror_bit;
1748 if (hmapx_contains(&mirror->dsts, bundle)) {
1749 bundle->dst_mirrors |= mirror_bit;
1751 bundle->dst_mirrors &= ~mirror_bit;
1754 if (mirror->out == bundle) {
1755 bundle->mirror_out |= mirror_bit;
1757 bundle->mirror_out &= ~mirror_bit;
1761 ofproto->need_revalidate = true;
1762 mac_learning_flush(ofproto->ml);
1768 mirror_destroy(struct ofmirror *mirror)
1770 struct ofproto_dpif *ofproto;
1771 mirror_mask_t mirror_bit;
1772 struct ofbundle *bundle;
1778 ofproto = mirror->ofproto;
1779 ofproto->need_revalidate = true;
1780 mac_learning_flush(ofproto->ml);
1782 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1783 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1784 bundle->src_mirrors &= ~mirror_bit;
1785 bundle->dst_mirrors &= ~mirror_bit;
1786 bundle->mirror_out &= ~mirror_bit;
1789 hmapx_destroy(&mirror->srcs);
1790 hmapx_destroy(&mirror->dsts);
1791 free(mirror->vlans);
1793 ofproto->mirrors[mirror->idx] = NULL;
1799 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1801 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1802 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1803 ofproto->need_revalidate = true;
1804 mac_learning_flush(ofproto->ml);
1810 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
1812 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1813 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1814 return bundle && bundle->mirror_out != 0;
1818 forward_bpdu_changed(struct ofproto *ofproto_)
1820 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1821 /* Revalidate cached flows whenever forward_bpdu option changes. */
1822 ofproto->need_revalidate = true;
1827 static struct ofport_dpif *
1828 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1830 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1831 return ofport ? ofport_dpif_cast(ofport) : NULL;
1834 static struct ofport_dpif *
1835 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1837 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1841 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1842 struct dpif_port *dpif_port)
1844 ofproto_port->name = dpif_port->name;
1845 ofproto_port->type = dpif_port->type;
1846 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1850 port_run(struct ofport_dpif *ofport)
1852 bool enable = netdev_get_carrier(ofport->up.netdev);
1855 cfm_run(ofport->cfm);
1857 if (cfm_should_send_ccm(ofport->cfm)) {
1858 struct ofpbuf packet;
1860 ofpbuf_init(&packet, 0);
1861 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1862 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1863 ofport->odp_port, &packet);
1864 ofpbuf_uninit(&packet);
1867 enable = enable && !cfm_get_fault(ofport->cfm)
1868 && cfm_get_opup(ofport->cfm);
1871 if (ofport->bundle) {
1872 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1875 if (ofport->may_enable != enable) {
1876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1878 if (ofproto->has_bundle_action) {
1879 ofproto->need_revalidate = true;
1883 ofport->may_enable = enable;
1887 port_wait(struct ofport_dpif *ofport)
1890 cfm_wait(ofport->cfm);
1895 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1896 struct ofproto_port *ofproto_port)
1898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1899 struct dpif_port dpif_port;
1902 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1904 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1910 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1916 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1918 *ofp_portp = odp_port_to_ofp_port(odp_port);
1924 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1926 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1929 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1931 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1933 /* The caller is going to close ofport->up.netdev. If this is a
1934 * bonded port, then the bond is using that netdev, so remove it
1935 * from the bond. The client will need to reconfigure everything
1936 * after deleting ports, so then the slave will get re-added. */
1937 bundle_remove(&ofport->up);
1943 struct port_dump_state {
1944 struct dpif_port_dump dump;
1949 port_dump_start(const struct ofproto *ofproto_, void **statep)
1951 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1952 struct port_dump_state *state;
1954 *statep = state = xmalloc(sizeof *state);
1955 dpif_port_dump_start(&state->dump, ofproto->dpif);
1956 state->done = false;
1961 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1962 struct ofproto_port *port)
1964 struct port_dump_state *state = state_;
1965 struct dpif_port dpif_port;
1967 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1968 ofproto_port_from_dpif_port(port, &dpif_port);
1971 int error = dpif_port_dump_done(&state->dump);
1973 return error ? error : EOF;
1978 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1980 struct port_dump_state *state = state_;
1983 dpif_port_dump_done(&state->dump);
1990 port_poll(const struct ofproto *ofproto_, char **devnamep)
1992 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1993 return dpif_port_poll(ofproto->dpif, devnamep);
1997 port_poll_wait(const struct ofproto *ofproto_)
1999 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2000 dpif_port_poll_wait(ofproto->dpif);
2004 port_is_lacp_current(const struct ofport *ofport_)
2006 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2007 return (ofport->bundle && ofport->bundle->lacp
2008 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2012 /* Upcall handling. */
2014 /* Flow miss batching.
2016 * Some dpifs implement operations faster when you hand them off in a batch.
2017 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2018 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2019 * more packets, plus possibly installing the flow in the dpif.
2021 * So far we only batch the operations that affect flow setup time the most.
2022 * It's possible to batch more than that, but the benefit might be minimal. */
2024 struct hmap_node hmap_node;
2026 const struct nlattr *key;
2028 struct list packets;
2031 struct flow_miss_op {
2032 union dpif_op dpif_op;
2033 struct facet *facet;
2036 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2037 * OpenFlow controller as necessary according to their individual
2040 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2041 * ownership is transferred to this function. */
2043 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2044 const struct flow *flow, bool clone)
2046 struct ofputil_packet_in pin;
2048 pin.packet = packet;
2049 pin.in_port = flow->in_port;
2050 pin.reason = OFPR_NO_MATCH;
2051 pin.buffer_id = 0; /* not yet known */
2052 pin.send_len = 0; /* not used for flow table misses */
2053 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2054 clone ? NULL : packet);
2057 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2058 * OpenFlow controller as necessary according to their individual
2061 * 'send_len' should be the number of bytes of 'packet' to send to the
2062 * controller, as specified in the action that caused the packet to be sent.
2064 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2065 * Otherwise, ownership is transferred to this function. */
2067 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2068 uint64_t userdata, const struct flow *flow, bool clone)
2070 struct ofputil_packet_in pin;
2071 struct user_action_cookie cookie;
2073 memcpy(&cookie, &userdata, sizeof(cookie));
2075 pin.packet = packet;
2076 pin.in_port = flow->in_port;
2077 pin.reason = OFPR_ACTION;
2078 pin.buffer_id = 0; /* not yet known */
2079 pin.send_len = cookie.data;
2080 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2081 clone ? NULL : packet);
2085 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2086 const struct ofpbuf *packet)
2088 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2094 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2096 cfm_process_heartbeat(ofport->cfm, packet);
2099 } else if (ofport->bundle && ofport->bundle->lacp
2100 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2102 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2105 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2107 stp_process_packet(ofport, packet);
2114 static struct flow_miss *
2115 flow_miss_create(struct hmap *todo, const struct flow *flow,
2116 const struct nlattr *key, size_t key_len)
2118 uint32_t hash = flow_hash(flow, 0);
2119 struct flow_miss *miss;
2121 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2122 if (flow_equal(&miss->flow, flow)) {
2127 miss = xmalloc(sizeof *miss);
2128 hmap_insert(todo, &miss->hmap_node, hash);
2131 miss->key_len = key_len;
2132 list_init(&miss->packets);
2137 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2138 struct flow_miss_op *ops, size_t *n_ops)
2140 const struct flow *flow = &miss->flow;
2141 struct ofpbuf *packet, *next_packet;
2142 struct facet *facet;
2144 facet = facet_lookup_valid(ofproto, flow);
2146 struct rule_dpif *rule;
2148 rule = rule_dpif_lookup(ofproto, flow, 0);
2150 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2151 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2153 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2154 COVERAGE_INC(ofproto_dpif_no_packet_in);
2155 /* XXX install 'drop' flow entry */
2159 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2163 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2165 list_remove(&packet->list_node);
2166 send_packet_in_miss(ofproto, packet, flow, false);
2172 facet = facet_create(rule, flow);
2175 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2176 list_remove(&packet->list_node);
2177 ofproto->n_matches++;
2179 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2181 * Extra-special case for fail-open mode.
2183 * We are in fail-open mode and the packet matched the fail-open
2184 * rule, but we are connected to a controller too. We should send
2185 * the packet up to the controller in the hope that it will try to
2186 * set up a flow and thereby allow us to exit fail-open.
2188 * See the top-level comment in fail-open.c for more information.
2190 send_packet_in_miss(ofproto, packet, flow, true);
2193 if (!facet->may_install) {
2194 facet_make_actions(ofproto, facet, packet);
2196 if (!execute_controller_action(ofproto, &facet->flow,
2197 facet->actions, facet->actions_len,
2199 struct flow_miss_op *op = &ops[(*n_ops)++];
2200 struct dpif_execute *execute = &op->dpif_op.execute;
2203 execute->type = DPIF_OP_EXECUTE;
2204 execute->key = miss->key;
2205 execute->key_len = miss->key_len;
2207 = (facet->may_install
2209 : xmemdup(facet->actions, facet->actions_len));
2210 execute->actions_len = facet->actions_len;
2211 execute->packet = packet;
2215 if (facet->may_install) {
2216 struct flow_miss_op *op = &ops[(*n_ops)++];
2217 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2220 put->type = DPIF_OP_FLOW_PUT;
2221 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2222 put->key = miss->key;
2223 put->key_len = miss->key_len;
2224 put->actions = facet->actions;
2225 put->actions_len = facet->actions_len;
2231 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2234 struct dpif_upcall *upcall;
2235 struct flow_miss *miss, *next_miss;
2236 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2237 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2246 /* Construct the to-do list.
2248 * This just amounts to extracting the flow from each packet and sticking
2249 * the packets that have the same flow in the same "flow_miss" structure so
2250 * that we can process them together. */
2252 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2253 struct flow_miss *miss;
2256 /* Obtain in_port and tun_id, at least, then set 'flow''s header
2258 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2259 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
2261 /* Handle 802.1ag, LACP, and STP specially. */
2262 if (process_special(ofproto, &flow, upcall->packet)) {
2263 ofpbuf_delete(upcall->packet);
2264 ofproto->n_matches++;
2268 /* Add other packets to a to-do list. */
2269 miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len);
2270 list_push_back(&miss->packets, &upcall->packet->list_node);
2273 /* Process each element in the to-do list, constructing the set of
2274 * operations to batch. */
2276 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2277 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2278 ofpbuf_list_delete(&miss->packets);
2279 hmap_remove(&todo, &miss->hmap_node);
2282 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2283 hmap_destroy(&todo);
2285 /* Execute batch. */
2286 for (i = 0; i < n_ops; i++) {
2287 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2289 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2291 /* Free memory and update facets. */
2292 for (i = 0; i < n_ops; i++) {
2293 struct flow_miss_op *op = &flow_miss_ops[i];
2294 struct dpif_execute *execute;
2295 struct dpif_flow_put *put;
2297 switch (op->dpif_op.type) {
2298 case DPIF_OP_EXECUTE:
2299 execute = &op->dpif_op.execute;
2300 if (op->facet->actions != execute->actions) {
2301 free((struct nlattr *) execute->actions);
2303 ofpbuf_delete((struct ofpbuf *) execute->packet);
2306 case DPIF_OP_FLOW_PUT:
2307 put = &op->dpif_op.flow_put;
2309 op->facet->installed = true;
2317 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2318 struct dpif_upcall *upcall)
2321 struct user_action_cookie cookie;
2323 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2325 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2326 if (ofproto->sflow) {
2327 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2328 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
2330 ofpbuf_delete(upcall->packet);
2332 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2333 COVERAGE_INC(ofproto_dpif_ctlr_action);
2334 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2335 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2338 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2343 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
2345 switch (upcall->type) {
2346 case DPIF_UC_ACTION:
2347 handle_userspace_upcall(ofproto, upcall);
2351 /* The caller handles these. */
2354 case DPIF_N_UC_TYPES:
2356 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
2361 /* Flow expiration. */
2363 static int facet_max_idle(const struct ofproto_dpif *);
2364 static void update_stats(struct ofproto_dpif *);
2365 static void rule_expire(struct rule_dpif *);
2366 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
2368 /* This function is called periodically by run(). Its job is to collect
2369 * updates for the flows that have been installed into the datapath, most
2370 * importantly when they last were used, and then use that information to
2371 * expire flows that have not been used recently.
2373 * Returns the number of milliseconds after which it should be called again. */
2375 expire(struct ofproto_dpif *ofproto)
2377 struct rule_dpif *rule, *next_rule;
2378 struct classifier *table;
2381 /* Update stats for each flow in the datapath. */
2382 update_stats(ofproto);
2384 /* Expire facets that have been idle too long. */
2385 dp_max_idle = facet_max_idle(ofproto);
2386 expire_facets(ofproto, dp_max_idle);
2388 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2389 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2390 struct cls_cursor cursor;
2392 cls_cursor_init(&cursor, table, NULL);
2393 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2398 /* All outstanding data in existing flows has been accounted, so it's a
2399 * good time to do bond rebalancing. */
2400 if (ofproto->has_bonded_bundles) {
2401 struct ofbundle *bundle;
2403 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2405 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2410 return MIN(dp_max_idle, 1000);
2413 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2415 * This function also pushes statistics updates to rules which each facet
2416 * resubmits into. Generally these statistics will be accurate. However, if a
2417 * facet changes the rule it resubmits into at some time in between
2418 * update_stats() runs, it is possible that statistics accrued to the
2419 * old rule will be incorrectly attributed to the new rule. This could be
2420 * avoided by calling update_stats() whenever rules are created or
2421 * deleted. However, the performance impact of making so many calls to the
2422 * datapath do not justify the benefit of having perfectly accurate statistics.
2425 update_stats(struct ofproto_dpif *p)
2427 const struct dpif_flow_stats *stats;
2428 struct dpif_flow_dump dump;
2429 const struct nlattr *key;
2432 dpif_flow_dump_start(&dump, p->dpif);
2433 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2434 struct facet *facet;
2437 if (odp_flow_key_to_flow(key, key_len, &flow)) {
2441 odp_flow_key_format(key, key_len, &s);
2442 VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
2448 facet = facet_find(p, &flow);
2450 if (facet && facet->installed) {
2452 if (stats->n_packets >= facet->dp_packet_count) {
2453 uint64_t extra = stats->n_packets - facet->dp_packet_count;
2454 facet->packet_count += extra;
2456 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2459 if (stats->n_bytes >= facet->dp_byte_count) {
2460 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
2462 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2465 facet->dp_packet_count = stats->n_packets;
2466 facet->dp_byte_count = stats->n_bytes;
2468 facet_update_time(p, facet, stats->used);
2469 facet_account(p, facet);
2470 facet_push_stats(facet);
2472 /* There's a flow in the datapath that we know nothing about.
2474 COVERAGE_INC(facet_unexpected);
2475 dpif_flow_del(p->dpif, key, key_len, NULL);
2478 dpif_flow_dump_done(&dump);
2481 /* Calculates and returns the number of milliseconds of idle time after which
2482 * facets should expire from the datapath and we should fold their statistics
2483 * into their parent rules in userspace. */
2485 facet_max_idle(const struct ofproto_dpif *ofproto)
2488 * Idle time histogram.
2490 * Most of the time a switch has a relatively small number of facets. When
2491 * this is the case we might as well keep statistics for all of them in
2492 * userspace and to cache them in the kernel datapath for performance as
2495 * As the number of facets increases, the memory required to maintain
2496 * statistics about them in userspace and in the kernel becomes
2497 * significant. However, with a large number of facets it is likely that
2498 * only a few of them are "heavy hitters" that consume a large amount of
2499 * bandwidth. At this point, only heavy hitters are worth caching in the
2500 * kernel and maintaining in userspaces; other facets we can discard.
2502 * The technique used to compute the idle time is to build a histogram with
2503 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
2504 * that is installed in the kernel gets dropped in the appropriate bucket.
2505 * After the histogram has been built, we compute the cutoff so that only
2506 * the most-recently-used 1% of facets (but at least
2507 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2508 * the most-recently-used bucket of facets is kept, so actually an
2509 * arbitrary number of facets can be kept in any given expiration run
2510 * (though the next run will delete most of those unless they receive
2513 * This requires a second pass through the facets, in addition to the pass
2514 * made by update_stats(), because the former function never looks
2515 * at uninstallable facets.
2517 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2518 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2519 int buckets[N_BUCKETS] = { 0 };
2520 int total, subtotal, bucket;
2521 struct facet *facet;
2525 total = hmap_count(&ofproto->facets);
2526 if (total <= ofproto->up.flow_eviction_threshold) {
2527 return N_BUCKETS * BUCKET_WIDTH;
2530 /* Build histogram. */
2532 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
2533 long long int idle = now - facet->used;
2534 int bucket = (idle <= 0 ? 0
2535 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2536 : (unsigned int) idle / BUCKET_WIDTH);
2540 /* Find the first bucket whose flows should be expired. */
2541 subtotal = bucket = 0;
2543 subtotal += buckets[bucket++];
2544 } while (bucket < N_BUCKETS &&
2545 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2547 if (VLOG_IS_DBG_ENABLED()) {
2551 ds_put_cstr(&s, "keep");
2552 for (i = 0; i < N_BUCKETS; i++) {
2554 ds_put_cstr(&s, ", drop");
2557 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2560 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2564 return bucket * BUCKET_WIDTH;
2568 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
2570 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
2571 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
2572 struct ofexpired expired;
2574 if (facet->installed) {
2575 struct dpif_flow_stats stats;
2577 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
2579 facet_update_stats(ofproto, facet, &stats);
2582 expired.flow = facet->flow;
2583 expired.packet_count = facet->packet_count;
2584 expired.byte_count = facet->byte_count;
2585 expired.used = facet->used;
2586 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2591 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
2593 long long int cutoff = time_msec() - dp_max_idle;
2594 struct facet *facet, *next_facet;
2596 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
2597 facet_active_timeout(ofproto, facet);
2598 if (facet->used < cutoff) {
2599 facet_remove(ofproto, facet);
2604 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2605 * then delete it entirely. */
2607 rule_expire(struct rule_dpif *rule)
2609 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2610 struct facet *facet, *next_facet;
2614 /* Has 'rule' expired? */
2616 if (rule->up.hard_timeout
2617 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
2618 reason = OFPRR_HARD_TIMEOUT;
2619 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2620 && now > rule->used + rule->up.idle_timeout * 1000) {
2621 reason = OFPRR_IDLE_TIMEOUT;
2626 COVERAGE_INC(ofproto_dpif_expired);
2628 /* Update stats. (This is a no-op if the rule expired due to an idle
2629 * timeout, because that only happens when the rule has no facets left.) */
2630 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2631 facet_remove(ofproto, facet);
2634 /* Get rid of the rule. */
2635 ofproto_rule_expire(&rule->up, reason);
2640 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
2642 * The caller must already have determined that no facet with an identical
2643 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2644 * the ofproto's classifier table.
2646 * The facet will initially have no ODP actions. The caller should fix that
2647 * by calling facet_make_actions(). */
2648 static struct facet *
2649 facet_create(struct rule_dpif *rule, const struct flow *flow)
2651 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2652 struct facet *facet;
2654 facet = xzalloc(sizeof *facet);
2655 facet->used = time_msec();
2656 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2657 list_push_back(&rule->facets, &facet->list_node);
2659 facet->flow = *flow;
2660 netflow_flow_init(&facet->nf_flow);
2661 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2667 facet_free(struct facet *facet)
2669 free(facet->actions);
2674 execute_controller_action(struct ofproto_dpif *ofproto,
2675 const struct flow *flow,
2676 const struct nlattr *odp_actions, size_t actions_len,
2677 struct ofpbuf *packet)
2680 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
2681 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
2682 /* As an optimization, avoid a round-trip from userspace to kernel to
2683 * userspace. This also avoids possibly filling up kernel packet
2684 * buffers along the way.
2686 * This optimization will not accidentally catch sFlow
2687 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
2688 * inside OVS_ACTION_ATTR_SAMPLE. */
2689 const struct nlattr *nla;
2691 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
2692 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
2700 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2701 * 'packet', which arrived on 'in_port'.
2703 * Takes ownership of 'packet'. */
2705 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2706 const struct nlattr *odp_actions, size_t actions_len,
2707 struct ofpbuf *packet)
2709 struct odputil_keybuf keybuf;
2713 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
2718 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2719 odp_flow_key_from_flow(&key, flow);
2721 error = dpif_execute(ofproto->dpif, key.data, key.size,
2722 odp_actions, actions_len, packet);
2724 ofpbuf_delete(packet);
2728 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2729 * statistics appropriately. 'packet' must have at least sizeof(struct
2730 * ofp_packet_in) bytes of headroom.
2732 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2733 * applying flow_extract() to 'packet' would yield the same flow as
2736 * 'facet' must have accurately composed datapath actions; that is, it must
2737 * not be in need of revalidation.
2739 * Takes ownership of 'packet'. */
2741 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2742 struct ofpbuf *packet)
2744 struct dpif_flow_stats stats;
2746 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2748 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2749 stats.used = time_msec();
2750 if (execute_odp_actions(ofproto, &facet->flow,
2751 facet->actions, facet->actions_len, packet)) {
2752 facet_update_stats(ofproto, facet, &stats);
2756 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2758 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2759 * rule's statistics, via facet_uninstall().
2761 * - Removes 'facet' from its rule and from ofproto->facets.
2764 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2766 facet_uninstall(ofproto, facet);
2767 facet_flush_stats(ofproto, facet);
2768 hmap_remove(&ofproto->facets, &facet->hmap_node);
2769 list_remove(&facet->list_node);
2773 /* Composes the datapath actions for 'facet' based on its rule's actions. */
2775 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2776 const struct ofpbuf *packet)
2778 const struct rule_dpif *rule = facet->rule;
2779 struct ofpbuf *odp_actions;
2780 struct action_xlate_ctx ctx;
2782 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2783 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2784 facet->tags = ctx.tags;
2785 facet->may_install = ctx.may_set_up_flow;
2786 facet->has_learn = ctx.has_learn;
2787 facet->has_normal = ctx.has_normal;
2788 facet->nf_flow.output_iface = ctx.nf_output_iface;
2790 if (facet->actions_len != odp_actions->size
2791 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2792 free(facet->actions);
2793 facet->actions_len = odp_actions->size;
2794 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2797 ofpbuf_delete(odp_actions);
2800 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2801 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2802 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2803 * since 'facet' was last updated.
2805 * Returns 0 if successful, otherwise a positive errno value.*/
2807 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2808 const struct nlattr *actions, size_t actions_len,
2809 struct dpif_flow_stats *stats)
2811 struct odputil_keybuf keybuf;
2812 enum dpif_flow_put_flags flags;
2816 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2818 flags |= DPIF_FP_ZERO_STATS;
2821 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2822 odp_flow_key_from_flow(&key, &facet->flow);
2824 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2825 actions, actions_len, stats);
2828 facet_reset_dp_stats(facet, stats);
2834 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2835 * 'zero_stats' is true, clears any existing statistics from the datapath for
2838 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2840 struct dpif_flow_stats stats;
2842 if (facet->may_install
2843 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2844 zero_stats ? &stats : NULL)) {
2845 facet->installed = true;
2850 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
2853 const struct nlattr *a;
2857 if (facet->byte_count <= facet->accounted_bytes) {
2860 n_bytes = facet->byte_count - facet->accounted_bytes;
2861 facet->accounted_bytes = facet->byte_count;
2863 /* Feed information from the active flows back into the learning table to
2864 * ensure that table is always in sync with what is actually flowing
2865 * through the datapath. */
2866 if (facet->has_learn || facet->has_normal) {
2867 struct action_xlate_ctx ctx;
2869 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2870 ctx.may_learn = true;
2871 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
2872 facet->rule->up.n_actions));
2875 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
2879 /* This loop feeds byte counters to bond_account() for rebalancing to use
2880 * as a basis. We also need to track the actual VLAN on which the packet
2881 * is going to be sent to ensure that it matches the one passed to
2882 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2884 vlan_tci = facet->flow.vlan_tci;
2885 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2886 struct ofport_dpif *port;
2888 switch (nl_attr_type(a)) {
2889 const struct nlattr *nested;
2890 case OVS_ACTION_ATTR_OUTPUT:
2891 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2892 if (port && port->bundle && port->bundle->bond) {
2893 bond_account(port->bundle->bond, &facet->flow,
2894 vlan_tci_to_vid(vlan_tci), n_bytes);
2898 case OVS_ACTION_ATTR_POP:
2899 if (nl_attr_get_u16(a) == OVS_KEY_ATTR_8021Q) {
2900 vlan_tci = htons(0);
2904 case OVS_ACTION_ATTR_PUSH:
2905 nested = nl_attr_get(a);
2906 if (nl_attr_type(nested) == OVS_KEY_ATTR_8021Q) {
2907 const struct ovs_key_8021q *q_key;
2909 q_key = nl_attr_get_unspec(nested, sizeof(*q_key));
2910 vlan_tci = q_key->q_tci;
2917 /* If 'rule' is installed in the datapath, uninstalls it. */
2919 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2921 if (facet->installed) {
2922 struct odputil_keybuf keybuf;
2923 struct dpif_flow_stats stats;
2927 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2928 odp_flow_key_from_flow(&key, &facet->flow);
2930 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2931 facet_reset_dp_stats(facet, &stats);
2933 facet_update_stats(p, facet, &stats);
2935 facet->installed = false;
2937 assert(facet->dp_packet_count == 0);
2938 assert(facet->dp_byte_count == 0);
2942 /* Returns true if the only action for 'facet' is to send to the controller.
2943 * (We don't report NetFlow expiration messages for such facets because they
2944 * are just part of the control logic for the network, not real traffic). */
2946 facet_is_controller_flow(struct facet *facet)
2949 && facet->rule->up.n_actions == 1
2950 && action_outputs_to_port(&facet->rule->up.actions[0],
2951 htons(OFPP_CONTROLLER)));
2954 /* Resets 'facet''s datapath statistics counters. This should be called when
2955 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2956 * it should contain the statistics returned by dpif when 'facet' was reset in
2957 * the datapath. 'stats' will be modified to only included statistics new
2958 * since 'facet' was last updated. */
2960 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2962 if (stats && facet->dp_packet_count <= stats->n_packets
2963 && facet->dp_byte_count <= stats->n_bytes) {
2964 stats->n_packets -= facet->dp_packet_count;
2965 stats->n_bytes -= facet->dp_byte_count;
2968 facet->dp_packet_count = 0;
2969 facet->dp_byte_count = 0;
2972 /* Folds all of 'facet''s statistics into its rule. Also updates the
2973 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2974 * 'facet''s statistics in the datapath should have been zeroed and folded into
2975 * its packet and byte counts before this function is called. */
2977 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2979 assert(!facet->dp_byte_count);
2980 assert(!facet->dp_packet_count);
2982 facet_push_stats(facet);
2983 facet_account(ofproto, facet);
2985 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2986 struct ofexpired expired;
2987 expired.flow = facet->flow;
2988 expired.packet_count = facet->packet_count;
2989 expired.byte_count = facet->byte_count;
2990 expired.used = facet->used;
2991 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2994 facet->rule->packet_count += facet->packet_count;
2995 facet->rule->byte_count += facet->byte_count;
2997 /* Reset counters to prevent double counting if 'facet' ever gets
2999 facet_reset_counters(facet);
3001 netflow_flow_clear(&facet->nf_flow);
3004 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3005 * Returns it if found, otherwise a null pointer.
3007 * The returned facet might need revalidation; use facet_lookup_valid()
3008 * instead if that is important. */
3009 static struct facet *
3010 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3012 struct facet *facet;
3014 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3016 if (flow_equal(flow, &facet->flow)) {
3024 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3025 * Returns it if found, otherwise a null pointer.
3027 * The returned facet is guaranteed to be valid. */
3028 static struct facet *
3029 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3031 struct facet *facet = facet_find(ofproto, flow);
3033 /* The facet we found might not be valid, since we could be in need of
3034 * revalidation. If it is not valid, don't return it. */
3036 && (ofproto->need_revalidate
3037 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3038 && !facet_revalidate(ofproto, facet)) {
3039 COVERAGE_INC(facet_invalidated);
3046 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3048 * - If the rule found is different from 'facet''s current rule, moves
3049 * 'facet' to the new rule and recompiles its actions.
3051 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3052 * where it is and recompiles its actions anyway.
3054 * - If there is none, destroys 'facet'.
3056 * Returns true if 'facet' still exists, false if it has been destroyed. */
3058 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3060 struct action_xlate_ctx ctx;
3061 struct ofpbuf *odp_actions;
3062 struct rule_dpif *new_rule;
3063 bool actions_changed;
3065 COVERAGE_INC(facet_revalidate);
3067 /* Determine the new rule. */
3068 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3070 /* No new rule, so delete the facet. */
3071 facet_remove(ofproto, facet);
3075 /* Calculate new datapath actions.
3077 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3078 * emit a NetFlow expiration and, if so, we need to have the old state
3079 * around to properly compose it. */
3080 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
3081 odp_actions = xlate_actions(&ctx,
3082 new_rule->up.actions, new_rule->up.n_actions);
3083 actions_changed = (facet->actions_len != odp_actions->size
3084 || memcmp(facet->actions, odp_actions->data,
3085 facet->actions_len));
3087 /* If the datapath actions changed or the installability changed,
3088 * then we need to talk to the datapath. */
3089 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
3090 if (ctx.may_set_up_flow) {
3091 struct dpif_flow_stats stats;
3093 facet_put__(ofproto, facet,
3094 odp_actions->data, odp_actions->size, &stats);
3095 facet_update_stats(ofproto, facet, &stats);
3097 facet_uninstall(ofproto, facet);
3100 /* The datapath flow is gone or has zeroed stats, so push stats out of
3101 * 'facet' into 'rule'. */
3102 facet_flush_stats(ofproto, facet);
3105 /* Update 'facet' now that we've taken care of all the old state. */
3106 facet->tags = ctx.tags;
3107 facet->nf_flow.output_iface = ctx.nf_output_iface;
3108 facet->may_install = ctx.may_set_up_flow;
3109 facet->has_learn = ctx.has_learn;
3110 facet->has_normal = ctx.has_normal;
3111 if (actions_changed) {
3112 free(facet->actions);
3113 facet->actions_len = odp_actions->size;
3114 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
3116 if (facet->rule != new_rule) {
3117 COVERAGE_INC(facet_changed_rule);
3118 list_remove(&facet->list_node);
3119 list_push_back(&new_rule->facets, &facet->list_node);
3120 facet->rule = new_rule;
3121 facet->used = new_rule->up.created;
3122 facet->rs_used = facet->used;
3125 ofpbuf_delete(odp_actions);
3130 /* Updates 'facet''s used time. Caller is responsible for calling
3131 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3133 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3136 if (used > facet->used) {
3138 if (used > facet->rule->used) {
3139 facet->rule->used = used;
3141 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3145 /* Folds the statistics from 'stats' into the counters in 'facet'.
3147 * Because of the meaning of a facet's counters, it only makes sense to do this
3148 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3149 * packet that was sent by hand or if it represents statistics that have been
3150 * cleared out of the datapath. */
3152 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
3153 const struct dpif_flow_stats *stats)
3155 if (stats->n_packets || stats->used > facet->used) {
3156 facet_update_time(ofproto, facet, stats->used);
3157 facet->packet_count += stats->n_packets;
3158 facet->byte_count += stats->n_bytes;
3159 facet_push_stats(facet);
3160 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3165 facet_reset_counters(struct facet *facet)
3167 facet->packet_count = 0;
3168 facet->byte_count = 0;
3169 facet->rs_packet_count = 0;
3170 facet->rs_byte_count = 0;
3171 facet->accounted_bytes = 0;
3175 facet_push_stats(struct facet *facet)
3177 uint64_t rs_packets, rs_bytes;
3179 assert(facet->packet_count >= facet->rs_packet_count);
3180 assert(facet->byte_count >= facet->rs_byte_count);
3181 assert(facet->used >= facet->rs_used);
3183 rs_packets = facet->packet_count - facet->rs_packet_count;
3184 rs_bytes = facet->byte_count - facet->rs_byte_count;
3186 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3187 facet->rs_packet_count = facet->packet_count;
3188 facet->rs_byte_count = facet->byte_count;
3189 facet->rs_used = facet->used;
3191 flow_push_stats(facet->rule, &facet->flow,
3192 rs_packets, rs_bytes, facet->used);
3196 struct ofproto_push {
3197 struct action_xlate_ctx ctx;
3204 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3206 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3209 rule->packet_count += push->packets;
3210 rule->byte_count += push->bytes;
3211 rule->used = MAX(push->used, rule->used);
3215 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3216 * 'rule''s actions. */
3218 flow_push_stats(const struct rule_dpif *rule,
3219 struct flow *flow, uint64_t packets, uint64_t bytes,
3222 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3223 struct ofproto_push push;
3225 push.packets = packets;
3229 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3230 push.ctx.resubmit_hook = push_resubmit;
3231 ofpbuf_delete(xlate_actions(&push.ctx,
3232 rule->up.actions, rule->up.n_actions));
3237 static struct rule_dpif *
3238 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3241 struct cls_rule *cls_rule;
3242 struct classifier *cls;
3244 if (table_id >= N_TABLES) {
3248 cls = &ofproto->up.tables[table_id];
3249 if (flow->tos_frag & FLOW_FRAG_ANY
3250 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3251 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3252 * are unavailable. */
3253 struct flow ofpc_normal_flow = *flow;
3254 ofpc_normal_flow.tp_src = htons(0);
3255 ofpc_normal_flow.tp_dst = htons(0);
3256 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3258 cls_rule = classifier_lookup(cls, flow);
3260 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3264 complete_operation(struct rule_dpif *rule)
3266 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3268 rule_invalidate(rule);
3270 struct dpif_completion *c = xmalloc(sizeof *c);
3271 c->op = rule->up.pending;
3272 list_push_back(&ofproto->completions, &c->list_node);
3274 ofoperation_complete(rule->up.pending, 0);
3278 static struct rule *
3281 struct rule_dpif *rule = xmalloc(sizeof *rule);
3286 rule_dealloc(struct rule *rule_)
3288 struct rule_dpif *rule = rule_dpif_cast(rule_);
3293 rule_construct(struct rule *rule_)
3295 struct rule_dpif *rule = rule_dpif_cast(rule_);
3296 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3297 struct rule_dpif *victim;
3301 error = validate_actions(rule->up.actions, rule->up.n_actions,
3302 &rule->up.cr.flow, ofproto->max_ports);
3307 rule->used = rule->up.created;
3308 rule->packet_count = 0;
3309 rule->byte_count = 0;
3311 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3312 if (victim && !list_is_empty(&victim->facets)) {
3313 struct facet *facet;
3315 rule->facets = victim->facets;
3316 list_moved(&rule->facets);
3317 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3318 /* XXX: We're only clearing our local counters here. It's possible
3319 * that quite a few packets are unaccounted for in the datapath
3320 * statistics. These will be accounted to the new rule instead of
3321 * cleared as required. This could be fixed by clearing out the
3322 * datapath statistics for this facet, but currently it doesn't
3324 facet_reset_counters(facet);
3328 /* Must avoid list_moved() in this case. */
3329 list_init(&rule->facets);
3332 table_id = rule->up.table_id;
3333 rule->tag = (victim ? victim->tag
3335 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3336 ofproto->tables[table_id].basis));
3338 complete_operation(rule);
3343 rule_destruct(struct rule *rule_)
3345 struct rule_dpif *rule = rule_dpif_cast(rule_);
3346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3347 struct facet *facet, *next_facet;
3349 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3350 facet_revalidate(ofproto, facet);
3353 complete_operation(rule);
3357 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3359 struct rule_dpif *rule = rule_dpif_cast(rule_);
3360 struct facet *facet;
3362 /* Start from historical data for 'rule' itself that are no longer tracked
3363 * in facets. This counts, for example, facets that have expired. */
3364 *packets = rule->packet_count;
3365 *bytes = rule->byte_count;
3367 /* Add any statistics that are tracked by facets. This includes
3368 * statistical data recently updated by ofproto_update_stats() as well as
3369 * stats for packets that were executed "by hand" via dpif_execute(). */
3370 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3371 *packets += facet->packet_count;
3372 *bytes += facet->byte_count;
3377 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
3379 struct rule_dpif *rule = rule_dpif_cast(rule_);
3380 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3381 struct action_xlate_ctx ctx;
3382 struct ofpbuf *odp_actions;
3383 struct facet *facet;
3386 /* First look for a related facet. If we find one, account it to that. */
3387 facet = facet_lookup_valid(ofproto, flow);
3388 if (facet && facet->rule == rule) {
3389 if (!facet->may_install) {
3390 facet_make_actions(ofproto, facet, packet);
3392 facet_execute(ofproto, facet, packet);
3396 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
3397 * create a new facet for it and use that. */
3398 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
3399 facet = facet_create(rule, flow);
3400 facet_make_actions(ofproto, facet, packet);
3401 facet_execute(ofproto, facet, packet);
3402 facet_install(ofproto, facet, true);
3406 /* We can't account anything to a facet. If we were to try, then that
3407 * facet would have a non-matching rule, busting our invariants. */
3408 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3409 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3410 size = packet->size;
3411 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3412 odp_actions->size, packet)) {
3413 rule->used = time_msec();
3414 rule->packet_count++;
3415 rule->byte_count += size;
3416 flow_push_stats(rule, flow, 1, size, rule->used);
3418 ofpbuf_delete(odp_actions);
3424 rule_modify_actions(struct rule *rule_)
3426 struct rule_dpif *rule = rule_dpif_cast(rule_);
3427 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3430 error = validate_actions(rule->up.actions, rule->up.n_actions,
3431 &rule->up.cr.flow, ofproto->max_ports);
3433 ofoperation_complete(rule->up.pending, error);
3437 complete_operation(rule);
3440 /* Sends 'packet' out of port 'odp_port' within 'ofproto'.
3441 * Returns 0 if successful, otherwise a positive errno value. */
3443 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
3444 const struct ofpbuf *packet)
3446 struct ofpbuf key, odp_actions;
3447 struct odputil_keybuf keybuf;
3451 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
3452 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3453 odp_flow_key_from_flow(&key, &flow);
3455 ofpbuf_init(&odp_actions, 32);
3456 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
3458 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3459 error = dpif_execute(ofproto->dpif,
3461 odp_actions.data, odp_actions.size,
3463 ofpbuf_uninit(&odp_actions);
3466 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
3467 ofproto->up.name, odp_port, strerror(error));
3472 /* OpenFlow to datapath action translation. */
3474 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
3475 struct action_xlate_ctx *ctx);
3476 static void xlate_normal(struct action_xlate_ctx *);
3479 put_userspace_action(const struct ofproto_dpif *ofproto,
3480 struct ofpbuf *odp_actions,
3481 const struct flow *flow,
3482 const struct user_action_cookie *cookie)
3487 pid = dpif_port_get_pid(ofproto->dpif,
3488 ofp_port_to_odp_port(flow->in_port));
3490 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
3491 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
3492 nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
3493 cookie, sizeof *cookie);
3494 nl_msg_end_nested(odp_actions, offset);
3496 return odp_actions->size - NLA_ALIGN(sizeof *cookie);
3499 /* Compose SAMPLE action for sFlow. */
3501 compose_sflow_action(const struct ofproto_dpif *ofproto,
3502 struct ofpbuf *odp_actions,
3503 const struct flow *flow,
3506 uint32_t port_ifindex;
3507 uint32_t probability;
3508 struct user_action_cookie cookie;
3509 size_t sample_offset, actions_offset;
3510 int cookie_offset, n_output;
3512 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
3516 if (odp_port == OVSP_NONE) {
3520 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
3524 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
3526 /* Number of packets out of UINT_MAX to sample. */
3527 probability = dpif_sflow_get_probability(ofproto->sflow);
3528 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
3530 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
3532 cookie.type = USER_ACTION_COOKIE_SFLOW;
3533 cookie.data = port_ifindex;
3534 cookie.n_output = n_output;
3535 cookie.vlan_tci = 0;
3536 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
3538 nl_msg_end_nested(odp_actions, actions_offset);
3539 nl_msg_end_nested(odp_actions, sample_offset);
3540 return cookie_offset;
3543 /* SAMPLE action must be first action in any given list of actions.
3544 * At this point we do not have all information required to build it. So try to
3545 * build sample action as complete as possible. */
3547 add_sflow_action(struct action_xlate_ctx *ctx)
3549 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
3551 &ctx->flow, OVSP_NONE);
3552 ctx->sflow_odp_port = 0;
3553 ctx->sflow_n_outputs = 0;
3556 /* Fix SAMPLE action according to data collected while composing ODP actions.
3557 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
3558 * USERSPACE action's user-cookie which is required for sflow. */
3560 fix_sflow_action(struct action_xlate_ctx *ctx)
3562 const struct flow *base = &ctx->base_flow;
3563 struct user_action_cookie *cookie;
3565 if (!ctx->user_cookie_offset) {
3569 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
3571 assert(cookie != NULL);
3572 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
3574 if (ctx->sflow_n_outputs) {
3575 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
3576 ctx->sflow_odp_port);
3578 if (ctx->sflow_n_outputs >= 255) {
3579 cookie->n_output = 255;
3581 cookie->n_output = ctx->sflow_n_outputs;
3583 cookie->vlan_tci = base->vlan_tci;
3587 commit_action__(struct ofpbuf *odp_actions,
3588 enum ovs_action_attr act_type,
3589 enum ovs_key_attr key_type,
3590 const void *key, size_t key_size)
3592 size_t offset = nl_msg_start_nested(odp_actions, act_type);
3594 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
3595 nl_msg_end_nested(odp_actions, offset);
3599 commit_set_tun_id_action(const struct flow *flow, struct flow *base,
3600 struct ofpbuf *odp_actions)
3602 if (base->tun_id == flow->tun_id) {
3605 base->tun_id = flow->tun_id;
3607 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3608 OVS_KEY_ATTR_TUN_ID, &base->tun_id, sizeof(base->tun_id));
3612 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
3613 struct ofpbuf *odp_actions)
3615 struct ovs_key_ethernet eth_key;
3617 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
3618 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
3622 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
3623 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
3625 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
3626 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
3628 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3629 OVS_KEY_ATTR_ETHERNET, ð_key, sizeof(eth_key));
3633 commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci)
3635 struct flow *base = &ctx->base_flow;
3637 if (base->vlan_tci == new_tci) {
3641 if (base->vlan_tci & htons(VLAN_CFI)) {
3642 nl_msg_put_u16(ctx->odp_actions, OVS_ACTION_ATTR_POP,
3643 OVS_KEY_ATTR_8021Q);
3646 if (new_tci & htons(VLAN_CFI)) {
3647 struct ovs_key_8021q q_key;
3649 q_key.q_tpid = htons(ETH_TYPE_VLAN);
3650 q_key.q_tci = new_tci & ~htons(VLAN_CFI);
3652 commit_action__(ctx->odp_actions, OVS_ACTION_ATTR_PUSH,
3653 OVS_KEY_ATTR_8021Q, &q_key, sizeof(q_key));
3655 base->vlan_tci = new_tci;
3659 commit_set_nw_action(const struct flow *flow, struct flow *base,
3660 struct ofpbuf *odp_actions)
3662 int frag = base->tos_frag & FLOW_FRAG_MASK;
3663 struct ovs_key_ipv4 ipv4_key;
3665 if (base->dl_type != htons(ETH_TYPE_IP) ||
3666 !base->nw_src || !base->nw_dst) {
3670 if (base->nw_src == flow->nw_src &&
3671 base->nw_dst == flow->nw_dst &&
3672 base->tos_frag == flow->tos_frag) {
3677 memset(&ipv4_key, 0, sizeof(ipv4_key));
3678 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
3679 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
3680 ipv4_key.ipv4_proto = base->nw_proto;
3681 ipv4_key.ipv4_tos = flow->tos_frag & IP_DSCP_MASK;
3682 ipv4_key.ipv4_frag = (frag == 0 ? OVS_FRAG_TYPE_NONE
3683 : frag == FLOW_FRAG_ANY ? OVS_FRAG_TYPE_FIRST
3684 : OVS_FRAG_TYPE_LATER);
3686 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3687 OVS_KEY_ATTR_IPV4, &ipv4_key, sizeof(ipv4_key));
3691 commit_set_port_action(const struct flow *flow, struct flow *base,
3692 struct ofpbuf *odp_actions)
3694 if (!base->tp_src || !base->tp_dst) {
3698 if (base->tp_src == flow->tp_src &&
3699 base->tp_dst == flow->tp_dst) {
3703 if (flow->nw_proto == IPPROTO_TCP) {
3704 struct ovs_key_tcp port_key;
3706 port_key.tcp_src = base->tp_src = flow->tp_src;
3707 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
3709 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3710 OVS_KEY_ATTR_TCP, &port_key, sizeof(port_key));
3712 } else if (flow->nw_proto == IPPROTO_UDP) {
3713 struct ovs_key_udp port_key;
3715 port_key.udp_src = base->tp_src = flow->tp_src;
3716 port_key.udp_dst = base->tp_dst = flow->tp_dst;
3718 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3719 OVS_KEY_ATTR_UDP, &port_key, sizeof(port_key));
3724 commit_priority_action(struct action_xlate_ctx *ctx)
3726 if (ctx->base_priority == ctx->priority) {
3730 if (ctx->priority) {
3731 nl_msg_put_u32(ctx->odp_actions,
3732 OVS_ACTION_ATTR_SET_PRIORITY, ctx->priority);
3734 nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_PRIORITY);
3736 ctx->base_priority = ctx->priority;
3740 commit_odp_actions(struct action_xlate_ctx *ctx)
3742 const struct flow *flow = &ctx->flow;
3743 struct flow *base = &ctx->base_flow;
3744 struct ofpbuf *odp_actions = ctx->odp_actions;
3746 commit_set_tun_id_action(flow, base, odp_actions);
3747 commit_set_ether_addr_action(flow, base, odp_actions);
3748 commit_vlan_action(ctx, flow->vlan_tci);
3749 commit_set_nw_action(flow, base, odp_actions);
3750 commit_set_port_action(flow, base, odp_actions);
3751 commit_priority_action(ctx);
3755 compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port)
3757 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3758 ctx->sflow_odp_port = odp_port;
3759 ctx->sflow_n_outputs++;
3763 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3765 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3766 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
3769 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
3770 || !stp_forward_in_state(ofport->stp_state)) {
3771 /* Forwarding disabled on port. */
3776 * We don't have an ofport record for this port, but it doesn't hurt to
3777 * allow forwarding to it anyhow. Maybe such a port will appear later
3778 * and we're pre-populating the flow table.
3782 commit_odp_actions(ctx);
3783 compose_output_action(ctx, odp_port);
3784 ctx->nf_output_iface = ofp_port;
3788 xlate_table_action(struct action_xlate_ctx *ctx,
3789 uint16_t in_port, uint8_t table_id)
3791 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
3792 struct ofproto_dpif *ofproto = ctx->ofproto;
3793 struct rule_dpif *rule;
3794 uint16_t old_in_port;
3795 uint8_t old_table_id;
3797 old_table_id = ctx->table_id;
3798 ctx->table_id = table_id;
3800 /* Look up a flow with 'in_port' as the input port. */
3801 old_in_port = ctx->flow.in_port;
3802 ctx->flow.in_port = in_port;
3803 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
3806 if (table_id > 0 && table_id < N_TABLES) {
3807 struct table_dpif *table = &ofproto->tables[table_id];
3808 if (table->other_table) {
3811 : rule_calculate_tag(&ctx->flow,
3812 &table->other_table->wc,
3817 /* Restore the original input port. Otherwise OFPP_NORMAL and
3818 * OFPP_IN_PORT will have surprising behavior. */
3819 ctx->flow.in_port = old_in_port;
3821 if (ctx->resubmit_hook) {
3822 ctx->resubmit_hook(ctx, rule);
3827 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
3831 ctx->table_id = old_table_id;
3833 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3835 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
3836 MAX_RESUBMIT_RECURSION);
3841 xlate_resubmit_table(struct action_xlate_ctx *ctx,
3842 const struct nx_action_resubmit *nar)
3847 in_port = (nar->in_port == htons(OFPP_IN_PORT)
3849 : ntohs(nar->in_port));
3850 table_id = nar->table == 255 ? ctx->table_id : nar->table;
3852 xlate_table_action(ctx, in_port, table_id);
3856 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
3858 struct ofport_dpif *ofport;
3860 commit_odp_actions(ctx);
3861 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
3862 uint16_t ofp_port = ofport->up.ofp_port;
3863 if (ofp_port != ctx->flow.in_port
3864 && !(ofport->up.opp.config & mask)
3865 && stp_forward_in_state(ofport->stp_state)) {
3866 compose_output_action(ctx, ofport->odp_port);
3870 ctx->nf_output_iface = NF_OUT_FLOOD;
3874 compose_controller_action(struct action_xlate_ctx *ctx, int len)
3876 struct user_action_cookie cookie;
3878 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
3880 cookie.n_output = 0;
3881 cookie.vlan_tci = 0;
3882 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
3886 xlate_output_action__(struct action_xlate_ctx *ctx,
3887 uint16_t port, uint16_t max_len)
3889 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
3891 ctx->nf_output_iface = NF_OUT_DROP;
3895 add_output_action(ctx, ctx->flow.in_port);
3898 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
3904 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
3907 flood_packets(ctx, htonl(0));
3909 case OFPP_CONTROLLER:
3910 commit_odp_actions(ctx);
3911 compose_controller_action(ctx, max_len);
3914 add_output_action(ctx, OFPP_LOCAL);
3919 if (port != ctx->flow.in_port) {
3920 add_output_action(ctx, port);
3925 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3926 ctx->nf_output_iface = NF_OUT_FLOOD;
3927 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3928 ctx->nf_output_iface = prev_nf_output_iface;
3929 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3930 ctx->nf_output_iface != NF_OUT_FLOOD) {
3931 ctx->nf_output_iface = NF_OUT_MULTI;
3936 xlate_output_reg_action(struct action_xlate_ctx *ctx,
3937 const struct nx_action_output_reg *naor)
3941 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
3943 if (ofp_port <= UINT16_MAX) {
3944 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
3949 xlate_output_action(struct action_xlate_ctx *ctx,
3950 const struct ofp_action_output *oao)
3952 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3956 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3957 const struct ofp_action_enqueue *oae)
3959 uint16_t ofp_port, odp_port;
3960 uint32_t ctx_priority, priority;
3963 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3966 /* Fall back to ordinary output action. */
3967 xlate_output_action__(ctx, ntohs(oae->port), 0);
3971 /* Figure out datapath output port. */
3972 ofp_port = ntohs(oae->port);
3973 if (ofp_port == OFPP_IN_PORT) {
3974 ofp_port = ctx->flow.in_port;
3975 } else if (ofp_port == ctx->flow.in_port) {
3978 odp_port = ofp_port_to_odp_port(ofp_port);
3980 /* Add datapath actions. */
3981 ctx_priority = ctx->priority;
3982 ctx->priority = priority;
3983 add_output_action(ctx, odp_port);
3984 ctx->priority = ctx_priority;
3986 /* Update NetFlow output port. */
3987 if (ctx->nf_output_iface == NF_OUT_DROP) {
3988 ctx->nf_output_iface = odp_port;
3989 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3990 ctx->nf_output_iface = NF_OUT_MULTI;
3995 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3996 const struct nx_action_set_queue *nasq)
4001 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4004 /* Couldn't translate queue to a priority, so ignore. A warning
4005 * has already been logged. */
4009 ctx->priority = priority;
4012 struct xlate_reg_state {
4018 xlate_autopath(struct action_xlate_ctx *ctx,
4019 const struct nx_action_autopath *naa)
4021 uint16_t ofp_port = ntohl(naa->id);
4022 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4024 if (!port || !port->bundle) {
4025 ofp_port = OFPP_NONE;
4026 } else if (port->bundle->bond) {
4027 /* Autopath does not support VLAN hashing. */
4028 struct ofport_dpif *slave = bond_choose_output_slave(
4029 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4031 ofp_port = slave->up.ofp_port;
4034 autopath_execute(naa, &ctx->flow, ofp_port);
4038 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4040 struct ofproto_dpif *ofproto = ofproto_;
4041 struct ofport_dpif *port;
4051 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4054 port = get_ofp_port(ofproto, ofp_port);
4055 return port ? port->may_enable : false;
4060 xlate_learn_action(struct action_xlate_ctx *ctx,
4061 const struct nx_action_learn *learn)
4063 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4064 struct ofputil_flow_mod fm;
4067 learn_execute(learn, &ctx->flow, &fm);
4069 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4070 if (error && !VLOG_DROP_WARN(&rl)) {
4071 char *msg = ofputil_error_to_string(error);
4072 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4080 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4082 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4083 ? htonl(OFPPC_NO_RECV_STP)
4084 : htonl(OFPPC_NO_RECV))) {
4088 /* Only drop packets here if both forwarding and learning are
4089 * disabled. If just learning is enabled, we need to have
4090 * OFPP_NORMAL and the learning action have a look at the packet
4091 * before we can drop it. */
4092 if (!stp_forward_in_state(port->stp_state)
4093 && !stp_learn_in_state(port->stp_state)) {
4101 do_xlate_actions(const union ofp_action *in, size_t n_in,
4102 struct action_xlate_ctx *ctx)
4104 const struct ofport_dpif *port;
4105 const union ofp_action *ia;
4108 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4109 if (port && !may_receive(port, ctx)) {
4110 /* Drop this flow. */
4114 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4115 const struct ofp_action_dl_addr *oada;
4116 const struct nx_action_resubmit *nar;
4117 const struct nx_action_set_tunnel *nast;
4118 const struct nx_action_set_queue *nasq;
4119 const struct nx_action_multipath *nam;
4120 const struct nx_action_autopath *naa;
4121 const struct nx_action_bundle *nab;
4122 const struct nx_action_output_reg *naor;
4123 enum ofputil_action_code code;
4126 code = ofputil_decode_action_unsafe(ia);
4128 case OFPUTIL_OFPAT_OUTPUT:
4129 xlate_output_action(ctx, &ia->output);
4132 case OFPUTIL_OFPAT_SET_VLAN_VID:
4133 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4134 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4137 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4138 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4139 ctx->flow.vlan_tci |= htons(
4140 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4143 case OFPUTIL_OFPAT_STRIP_VLAN:
4144 ctx->flow.vlan_tci = htons(0);
4147 case OFPUTIL_OFPAT_SET_DL_SRC:
4148 oada = ((struct ofp_action_dl_addr *) ia);
4149 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4152 case OFPUTIL_OFPAT_SET_DL_DST:
4153 oada = ((struct ofp_action_dl_addr *) ia);
4154 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4157 case OFPUTIL_OFPAT_SET_NW_SRC:
4158 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4161 case OFPUTIL_OFPAT_SET_NW_DST:
4162 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4165 case OFPUTIL_OFPAT_SET_NW_TOS:
4166 ctx->flow.tos_frag &= ~IP_DSCP_MASK;
4167 ctx->flow.tos_frag |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4170 case OFPUTIL_OFPAT_SET_TP_SRC:
4171 ctx->flow.tp_src = ia->tp_port.tp_port;
4174 case OFPUTIL_OFPAT_SET_TP_DST:
4175 ctx->flow.tp_dst = ia->tp_port.tp_port;
4178 case OFPUTIL_OFPAT_ENQUEUE:
4179 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4182 case OFPUTIL_NXAST_RESUBMIT:
4183 nar = (const struct nx_action_resubmit *) ia;
4184 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4187 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4188 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4191 case OFPUTIL_NXAST_SET_TUNNEL:
4192 nast = (const struct nx_action_set_tunnel *) ia;
4193 tun_id = htonll(ntohl(nast->tun_id));
4194 ctx->flow.tun_id = tun_id;
4197 case OFPUTIL_NXAST_SET_QUEUE:
4198 nasq = (const struct nx_action_set_queue *) ia;
4199 xlate_set_queue_action(ctx, nasq);
4202 case OFPUTIL_NXAST_POP_QUEUE:
4206 case OFPUTIL_NXAST_REG_MOVE:
4207 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4211 case OFPUTIL_NXAST_REG_LOAD:
4212 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4216 case OFPUTIL_NXAST_NOTE:
4217 /* Nothing to do. */
4220 case OFPUTIL_NXAST_SET_TUNNEL64:
4221 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4222 ctx->flow.tun_id = tun_id;
4225 case OFPUTIL_NXAST_MULTIPATH:
4226 nam = (const struct nx_action_multipath *) ia;
4227 multipath_execute(nam, &ctx->flow);
4230 case OFPUTIL_NXAST_AUTOPATH:
4231 naa = (const struct nx_action_autopath *) ia;
4232 xlate_autopath(ctx, naa);
4235 case OFPUTIL_NXAST_BUNDLE:
4236 ctx->ofproto->has_bundle_action = true;
4237 nab = (const struct nx_action_bundle *) ia;
4238 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4243 case OFPUTIL_NXAST_BUNDLE_LOAD:
4244 ctx->ofproto->has_bundle_action = true;
4245 nab = (const struct nx_action_bundle *) ia;
4246 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4250 case OFPUTIL_NXAST_OUTPUT_REG:
4251 naor = (const struct nx_action_output_reg *) ia;
4252 xlate_output_reg_action(ctx, naor);
4255 case OFPUTIL_NXAST_LEARN:
4256 ctx->has_learn = true;
4257 if (ctx->may_learn) {
4258 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4264 /* We've let OFPP_NORMAL and the learning action look at the packet,
4265 * so drop it now if forwarding is disabled. */
4266 if (port && !stp_forward_in_state(port->stp_state)) {
4267 ofpbuf_clear(ctx->odp_actions);
4268 add_sflow_action(ctx);
4273 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4274 struct ofproto_dpif *ofproto, const struct flow *flow,
4275 const struct ofpbuf *packet)
4277 ctx->ofproto = ofproto;
4279 ctx->packet = packet;
4280 ctx->may_learn = packet != NULL;
4281 ctx->resubmit_hook = NULL;
4284 static struct ofpbuf *
4285 xlate_actions(struct action_xlate_ctx *ctx,
4286 const union ofp_action *in, size_t n_in)
4288 COVERAGE_INC(ofproto_dpif_xlate);
4290 ctx->odp_actions = ofpbuf_new(512);
4291 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4293 ctx->may_set_up_flow = true;
4294 ctx->has_learn = false;
4295 ctx->has_normal = false;
4296 ctx->nf_output_iface = NF_OUT_DROP;
4299 ctx->base_priority = 0;
4300 ctx->base_flow = ctx->flow;
4301 ctx->base_flow.tun_id = 0;
4304 if (ctx->flow.tos_frag & FLOW_FRAG_ANY) {
4305 switch (ctx->ofproto->up.frag_handling) {
4306 case OFPC_FRAG_NORMAL:
4307 /* We must pretend that transport ports are unavailable. */
4308 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4309 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4312 case OFPC_FRAG_DROP:
4313 return ctx->odp_actions;
4315 case OFPC_FRAG_REASM:
4318 case OFPC_FRAG_NX_MATCH:
4319 /* Nothing to do. */
4324 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4325 ctx->may_set_up_flow = false;
4326 return ctx->odp_actions;
4328 add_sflow_action(ctx);
4329 do_xlate_actions(in, n_in, ctx);
4331 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4332 ctx->odp_actions->data,
4333 ctx->odp_actions->size)) {
4334 ctx->may_set_up_flow = false;
4336 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4338 compose_output_action(ctx, OVSP_LOCAL);
4341 fix_sflow_action(ctx);
4344 return ctx->odp_actions;
4347 /* OFPP_NORMAL implementation. */
4350 struct ofport_dpif *port;
4355 struct dst builtin[32];
4357 size_t n, allocated;
4360 static void dst_set_init(struct dst_set *);
4361 static void dst_set_add(struct dst_set *, const struct dst *);
4362 static void dst_set_free(struct dst_set *);
4364 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4366 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4367 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4368 * the bundle on which the packet was received, returns the VLAN to which the
4371 * Both 'vid' and the return value are in the range 0...4095. */
4373 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4375 switch (in_bundle->vlan_mode) {
4376 case PORT_VLAN_ACCESS:
4377 return in_bundle->vlan;
4380 case PORT_VLAN_TRUNK:
4383 case PORT_VLAN_NATIVE_UNTAGGED:
4384 case PORT_VLAN_NATIVE_TAGGED:
4385 return vid ? vid : in_bundle->vlan;
4392 /* Given 'vlan', the VLAN that a packet belongs to, and
4393 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4394 * that should be included in the 802.1Q header. (If the return value is 0,
4395 * then the 802.1Q header should only be included in the packet if there is a
4398 * Both 'vlan' and the return value are in the range 0...4095. */
4400 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4402 switch (out_bundle->vlan_mode) {
4403 case PORT_VLAN_ACCESS:
4406 case PORT_VLAN_TRUNK:
4407 case PORT_VLAN_NATIVE_TAGGED:
4410 case PORT_VLAN_NATIVE_UNTAGGED:
4411 return vlan == out_bundle->vlan ? 0 : vlan;
4419 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
4420 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
4424 vlan = input_vid_to_vlan(in_bundle, vlan_tci_to_vid(ctx->flow.vlan_tci));
4425 dst->vid = output_vlan_to_vid(out_bundle, vlan);
4427 dst->port = (!out_bundle->bond
4428 ? ofbundle_get_a_port(out_bundle)
4429 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4430 dst->vid, &ctx->tags));
4431 return dst->port != NULL;
4435 mirror_mask_ffs(mirror_mask_t mask)
4437 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4442 dst_set_init(struct dst_set *set)
4444 set->dsts = set->builtin;
4446 set->allocated = ARRAY_SIZE(set->builtin);
4450 dst_set_add(struct dst_set *set, const struct dst *dst)
4452 if (set->n >= set->allocated) {
4453 size_t new_allocated;
4454 struct dst *new_dsts;
4456 new_allocated = set->allocated * 2;
4457 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
4458 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
4462 set->dsts = new_dsts;
4463 set->allocated = new_allocated;
4465 set->dsts[set->n++] = *dst;
4469 dst_set_free(struct dst_set *set)
4471 if (set->dsts != set->builtin) {
4477 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
4480 for (i = 0; i < set->n; i++) {
4481 if (set->dsts[i].vid == test->vid
4482 && set->dsts[i].port == test->port) {
4490 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4492 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4493 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4497 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4499 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4502 /* Returns an arbitrary interface within 'bundle'. */
4503 static struct ofport_dpif *
4504 ofbundle_get_a_port(const struct ofbundle *bundle)
4506 return CONTAINER_OF(list_front(&bundle->ports),
4507 struct ofport_dpif, bundle_node);
4511 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
4512 const struct ofbundle *in_bundle,
4513 const struct ofbundle *out_bundle, struct dst_set *set)
4517 if (out_bundle == OFBUNDLE_FLOOD) {
4518 struct ofbundle *bundle;
4520 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
4521 if (bundle != in_bundle
4522 && ofbundle_includes_vlan(bundle, vlan)
4523 && bundle->floodable
4524 && !bundle->mirror_out
4525 && set_dst(ctx, &dst, in_bundle, bundle)) {
4526 dst_set_add(set, &dst);
4529 ctx->nf_output_iface = NF_OUT_FLOOD;
4530 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
4531 dst_set_add(set, &dst);
4532 ctx->nf_output_iface = dst.port->odp_port;
4537 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4539 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4542 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4543 * to a VLAN. In general most packets may be mirrored but we want to drop
4544 * protocols that may confuse switches. */
4546 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4548 /* If you change this function's behavior, please update corresponding
4549 * documentation in vswitch.xml at the same time. */
4550 if (dst[0] != 0x01) {
4551 /* All the currently banned MACs happen to start with 01 currently, so
4552 * this is a quick way to eliminate most of the good ones. */
4554 if (eth_addr_is_reserved(dst)) {
4555 /* Drop STP, IEEE pause frames, and other reserved protocols
4556 * (01-80-c2-00-00-0x). */
4560 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4562 if ((dst[3] & 0xfe) == 0xcc &&
4563 (dst[4] & 0xfe) == 0xcc &&
4564 (dst[5] & 0xfe) == 0xcc) {
4565 /* Drop the following protocols plus others following the same
4568 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4569 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4570 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4574 if (!(dst[3] | dst[4] | dst[5])) {
4575 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4584 compose_mirror_dsts(struct action_xlate_ctx *ctx,
4585 uint16_t vlan, const struct ofbundle *in_bundle,
4586 struct dst_set *set)
4588 struct ofproto_dpif *ofproto = ctx->ofproto;
4589 mirror_mask_t mirrors;
4593 mirrors = in_bundle->src_mirrors;
4594 for (i = 0; i < set->n; i++) {
4595 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
4602 flow_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4604 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
4605 if (vlan_is_mirrored(m, vlan)) {
4609 if (set_dst(ctx, &dst, in_bundle, m->out)
4610 && !dst_is_duplicate(set, &dst)) {
4611 dst_set_add(set, &dst);
4613 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
4614 struct ofbundle *bundle;
4616 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4617 if (ofbundle_includes_vlan(bundle, m->out_vlan)
4618 && set_dst(ctx, &dst, in_bundle, bundle))
4620 /* set_dst() got dst->vid from the input packet's VLAN,
4621 * not from m->out_vlan, so recompute it. */
4622 dst.vid = output_vlan_to_vid(bundle, m->out_vlan);
4624 if (dst_is_duplicate(set, &dst)) {
4628 if (bundle == in_bundle && dst.vid == flow_vid) {
4629 /* Don't send out input port on same VLAN. */
4632 dst_set_add(set, &dst);
4637 mirrors &= mirrors - 1;
4642 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
4643 const struct ofbundle *in_bundle,
4644 const struct ofbundle *out_bundle)
4646 uint16_t initial_vid, cur_vid;
4647 const struct dst *dst;
4651 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
4652 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
4658 /* Output all the packets we can without having to change the VLAN. */
4659 commit_odp_actions(ctx);
4660 initial_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4661 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
4662 if (dst->vid != initial_vid) {
4665 compose_output_action(ctx, dst->port->odp_port);
4668 /* Then output the rest. */
4669 cur_vid = initial_vid;
4670 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
4671 if (dst->vid == initial_vid) {
4674 if (dst->vid != cur_vid) {
4677 tci = htons(dst->vid);
4678 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4680 tci |= htons(VLAN_CFI);
4682 commit_vlan_action(ctx, tci);
4686 compose_output_action(ctx, dst->port->odp_port);
4692 /* Returns the effective vlan of a packet, taking into account both the
4693 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
4694 * the packet is untagged and -1 indicates it has an invalid header and
4695 * should be dropped. */
4697 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
4698 struct ofbundle *in_bundle, bool have_packet)
4700 int vlan = vlan_tci_to_vid(flow->vlan_tci);
4702 if (in_bundle->vlan_mode == PORT_VLAN_ACCESS) {
4703 /* Drop tagged packet on access port */
4705 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4706 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
4707 "packet received on port %s configured with "
4708 "implicit VLAN %"PRIu16,
4709 ofproto->up.name, vlan,
4710 in_bundle->name, in_bundle->vlan);
4713 } else if (ofbundle_includes_vlan(in_bundle, vlan)) {
4716 /* Drop packets from a VLAN not member of the trunk */
4718 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4719 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
4720 "packet received on port %s not configured for "
4722 ofproto->up.name, vlan, in_bundle->name, vlan);
4727 if (in_bundle->vlan_mode != PORT_VLAN_TRUNK) {
4728 return in_bundle->vlan;
4730 return ofbundle_includes_vlan(in_bundle, 0) ? 0 : -1;
4735 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
4736 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
4737 * indicate this; newer upstream kernels use gratuitous ARP requests. */
4739 is_gratuitous_arp(const struct flow *flow)
4741 return (flow->dl_type == htons(ETH_TYPE_ARP)
4742 && eth_addr_is_broadcast(flow->dl_dst)
4743 && (flow->nw_proto == ARP_OP_REPLY
4744 || (flow->nw_proto == ARP_OP_REQUEST
4745 && flow->nw_src == flow->nw_dst)));
4749 update_learning_table(struct ofproto_dpif *ofproto,
4750 const struct flow *flow, int vlan,
4751 struct ofbundle *in_bundle)
4753 struct mac_entry *mac;
4755 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
4759 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
4760 if (is_gratuitous_arp(flow)) {
4761 /* We don't want to learn from gratuitous ARP packets that are
4762 * reflected back over bond slaves so we lock the learning table. */
4763 if (!in_bundle->bond) {
4764 mac_entry_set_grat_arp_lock(mac);
4765 } else if (mac_entry_is_grat_arp_locked(mac)) {
4770 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
4771 /* The log messages here could actually be useful in debugging,
4772 * so keep the rate limit relatively high. */
4773 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4774 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
4775 "on port %s in VLAN %d",
4776 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
4777 in_bundle->name, vlan);
4779 mac->port.p = in_bundle;
4780 tag_set_add(&ofproto->revalidate_set,
4781 mac_learning_changed(ofproto->ml, mac));
4785 /* Determines whether packets in 'flow' within 'br' should be forwarded or
4786 * dropped. Returns true if they may be forwarded, false if they should be
4789 * If 'have_packet' is true, it indicates that the caller is processing a
4790 * received packet. If 'have_packet' is false, then the caller is just
4791 * revalidating an existing flow because configuration has changed. Either
4792 * way, 'have_packet' only affects logging (there is no point in logging errors
4793 * during revalidation).
4795 * Sets '*in_portp' to the input port. This will be a null pointer if
4796 * flow->in_port does not designate a known input port (in which case
4797 * is_admissible() returns false).
4799 * When returning true, sets '*vlanp' to the effective VLAN of the input
4800 * packet, as returned by flow_get_vlan().
4802 * May also add tags to '*tags', although the current implementation only does
4803 * so in one special case.
4806 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
4808 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
4810 struct ofport_dpif *in_port;
4811 struct ofbundle *in_bundle;
4814 /* Find the port and bundle for the received packet. */
4815 in_port = get_ofp_port(ofproto, flow->in_port);
4816 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
4817 if (!in_port || !in_bundle) {
4818 /* No interface? Something fishy... */
4820 /* Odd. A few possible reasons here:
4822 * - We deleted a port but there are still a few packets queued up
4825 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
4826 * we don't know about.
4828 * - Packet arrived on the local port but the local port is not
4831 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4833 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
4835 ofproto->up.name, flow->in_port);
4840 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
4845 /* Drop frames for reserved multicast addresses only if forward_bpdu
4846 * option is absent. */
4847 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
4851 /* Drop frames on bundles reserved for mirroring. */
4852 if (in_bundle->mirror_out) {
4854 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4855 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4856 "%s, which is reserved exclusively for mirroring",
4857 ofproto->up.name, in_bundle->name);
4862 if (in_bundle->bond) {
4863 struct mac_entry *mac;
4865 switch (bond_check_admissibility(in_bundle->bond, in_port,
4866 flow->dl_dst, tags)) {
4873 case BV_DROP_IF_MOVED:
4874 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
4875 if (mac && mac->port.p != in_bundle &&
4876 (!is_gratuitous_arp(flow)
4877 || mac_entry_is_grat_arp_locked(mac))) {
4888 xlate_normal(struct action_xlate_ctx *ctx)
4890 struct ofbundle *in_bundle;
4891 struct ofbundle *out_bundle;
4892 struct mac_entry *mac;
4895 ctx->has_normal = true;
4897 /* Check whether we should drop packets in this flow. */
4898 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
4899 &ctx->tags, &vlan, &in_bundle)) {
4904 /* Learn source MAC. */
4905 if (ctx->may_learn) {
4906 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
4909 /* Determine output bundle. */
4910 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
4913 out_bundle = mac->port.p;
4914 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
4915 /* If we are revalidating but don't have a learning entry then eject
4916 * the flow. Installing a flow that floods packets opens up a window
4917 * of time where we could learn from a packet reflected on a bond and
4918 * blackhole packets before the learning table is updated to reflect
4919 * the correct port. */
4920 ctx->may_set_up_flow = false;
4923 out_bundle = OFBUNDLE_FLOOD;
4926 /* Don't send packets out their input bundles. */
4927 if (in_bundle == out_bundle) {
4933 compose_actions(ctx, vlan, in_bundle, out_bundle);
4937 /* Optimized flow revalidation.
4939 * It's a difficult problem, in general, to tell which facets need to have
4940 * their actions recalculated whenever the OpenFlow flow table changes. We
4941 * don't try to solve that general problem: for most kinds of OpenFlow flow
4942 * table changes, we recalculate the actions for every facet. This is
4943 * relatively expensive, but it's good enough if the OpenFlow flow table
4944 * doesn't change very often.
4946 * However, we can expect one particular kind of OpenFlow flow table change to
4947 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
4948 * of CPU on revalidating every facet whenever MAC learning modifies the flow
4949 * table, we add a special case that applies to flow tables in which every rule
4950 * has the same form (that is, the same wildcards), except that the table is
4951 * also allowed to have a single "catch-all" flow that matches all packets. We
4952 * optimize this case by tagging all of the facets that resubmit into the table
4953 * and invalidating the same tag whenever a flow changes in that table. The
4954 * end result is that we revalidate just the facets that need it (and sometimes
4955 * a few more, but not all of the facets or even all of the facets that
4956 * resubmit to the table modified by MAC learning). */
4958 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
4959 * into an OpenFlow table with the given 'basis'. */
4961 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
4964 if (flow_wildcards_is_catchall(wc)) {
4967 struct flow tag_flow = *flow;
4968 flow_zero_wildcards(&tag_flow, wc);
4969 return tag_create_deterministic(flow_hash(&tag_flow, secret));
4973 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
4974 * taggability of that table.
4976 * This function must be called after *each* change to a flow table. If you
4977 * skip calling it on some changes then the pointer comparisons at the end can
4978 * be invalid if you get unlucky. For example, if a flow removal causes a
4979 * cls_table to be destroyed and then a flow insertion causes a cls_table with
4980 * different wildcards to be created with the same address, then this function
4981 * will incorrectly skip revalidation. */
4983 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
4985 struct table_dpif *table = &ofproto->tables[table_id];
4986 const struct classifier *cls = &ofproto->up.tables[table_id];
4987 struct cls_table *catchall, *other;
4988 struct cls_table *t;
4990 catchall = other = NULL;
4992 switch (hmap_count(&cls->tables)) {
4994 /* We could tag this OpenFlow table but it would make the logic a
4995 * little harder and it's a corner case that doesn't seem worth it
5001 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5002 if (cls_table_is_catchall(t)) {
5004 } else if (!other) {
5007 /* Indicate that we can't tag this by setting both tables to
5008 * NULL. (We know that 'catchall' is already NULL.) */
5015 /* Can't tag this table. */
5019 if (table->catchall_table != catchall || table->other_table != other) {
5020 table->catchall_table = catchall;
5021 table->other_table = other;
5022 ofproto->need_revalidate = true;
5026 /* Given 'rule' that has changed in some way (either it is a rule being
5027 * inserted, a rule being deleted, or a rule whose actions are being
5028 * modified), marks facets for revalidation to ensure that packets will be
5029 * forwarded correctly according to the new state of the flow table.
5031 * This function must be called after *each* change to a flow table. See
5032 * the comment on table_update_taggable() for more information. */
5034 rule_invalidate(const struct rule_dpif *rule)
5036 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5038 table_update_taggable(ofproto, rule->up.table_id);
5040 if (!ofproto->need_revalidate) {
5041 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5043 if (table->other_table && rule->tag) {
5044 tag_set_add(&ofproto->revalidate_set, rule->tag);
5046 ofproto->need_revalidate = true;
5052 set_frag_handling(struct ofproto *ofproto_,
5053 enum ofp_config_flags frag_handling)
5055 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5057 if (frag_handling != OFPC_FRAG_REASM) {
5058 ofproto->need_revalidate = true;
5066 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5067 const struct flow *flow,
5068 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5070 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5073 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5074 ofproto->max_ports);
5076 struct odputil_keybuf keybuf;
5077 struct action_xlate_ctx ctx;
5078 struct ofpbuf *odp_actions;
5081 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5082 odp_flow_key_from_flow(&key, flow);
5084 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
5085 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5086 dpif_execute(ofproto->dpif, key.data, key.size,
5087 odp_actions->data, odp_actions->size, packet);
5088 ofpbuf_delete(odp_actions);
5094 get_netflow_ids(const struct ofproto *ofproto_,
5095 uint8_t *engine_type, uint8_t *engine_id)
5097 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5099 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5102 static struct ofproto_dpif *
5103 ofproto_dpif_lookup(const char *name)
5105 struct ofproto *ofproto = ofproto_lookup(name);
5106 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
5107 ? ofproto_dpif_cast(ofproto)
5112 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5113 const char *args, void *aux OVS_UNUSED)
5115 struct ds ds = DS_EMPTY_INITIALIZER;
5116 const struct ofproto_dpif *ofproto;
5117 const struct mac_entry *e;
5119 ofproto = ofproto_dpif_lookup(args);
5121 unixctl_command_reply(conn, 501, "no such bridge");
5125 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5126 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5127 struct ofbundle *bundle = e->port.p;
5128 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5129 ofbundle_get_a_port(bundle)->odp_port,
5130 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5132 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5136 struct ofproto_trace {
5137 struct action_xlate_ctx ctx;
5143 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5144 const struct rule_dpif *rule)
5146 ds_put_char_multiple(result, '\t', level);
5148 ds_put_cstr(result, "No match\n");
5152 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5153 table_id, ntohll(rule->up.flow_cookie));
5154 cls_rule_format(&rule->up.cr, result);
5155 ds_put_char(result, '\n');
5157 ds_put_char_multiple(result, '\t', level);
5158 ds_put_cstr(result, "OpenFlow ");
5159 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5160 ds_put_char(result, '\n');
5164 trace_format_flow(struct ds *result, int level, const char *title,
5165 struct ofproto_trace *trace)
5167 ds_put_char_multiple(result, '\t', level);
5168 ds_put_format(result, "%s: ", title);
5169 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5170 ds_put_cstr(result, "unchanged");
5172 flow_format(result, &trace->ctx.flow);
5173 trace->flow = trace->ctx.flow;
5175 ds_put_char(result, '\n');
5179 trace_format_regs(struct ds *result, int level, const char *title,
5180 struct ofproto_trace *trace)
5184 ds_put_char_multiple(result, '\t', level);
5185 ds_put_format(result, "%s:", title);
5186 for (i = 0; i < FLOW_N_REGS; i++) {
5187 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5189 ds_put_char(result, '\n');
5193 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5195 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5196 struct ds *result = trace->result;
5198 ds_put_char(result, '\n');
5199 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5200 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5201 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5205 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5206 void *aux OVS_UNUSED)
5208 char *dpname, *arg1, *arg2, *arg3;
5209 char *args = xstrdup(args_);
5210 char *save_ptr = NULL;
5211 struct ofproto_dpif *ofproto;
5212 struct ofpbuf odp_key;
5213 struct ofpbuf *packet;
5214 struct rule_dpif *rule;
5220 ofpbuf_init(&odp_key, 0);
5223 dpname = strtok_r(args, " ", &save_ptr);
5224 arg1 = strtok_r(NULL, " ", &save_ptr);
5225 arg2 = strtok_r(NULL, " ", &save_ptr);
5226 arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5227 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5228 /* ofproto/trace dpname flow [-generate] */
5231 /* Convert string to datapath key. */
5232 ofpbuf_init(&odp_key, 0);
5233 error = odp_flow_key_from_string(arg1, &odp_key);
5235 unixctl_command_reply(conn, 501, "Bad flow syntax");
5239 /* Convert odp_key to flow. */
5240 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
5242 unixctl_command_reply(conn, 501, "Invalid flow");
5246 /* Generate a packet, if requested. */
5248 packet = ofpbuf_new(0);
5249 flow_compose(packet, &flow);
5251 } else if (dpname && arg1 && arg2 && arg3) {
5252 /* ofproto/trace dpname tun_id in_port packet */
5256 tun_id = htonll(strtoull(arg1, NULL, 0));
5257 in_port = ofp_port_to_odp_port(atoi(arg2));
5259 packet = ofpbuf_new(strlen(args) / 2);
5260 arg3 = ofpbuf_put_hex(packet, arg3, NULL);
5261 arg3 += strspn(arg3, " ");
5262 if (*arg3 != '\0') {
5263 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5266 if (packet->size < ETH_HEADER_LEN) {
5267 unixctl_command_reply(conn, 501,
5268 "Packet data too short for Ethernet");
5272 ds_put_cstr(&result, "Packet: ");
5273 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5274 ds_put_cstr(&result, s);
5277 flow_extract(packet, tun_id, in_port, &flow);
5279 unixctl_command_reply(conn, 501, "Bad command syntax");
5283 ofproto = ofproto_dpif_lookup(dpname);
5285 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5290 ds_put_cstr(&result, "Flow: ");
5291 flow_format(&result, &flow);
5292 ds_put_char(&result, '\n');
5294 rule = rule_dpif_lookup(ofproto, &flow, 0);
5295 trace_format_rule(&result, 0, 0, rule);
5297 struct ofproto_trace trace;
5298 struct ofpbuf *odp_actions;
5300 trace.result = &result;
5302 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
5303 trace.ctx.resubmit_hook = trace_resubmit;
5304 odp_actions = xlate_actions(&trace.ctx,
5305 rule->up.actions, rule->up.n_actions);
5307 ds_put_char(&result, '\n');
5308 trace_format_flow(&result, 0, "Final flow", &trace);
5309 ds_put_cstr(&result, "Datapath actions: ");
5310 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5311 ofpbuf_delete(odp_actions);
5313 if (!trace.ctx.may_set_up_flow) {
5315 ds_put_cstr(&result, "\nThis flow is not cachable.");
5317 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5318 "for complete actions, please supply a packet.");
5323 unixctl_command_reply(conn, 200, ds_cstr(&result));
5326 ds_destroy(&result);
5327 ofpbuf_delete(packet);
5328 ofpbuf_uninit(&odp_key);
5333 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5334 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5337 unixctl_command_reply(conn, 200, NULL);
5341 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5342 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5345 unixctl_command_reply(conn, 200, NULL);
5349 ofproto_dpif_unixctl_init(void)
5351 static bool registered;
5357 unixctl_command_register("ofproto/trace",
5358 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5359 ofproto_unixctl_trace, NULL);
5360 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5362 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5363 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5366 const struct ofproto_class ofproto_dpif_class = {
5393 port_is_lacp_current,
5394 NULL, /* rule_choose_table */
5401 rule_modify_actions,
5409 get_cfm_remote_mpids,
5413 get_stp_port_status,
5418 is_mirror_output_bundle,
5419 forward_bpdu_changed,