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 int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch);
420 /* Flow expiration. */
421 static int expire(struct ofproto_dpif *);
424 static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
425 const struct ofpbuf *packet);
427 compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions,
428 const struct flow *, uint32_t odp_port);
429 /* Global variables. */
430 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
432 /* Factory functions. */
435 enumerate_types(struct sset *types)
437 dp_enumerate_types(types);
441 enumerate_names(const char *type, struct sset *names)
443 return dp_enumerate_names(type, names);
447 del(const char *type, const char *name)
452 error = dpif_open(name, type, &dpif);
454 error = dpif_delete(dpif);
460 /* Basic life-cycle. */
462 static struct ofproto *
465 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
470 dealloc(struct ofproto *ofproto_)
472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
477 construct(struct ofproto *ofproto_, int *n_tablesp)
479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
480 const char *name = ofproto->up.name;
484 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
486 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
490 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
491 ofproto->n_matches = 0;
493 dpif_flow_flush(ofproto->dpif);
494 dpif_recv_purge(ofproto->dpif);
496 error = dpif_recv_set_mask(ofproto->dpif,
497 ((1u << DPIF_UC_MISS) |
498 (1u << DPIF_UC_ACTION)));
500 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
501 dpif_close(ofproto->dpif);
505 ofproto->netflow = NULL;
506 ofproto->sflow = NULL;
508 hmap_init(&ofproto->bundles);
509 ofproto->ml = mac_learning_create();
510 for (i = 0; i < MAX_MIRRORS; i++) {
511 ofproto->mirrors[i] = NULL;
513 ofproto->has_bonded_bundles = false;
515 timer_set_duration(&ofproto->next_expiration, 1000);
517 hmap_init(&ofproto->facets);
519 for (i = 0; i < N_TABLES; i++) {
520 struct table_dpif *table = &ofproto->tables[i];
522 table->catchall_table = NULL;
523 table->other_table = NULL;
524 table->basis = random_uint32();
526 ofproto->need_revalidate = false;
527 tag_set_init(&ofproto->revalidate_set);
529 list_init(&ofproto->completions);
531 ofproto_dpif_unixctl_init();
533 ofproto->has_bundle_action = false;
535 *n_tablesp = N_TABLES;
540 complete_operations(struct ofproto_dpif *ofproto)
542 struct dpif_completion *c, *next;
544 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
545 ofoperation_complete(c->op, 0);
546 list_remove(&c->list_node);
552 destruct(struct ofproto *ofproto_)
554 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
555 struct rule_dpif *rule, *next_rule;
556 struct classifier *table;
559 complete_operations(ofproto);
561 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
562 struct cls_cursor cursor;
564 cls_cursor_init(&cursor, table, NULL);
565 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
566 ofproto_rule_destroy(&rule->up);
570 for (i = 0; i < MAX_MIRRORS; i++) {
571 mirror_destroy(ofproto->mirrors[i]);
574 netflow_destroy(ofproto->netflow);
575 dpif_sflow_destroy(ofproto->sflow);
576 hmap_destroy(&ofproto->bundles);
577 mac_learning_destroy(ofproto->ml);
579 hmap_destroy(&ofproto->facets);
581 dpif_close(ofproto->dpif);
585 run(struct ofproto *ofproto_)
587 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
588 struct ofport_dpif *ofport;
589 struct ofbundle *bundle;
593 complete_operations(ofproto);
595 dpif_run(ofproto->dpif);
597 /* Handle one or more batches of upcalls, until there's nothing left to do
598 * or until we do a fixed total amount of work.
600 * We do work in batches because it can be much cheaper to set up a number
601 * of flows and fire off their patches all at once. We do multiple batches
602 * because in some cases handling a packet can cause another packet to be
603 * queued almost immediately as part of the return flow. Both
604 * optimizations can make major improvements on some benchmarks and
605 * presumably for real traffic as well. */
607 while (work < FLOW_MISS_MAX_BATCH) {
608 int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work);
611 } else if (!retval) {
618 if (timer_expired(&ofproto->next_expiration)) {
619 int delay = expire(ofproto);
620 timer_set_duration(&ofproto->next_expiration, delay);
623 if (ofproto->netflow) {
624 netflow_run(ofproto->netflow);
626 if (ofproto->sflow) {
627 dpif_sflow_run(ofproto->sflow);
630 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
633 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
638 mac_learning_run(ofproto->ml, &ofproto->revalidate_set);
640 /* Now revalidate if there's anything to do. */
641 if (ofproto->need_revalidate
642 || !tag_set_is_empty(&ofproto->revalidate_set)) {
643 struct tag_set revalidate_set = ofproto->revalidate_set;
644 bool revalidate_all = ofproto->need_revalidate;
645 struct facet *facet, *next;
647 /* Clear the revalidation flags. */
648 tag_set_init(&ofproto->revalidate_set);
649 ofproto->need_revalidate = false;
651 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
653 || tag_set_intersects(&revalidate_set, facet->tags)) {
654 facet_revalidate(ofproto, facet);
663 wait(struct ofproto *ofproto_)
665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
666 struct ofport_dpif *ofport;
667 struct ofbundle *bundle;
669 if (!clogged && !list_is_empty(&ofproto->completions)) {
670 poll_immediate_wake();
673 dpif_wait(ofproto->dpif);
674 dpif_recv_wait(ofproto->dpif);
675 if (ofproto->sflow) {
676 dpif_sflow_wait(ofproto->sflow);
678 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
679 poll_immediate_wake();
681 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
684 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
687 mac_learning_wait(ofproto->ml);
689 if (ofproto->need_revalidate) {
690 /* Shouldn't happen, but if it does just go around again. */
691 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
692 poll_immediate_wake();
694 timer_wait(&ofproto->next_expiration);
699 flush(struct ofproto *ofproto_)
701 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
702 struct facet *facet, *next_facet;
704 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
705 /* Mark the facet as not installed so that facet_remove() doesn't
706 * bother trying to uninstall it. There is no point in uninstalling it
707 * individually since we are about to blow away all the facets with
708 * dpif_flow_flush(). */
709 facet->installed = false;
710 facet->dp_packet_count = 0;
711 facet->dp_byte_count = 0;
712 facet_remove(ofproto, facet);
714 dpif_flow_flush(ofproto->dpif);
718 get_features(struct ofproto *ofproto_ OVS_UNUSED,
719 bool *arp_match_ip, uint32_t *actions)
721 *arp_match_ip = true;
722 *actions = ((1u << OFPAT_OUTPUT) |
723 (1u << OFPAT_SET_VLAN_VID) |
724 (1u << OFPAT_SET_VLAN_PCP) |
725 (1u << OFPAT_STRIP_VLAN) |
726 (1u << OFPAT_SET_DL_SRC) |
727 (1u << OFPAT_SET_DL_DST) |
728 (1u << OFPAT_SET_NW_SRC) |
729 (1u << OFPAT_SET_NW_DST) |
730 (1u << OFPAT_SET_NW_TOS) |
731 (1u << OFPAT_SET_TP_SRC) |
732 (1u << OFPAT_SET_TP_DST) |
733 (1u << OFPAT_ENQUEUE));
737 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
739 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
740 struct dpif_dp_stats s;
742 strcpy(ots->name, "classifier");
744 dpif_get_dp_stats(ofproto->dpif, &s);
745 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
746 put_32aligned_be64(&ots->matched_count,
747 htonll(s.n_hit + ofproto->n_matches));
751 set_netflow(struct ofproto *ofproto_,
752 const struct netflow_options *netflow_options)
754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
756 if (netflow_options) {
757 if (!ofproto->netflow) {
758 ofproto->netflow = netflow_create();
760 return netflow_set_options(ofproto->netflow, netflow_options);
762 netflow_destroy(ofproto->netflow);
763 ofproto->netflow = NULL;
768 static struct ofport *
771 struct ofport_dpif *port = xmalloc(sizeof *port);
776 port_dealloc(struct ofport *port_)
778 struct ofport_dpif *port = ofport_dpif_cast(port_);
783 port_construct(struct ofport *port_)
785 struct ofport_dpif *port = ofport_dpif_cast(port_);
786 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
788 ofproto->need_revalidate = true;
789 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
792 port->tag = tag_create_random();
793 port->may_enable = true;
794 port->stp_port = NULL;
795 port->stp_state = STP_DISABLED;
797 if (ofproto->sflow) {
798 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
799 netdev_get_name(port->up.netdev));
806 port_destruct(struct ofport *port_)
808 struct ofport_dpif *port = ofport_dpif_cast(port_);
809 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
811 ofproto->need_revalidate = true;
812 bundle_remove(port_);
813 set_cfm(port_, NULL);
814 if (ofproto->sflow) {
815 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
820 port_modified(struct ofport *port_)
822 struct ofport_dpif *port = ofport_dpif_cast(port_);
824 if (port->bundle && port->bundle->bond) {
825 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
830 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
832 struct ofport_dpif *port = ofport_dpif_cast(port_);
833 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
834 ovs_be32 changed = old_config ^ port->up.opp.config;
836 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
837 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
838 ofproto->need_revalidate = true;
840 if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) {
841 bundle_update(port->bundle);
847 set_sflow(struct ofproto *ofproto_,
848 const struct ofproto_sflow_options *sflow_options)
850 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
851 struct dpif_sflow *ds = ofproto->sflow;
855 struct ofport_dpif *ofport;
857 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
858 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
859 dpif_sflow_add_port(ds, ofport->odp_port,
860 netdev_get_name(ofport->up.netdev));
862 ofproto->need_revalidate = true;
864 dpif_sflow_set_options(ds, sflow_options);
867 dpif_sflow_destroy(ds);
868 ofproto->need_revalidate = true;
869 ofproto->sflow = NULL;
876 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
878 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
885 struct ofproto_dpif *ofproto;
887 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
888 ofproto->need_revalidate = true;
889 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
892 if (cfm_configure(ofport->cfm, s)) {
898 cfm_destroy(ofport->cfm);
904 get_cfm_fault(const struct ofport *ofport_)
906 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
908 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
912 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
915 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
918 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
928 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
930 struct ofproto_dpif *ofproto = ofproto_;
931 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
932 struct ofport_dpif *ofport;
934 ofport = stp_port_get_aux(sp);
936 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
937 ofproto->up.name, port_num);
939 struct eth_header *eth = pkt->l2;
941 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
942 if (eth_addr_is_zero(eth->eth_src)) {
943 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
944 "with unknown MAC", ofproto->up.name, port_num);
946 int error = netdev_send(ofport->up.netdev, pkt);
948 VLOG_WARN_RL(&rl, "%s: sending BPDU on port %s failed (%s)",
950 netdev_get_name(ofport->up.netdev),
958 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
960 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
962 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
964 /* Only revalidate flows if the configuration changed. */
965 if (!s != !ofproto->stp) {
966 ofproto->need_revalidate = true;
971 ofproto->stp = stp_create(ofproto_->name, s->system_id,
972 send_bpdu_cb, ofproto);
973 ofproto->stp_last_tick = time_msec();
976 stp_set_bridge_id(ofproto->stp, s->system_id);
977 stp_set_bridge_priority(ofproto->stp, s->priority);
978 stp_set_hello_time(ofproto->stp, s->hello_time);
979 stp_set_max_age(ofproto->stp, s->max_age);
980 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
982 stp_destroy(ofproto->stp);
990 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
992 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
996 s->bridge_id = stp_get_bridge_id(ofproto->stp);
997 s->designated_root = stp_get_designated_root(ofproto->stp);
998 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1007 update_stp_port_state(struct ofport_dpif *ofport)
1009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1010 enum stp_state state;
1012 /* Figure out new state. */
1013 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1017 if (ofport->stp_state != state) {
1021 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1022 netdev_get_name(ofport->up.netdev),
1023 stp_state_name(ofport->stp_state),
1024 stp_state_name(state));
1025 if (stp_learn_in_state(ofport->stp_state)
1026 != stp_learn_in_state(state)) {
1027 /* xxx Learning action flows should also be flushed. */
1028 mac_learning_flush(ofproto->ml);
1030 fwd_change = stp_forward_in_state(ofport->stp_state)
1031 != stp_forward_in_state(state);
1033 ofproto->need_revalidate = true;
1034 ofport->stp_state = state;
1035 ofport->stp_state_entered = time_msec();
1037 if (fwd_change && ofport->bundle) {
1038 bundle_update(ofport->bundle);
1041 /* Update the STP state bits in the OpenFlow port description. */
1042 of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK))
1043 | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN
1044 : state == STP_LEARNING ? OFPPS_STP_LEARN
1045 : state == STP_FORWARDING ? OFPPS_STP_FORWARD
1046 : state == STP_BLOCKING ? OFPPS_STP_BLOCK
1048 ofproto_port_set_state(&ofport->up, of_state);
1052 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1053 * caller is responsible for assigning STP port numbers and ensuring
1054 * there are no duplicates. */
1056 set_stp_port(struct ofport *ofport_,
1057 const struct ofproto_port_stp_settings *s)
1059 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1060 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1061 struct stp_port *sp = ofport->stp_port;
1063 if (!s || !s->enable) {
1065 ofport->stp_port = NULL;
1066 stp_port_disable(sp);
1067 update_stp_port_state(ofport);
1070 } else if (sp && stp_port_no(sp) != s->port_num
1071 && ofport == stp_port_get_aux(sp)) {
1072 /* The port-id changed, so disable the old one if it's not
1073 * already in use by another port. */
1074 stp_port_disable(sp);
1077 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1078 stp_port_enable(sp);
1080 stp_port_set_aux(sp, ofport);
1081 stp_port_set_priority(sp, s->priority);
1082 stp_port_set_path_cost(sp, s->path_cost);
1084 update_stp_port_state(ofport);
1090 get_stp_port_status(struct ofport *ofport_,
1091 struct ofproto_port_stp_status *s)
1093 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1094 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1095 struct stp_port *sp = ofport->stp_port;
1097 if (!ofproto->stp || !sp) {
1103 s->port_id = stp_port_get_id(sp);
1104 s->state = stp_port_get_state(sp);
1105 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1106 s->role = stp_port_get_role(sp);
1107 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1113 stp_run(struct ofproto_dpif *ofproto)
1116 long long int now = time_msec();
1117 long long int elapsed = now - ofproto->stp_last_tick;
1118 struct stp_port *sp;
1121 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1122 ofproto->stp_last_tick = now;
1124 while (stp_get_changed_port(ofproto->stp, &sp)) {
1125 struct ofport_dpif *ofport = stp_port_get_aux(sp);
1128 update_stp_port_state(ofport);
1135 stp_wait(struct ofproto_dpif *ofproto)
1138 poll_timer_wait(1000);
1142 /* Returns true if STP should process 'flow'. */
1144 stp_should_process_flow(const struct flow *flow)
1146 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
1150 stp_process_packet(const struct ofport_dpif *ofport,
1151 const struct ofpbuf *packet)
1153 struct ofpbuf payload = *packet;
1154 struct eth_header *eth = payload.data;
1155 struct stp_port *sp = ofport->stp_port;
1157 /* Sink packets on ports that have STP disabled when the bridge has
1159 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1163 /* Trim off padding on payload. */
1164 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1165 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
1168 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1169 stp_received_bpdu(sp, payload.data, payload.size);
1175 /* Expires all MAC learning entries associated with 'port' and forces ofproto
1176 * to revalidate every flow. */
1178 bundle_flush_macs(struct ofbundle *bundle)
1180 struct ofproto_dpif *ofproto = bundle->ofproto;
1181 struct mac_learning *ml = ofproto->ml;
1182 struct mac_entry *mac, *next_mac;
1184 ofproto->need_revalidate = true;
1185 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1186 if (mac->port.p == bundle) {
1187 mac_learning_expire(ml, mac);
1192 static struct ofbundle *
1193 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1195 struct ofbundle *bundle;
1197 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1198 &ofproto->bundles) {
1199 if (bundle->aux == aux) {
1206 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
1207 * ones that are found to 'bundles'. */
1209 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
1210 void **auxes, size_t n_auxes,
1211 struct hmapx *bundles)
1215 hmapx_init(bundles);
1216 for (i = 0; i < n_auxes; i++) {
1217 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
1219 hmapx_add(bundles, bundle);
1225 bundle_update(struct ofbundle *bundle)
1227 struct ofport_dpif *port;
1229 bundle->floodable = true;
1230 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1231 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1232 || !stp_forward_in_state(port->stp_state)) {
1233 bundle->floodable = false;
1240 bundle_del_port(struct ofport_dpif *port)
1242 struct ofbundle *bundle = port->bundle;
1244 bundle->ofproto->need_revalidate = true;
1246 list_remove(&port->bundle_node);
1247 port->bundle = NULL;
1250 lacp_slave_unregister(bundle->lacp, port);
1253 bond_slave_unregister(bundle->bond, port);
1256 bundle_update(bundle);
1260 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
1261 struct lacp_slave_settings *lacp,
1262 uint32_t bond_stable_id)
1264 struct ofport_dpif *port;
1266 port = get_ofp_port(bundle->ofproto, ofp_port);
1271 if (port->bundle != bundle) {
1272 bundle->ofproto->need_revalidate = true;
1274 bundle_del_port(port);
1277 port->bundle = bundle;
1278 list_push_back(&bundle->ports, &port->bundle_node);
1279 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)
1280 || !stp_forward_in_state(port->stp_state)) {
1281 bundle->floodable = false;
1285 port->bundle->ofproto->need_revalidate = true;
1286 lacp_slave_register(bundle->lacp, port, lacp);
1289 port->bond_stable_id = bond_stable_id;
1295 bundle_destroy(struct ofbundle *bundle)
1297 struct ofproto_dpif *ofproto;
1298 struct ofport_dpif *port, *next_port;
1305 ofproto = bundle->ofproto;
1306 for (i = 0; i < MAX_MIRRORS; i++) {
1307 struct ofmirror *m = ofproto->mirrors[i];
1309 if (m->out == bundle) {
1311 } else if (hmapx_find_and_delete(&m->srcs, bundle)
1312 || hmapx_find_and_delete(&m->dsts, bundle)) {
1313 ofproto->need_revalidate = true;
1318 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1319 bundle_del_port(port);
1322 bundle_flush_macs(bundle);
1323 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
1325 free(bundle->trunks);
1326 lacp_destroy(bundle->lacp);
1327 bond_destroy(bundle->bond);
1332 bundle_set(struct ofproto *ofproto_, void *aux,
1333 const struct ofproto_bundle_settings *s)
1335 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1336 bool need_flush = false;
1337 struct ofport_dpif *port;
1338 struct ofbundle *bundle;
1339 unsigned long *trunks;
1345 bundle_destroy(bundle_lookup(ofproto, aux));
1349 assert(s->n_slaves == 1 || s->bond != NULL);
1350 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
1352 bundle = bundle_lookup(ofproto, aux);
1354 bundle = xmalloc(sizeof *bundle);
1356 bundle->ofproto = ofproto;
1357 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
1358 hash_pointer(aux, 0));
1360 bundle->name = NULL;
1362 list_init(&bundle->ports);
1363 bundle->vlan_mode = PORT_VLAN_TRUNK;
1365 bundle->trunks = NULL;
1366 bundle->lacp = NULL;
1367 bundle->bond = NULL;
1369 bundle->floodable = true;
1371 bundle->src_mirrors = 0;
1372 bundle->dst_mirrors = 0;
1373 bundle->mirror_out = 0;
1376 if (!bundle->name || strcmp(s->name, bundle->name)) {
1378 bundle->name = xstrdup(s->name);
1383 if (!bundle->lacp) {
1384 ofproto->need_revalidate = true;
1385 bundle->lacp = lacp_create();
1387 lacp_configure(bundle->lacp, s->lacp);
1389 lacp_destroy(bundle->lacp);
1390 bundle->lacp = NULL;
1393 /* Update set of ports. */
1395 for (i = 0; i < s->n_slaves; i++) {
1396 if (!bundle_add_port(bundle, s->slaves[i],
1397 s->lacp ? &s->lacp_slaves[i] : NULL,
1398 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1402 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1403 struct ofport_dpif *next_port;
1405 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1406 for (i = 0; i < s->n_slaves; i++) {
1407 if (s->slaves[i] == port->up.ofp_port) {
1412 bundle_del_port(port);
1416 assert(list_size(&bundle->ports) <= s->n_slaves);
1418 if (list_is_empty(&bundle->ports)) {
1419 bundle_destroy(bundle);
1423 /* Set VLAN tagging mode */
1424 if (s->vlan_mode != bundle->vlan_mode) {
1425 bundle->vlan_mode = s->vlan_mode;
1430 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
1431 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
1433 if (vlan != bundle->vlan) {
1434 bundle->vlan = vlan;
1438 /* Get trunked VLANs. */
1439 switch (s->vlan_mode) {
1440 case PORT_VLAN_ACCESS:
1444 case PORT_VLAN_TRUNK:
1445 trunks = (unsigned long *) s->trunks;
1448 case PORT_VLAN_NATIVE_UNTAGGED:
1449 case PORT_VLAN_NATIVE_TAGGED:
1450 if (vlan != 0 && (!s->trunks
1451 || !bitmap_is_set(s->trunks, vlan)
1452 || bitmap_is_set(s->trunks, 0))) {
1453 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
1455 trunks = bitmap_clone(s->trunks, 4096);
1457 trunks = bitmap_allocate1(4096);
1459 bitmap_set1(trunks, vlan);
1460 bitmap_set0(trunks, 0);
1462 trunks = (unsigned long *) s->trunks;
1469 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1470 free(bundle->trunks);
1471 if (trunks == s->trunks) {
1472 bundle->trunks = vlan_bitmap_clone(trunks);
1474 bundle->trunks = trunks;
1479 if (trunks != s->trunks) {
1484 if (!list_is_short(&bundle->ports)) {
1485 bundle->ofproto->has_bonded_bundles = true;
1487 if (bond_reconfigure(bundle->bond, s->bond)) {
1488 ofproto->need_revalidate = true;
1491 bundle->bond = bond_create(s->bond);
1492 ofproto->need_revalidate = true;
1495 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1496 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1500 bond_destroy(bundle->bond);
1501 bundle->bond = NULL;
1504 /* If we changed something that would affect MAC learning, un-learn
1505 * everything on this port and force flow revalidation. */
1507 bundle_flush_macs(bundle);
1514 bundle_remove(struct ofport *port_)
1516 struct ofport_dpif *port = ofport_dpif_cast(port_);
1517 struct ofbundle *bundle = port->bundle;
1520 bundle_del_port(port);
1521 if (list_is_empty(&bundle->ports)) {
1522 bundle_destroy(bundle);
1523 } else if (list_is_short(&bundle->ports)) {
1524 bond_destroy(bundle->bond);
1525 bundle->bond = NULL;
1531 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
1533 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1534 struct ofport_dpif *port = port_;
1535 uint8_t ea[ETH_ADDR_LEN];
1538 error = netdev_get_etheraddr(port->up.netdev, ea);
1540 struct ofpbuf packet;
1543 ofpbuf_init(&packet, 0);
1544 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1546 memcpy(packet_pdu, pdu, pdu_size);
1548 error = netdev_send(port->up.netdev, &packet);
1550 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1551 "(%s)", port->bundle->name,
1552 netdev_get_name(port->up.netdev), strerror(error));
1554 ofpbuf_uninit(&packet);
1556 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1557 "%s (%s)", port->bundle->name,
1558 netdev_get_name(port->up.netdev), strerror(error));
1563 bundle_send_learning_packets(struct ofbundle *bundle)
1565 struct ofproto_dpif *ofproto = bundle->ofproto;
1566 int error, n_packets, n_errors;
1567 struct mac_entry *e;
1569 error = n_packets = n_errors = 0;
1570 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1571 if (e->port.p != bundle) {
1572 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1582 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1583 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1584 "packets, last error was: %s",
1585 bundle->name, n_errors, n_packets, strerror(error));
1587 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1588 bundle->name, n_packets);
1593 bundle_run(struct ofbundle *bundle)
1596 lacp_run(bundle->lacp, send_pdu_cb);
1599 struct ofport_dpif *port;
1601 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1602 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1605 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1606 lacp_negotiated(bundle->lacp));
1607 if (bond_should_send_learning_packets(bundle->bond)) {
1608 bundle_send_learning_packets(bundle);
1614 bundle_wait(struct ofbundle *bundle)
1617 lacp_wait(bundle->lacp);
1620 bond_wait(bundle->bond);
1627 mirror_scan(struct ofproto_dpif *ofproto)
1631 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1632 if (!ofproto->mirrors[idx]) {
1639 static struct ofmirror *
1640 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1644 for (i = 0; i < MAX_MIRRORS; i++) {
1645 struct ofmirror *mirror = ofproto->mirrors[i];
1646 if (mirror && mirror->aux == aux) {
1655 mirror_set(struct ofproto *ofproto_, void *aux,
1656 const struct ofproto_mirror_settings *s)
1658 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1659 mirror_mask_t mirror_bit;
1660 struct ofbundle *bundle;
1661 struct ofmirror *mirror;
1662 struct ofbundle *out;
1663 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1664 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1667 mirror = mirror_lookup(ofproto, aux);
1669 mirror_destroy(mirror);
1675 idx = mirror_scan(ofproto);
1677 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1679 ofproto->up.name, MAX_MIRRORS, s->name);
1683 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1684 mirror->ofproto = ofproto;
1687 mirror->out_vlan = -1;
1688 mirror->name = NULL;
1691 if (!mirror->name || strcmp(s->name, mirror->name)) {
1693 mirror->name = xstrdup(s->name);
1696 /* Get the new configuration. */
1697 if (s->out_bundle) {
1698 out = bundle_lookup(ofproto, s->out_bundle);
1700 mirror_destroy(mirror);
1706 out_vlan = s->out_vlan;
1708 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1709 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1711 /* If the configuration has not changed, do nothing. */
1712 if (hmapx_equals(&srcs, &mirror->srcs)
1713 && hmapx_equals(&dsts, &mirror->dsts)
1714 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1715 && mirror->out == out
1716 && mirror->out_vlan == out_vlan)
1718 hmapx_destroy(&srcs);
1719 hmapx_destroy(&dsts);
1723 hmapx_swap(&srcs, &mirror->srcs);
1724 hmapx_destroy(&srcs);
1726 hmapx_swap(&dsts, &mirror->dsts);
1727 hmapx_destroy(&dsts);
1729 free(mirror->vlans);
1730 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1733 mirror->out_vlan = out_vlan;
1735 /* Update bundles. */
1736 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1737 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1738 if (hmapx_contains(&mirror->srcs, bundle)) {
1739 bundle->src_mirrors |= mirror_bit;
1741 bundle->src_mirrors &= ~mirror_bit;
1744 if (hmapx_contains(&mirror->dsts, bundle)) {
1745 bundle->dst_mirrors |= mirror_bit;
1747 bundle->dst_mirrors &= ~mirror_bit;
1750 if (mirror->out == bundle) {
1751 bundle->mirror_out |= mirror_bit;
1753 bundle->mirror_out &= ~mirror_bit;
1757 ofproto->need_revalidate = true;
1758 mac_learning_flush(ofproto->ml);
1764 mirror_destroy(struct ofmirror *mirror)
1766 struct ofproto_dpif *ofproto;
1767 mirror_mask_t mirror_bit;
1768 struct ofbundle *bundle;
1774 ofproto = mirror->ofproto;
1775 ofproto->need_revalidate = true;
1776 mac_learning_flush(ofproto->ml);
1778 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1779 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1780 bundle->src_mirrors &= ~mirror_bit;
1781 bundle->dst_mirrors &= ~mirror_bit;
1782 bundle->mirror_out &= ~mirror_bit;
1785 hmapx_destroy(&mirror->srcs);
1786 hmapx_destroy(&mirror->dsts);
1787 free(mirror->vlans);
1789 ofproto->mirrors[mirror->idx] = NULL;
1795 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1797 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1798 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1799 ofproto->need_revalidate = true;
1800 mac_learning_flush(ofproto->ml);
1806 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
1808 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1809 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1810 return bundle && bundle->mirror_out != 0;
1814 forward_bpdu_changed(struct ofproto *ofproto_)
1816 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1817 /* Revalidate cached flows whenever forward_bpdu option changes. */
1818 ofproto->need_revalidate = true;
1823 static struct ofport_dpif *
1824 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1826 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1827 return ofport ? ofport_dpif_cast(ofport) : NULL;
1830 static struct ofport_dpif *
1831 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1833 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1837 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1838 struct dpif_port *dpif_port)
1840 ofproto_port->name = dpif_port->name;
1841 ofproto_port->type = dpif_port->type;
1842 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1846 port_run(struct ofport_dpif *ofport)
1848 bool enable = netdev_get_carrier(ofport->up.netdev);
1851 cfm_run(ofport->cfm);
1853 if (cfm_should_send_ccm(ofport->cfm)) {
1854 struct ofpbuf packet;
1856 ofpbuf_init(&packet, 0);
1857 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1858 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1859 ofport->odp_port, &packet);
1860 ofpbuf_uninit(&packet);
1863 enable = enable && !cfm_get_fault(ofport->cfm)
1864 && cfm_get_opup(ofport->cfm);
1867 if (ofport->bundle) {
1868 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1871 if (ofport->may_enable != enable) {
1872 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1874 if (ofproto->has_bundle_action) {
1875 ofproto->need_revalidate = true;
1879 ofport->may_enable = enable;
1883 port_wait(struct ofport_dpif *ofport)
1886 cfm_wait(ofport->cfm);
1891 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1892 struct ofproto_port *ofproto_port)
1894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1895 struct dpif_port dpif_port;
1898 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1900 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1906 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1908 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1912 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1914 *ofp_portp = odp_port_to_ofp_port(odp_port);
1920 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1922 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1925 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1927 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1929 /* The caller is going to close ofport->up.netdev. If this is a
1930 * bonded port, then the bond is using that netdev, so remove it
1931 * from the bond. The client will need to reconfigure everything
1932 * after deleting ports, so then the slave will get re-added. */
1933 bundle_remove(&ofport->up);
1939 struct port_dump_state {
1940 struct dpif_port_dump dump;
1945 port_dump_start(const struct ofproto *ofproto_, void **statep)
1947 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1948 struct port_dump_state *state;
1950 *statep = state = xmalloc(sizeof *state);
1951 dpif_port_dump_start(&state->dump, ofproto->dpif);
1952 state->done = false;
1957 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1958 struct ofproto_port *port)
1960 struct port_dump_state *state = state_;
1961 struct dpif_port dpif_port;
1963 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1964 ofproto_port_from_dpif_port(port, &dpif_port);
1967 int error = dpif_port_dump_done(&state->dump);
1969 return error ? error : EOF;
1974 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1976 struct port_dump_state *state = state_;
1979 dpif_port_dump_done(&state->dump);
1986 port_poll(const struct ofproto *ofproto_, char **devnamep)
1988 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1989 return dpif_port_poll(ofproto->dpif, devnamep);
1993 port_poll_wait(const struct ofproto *ofproto_)
1995 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1996 dpif_port_poll_wait(ofproto->dpif);
2000 port_is_lacp_current(const struct ofport *ofport_)
2002 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2003 return (ofport->bundle && ofport->bundle->lacp
2004 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2008 /* Upcall handling. */
2010 /* Flow miss batching.
2012 * Some dpifs implement operations faster when you hand them off in a batch.
2013 * To allow batching, "struct flow_miss" queues the dpif-related work needed
2014 * for a given flow. Each "struct flow_miss" corresponds to sending one or
2015 * more packets, plus possibly installing the flow in the dpif.
2017 * So far we only batch the operations that affect flow setup time the most.
2018 * It's possible to batch more than that, but the benefit might be minimal. */
2020 struct hmap_node hmap_node;
2022 const struct nlattr *key;
2024 struct list packets;
2027 struct flow_miss_op {
2028 union dpif_op dpif_op;
2029 struct facet *facet;
2032 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
2033 * OpenFlow controller as necessary according to their individual
2036 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
2037 * ownership is transferred to this function. */
2039 send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2040 const struct flow *flow, bool clone)
2042 struct ofputil_packet_in pin;
2044 pin.packet = packet;
2045 pin.in_port = flow->in_port;
2046 pin.reason = OFPR_NO_MATCH;
2047 pin.buffer_id = 0; /* not yet known */
2048 pin.send_len = 0; /* not used for flow table misses */
2049 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2050 clone ? NULL : packet);
2053 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each
2054 * OpenFlow controller as necessary according to their individual
2057 * 'send_len' should be the number of bytes of 'packet' to send to the
2058 * controller, as specified in the action that caused the packet to be sent.
2060 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
2061 * Otherwise, ownership is transferred to this function. */
2063 send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet,
2064 uint64_t userdata, const struct flow *flow, bool clone)
2066 struct ofputil_packet_in pin;
2067 struct user_action_cookie cookie;
2069 memcpy(&cookie, &userdata, sizeof(cookie));
2071 pin.packet = packet;
2072 pin.in_port = flow->in_port;
2073 pin.reason = OFPR_ACTION;
2074 pin.buffer_id = 0; /* not yet known */
2075 pin.send_len = cookie.data;
2076 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
2077 clone ? NULL : packet);
2081 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
2082 const struct ofpbuf *packet)
2084 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
2090 if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
2092 cfm_process_heartbeat(ofport->cfm, packet);
2095 } else if (ofport->bundle && ofport->bundle->lacp
2096 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2098 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
2101 } else if (ofproto->stp && stp_should_process_flow(flow)) {
2103 stp_process_packet(ofport, packet);
2110 static struct flow_miss *
2111 flow_miss_create(struct hmap *todo, const struct flow *flow,
2112 const struct nlattr *key, size_t key_len)
2114 uint32_t hash = flow_hash(flow, 0);
2115 struct flow_miss *miss;
2117 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
2118 if (flow_equal(&miss->flow, flow)) {
2123 miss = xmalloc(sizeof *miss);
2124 hmap_insert(todo, &miss->hmap_node, hash);
2127 miss->key_len = key_len;
2128 list_init(&miss->packets);
2133 handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss,
2134 struct flow_miss_op *ops, size_t *n_ops)
2136 const struct flow *flow = &miss->flow;
2137 struct ofpbuf *packet, *next_packet;
2138 struct facet *facet;
2140 facet = facet_lookup_valid(ofproto, flow);
2142 struct rule_dpif *rule;
2144 rule = rule_dpif_lookup(ofproto, flow, 0);
2146 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
2147 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
2149 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
2150 COVERAGE_INC(ofproto_dpif_no_packet_in);
2151 /* XXX install 'drop' flow entry */
2155 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
2159 LIST_FOR_EACH_SAFE (packet, next_packet, list_node,
2161 list_remove(&packet->list_node);
2162 send_packet_in_miss(ofproto, packet, flow, false);
2168 facet = facet_create(rule, flow);
2171 LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) {
2172 list_remove(&packet->list_node);
2173 ofproto->n_matches++;
2175 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
2177 * Extra-special case for fail-open mode.
2179 * We are in fail-open mode and the packet matched the fail-open
2180 * rule, but we are connected to a controller too. We should send
2181 * the packet up to the controller in the hope that it will try to
2182 * set up a flow and thereby allow us to exit fail-open.
2184 * See the top-level comment in fail-open.c for more information.
2186 send_packet_in_miss(ofproto, packet, flow, true);
2189 if (!facet->may_install) {
2190 facet_make_actions(ofproto, facet, packet);
2192 if (!execute_controller_action(ofproto, &facet->flow,
2193 facet->actions, facet->actions_len,
2195 struct flow_miss_op *op = &ops[(*n_ops)++];
2196 struct dpif_execute *execute = &op->dpif_op.execute;
2199 execute->type = DPIF_OP_EXECUTE;
2200 execute->key = miss->key;
2201 execute->key_len = miss->key_len;
2203 = (facet->may_install
2205 : xmemdup(facet->actions, facet->actions_len));
2206 execute->actions_len = facet->actions_len;
2207 execute->packet = packet;
2211 if (facet->may_install) {
2212 struct flow_miss_op *op = &ops[(*n_ops)++];
2213 struct dpif_flow_put *put = &op->dpif_op.flow_put;
2216 put->type = DPIF_OP_FLOW_PUT;
2217 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2218 put->key = miss->key;
2219 put->key_len = miss->key_len;
2220 put->actions = facet->actions;
2221 put->actions_len = facet->actions_len;
2227 handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls,
2230 struct dpif_upcall *upcall;
2231 struct flow_miss *miss, *next_miss;
2232 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
2233 union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
2242 /* Construct the to-do list.
2244 * This just amounts to extracting the flow from each packet and sticking
2245 * the packets that have the same flow in the same "flow_miss" structure so
2246 * that we can process them together. */
2248 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
2249 struct flow_miss *miss;
2252 /* Obtain in_port and tun_id, at least, then set 'flow''s header
2254 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2255 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
2257 /* Handle 802.1ag, LACP, and STP specially. */
2258 if (process_special(ofproto, &flow, upcall->packet)) {
2259 ofpbuf_delete(upcall->packet);
2260 ofproto->n_matches++;
2264 /* Add other packets to a to-do list. */
2265 miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len);
2266 list_push_back(&miss->packets, &upcall->packet->list_node);
2269 /* Process each element in the to-do list, constructing the set of
2270 * operations to batch. */
2272 HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) {
2273 handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops);
2274 ofpbuf_list_delete(&miss->packets);
2275 hmap_remove(&todo, &miss->hmap_node);
2278 assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
2279 hmap_destroy(&todo);
2281 /* Execute batch. */
2282 for (i = 0; i < n_ops; i++) {
2283 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
2285 dpif_operate(ofproto->dpif, dpif_ops, n_ops);
2287 /* Free memory and update facets. */
2288 for (i = 0; i < n_ops; i++) {
2289 struct flow_miss_op *op = &flow_miss_ops[i];
2290 struct dpif_execute *execute;
2291 struct dpif_flow_put *put;
2293 switch (op->dpif_op.type) {
2294 case DPIF_OP_EXECUTE:
2295 execute = &op->dpif_op.execute;
2296 if (op->facet->actions != execute->actions) {
2297 free((struct nlattr *) execute->actions);
2299 ofpbuf_delete((struct ofpbuf *) execute->packet);
2302 case DPIF_OP_FLOW_PUT:
2303 put = &op->dpif_op.flow_put;
2305 op->facet->installed = true;
2313 handle_userspace_upcall(struct ofproto_dpif *ofproto,
2314 struct dpif_upcall *upcall)
2317 struct user_action_cookie cookie;
2319 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
2321 if (cookie.type == USER_ACTION_COOKIE_SFLOW) {
2322 if (ofproto->sflow) {
2323 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2324 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie);
2326 ofpbuf_delete(upcall->packet);
2328 } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) {
2329 COVERAGE_INC(ofproto_dpif_ctlr_action);
2330 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
2331 send_packet_in_action(ofproto, upcall->packet, upcall->userdata,
2334 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
2339 handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch)
2341 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
2345 assert (max_batch <= FLOW_MISS_MAX_BATCH);
2348 for (i = 0; i < max_batch; i++) {
2349 struct dpif_upcall *upcall = &misses[n_misses];
2352 error = dpif_recv(ofproto->dpif, upcall);
2354 if (error == ENODEV && n_misses == 0) {
2360 switch (upcall->type) {
2361 case DPIF_UC_ACTION:
2362 handle_userspace_upcall(ofproto, upcall);
2366 /* Handle it later. */
2370 case DPIF_N_UC_TYPES:
2372 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32,
2378 handle_miss_upcalls(ofproto, misses, n_misses);
2383 /* Flow expiration. */
2385 static int facet_max_idle(const struct ofproto_dpif *);
2386 static void update_stats(struct ofproto_dpif *);
2387 static void rule_expire(struct rule_dpif *);
2388 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
2390 /* This function is called periodically by run(). Its job is to collect
2391 * updates for the flows that have been installed into the datapath, most
2392 * importantly when they last were used, and then use that information to
2393 * expire flows that have not been used recently.
2395 * Returns the number of milliseconds after which it should be called again. */
2397 expire(struct ofproto_dpif *ofproto)
2399 struct rule_dpif *rule, *next_rule;
2400 struct classifier *table;
2403 /* Update stats for each flow in the datapath. */
2404 update_stats(ofproto);
2406 /* Expire facets that have been idle too long. */
2407 dp_max_idle = facet_max_idle(ofproto);
2408 expire_facets(ofproto, dp_max_idle);
2410 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
2411 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
2412 struct cls_cursor cursor;
2414 cls_cursor_init(&cursor, table, NULL);
2415 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
2420 /* All outstanding data in existing flows has been accounted, so it's a
2421 * good time to do bond rebalancing. */
2422 if (ofproto->has_bonded_bundles) {
2423 struct ofbundle *bundle;
2425 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2427 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
2432 return MIN(dp_max_idle, 1000);
2435 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
2437 * This function also pushes statistics updates to rules which each facet
2438 * resubmits into. Generally these statistics will be accurate. However, if a
2439 * facet changes the rule it resubmits into at some time in between
2440 * update_stats() runs, it is possible that statistics accrued to the
2441 * old rule will be incorrectly attributed to the new rule. This could be
2442 * avoided by calling update_stats() whenever rules are created or
2443 * deleted. However, the performance impact of making so many calls to the
2444 * datapath do not justify the benefit of having perfectly accurate statistics.
2447 update_stats(struct ofproto_dpif *p)
2449 const struct dpif_flow_stats *stats;
2450 struct dpif_flow_dump dump;
2451 const struct nlattr *key;
2454 dpif_flow_dump_start(&dump, p->dpif);
2455 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
2456 struct facet *facet;
2459 if (odp_flow_key_to_flow(key, key_len, &flow)) {
2463 odp_flow_key_format(key, key_len, &s);
2464 VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s",
2470 facet = facet_find(p, &flow);
2472 if (facet && facet->installed) {
2474 if (stats->n_packets >= facet->dp_packet_count) {
2475 uint64_t extra = stats->n_packets - facet->dp_packet_count;
2476 facet->packet_count += extra;
2478 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
2481 if (stats->n_bytes >= facet->dp_byte_count) {
2482 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
2484 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
2487 facet->dp_packet_count = stats->n_packets;
2488 facet->dp_byte_count = stats->n_bytes;
2490 facet_update_time(p, facet, stats->used);
2491 facet_account(p, facet);
2492 facet_push_stats(facet);
2494 /* There's a flow in the datapath that we know nothing about.
2496 COVERAGE_INC(facet_unexpected);
2497 dpif_flow_del(p->dpif, key, key_len, NULL);
2500 dpif_flow_dump_done(&dump);
2503 /* Calculates and returns the number of milliseconds of idle time after which
2504 * facets should expire from the datapath and we should fold their statistics
2505 * into their parent rules in userspace. */
2507 facet_max_idle(const struct ofproto_dpif *ofproto)
2510 * Idle time histogram.
2512 * Most of the time a switch has a relatively small number of facets. When
2513 * this is the case we might as well keep statistics for all of them in
2514 * userspace and to cache them in the kernel datapath for performance as
2517 * As the number of facets increases, the memory required to maintain
2518 * statistics about them in userspace and in the kernel becomes
2519 * significant. However, with a large number of facets it is likely that
2520 * only a few of them are "heavy hitters" that consume a large amount of
2521 * bandwidth. At this point, only heavy hitters are worth caching in the
2522 * kernel and maintaining in userspaces; other facets we can discard.
2524 * The technique used to compute the idle time is to build a histogram with
2525 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
2526 * that is installed in the kernel gets dropped in the appropriate bucket.
2527 * After the histogram has been built, we compute the cutoff so that only
2528 * the most-recently-used 1% of facets (but at least
2529 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
2530 * the most-recently-used bucket of facets is kept, so actually an
2531 * arbitrary number of facets can be kept in any given expiration run
2532 * (though the next run will delete most of those unless they receive
2535 * This requires a second pass through the facets, in addition to the pass
2536 * made by update_stats(), because the former function never looks
2537 * at uninstallable facets.
2539 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
2540 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
2541 int buckets[N_BUCKETS] = { 0 };
2542 int total, subtotal, bucket;
2543 struct facet *facet;
2547 total = hmap_count(&ofproto->facets);
2548 if (total <= ofproto->up.flow_eviction_threshold) {
2549 return N_BUCKETS * BUCKET_WIDTH;
2552 /* Build histogram. */
2554 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
2555 long long int idle = now - facet->used;
2556 int bucket = (idle <= 0 ? 0
2557 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
2558 : (unsigned int) idle / BUCKET_WIDTH);
2562 /* Find the first bucket whose flows should be expired. */
2563 subtotal = bucket = 0;
2565 subtotal += buckets[bucket++];
2566 } while (bucket < N_BUCKETS &&
2567 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
2569 if (VLOG_IS_DBG_ENABLED()) {
2573 ds_put_cstr(&s, "keep");
2574 for (i = 0; i < N_BUCKETS; i++) {
2576 ds_put_cstr(&s, ", drop");
2579 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
2582 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
2586 return bucket * BUCKET_WIDTH;
2590 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
2592 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
2593 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
2594 struct ofexpired expired;
2596 if (facet->installed) {
2597 struct dpif_flow_stats stats;
2599 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
2601 facet_update_stats(ofproto, facet, &stats);
2604 expired.flow = facet->flow;
2605 expired.packet_count = facet->packet_count;
2606 expired.byte_count = facet->byte_count;
2607 expired.used = facet->used;
2608 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2613 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
2615 long long int cutoff = time_msec() - dp_max_idle;
2616 struct facet *facet, *next_facet;
2618 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
2619 facet_active_timeout(ofproto, facet);
2620 if (facet->used < cutoff) {
2621 facet_remove(ofproto, facet);
2626 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2627 * then delete it entirely. */
2629 rule_expire(struct rule_dpif *rule)
2631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2632 struct facet *facet, *next_facet;
2636 /* Has 'rule' expired? */
2638 if (rule->up.hard_timeout
2639 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
2640 reason = OFPRR_HARD_TIMEOUT;
2641 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
2642 && now > rule->used + rule->up.idle_timeout * 1000) {
2643 reason = OFPRR_IDLE_TIMEOUT;
2648 COVERAGE_INC(ofproto_dpif_expired);
2650 /* Update stats. (This is a no-op if the rule expired due to an idle
2651 * timeout, because that only happens when the rule has no facets left.) */
2652 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2653 facet_remove(ofproto, facet);
2656 /* Get rid of the rule. */
2657 ofproto_rule_expire(&rule->up, reason);
2662 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
2664 * The caller must already have determined that no facet with an identical
2665 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2666 * the ofproto's classifier table.
2668 * The facet will initially have no ODP actions. The caller should fix that
2669 * by calling facet_make_actions(). */
2670 static struct facet *
2671 facet_create(struct rule_dpif *rule, const struct flow *flow)
2673 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2674 struct facet *facet;
2676 facet = xzalloc(sizeof *facet);
2677 facet->used = time_msec();
2678 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2679 list_push_back(&rule->facets, &facet->list_node);
2681 facet->flow = *flow;
2682 netflow_flow_init(&facet->nf_flow);
2683 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2689 facet_free(struct facet *facet)
2691 free(facet->actions);
2696 execute_controller_action(struct ofproto_dpif *ofproto,
2697 const struct flow *flow,
2698 const struct nlattr *odp_actions, size_t actions_len,
2699 struct ofpbuf *packet)
2702 && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE
2703 && NLA_ALIGN(odp_actions->nla_len) == actions_len) {
2704 /* As an optimization, avoid a round-trip from userspace to kernel to
2705 * userspace. This also avoids possibly filling up kernel packet
2706 * buffers along the way.
2708 * This optimization will not accidentally catch sFlow
2709 * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated
2710 * inside OVS_ACTION_ATTR_SAMPLE. */
2711 const struct nlattr *nla;
2713 nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA);
2714 send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow,
2722 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2723 * 'packet', which arrived on 'in_port'.
2725 * Takes ownership of 'packet'. */
2727 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2728 const struct nlattr *odp_actions, size_t actions_len,
2729 struct ofpbuf *packet)
2731 struct odputil_keybuf keybuf;
2735 if (execute_controller_action(ofproto, flow, odp_actions, actions_len,
2740 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2741 odp_flow_key_from_flow(&key, flow);
2743 error = dpif_execute(ofproto->dpif, key.data, key.size,
2744 odp_actions, actions_len, packet);
2746 ofpbuf_delete(packet);
2750 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2751 * statistics appropriately. 'packet' must have at least sizeof(struct
2752 * ofp_packet_in) bytes of headroom.
2754 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2755 * applying flow_extract() to 'packet' would yield the same flow as
2758 * 'facet' must have accurately composed datapath actions; that is, it must
2759 * not be in need of revalidation.
2761 * Takes ownership of 'packet'. */
2763 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2764 struct ofpbuf *packet)
2766 struct dpif_flow_stats stats;
2768 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2770 dpif_flow_stats_extract(&facet->flow, packet, &stats);
2771 stats.used = time_msec();
2772 if (execute_odp_actions(ofproto, &facet->flow,
2773 facet->actions, facet->actions_len, packet)) {
2774 facet_update_stats(ofproto, facet, &stats);
2778 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2780 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2781 * rule's statistics, via facet_uninstall().
2783 * - Removes 'facet' from its rule and from ofproto->facets.
2786 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2788 facet_uninstall(ofproto, facet);
2789 facet_flush_stats(ofproto, facet);
2790 hmap_remove(&ofproto->facets, &facet->hmap_node);
2791 list_remove(&facet->list_node);
2795 /* Composes the datapath actions for 'facet' based on its rule's actions. */
2797 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2798 const struct ofpbuf *packet)
2800 const struct rule_dpif *rule = facet->rule;
2801 struct ofpbuf *odp_actions;
2802 struct action_xlate_ctx ctx;
2804 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2805 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2806 facet->tags = ctx.tags;
2807 facet->may_install = ctx.may_set_up_flow;
2808 facet->has_learn = ctx.has_learn;
2809 facet->has_normal = ctx.has_normal;
2810 facet->nf_flow.output_iface = ctx.nf_output_iface;
2812 if (facet->actions_len != odp_actions->size
2813 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2814 free(facet->actions);
2815 facet->actions_len = odp_actions->size;
2816 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2819 ofpbuf_delete(odp_actions);
2822 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2823 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2824 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2825 * since 'facet' was last updated.
2827 * Returns 0 if successful, otherwise a positive errno value.*/
2829 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2830 const struct nlattr *actions, size_t actions_len,
2831 struct dpif_flow_stats *stats)
2833 struct odputil_keybuf keybuf;
2834 enum dpif_flow_put_flags flags;
2838 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2840 flags |= DPIF_FP_ZERO_STATS;
2843 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2844 odp_flow_key_from_flow(&key, &facet->flow);
2846 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2847 actions, actions_len, stats);
2850 facet_reset_dp_stats(facet, stats);
2856 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2857 * 'zero_stats' is true, clears any existing statistics from the datapath for
2860 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2862 struct dpif_flow_stats stats;
2864 if (facet->may_install
2865 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2866 zero_stats ? &stats : NULL)) {
2867 facet->installed = true;
2872 facet_account(struct ofproto_dpif *ofproto, struct facet *facet)
2875 const struct nlattr *a;
2879 if (facet->byte_count <= facet->accounted_bytes) {
2882 n_bytes = facet->byte_count - facet->accounted_bytes;
2883 facet->accounted_bytes = facet->byte_count;
2885 /* Feed information from the active flows back into the learning table to
2886 * ensure that table is always in sync with what is actually flowing
2887 * through the datapath. */
2888 if (facet->has_learn || facet->has_normal) {
2889 struct action_xlate_ctx ctx;
2891 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2892 ctx.may_learn = true;
2893 ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions,
2894 facet->rule->up.n_actions));
2897 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
2901 /* This loop feeds byte counters to bond_account() for rebalancing to use
2902 * as a basis. We also need to track the actual VLAN on which the packet
2903 * is going to be sent to ensure that it matches the one passed to
2904 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2906 vlan_tci = facet->flow.vlan_tci;
2907 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2908 struct ofport_dpif *port;
2910 switch (nl_attr_type(a)) {
2911 const struct nlattr *nested;
2912 case OVS_ACTION_ATTR_OUTPUT:
2913 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2914 if (port && port->bundle && port->bundle->bond) {
2915 bond_account(port->bundle->bond, &facet->flow,
2916 vlan_tci_to_vid(vlan_tci), n_bytes);
2920 case OVS_ACTION_ATTR_POP:
2921 if (nl_attr_get_u16(a) == OVS_KEY_ATTR_8021Q) {
2922 vlan_tci = htons(0);
2926 case OVS_ACTION_ATTR_PUSH:
2927 nested = nl_attr_get(a);
2928 if (nl_attr_type(nested) == OVS_KEY_ATTR_8021Q) {
2929 const struct ovs_key_8021q *q_key;
2931 q_key = nl_attr_get_unspec(nested, sizeof(*q_key));
2932 vlan_tci = q_key->q_tci;
2939 /* If 'rule' is installed in the datapath, uninstalls it. */
2941 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2943 if (facet->installed) {
2944 struct odputil_keybuf keybuf;
2945 struct dpif_flow_stats stats;
2949 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2950 odp_flow_key_from_flow(&key, &facet->flow);
2952 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2953 facet_reset_dp_stats(facet, &stats);
2955 facet_update_stats(p, facet, &stats);
2957 facet->installed = false;
2959 assert(facet->dp_packet_count == 0);
2960 assert(facet->dp_byte_count == 0);
2964 /* Returns true if the only action for 'facet' is to send to the controller.
2965 * (We don't report NetFlow expiration messages for such facets because they
2966 * are just part of the control logic for the network, not real traffic). */
2968 facet_is_controller_flow(struct facet *facet)
2971 && facet->rule->up.n_actions == 1
2972 && action_outputs_to_port(&facet->rule->up.actions[0],
2973 htons(OFPP_CONTROLLER)));
2976 /* Resets 'facet''s datapath statistics counters. This should be called when
2977 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2978 * it should contain the statistics returned by dpif when 'facet' was reset in
2979 * the datapath. 'stats' will be modified to only included statistics new
2980 * since 'facet' was last updated. */
2982 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2984 if (stats && facet->dp_packet_count <= stats->n_packets
2985 && facet->dp_byte_count <= stats->n_bytes) {
2986 stats->n_packets -= facet->dp_packet_count;
2987 stats->n_bytes -= facet->dp_byte_count;
2990 facet->dp_packet_count = 0;
2991 facet->dp_byte_count = 0;
2994 /* Folds all of 'facet''s statistics into its rule. Also updates the
2995 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2996 * 'facet''s statistics in the datapath should have been zeroed and folded into
2997 * its packet and byte counts before this function is called. */
2999 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
3001 assert(!facet->dp_byte_count);
3002 assert(!facet->dp_packet_count);
3004 facet_push_stats(facet);
3005 facet_account(ofproto, facet);
3007 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
3008 struct ofexpired expired;
3009 expired.flow = facet->flow;
3010 expired.packet_count = facet->packet_count;
3011 expired.byte_count = facet->byte_count;
3012 expired.used = facet->used;
3013 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
3016 facet->rule->packet_count += facet->packet_count;
3017 facet->rule->byte_count += facet->byte_count;
3019 /* Reset counters to prevent double counting if 'facet' ever gets
3021 facet_reset_counters(facet);
3023 netflow_flow_clear(&facet->nf_flow);
3026 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3027 * Returns it if found, otherwise a null pointer.
3029 * The returned facet might need revalidation; use facet_lookup_valid()
3030 * instead if that is important. */
3031 static struct facet *
3032 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
3034 struct facet *facet;
3036 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
3038 if (flow_equal(flow, &facet->flow)) {
3046 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
3047 * Returns it if found, otherwise a null pointer.
3049 * The returned facet is guaranteed to be valid. */
3050 static struct facet *
3051 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
3053 struct facet *facet = facet_find(ofproto, flow);
3055 /* The facet we found might not be valid, since we could be in need of
3056 * revalidation. If it is not valid, don't return it. */
3058 && (ofproto->need_revalidate
3059 || tag_set_intersects(&ofproto->revalidate_set, facet->tags))
3060 && !facet_revalidate(ofproto, facet)) {
3061 COVERAGE_INC(facet_invalidated);
3068 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
3070 * - If the rule found is different from 'facet''s current rule, moves
3071 * 'facet' to the new rule and recompiles its actions.
3073 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
3074 * where it is and recompiles its actions anyway.
3076 * - If there is none, destroys 'facet'.
3078 * Returns true if 'facet' still exists, false if it has been destroyed. */
3080 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
3082 struct action_xlate_ctx ctx;
3083 struct ofpbuf *odp_actions;
3084 struct rule_dpif *new_rule;
3085 bool actions_changed;
3087 COVERAGE_INC(facet_revalidate);
3089 /* Determine the new rule. */
3090 new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0);
3092 /* No new rule, so delete the facet. */
3093 facet_remove(ofproto, facet);
3097 /* Calculate new datapath actions.
3099 * We do not modify any 'facet' state yet, because we might need to, e.g.,
3100 * emit a NetFlow expiration and, if so, we need to have the old state
3101 * around to properly compose it. */
3102 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
3103 odp_actions = xlate_actions(&ctx,
3104 new_rule->up.actions, new_rule->up.n_actions);
3105 actions_changed = (facet->actions_len != odp_actions->size
3106 || memcmp(facet->actions, odp_actions->data,
3107 facet->actions_len));
3109 /* If the datapath actions changed or the installability changed,
3110 * then we need to talk to the datapath. */
3111 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
3112 if (ctx.may_set_up_flow) {
3113 struct dpif_flow_stats stats;
3115 facet_put__(ofproto, facet,
3116 odp_actions->data, odp_actions->size, &stats);
3117 facet_update_stats(ofproto, facet, &stats);
3119 facet_uninstall(ofproto, facet);
3122 /* The datapath flow is gone or has zeroed stats, so push stats out of
3123 * 'facet' into 'rule'. */
3124 facet_flush_stats(ofproto, facet);
3127 /* Update 'facet' now that we've taken care of all the old state. */
3128 facet->tags = ctx.tags;
3129 facet->nf_flow.output_iface = ctx.nf_output_iface;
3130 facet->may_install = ctx.may_set_up_flow;
3131 facet->has_learn = ctx.has_learn;
3132 facet->has_normal = ctx.has_normal;
3133 if (actions_changed) {
3134 free(facet->actions);
3135 facet->actions_len = odp_actions->size;
3136 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
3138 if (facet->rule != new_rule) {
3139 COVERAGE_INC(facet_changed_rule);
3140 list_remove(&facet->list_node);
3141 list_push_back(&new_rule->facets, &facet->list_node);
3142 facet->rule = new_rule;
3143 facet->used = new_rule->up.created;
3144 facet->rs_used = facet->used;
3147 ofpbuf_delete(odp_actions);
3152 /* Updates 'facet''s used time. Caller is responsible for calling
3153 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3155 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
3158 if (used > facet->used) {
3160 if (used > facet->rule->used) {
3161 facet->rule->used = used;
3163 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3167 /* Folds the statistics from 'stats' into the counters in 'facet'.
3169 * Because of the meaning of a facet's counters, it only makes sense to do this
3170 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3171 * packet that was sent by hand or if it represents statistics that have been
3172 * cleared out of the datapath. */
3174 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
3175 const struct dpif_flow_stats *stats)
3177 if (stats->n_packets || stats->used > facet->used) {
3178 facet_update_time(ofproto, facet, stats->used);
3179 facet->packet_count += stats->n_packets;
3180 facet->byte_count += stats->n_bytes;
3181 facet_push_stats(facet);
3182 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3187 facet_reset_counters(struct facet *facet)
3189 facet->packet_count = 0;
3190 facet->byte_count = 0;
3191 facet->rs_packet_count = 0;
3192 facet->rs_byte_count = 0;
3193 facet->accounted_bytes = 0;
3197 facet_push_stats(struct facet *facet)
3199 uint64_t rs_packets, rs_bytes;
3201 assert(facet->packet_count >= facet->rs_packet_count);
3202 assert(facet->byte_count >= facet->rs_byte_count);
3203 assert(facet->used >= facet->rs_used);
3205 rs_packets = facet->packet_count - facet->rs_packet_count;
3206 rs_bytes = facet->byte_count - facet->rs_byte_count;
3208 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3209 facet->rs_packet_count = facet->packet_count;
3210 facet->rs_byte_count = facet->byte_count;
3211 facet->rs_used = facet->used;
3213 flow_push_stats(facet->rule, &facet->flow,
3214 rs_packets, rs_bytes, facet->used);
3218 struct ofproto_push {
3219 struct action_xlate_ctx ctx;
3226 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3228 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3231 rule->packet_count += push->packets;
3232 rule->byte_count += push->bytes;
3233 rule->used = MAX(push->used, rule->used);
3237 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3238 * 'rule''s actions. */
3240 flow_push_stats(const struct rule_dpif *rule,
3241 struct flow *flow, uint64_t packets, uint64_t bytes,
3244 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3245 struct ofproto_push push;
3247 push.packets = packets;
3251 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
3252 push.ctx.resubmit_hook = push_resubmit;
3253 ofpbuf_delete(xlate_actions(&push.ctx,
3254 rule->up.actions, rule->up.n_actions));
3259 static struct rule_dpif *
3260 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
3263 struct cls_rule *cls_rule;
3264 struct classifier *cls;
3266 if (table_id >= N_TABLES) {
3270 cls = &ofproto->up.tables[table_id];
3271 if (flow->tos_frag & FLOW_FRAG_ANY
3272 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3273 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
3274 * are unavailable. */
3275 struct flow ofpc_normal_flow = *flow;
3276 ofpc_normal_flow.tp_src = htons(0);
3277 ofpc_normal_flow.tp_dst = htons(0);
3278 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
3280 cls_rule = classifier_lookup(cls, flow);
3282 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
3286 complete_operation(struct rule_dpif *rule)
3288 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3290 rule_invalidate(rule);
3292 struct dpif_completion *c = xmalloc(sizeof *c);
3293 c->op = rule->up.pending;
3294 list_push_back(&ofproto->completions, &c->list_node);
3296 ofoperation_complete(rule->up.pending, 0);
3300 static struct rule *
3303 struct rule_dpif *rule = xmalloc(sizeof *rule);
3308 rule_dealloc(struct rule *rule_)
3310 struct rule_dpif *rule = rule_dpif_cast(rule_);
3315 rule_construct(struct rule *rule_)
3317 struct rule_dpif *rule = rule_dpif_cast(rule_);
3318 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3319 struct rule_dpif *victim;
3323 error = validate_actions(rule->up.actions, rule->up.n_actions,
3324 &rule->up.cr.flow, ofproto->max_ports);
3329 rule->used = rule->up.created;
3330 rule->packet_count = 0;
3331 rule->byte_count = 0;
3333 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
3334 if (victim && !list_is_empty(&victim->facets)) {
3335 struct facet *facet;
3337 rule->facets = victim->facets;
3338 list_moved(&rule->facets);
3339 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3340 /* XXX: We're only clearing our local counters here. It's possible
3341 * that quite a few packets are unaccounted for in the datapath
3342 * statistics. These will be accounted to the new rule instead of
3343 * cleared as required. This could be fixed by clearing out the
3344 * datapath statistics for this facet, but currently it doesn't
3346 facet_reset_counters(facet);
3350 /* Must avoid list_moved() in this case. */
3351 list_init(&rule->facets);
3354 table_id = rule->up.table_id;
3355 rule->tag = (victim ? victim->tag
3357 : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc,
3358 ofproto->tables[table_id].basis));
3360 complete_operation(rule);
3365 rule_destruct(struct rule *rule_)
3367 struct rule_dpif *rule = rule_dpif_cast(rule_);
3368 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3369 struct facet *facet, *next_facet;
3371 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
3372 facet_revalidate(ofproto, facet);
3375 complete_operation(rule);
3379 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3381 struct rule_dpif *rule = rule_dpif_cast(rule_);
3382 struct facet *facet;
3384 /* Start from historical data for 'rule' itself that are no longer tracked
3385 * in facets. This counts, for example, facets that have expired. */
3386 *packets = rule->packet_count;
3387 *bytes = rule->byte_count;
3389 /* Add any statistics that are tracked by facets. This includes
3390 * statistical data recently updated by ofproto_update_stats() as well as
3391 * stats for packets that were executed "by hand" via dpif_execute(). */
3392 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3393 *packets += facet->packet_count;
3394 *bytes += facet->byte_count;
3399 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
3401 struct rule_dpif *rule = rule_dpif_cast(rule_);
3402 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3403 struct action_xlate_ctx ctx;
3404 struct ofpbuf *odp_actions;
3405 struct facet *facet;
3408 /* First look for a related facet. If we find one, account it to that. */
3409 facet = facet_lookup_valid(ofproto, flow);
3410 if (facet && facet->rule == rule) {
3411 if (!facet->may_install) {
3412 facet_make_actions(ofproto, facet, packet);
3414 facet_execute(ofproto, facet, packet);
3418 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
3419 * create a new facet for it and use that. */
3420 if (rule_dpif_lookup(ofproto, flow, 0) == rule) {
3421 facet = facet_create(rule, flow);
3422 facet_make_actions(ofproto, facet, packet);
3423 facet_execute(ofproto, facet, packet);
3424 facet_install(ofproto, facet, true);
3428 /* We can't account anything to a facet. If we were to try, then that
3429 * facet would have a non-matching rule, busting our invariants. */
3430 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3431 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
3432 size = packet->size;
3433 if (execute_odp_actions(ofproto, flow, odp_actions->data,
3434 odp_actions->size, packet)) {
3435 rule->used = time_msec();
3436 rule->packet_count++;
3437 rule->byte_count += size;
3438 flow_push_stats(rule, flow, 1, size, rule->used);
3440 ofpbuf_delete(odp_actions);
3446 rule_modify_actions(struct rule *rule_)
3448 struct rule_dpif *rule = rule_dpif_cast(rule_);
3449 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3452 error = validate_actions(rule->up.actions, rule->up.n_actions,
3453 &rule->up.cr.flow, ofproto->max_ports);
3455 ofoperation_complete(rule->up.pending, error);
3459 complete_operation(rule);
3462 /* Sends 'packet' out of port 'odp_port' within 'ofproto'.
3463 * Returns 0 if successful, otherwise a positive errno value. */
3465 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
3466 const struct ofpbuf *packet)
3468 struct ofpbuf key, odp_actions;
3469 struct odputil_keybuf keybuf;
3473 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
3474 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3475 odp_flow_key_from_flow(&key, &flow);
3477 ofpbuf_init(&odp_actions, 32);
3478 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
3480 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3481 error = dpif_execute(ofproto->dpif,
3483 odp_actions.data, odp_actions.size,
3485 ofpbuf_uninit(&odp_actions);
3488 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
3489 ofproto->up.name, odp_port, strerror(error));
3494 /* OpenFlow to datapath action translation. */
3496 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
3497 struct action_xlate_ctx *ctx);
3498 static void xlate_normal(struct action_xlate_ctx *);
3501 put_userspace_action(const struct ofproto_dpif *ofproto,
3502 struct ofpbuf *odp_actions,
3503 const struct flow *flow,
3504 const struct user_action_cookie *cookie)
3509 pid = dpif_port_get_pid(ofproto->dpif,
3510 ofp_port_to_odp_port(flow->in_port));
3512 offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE);
3513 nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid);
3514 nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA,
3515 cookie, sizeof *cookie);
3516 nl_msg_end_nested(odp_actions, offset);
3518 return odp_actions->size - NLA_ALIGN(sizeof *cookie);
3521 /* Compose SAMPLE action for sFlow. */
3523 compose_sflow_action(const struct ofproto_dpif *ofproto,
3524 struct ofpbuf *odp_actions,
3525 const struct flow *flow,
3528 uint32_t port_ifindex;
3529 uint32_t probability;
3530 struct user_action_cookie cookie;
3531 size_t sample_offset, actions_offset;
3532 int cookie_offset, n_output;
3534 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
3538 if (odp_port == OVSP_NONE) {
3542 port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
3546 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
3548 /* Number of packets out of UINT_MAX to sample. */
3549 probability = dpif_sflow_get_probability(ofproto->sflow);
3550 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
3552 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
3554 cookie.type = USER_ACTION_COOKIE_SFLOW;
3555 cookie.data = port_ifindex;
3556 cookie.n_output = n_output;
3557 cookie.vlan_tci = 0;
3558 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
3560 nl_msg_end_nested(odp_actions, actions_offset);
3561 nl_msg_end_nested(odp_actions, sample_offset);
3562 return cookie_offset;
3565 /* SAMPLE action must be first action in any given list of actions.
3566 * At this point we do not have all information required to build it. So try to
3567 * build sample action as complete as possible. */
3569 add_sflow_action(struct action_xlate_ctx *ctx)
3571 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
3573 &ctx->flow, OVSP_NONE);
3574 ctx->sflow_odp_port = 0;
3575 ctx->sflow_n_outputs = 0;
3578 /* Fix SAMPLE action according to data collected while composing ODP actions.
3579 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
3580 * USERSPACE action's user-cookie which is required for sflow. */
3582 fix_sflow_action(struct action_xlate_ctx *ctx)
3584 const struct flow *base = &ctx->base_flow;
3585 struct user_action_cookie *cookie;
3587 if (!ctx->user_cookie_offset) {
3591 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
3593 assert(cookie != NULL);
3594 assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
3596 if (ctx->sflow_n_outputs) {
3597 cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow,
3598 ctx->sflow_odp_port);
3600 if (ctx->sflow_n_outputs >= 255) {
3601 cookie->n_output = 255;
3603 cookie->n_output = ctx->sflow_n_outputs;
3605 cookie->vlan_tci = base->vlan_tci;
3609 commit_action__(struct ofpbuf *odp_actions,
3610 enum ovs_action_attr act_type,
3611 enum ovs_key_attr key_type,
3612 const void *key, size_t key_size)
3614 size_t offset = nl_msg_start_nested(odp_actions, act_type);
3616 nl_msg_put_unspec(odp_actions, key_type, key, key_size);
3617 nl_msg_end_nested(odp_actions, offset);
3621 commit_set_tun_id_action(const struct flow *flow, struct flow *base,
3622 struct ofpbuf *odp_actions)
3624 if (base->tun_id == flow->tun_id) {
3627 base->tun_id = flow->tun_id;
3629 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3630 OVS_KEY_ATTR_TUN_ID, &base->tun_id, sizeof(base->tun_id));
3634 commit_set_ether_addr_action(const struct flow *flow, struct flow *base,
3635 struct ofpbuf *odp_actions)
3637 struct ovs_key_ethernet eth_key;
3639 if (eth_addr_equals(base->dl_src, flow->dl_src) &&
3640 eth_addr_equals(base->dl_dst, flow->dl_dst)) {
3644 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
3645 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
3647 memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN);
3648 memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN);
3650 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3651 OVS_KEY_ATTR_ETHERNET, ð_key, sizeof(eth_key));
3655 commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci)
3657 struct flow *base = &ctx->base_flow;
3659 if (base->vlan_tci == new_tci) {
3663 if (base->vlan_tci & htons(VLAN_CFI)) {
3664 nl_msg_put_u16(ctx->odp_actions, OVS_ACTION_ATTR_POP,
3665 OVS_KEY_ATTR_8021Q);
3668 if (new_tci & htons(VLAN_CFI)) {
3669 struct ovs_key_8021q q_key;
3671 q_key.q_tpid = htons(ETH_TYPE_VLAN);
3672 q_key.q_tci = new_tci & ~htons(VLAN_CFI);
3674 commit_action__(ctx->odp_actions, OVS_ACTION_ATTR_PUSH,
3675 OVS_KEY_ATTR_8021Q, &q_key, sizeof(q_key));
3677 base->vlan_tci = new_tci;
3681 commit_set_nw_action(const struct flow *flow, struct flow *base,
3682 struct ofpbuf *odp_actions)
3684 int frag = base->tos_frag & FLOW_FRAG_MASK;
3685 struct ovs_key_ipv4 ipv4_key;
3687 if (base->dl_type != htons(ETH_TYPE_IP) ||
3688 !base->nw_src || !base->nw_dst) {
3692 if (base->nw_src == flow->nw_src &&
3693 base->nw_dst == flow->nw_dst &&
3694 base->tos_frag == flow->tos_frag) {
3699 memset(&ipv4_key, 0, sizeof(ipv4_key));
3700 ipv4_key.ipv4_src = base->nw_src = flow->nw_src;
3701 ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst;
3702 ipv4_key.ipv4_proto = base->nw_proto;
3703 ipv4_key.ipv4_tos = flow->tos_frag & IP_DSCP_MASK;
3704 ipv4_key.ipv4_frag = (frag == 0 ? OVS_FRAG_TYPE_NONE
3705 : frag == FLOW_FRAG_ANY ? OVS_FRAG_TYPE_FIRST
3706 : OVS_FRAG_TYPE_LATER);
3708 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3709 OVS_KEY_ATTR_IPV4, &ipv4_key, sizeof(ipv4_key));
3713 commit_set_port_action(const struct flow *flow, struct flow *base,
3714 struct ofpbuf *odp_actions)
3716 if (!base->tp_src || !base->tp_dst) {
3720 if (base->tp_src == flow->tp_src &&
3721 base->tp_dst == flow->tp_dst) {
3725 if (flow->nw_proto == IPPROTO_TCP) {
3726 struct ovs_key_tcp port_key;
3728 port_key.tcp_src = base->tp_src = flow->tp_src;
3729 port_key.tcp_dst = base->tp_dst = flow->tp_dst;
3731 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3732 OVS_KEY_ATTR_TCP, &port_key, sizeof(port_key));
3734 } else if (flow->nw_proto == IPPROTO_UDP) {
3735 struct ovs_key_udp port_key;
3737 port_key.udp_src = base->tp_src = flow->tp_src;
3738 port_key.udp_dst = base->tp_dst = flow->tp_dst;
3740 commit_action__(odp_actions, OVS_ACTION_ATTR_SET,
3741 OVS_KEY_ATTR_UDP, &port_key, sizeof(port_key));
3746 commit_priority_action(struct action_xlate_ctx *ctx)
3748 if (ctx->base_priority == ctx->priority) {
3752 if (ctx->priority) {
3753 nl_msg_put_u32(ctx->odp_actions,
3754 OVS_ACTION_ATTR_SET_PRIORITY, ctx->priority);
3756 nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_PRIORITY);
3758 ctx->base_priority = ctx->priority;
3762 commit_odp_actions(struct action_xlate_ctx *ctx)
3764 const struct flow *flow = &ctx->flow;
3765 struct flow *base = &ctx->base_flow;
3766 struct ofpbuf *odp_actions = ctx->odp_actions;
3768 commit_set_tun_id_action(flow, base, odp_actions);
3769 commit_set_ether_addr_action(flow, base, odp_actions);
3770 commit_vlan_action(ctx, flow->vlan_tci);
3771 commit_set_nw_action(flow, base, odp_actions);
3772 commit_set_port_action(flow, base, odp_actions);
3773 commit_priority_action(ctx);
3777 compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port)
3779 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
3780 ctx->sflow_odp_port = odp_port;
3781 ctx->sflow_n_outputs++;
3785 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
3787 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
3788 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
3791 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)
3792 || !stp_forward_in_state(ofport->stp_state)) {
3793 /* Forwarding disabled on port. */
3798 * We don't have an ofport record for this port, but it doesn't hurt to
3799 * allow forwarding to it anyhow. Maybe such a port will appear later
3800 * and we're pre-populating the flow table.
3804 commit_odp_actions(ctx);
3805 compose_output_action(ctx, odp_port);
3806 ctx->nf_output_iface = ofp_port;
3810 xlate_table_action(struct action_xlate_ctx *ctx,
3811 uint16_t in_port, uint8_t table_id)
3813 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
3814 struct ofproto_dpif *ofproto = ctx->ofproto;
3815 struct rule_dpif *rule;
3816 uint16_t old_in_port;
3817 uint8_t old_table_id;
3819 old_table_id = ctx->table_id;
3820 ctx->table_id = table_id;
3822 /* Look up a flow with 'in_port' as the input port. */
3823 old_in_port = ctx->flow.in_port;
3824 ctx->flow.in_port = in_port;
3825 rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id);
3828 if (table_id > 0 && table_id < N_TABLES) {
3829 struct table_dpif *table = &ofproto->tables[table_id];
3830 if (table->other_table) {
3833 : rule_calculate_tag(&ctx->flow,
3834 &table->other_table->wc,
3839 /* Restore the original input port. Otherwise OFPP_NORMAL and
3840 * OFPP_IN_PORT will have surprising behavior. */
3841 ctx->flow.in_port = old_in_port;
3843 if (ctx->resubmit_hook) {
3844 ctx->resubmit_hook(ctx, rule);
3849 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
3853 ctx->table_id = old_table_id;
3855 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3857 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
3858 MAX_RESUBMIT_RECURSION);
3863 xlate_resubmit_table(struct action_xlate_ctx *ctx,
3864 const struct nx_action_resubmit *nar)
3869 in_port = (nar->in_port == htons(OFPP_IN_PORT)
3871 : ntohs(nar->in_port));
3872 table_id = nar->table == 255 ? ctx->table_id : nar->table;
3874 xlate_table_action(ctx, in_port, table_id);
3878 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
3880 struct ofport_dpif *ofport;
3882 commit_odp_actions(ctx);
3883 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
3884 uint16_t ofp_port = ofport->up.ofp_port;
3885 if (ofp_port != ctx->flow.in_port
3886 && !(ofport->up.opp.config & mask)
3887 && stp_forward_in_state(ofport->stp_state)) {
3888 compose_output_action(ctx, ofport->odp_port);
3892 ctx->nf_output_iface = NF_OUT_FLOOD;
3896 compose_controller_action(struct action_xlate_ctx *ctx, int len)
3898 struct user_action_cookie cookie;
3900 cookie.type = USER_ACTION_COOKIE_CONTROLLER;
3902 cookie.n_output = 0;
3903 cookie.vlan_tci = 0;
3904 put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie);
3908 xlate_output_action__(struct action_xlate_ctx *ctx,
3909 uint16_t port, uint16_t max_len)
3911 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
3913 ctx->nf_output_iface = NF_OUT_DROP;
3917 add_output_action(ctx, ctx->flow.in_port);
3920 xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id);
3926 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
3929 flood_packets(ctx, htonl(0));
3931 case OFPP_CONTROLLER:
3932 commit_odp_actions(ctx);
3933 compose_controller_action(ctx, max_len);
3936 add_output_action(ctx, OFPP_LOCAL);
3941 if (port != ctx->flow.in_port) {
3942 add_output_action(ctx, port);
3947 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3948 ctx->nf_output_iface = NF_OUT_FLOOD;
3949 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3950 ctx->nf_output_iface = prev_nf_output_iface;
3951 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3952 ctx->nf_output_iface != NF_OUT_FLOOD) {
3953 ctx->nf_output_iface = NF_OUT_MULTI;
3958 xlate_output_reg_action(struct action_xlate_ctx *ctx,
3959 const struct nx_action_output_reg *naor)
3963 ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow);
3965 if (ofp_port <= UINT16_MAX) {
3966 xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len));
3971 xlate_output_action(struct action_xlate_ctx *ctx,
3972 const struct ofp_action_output *oao)
3974 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
3978 xlate_enqueue_action(struct action_xlate_ctx *ctx,
3979 const struct ofp_action_enqueue *oae)
3981 uint16_t ofp_port, odp_port;
3982 uint32_t ctx_priority, priority;
3985 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3988 /* Fall back to ordinary output action. */
3989 xlate_output_action__(ctx, ntohs(oae->port), 0);
3993 /* Figure out datapath output port. */
3994 ofp_port = ntohs(oae->port);
3995 if (ofp_port == OFPP_IN_PORT) {
3996 ofp_port = ctx->flow.in_port;
3997 } else if (ofp_port == ctx->flow.in_port) {
4000 odp_port = ofp_port_to_odp_port(ofp_port);
4002 /* Add datapath actions. */
4003 ctx_priority = ctx->priority;
4004 ctx->priority = priority;
4005 add_output_action(ctx, odp_port);
4006 ctx->priority = ctx_priority;
4008 /* Update NetFlow output port. */
4009 if (ctx->nf_output_iface == NF_OUT_DROP) {
4010 ctx->nf_output_iface = odp_port;
4011 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4012 ctx->nf_output_iface = NF_OUT_MULTI;
4017 xlate_set_queue_action(struct action_xlate_ctx *ctx,
4018 const struct nx_action_set_queue *nasq)
4023 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
4026 /* Couldn't translate queue to a priority, so ignore. A warning
4027 * has already been logged. */
4031 ctx->priority = priority;
4034 struct xlate_reg_state {
4040 xlate_autopath(struct action_xlate_ctx *ctx,
4041 const struct nx_action_autopath *naa)
4043 uint16_t ofp_port = ntohl(naa->id);
4044 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
4046 if (!port || !port->bundle) {
4047 ofp_port = OFPP_NONE;
4048 } else if (port->bundle->bond) {
4049 /* Autopath does not support VLAN hashing. */
4050 struct ofport_dpif *slave = bond_choose_output_slave(
4051 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
4053 ofp_port = slave->up.ofp_port;
4056 autopath_execute(naa, &ctx->flow, ofp_port);
4060 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
4062 struct ofproto_dpif *ofproto = ofproto_;
4063 struct ofport_dpif *port;
4073 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4076 port = get_ofp_port(ofproto, ofp_port);
4077 return port ? port->may_enable : false;
4082 xlate_learn_action(struct action_xlate_ctx *ctx,
4083 const struct nx_action_learn *learn)
4085 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
4086 struct ofputil_flow_mod fm;
4089 learn_execute(learn, &ctx->flow, &fm);
4091 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
4092 if (error && !VLOG_DROP_WARN(&rl)) {
4093 char *msg = ofputil_error_to_string(error);
4094 VLOG_WARN("learning action failed to modify flow table (%s)", msg);
4102 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
4104 if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
4105 ? htonl(OFPPC_NO_RECV_STP)
4106 : htonl(OFPPC_NO_RECV))) {
4110 /* Only drop packets here if both forwarding and learning are
4111 * disabled. If just learning is enabled, we need to have
4112 * OFPP_NORMAL and the learning action have a look at the packet
4113 * before we can drop it. */
4114 if (!stp_forward_in_state(port->stp_state)
4115 && !stp_learn_in_state(port->stp_state)) {
4123 do_xlate_actions(const union ofp_action *in, size_t n_in,
4124 struct action_xlate_ctx *ctx)
4126 const struct ofport_dpif *port;
4127 const union ofp_action *ia;
4130 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
4131 if (port && !may_receive(port, ctx)) {
4132 /* Drop this flow. */
4136 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
4137 const struct ofp_action_dl_addr *oada;
4138 const struct nx_action_resubmit *nar;
4139 const struct nx_action_set_tunnel *nast;
4140 const struct nx_action_set_queue *nasq;
4141 const struct nx_action_multipath *nam;
4142 const struct nx_action_autopath *naa;
4143 const struct nx_action_bundle *nab;
4144 const struct nx_action_output_reg *naor;
4145 enum ofputil_action_code code;
4148 code = ofputil_decode_action_unsafe(ia);
4150 case OFPUTIL_OFPAT_OUTPUT:
4151 xlate_output_action(ctx, &ia->output);
4154 case OFPUTIL_OFPAT_SET_VLAN_VID:
4155 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
4156 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
4159 case OFPUTIL_OFPAT_SET_VLAN_PCP:
4160 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
4161 ctx->flow.vlan_tci |= htons(
4162 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
4165 case OFPUTIL_OFPAT_STRIP_VLAN:
4166 ctx->flow.vlan_tci = htons(0);
4169 case OFPUTIL_OFPAT_SET_DL_SRC:
4170 oada = ((struct ofp_action_dl_addr *) ia);
4171 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
4174 case OFPUTIL_OFPAT_SET_DL_DST:
4175 oada = ((struct ofp_action_dl_addr *) ia);
4176 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
4179 case OFPUTIL_OFPAT_SET_NW_SRC:
4180 ctx->flow.nw_src = ia->nw_addr.nw_addr;
4183 case OFPUTIL_OFPAT_SET_NW_DST:
4184 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
4187 case OFPUTIL_OFPAT_SET_NW_TOS:
4188 ctx->flow.tos_frag &= ~IP_DSCP_MASK;
4189 ctx->flow.tos_frag |= ia->nw_tos.nw_tos & IP_DSCP_MASK;
4192 case OFPUTIL_OFPAT_SET_TP_SRC:
4193 ctx->flow.tp_src = ia->tp_port.tp_port;
4196 case OFPUTIL_OFPAT_SET_TP_DST:
4197 ctx->flow.tp_dst = ia->tp_port.tp_port;
4200 case OFPUTIL_OFPAT_ENQUEUE:
4201 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
4204 case OFPUTIL_NXAST_RESUBMIT:
4205 nar = (const struct nx_action_resubmit *) ia;
4206 xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id);
4209 case OFPUTIL_NXAST_RESUBMIT_TABLE:
4210 xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia);
4213 case OFPUTIL_NXAST_SET_TUNNEL:
4214 nast = (const struct nx_action_set_tunnel *) ia;
4215 tun_id = htonll(ntohl(nast->tun_id));
4216 ctx->flow.tun_id = tun_id;
4219 case OFPUTIL_NXAST_SET_QUEUE:
4220 nasq = (const struct nx_action_set_queue *) ia;
4221 xlate_set_queue_action(ctx, nasq);
4224 case OFPUTIL_NXAST_POP_QUEUE:
4228 case OFPUTIL_NXAST_REG_MOVE:
4229 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
4233 case OFPUTIL_NXAST_REG_LOAD:
4234 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
4238 case OFPUTIL_NXAST_NOTE:
4239 /* Nothing to do. */
4242 case OFPUTIL_NXAST_SET_TUNNEL64:
4243 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
4244 ctx->flow.tun_id = tun_id;
4247 case OFPUTIL_NXAST_MULTIPATH:
4248 nam = (const struct nx_action_multipath *) ia;
4249 multipath_execute(nam, &ctx->flow);
4252 case OFPUTIL_NXAST_AUTOPATH:
4253 naa = (const struct nx_action_autopath *) ia;
4254 xlate_autopath(ctx, naa);
4257 case OFPUTIL_NXAST_BUNDLE:
4258 ctx->ofproto->has_bundle_action = true;
4259 nab = (const struct nx_action_bundle *) ia;
4260 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
4265 case OFPUTIL_NXAST_BUNDLE_LOAD:
4266 ctx->ofproto->has_bundle_action = true;
4267 nab = (const struct nx_action_bundle *) ia;
4268 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
4272 case OFPUTIL_NXAST_OUTPUT_REG:
4273 naor = (const struct nx_action_output_reg *) ia;
4274 xlate_output_reg_action(ctx, naor);
4277 case OFPUTIL_NXAST_LEARN:
4278 ctx->has_learn = true;
4279 if (ctx->may_learn) {
4280 xlate_learn_action(ctx, (const struct nx_action_learn *) ia);
4286 /* We've let OFPP_NORMAL and the learning action look at the packet,
4287 * so drop it now if forwarding is disabled. */
4288 if (port && !stp_forward_in_state(port->stp_state)) {
4289 ofpbuf_clear(ctx->odp_actions);
4290 add_sflow_action(ctx);
4295 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
4296 struct ofproto_dpif *ofproto, const struct flow *flow,
4297 const struct ofpbuf *packet)
4299 ctx->ofproto = ofproto;
4301 ctx->packet = packet;
4302 ctx->may_learn = packet != NULL;
4303 ctx->resubmit_hook = NULL;
4306 static struct ofpbuf *
4307 xlate_actions(struct action_xlate_ctx *ctx,
4308 const union ofp_action *in, size_t n_in)
4310 COVERAGE_INC(ofproto_dpif_xlate);
4312 ctx->odp_actions = ofpbuf_new(512);
4313 ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE);
4315 ctx->may_set_up_flow = true;
4316 ctx->has_learn = false;
4317 ctx->has_normal = false;
4318 ctx->nf_output_iface = NF_OUT_DROP;
4321 ctx->base_priority = 0;
4322 ctx->base_flow = ctx->flow;
4323 ctx->base_flow.tun_id = 0;
4326 if (ctx->flow.tos_frag & FLOW_FRAG_ANY) {
4327 switch (ctx->ofproto->up.frag_handling) {
4328 case OFPC_FRAG_NORMAL:
4329 /* We must pretend that transport ports are unavailable. */
4330 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
4331 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
4334 case OFPC_FRAG_DROP:
4335 return ctx->odp_actions;
4337 case OFPC_FRAG_REASM:
4340 case OFPC_FRAG_NX_MATCH:
4341 /* Nothing to do. */
4346 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
4347 ctx->may_set_up_flow = false;
4348 return ctx->odp_actions;
4350 add_sflow_action(ctx);
4351 do_xlate_actions(in, n_in, ctx);
4353 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
4354 ctx->odp_actions->data,
4355 ctx->odp_actions->size)) {
4356 ctx->may_set_up_flow = false;
4358 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
4360 compose_output_action(ctx, OVSP_LOCAL);
4363 fix_sflow_action(ctx);
4366 return ctx->odp_actions;
4369 /* OFPP_NORMAL implementation. */
4372 struct ofport_dpif *port;
4377 struct dst builtin[32];
4379 size_t n, allocated;
4382 static void dst_set_init(struct dst_set *);
4383 static void dst_set_add(struct dst_set *, const struct dst *);
4384 static void dst_set_free(struct dst_set *);
4386 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
4388 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
4389 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
4390 * the bundle on which the packet was received, returns the VLAN to which the
4393 * Both 'vid' and the return value are in the range 0...4095. */
4395 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
4397 switch (in_bundle->vlan_mode) {
4398 case PORT_VLAN_ACCESS:
4399 return in_bundle->vlan;
4402 case PORT_VLAN_TRUNK:
4405 case PORT_VLAN_NATIVE_UNTAGGED:
4406 case PORT_VLAN_NATIVE_TAGGED:
4407 return vid ? vid : in_bundle->vlan;
4414 /* Given 'vlan', the VLAN that a packet belongs to, and
4415 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
4416 * that should be included in the 802.1Q header. (If the return value is 0,
4417 * then the 802.1Q header should only be included in the packet if there is a
4420 * Both 'vlan' and the return value are in the range 0...4095. */
4422 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
4424 switch (out_bundle->vlan_mode) {
4425 case PORT_VLAN_ACCESS:
4428 case PORT_VLAN_TRUNK:
4429 case PORT_VLAN_NATIVE_TAGGED:
4432 case PORT_VLAN_NATIVE_UNTAGGED:
4433 return vlan == out_bundle->vlan ? 0 : vlan;
4441 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
4442 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
4446 vlan = input_vid_to_vlan(in_bundle, vlan_tci_to_vid(ctx->flow.vlan_tci));
4447 dst->vid = output_vlan_to_vid(out_bundle, vlan);
4449 dst->port = (!out_bundle->bond
4450 ? ofbundle_get_a_port(out_bundle)
4451 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
4452 dst->vid, &ctx->tags));
4453 return dst->port != NULL;
4457 mirror_mask_ffs(mirror_mask_t mask)
4459 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
4464 dst_set_init(struct dst_set *set)
4466 set->dsts = set->builtin;
4468 set->allocated = ARRAY_SIZE(set->builtin);
4472 dst_set_add(struct dst_set *set, const struct dst *dst)
4474 if (set->n >= set->allocated) {
4475 size_t new_allocated;
4476 struct dst *new_dsts;
4478 new_allocated = set->allocated * 2;
4479 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
4480 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
4484 set->dsts = new_dsts;
4485 set->allocated = new_allocated;
4487 set->dsts[set->n++] = *dst;
4491 dst_set_free(struct dst_set *set)
4493 if (set->dsts != set->builtin) {
4499 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
4502 for (i = 0; i < set->n; i++) {
4503 if (set->dsts[i].vid == test->vid
4504 && set->dsts[i].port == test->port) {
4512 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
4514 return (bundle->vlan_mode != PORT_VLAN_ACCESS
4515 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
4519 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
4521 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
4524 /* Returns an arbitrary interface within 'bundle'. */
4525 static struct ofport_dpif *
4526 ofbundle_get_a_port(const struct ofbundle *bundle)
4528 return CONTAINER_OF(list_front(&bundle->ports),
4529 struct ofport_dpif, bundle_node);
4533 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
4534 const struct ofbundle *in_bundle,
4535 const struct ofbundle *out_bundle, struct dst_set *set)
4539 if (out_bundle == OFBUNDLE_FLOOD) {
4540 struct ofbundle *bundle;
4542 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
4543 if (bundle != in_bundle
4544 && ofbundle_includes_vlan(bundle, vlan)
4545 && bundle->floodable
4546 && !bundle->mirror_out
4547 && set_dst(ctx, &dst, in_bundle, bundle)) {
4548 dst_set_add(set, &dst);
4551 ctx->nf_output_iface = NF_OUT_FLOOD;
4552 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
4553 dst_set_add(set, &dst);
4554 ctx->nf_output_iface = dst.port->odp_port;
4559 vlan_is_mirrored(const struct ofmirror *m, int vlan)
4561 return !m->vlans || bitmap_is_set(m->vlans, vlan);
4564 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
4565 * to a VLAN. In general most packets may be mirrored but we want to drop
4566 * protocols that may confuse switches. */
4568 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
4570 /* If you change this function's behavior, please update corresponding
4571 * documentation in vswitch.xml at the same time. */
4572 if (dst[0] != 0x01) {
4573 /* All the currently banned MACs happen to start with 01 currently, so
4574 * this is a quick way to eliminate most of the good ones. */
4576 if (eth_addr_is_reserved(dst)) {
4577 /* Drop STP, IEEE pause frames, and other reserved protocols
4578 * (01-80-c2-00-00-0x). */
4582 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
4584 if ((dst[3] & 0xfe) == 0xcc &&
4585 (dst[4] & 0xfe) == 0xcc &&
4586 (dst[5] & 0xfe) == 0xcc) {
4587 /* Drop the following protocols plus others following the same
4590 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
4591 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
4592 STP Uplink Fast (01-00-0c-cd-cd-cd) */
4596 if (!(dst[3] | dst[4] | dst[5])) {
4597 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
4606 compose_mirror_dsts(struct action_xlate_ctx *ctx,
4607 uint16_t vlan, const struct ofbundle *in_bundle,
4608 struct dst_set *set)
4610 struct ofproto_dpif *ofproto = ctx->ofproto;
4611 mirror_mask_t mirrors;
4615 mirrors = in_bundle->src_mirrors;
4616 for (i = 0; i < set->n; i++) {
4617 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
4624 flow_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4626 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
4627 if (vlan_is_mirrored(m, vlan)) {
4631 if (set_dst(ctx, &dst, in_bundle, m->out)
4632 && !dst_is_duplicate(set, &dst)) {
4633 dst_set_add(set, &dst);
4635 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
4636 struct ofbundle *bundle;
4638 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4639 if (ofbundle_includes_vlan(bundle, m->out_vlan)
4640 && set_dst(ctx, &dst, in_bundle, bundle))
4642 /* set_dst() got dst->vid from the input packet's VLAN,
4643 * not from m->out_vlan, so recompute it. */
4644 dst.vid = output_vlan_to_vid(bundle, m->out_vlan);
4646 if (dst_is_duplicate(set, &dst)) {
4650 if (bundle == in_bundle && dst.vid == flow_vid) {
4651 /* Don't send out input port on same VLAN. */
4654 dst_set_add(set, &dst);
4659 mirrors &= mirrors - 1;
4664 compose_dst_output_action(struct action_xlate_ctx *ctx, const struct dst *dst)
4668 tci = htons(dst->vid);
4670 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
4671 tci |= htons(VLAN_CFI);
4673 commit_vlan_action(ctx, tci);
4675 compose_output_action(ctx, dst->port->odp_port);
4679 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
4680 const struct ofbundle *in_bundle,
4681 const struct ofbundle *out_bundle)
4683 uint16_t initial_vid;
4684 const struct dst *dst;
4688 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
4689 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
4695 /* Output all the packets we can without having to change the VLAN. */
4696 commit_odp_actions(ctx);
4697 initial_vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
4698 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
4699 if (dst->vid == initial_vid) {
4700 compose_dst_output_action(ctx, dst);
4704 /* Then output the rest. */
4705 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
4706 if (dst->vid != initial_vid) {
4707 compose_dst_output_action(ctx, dst);
4714 /* Returns the effective vlan of a packet, taking into account both the
4715 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
4716 * the packet is untagged and -1 indicates it has an invalid header and
4717 * should be dropped. */
4719 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
4720 struct ofbundle *in_bundle, bool have_packet)
4722 int vlan = vlan_tci_to_vid(flow->vlan_tci);
4724 if (in_bundle->vlan_mode == PORT_VLAN_ACCESS) {
4725 /* Drop tagged packet on access port */
4727 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4728 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
4729 "packet received on port %s configured with "
4730 "implicit VLAN %"PRIu16,
4731 ofproto->up.name, vlan,
4732 in_bundle->name, in_bundle->vlan);
4735 } else if (ofbundle_includes_vlan(in_bundle, vlan)) {
4738 /* Drop packets from a VLAN not member of the trunk */
4740 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4741 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
4742 "packet received on port %s not configured for "
4744 ofproto->up.name, vlan, in_bundle->name, vlan);
4749 if (in_bundle->vlan_mode != PORT_VLAN_TRUNK) {
4750 return in_bundle->vlan;
4752 return ofbundle_includes_vlan(in_bundle, 0) ? 0 : -1;
4757 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
4758 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
4759 * indicate this; newer upstream kernels use gratuitous ARP requests. */
4761 is_gratuitous_arp(const struct flow *flow)
4763 return (flow->dl_type == htons(ETH_TYPE_ARP)
4764 && eth_addr_is_broadcast(flow->dl_dst)
4765 && (flow->nw_proto == ARP_OP_REPLY
4766 || (flow->nw_proto == ARP_OP_REQUEST
4767 && flow->nw_src == flow->nw_dst)));
4771 update_learning_table(struct ofproto_dpif *ofproto,
4772 const struct flow *flow, int vlan,
4773 struct ofbundle *in_bundle)
4775 struct mac_entry *mac;
4777 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
4781 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
4782 if (is_gratuitous_arp(flow)) {
4783 /* We don't want to learn from gratuitous ARP packets that are
4784 * reflected back over bond slaves so we lock the learning table. */
4785 if (!in_bundle->bond) {
4786 mac_entry_set_grat_arp_lock(mac);
4787 } else if (mac_entry_is_grat_arp_locked(mac)) {
4792 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
4793 /* The log messages here could actually be useful in debugging,
4794 * so keep the rate limit relatively high. */
4795 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4796 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
4797 "on port %s in VLAN %d",
4798 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
4799 in_bundle->name, vlan);
4801 mac->port.p = in_bundle;
4802 tag_set_add(&ofproto->revalidate_set,
4803 mac_learning_changed(ofproto->ml, mac));
4807 /* Determines whether packets in 'flow' within 'br' should be forwarded or
4808 * dropped. Returns true if they may be forwarded, false if they should be
4811 * If 'have_packet' is true, it indicates that the caller is processing a
4812 * received packet. If 'have_packet' is false, then the caller is just
4813 * revalidating an existing flow because configuration has changed. Either
4814 * way, 'have_packet' only affects logging (there is no point in logging errors
4815 * during revalidation).
4817 * Sets '*in_portp' to the input port. This will be a null pointer if
4818 * flow->in_port does not designate a known input port (in which case
4819 * is_admissible() returns false).
4821 * When returning true, sets '*vlanp' to the effective VLAN of the input
4822 * packet, as returned by flow_get_vlan().
4824 * May also add tags to '*tags', although the current implementation only does
4825 * so in one special case.
4828 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
4830 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
4832 struct ofport_dpif *in_port;
4833 struct ofbundle *in_bundle;
4836 /* Find the port and bundle for the received packet. */
4837 in_port = get_ofp_port(ofproto, flow->in_port);
4838 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
4839 if (!in_port || !in_bundle) {
4840 /* No interface? Something fishy... */
4842 /* Odd. A few possible reasons here:
4844 * - We deleted a port but there are still a few packets queued up
4847 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
4848 * we don't know about.
4850 * - Packet arrived on the local port but the local port is not
4853 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4855 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
4857 ofproto->up.name, flow->in_port);
4862 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
4867 /* Drop frames for reserved multicast addresses only if forward_bpdu
4868 * option is absent. */
4869 if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) {
4873 /* Drop frames on bundles reserved for mirroring. */
4874 if (in_bundle->mirror_out) {
4876 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4877 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
4878 "%s, which is reserved exclusively for mirroring",
4879 ofproto->up.name, in_bundle->name);
4884 if (in_bundle->bond) {
4885 struct mac_entry *mac;
4887 switch (bond_check_admissibility(in_bundle->bond, in_port,
4888 flow->dl_dst, tags)) {
4895 case BV_DROP_IF_MOVED:
4896 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
4897 if (mac && mac->port.p != in_bundle &&
4898 (!is_gratuitous_arp(flow)
4899 || mac_entry_is_grat_arp_locked(mac))) {
4910 xlate_normal(struct action_xlate_ctx *ctx)
4912 struct ofbundle *in_bundle;
4913 struct ofbundle *out_bundle;
4914 struct mac_entry *mac;
4917 ctx->has_normal = true;
4919 /* Check whether we should drop packets in this flow. */
4920 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
4921 &ctx->tags, &vlan, &in_bundle)) {
4926 /* Learn source MAC. */
4927 if (ctx->may_learn) {
4928 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
4931 /* Determine output bundle. */
4932 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
4935 out_bundle = mac->port.p;
4936 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
4937 /* If we are revalidating but don't have a learning entry then eject
4938 * the flow. Installing a flow that floods packets opens up a window
4939 * of time where we could learn from a packet reflected on a bond and
4940 * blackhole packets before the learning table is updated to reflect
4941 * the correct port. */
4942 ctx->may_set_up_flow = false;
4945 out_bundle = OFBUNDLE_FLOOD;
4948 /* Don't send packets out their input bundles. */
4949 if (in_bundle == out_bundle) {
4955 compose_actions(ctx, vlan, in_bundle, out_bundle);
4959 /* Optimized flow revalidation.
4961 * It's a difficult problem, in general, to tell which facets need to have
4962 * their actions recalculated whenever the OpenFlow flow table changes. We
4963 * don't try to solve that general problem: for most kinds of OpenFlow flow
4964 * table changes, we recalculate the actions for every facet. This is
4965 * relatively expensive, but it's good enough if the OpenFlow flow table
4966 * doesn't change very often.
4968 * However, we can expect one particular kind of OpenFlow flow table change to
4969 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
4970 * of CPU on revalidating every facet whenever MAC learning modifies the flow
4971 * table, we add a special case that applies to flow tables in which every rule
4972 * has the same form (that is, the same wildcards), except that the table is
4973 * also allowed to have a single "catch-all" flow that matches all packets. We
4974 * optimize this case by tagging all of the facets that resubmit into the table
4975 * and invalidating the same tag whenever a flow changes in that table. The
4976 * end result is that we revalidate just the facets that need it (and sometimes
4977 * a few more, but not all of the facets or even all of the facets that
4978 * resubmit to the table modified by MAC learning). */
4980 /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted
4981 * into an OpenFlow table with the given 'basis'. */
4983 rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc,
4986 if (flow_wildcards_is_catchall(wc)) {
4989 struct flow tag_flow = *flow;
4990 flow_zero_wildcards(&tag_flow, wc);
4991 return tag_create_deterministic(flow_hash(&tag_flow, secret));
4995 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
4996 * taggability of that table.
4998 * This function must be called after *each* change to a flow table. If you
4999 * skip calling it on some changes then the pointer comparisons at the end can
5000 * be invalid if you get unlucky. For example, if a flow removal causes a
5001 * cls_table to be destroyed and then a flow insertion causes a cls_table with
5002 * different wildcards to be created with the same address, then this function
5003 * will incorrectly skip revalidation. */
5005 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
5007 struct table_dpif *table = &ofproto->tables[table_id];
5008 const struct classifier *cls = &ofproto->up.tables[table_id];
5009 struct cls_table *catchall, *other;
5010 struct cls_table *t;
5012 catchall = other = NULL;
5014 switch (hmap_count(&cls->tables)) {
5016 /* We could tag this OpenFlow table but it would make the logic a
5017 * little harder and it's a corner case that doesn't seem worth it
5023 HMAP_FOR_EACH (t, hmap_node, &cls->tables) {
5024 if (cls_table_is_catchall(t)) {
5026 } else if (!other) {
5029 /* Indicate that we can't tag this by setting both tables to
5030 * NULL. (We know that 'catchall' is already NULL.) */
5037 /* Can't tag this table. */
5041 if (table->catchall_table != catchall || table->other_table != other) {
5042 table->catchall_table = catchall;
5043 table->other_table = other;
5044 ofproto->need_revalidate = true;
5048 /* Given 'rule' that has changed in some way (either it is a rule being
5049 * inserted, a rule being deleted, or a rule whose actions are being
5050 * modified), marks facets for revalidation to ensure that packets will be
5051 * forwarded correctly according to the new state of the flow table.
5053 * This function must be called after *each* change to a flow table. See
5054 * the comment on table_update_taggable() for more information. */
5056 rule_invalidate(const struct rule_dpif *rule)
5058 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5060 table_update_taggable(ofproto, rule->up.table_id);
5062 if (!ofproto->need_revalidate) {
5063 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
5065 if (table->other_table && rule->tag) {
5066 tag_set_add(&ofproto->revalidate_set, rule->tag);
5068 ofproto->need_revalidate = true;
5074 set_frag_handling(struct ofproto *ofproto_,
5075 enum ofp_config_flags frag_handling)
5077 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5079 if (frag_handling != OFPC_FRAG_REASM) {
5080 ofproto->need_revalidate = true;
5088 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5089 const struct flow *flow,
5090 const union ofp_action *ofp_actions, size_t n_ofp_actions)
5092 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5095 error = validate_actions(ofp_actions, n_ofp_actions, flow,
5096 ofproto->max_ports);
5098 struct odputil_keybuf keybuf;
5099 struct action_xlate_ctx ctx;
5100 struct ofpbuf *odp_actions;
5103 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5104 odp_flow_key_from_flow(&key, flow);
5106 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
5107 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
5108 dpif_execute(ofproto->dpif, key.data, key.size,
5109 odp_actions->data, odp_actions->size, packet);
5110 ofpbuf_delete(odp_actions);
5116 get_netflow_ids(const struct ofproto *ofproto_,
5117 uint8_t *engine_type, uint8_t *engine_id)
5119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5121 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
5124 static struct ofproto_dpif *
5125 ofproto_dpif_lookup(const char *name)
5127 struct ofproto *ofproto = ofproto_lookup(name);
5128 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
5129 ? ofproto_dpif_cast(ofproto)
5134 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
5135 const char *args, void *aux OVS_UNUSED)
5137 struct ds ds = DS_EMPTY_INITIALIZER;
5138 const struct ofproto_dpif *ofproto;
5139 const struct mac_entry *e;
5141 ofproto = ofproto_dpif_lookup(args);
5143 unixctl_command_reply(conn, 501, "no such bridge");
5147 ds_put_cstr(&ds, " port VLAN MAC Age\n");
5148 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5149 struct ofbundle *bundle = e->port.p;
5150 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
5151 ofbundle_get_a_port(bundle)->odp_port,
5152 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
5154 unixctl_command_reply(conn, 200, ds_cstr(&ds));
5158 struct ofproto_trace {
5159 struct action_xlate_ctx ctx;
5165 trace_format_rule(struct ds *result, uint8_t table_id, int level,
5166 const struct rule_dpif *rule)
5168 ds_put_char_multiple(result, '\t', level);
5170 ds_put_cstr(result, "No match\n");
5174 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5175 table_id, ntohll(rule->up.flow_cookie));
5176 cls_rule_format(&rule->up.cr, result);
5177 ds_put_char(result, '\n');
5179 ds_put_char_multiple(result, '\t', level);
5180 ds_put_cstr(result, "OpenFlow ");
5181 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
5182 ds_put_char(result, '\n');
5186 trace_format_flow(struct ds *result, int level, const char *title,
5187 struct ofproto_trace *trace)
5189 ds_put_char_multiple(result, '\t', level);
5190 ds_put_format(result, "%s: ", title);
5191 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5192 ds_put_cstr(result, "unchanged");
5194 flow_format(result, &trace->ctx.flow);
5195 trace->flow = trace->ctx.flow;
5197 ds_put_char(result, '\n');
5201 trace_format_regs(struct ds *result, int level, const char *title,
5202 struct ofproto_trace *trace)
5206 ds_put_char_multiple(result, '\t', level);
5207 ds_put_format(result, "%s:", title);
5208 for (i = 0; i < FLOW_N_REGS; i++) {
5209 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5211 ds_put_char(result, '\n');
5215 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5217 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5218 struct ds *result = trace->result;
5220 ds_put_char(result, '\n');
5221 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5222 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
5223 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
5227 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5228 void *aux OVS_UNUSED)
5230 char *dpname, *arg1, *arg2, *arg3;
5231 char *args = xstrdup(args_);
5232 char *save_ptr = NULL;
5233 struct ofproto_dpif *ofproto;
5234 struct ofpbuf odp_key;
5235 struct ofpbuf *packet;
5236 struct rule_dpif *rule;
5242 ofpbuf_init(&odp_key, 0);
5245 dpname = strtok_r(args, " ", &save_ptr);
5246 arg1 = strtok_r(NULL, " ", &save_ptr);
5247 arg2 = strtok_r(NULL, " ", &save_ptr);
5248 arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5249 if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) {
5250 /* ofproto/trace dpname flow [-generate] */
5253 /* Convert string to datapath key. */
5254 ofpbuf_init(&odp_key, 0);
5255 error = odp_flow_key_from_string(arg1, &odp_key);
5257 unixctl_command_reply(conn, 501, "Bad flow syntax");
5261 /* Convert odp_key to flow. */
5262 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
5264 unixctl_command_reply(conn, 501, "Invalid flow");
5268 /* Generate a packet, if requested. */
5270 packet = ofpbuf_new(0);
5271 flow_compose(packet, &flow);
5273 } else if (dpname && arg1 && arg2 && arg3) {
5274 /* ofproto/trace dpname tun_id in_port packet */
5278 tun_id = htonll(strtoull(arg1, NULL, 0));
5279 in_port = ofp_port_to_odp_port(atoi(arg2));
5281 packet = ofpbuf_new(strlen(args) / 2);
5282 arg3 = ofpbuf_put_hex(packet, arg3, NULL);
5283 arg3 += strspn(arg3, " ");
5284 if (*arg3 != '\0') {
5285 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5288 if (packet->size < ETH_HEADER_LEN) {
5289 unixctl_command_reply(conn, 501,
5290 "Packet data too short for Ethernet");
5294 ds_put_cstr(&result, "Packet: ");
5295 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
5296 ds_put_cstr(&result, s);
5299 flow_extract(packet, tun_id, in_port, &flow);
5301 unixctl_command_reply(conn, 501, "Bad command syntax");
5305 ofproto = ofproto_dpif_lookup(dpname);
5307 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5312 ds_put_cstr(&result, "Flow: ");
5313 flow_format(&result, &flow);
5314 ds_put_char(&result, '\n');
5316 rule = rule_dpif_lookup(ofproto, &flow, 0);
5317 trace_format_rule(&result, 0, 0, rule);
5319 struct ofproto_trace trace;
5320 struct ofpbuf *odp_actions;
5322 trace.result = &result;
5324 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
5325 trace.ctx.resubmit_hook = trace_resubmit;
5326 odp_actions = xlate_actions(&trace.ctx,
5327 rule->up.actions, rule->up.n_actions);
5329 ds_put_char(&result, '\n');
5330 trace_format_flow(&result, 0, "Final flow", &trace);
5331 ds_put_cstr(&result, "Datapath actions: ");
5332 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5333 ofpbuf_delete(odp_actions);
5335 if (!trace.ctx.may_set_up_flow) {
5337 ds_put_cstr(&result, "\nThis flow is not cachable.");
5339 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
5340 "for complete actions, please supply a packet.");
5345 unixctl_command_reply(conn, 200, ds_cstr(&result));
5348 ds_destroy(&result);
5349 ofpbuf_delete(packet);
5350 ofpbuf_uninit(&odp_key);
5355 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
5356 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5359 unixctl_command_reply(conn, 200, NULL);
5363 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
5364 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
5367 unixctl_command_reply(conn, 200, NULL);
5371 ofproto_dpif_unixctl_init(void)
5373 static bool registered;
5379 unixctl_command_register("ofproto/trace",
5380 "bridge {tun_id in_port packet | odp_flow [-generate]}",
5381 ofproto_unixctl_trace, NULL);
5382 unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show,
5384 unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL);
5385 unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL);
5388 const struct ofproto_class ofproto_dpif_class = {
5415 port_is_lacp_current,
5416 NULL, /* rule_choose_table */
5423 rule_modify_actions,
5431 get_cfm_remote_mpids,
5435 get_stp_port_status,
5440 is_mirror_output_bundle,
5441 forward_bpdu_changed,