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"
35 #include "mac-learning.h"
36 #include "multipath.h"
43 #include "ofp-print.h"
44 #include "ofproto-dpif-sflow.h"
45 #include "poll-loop.h"
47 #include "unaligned.h"
49 #include "vlan-bitmap.h"
52 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
54 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
55 COVERAGE_DEFINE(ofproto_dpif_expired);
56 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
57 COVERAGE_DEFINE(ofproto_dpif_xlate);
58 COVERAGE_DEFINE(facet_changed_rule);
59 COVERAGE_DEFINE(facet_invalidated);
60 COVERAGE_DEFINE(facet_revalidate);
61 COVERAGE_DEFINE(facet_unexpected);
63 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
64 * flow translation. */
65 #define MAX_RESUBMIT_RECURSION 16
73 long long int used; /* Time last used; time created if not used. */
77 * - Do include packets and bytes from facets that have been deleted or
78 * whose own statistics have been folded into the rule.
80 * - Do include packets and bytes sent "by hand" that were accounted to
81 * the rule without any facet being involved (this is a rare corner
82 * case in rule_execute()).
84 * - Do not include packet or bytes that can be obtained from any facet's
85 * packet_count or byte_count member or that can be obtained from the
86 * datapath by, e.g., dpif_flow_get() for any facet.
88 uint64_t packet_count; /* Number of packets received. */
89 uint64_t byte_count; /* Number of bytes received. */
91 struct list facets; /* List of "struct facet"s. */
94 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
96 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
99 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *ofproto,
100 const struct flow *flow);
102 #define MAX_MIRRORS 32
103 typedef uint32_t mirror_mask_t;
104 #define MIRROR_MASK_C(X) UINT32_C(X)
105 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
107 struct ofproto_dpif *ofproto; /* Owning ofproto. */
108 size_t idx; /* In ofproto's "mirrors" array. */
109 void *aux; /* Key supplied by ofproto's client. */
110 char *name; /* Identifier for log messages. */
112 /* Selection criteria. */
113 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
114 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
115 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
117 /* Output (mutually exclusive). */
118 struct ofbundle *out; /* Output port or NULL. */
119 int out_vlan; /* Output VLAN or -1. */
122 static void mirror_destroy(struct ofmirror *);
124 /* A group of one or more OpenFlow ports. */
125 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
127 struct ofproto_dpif *ofproto; /* Owning ofproto. */
128 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
129 void *aux; /* Key supplied by ofproto's client. */
130 char *name; /* Identifier for log messages. */
133 struct list ports; /* Contains "struct ofport"s. */
134 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
135 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
136 * NULL if all VLANs are trunked. */
137 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
138 struct bond *bond; /* Nonnull iff more than one port. */
141 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
143 /* Port mirroring info. */
144 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
145 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
146 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
149 static void bundle_remove(struct ofport *);
150 static void bundle_destroy(struct ofbundle *);
151 static void bundle_del_port(struct ofport_dpif *);
152 static void bundle_run(struct ofbundle *);
153 static void bundle_wait(struct ofbundle *);
155 struct action_xlate_ctx {
156 /* action_xlate_ctx_init() initializes these members. */
159 struct ofproto_dpif *ofproto;
161 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
162 * this flow when actions change header fields. */
165 /* The packet corresponding to 'flow', or a null pointer if we are
166 * revalidating without a packet to refer to. */
167 const struct ofpbuf *packet;
169 /* If nonnull, called just before executing a resubmit action.
171 * This is normally null so the client has to set it manually after
172 * calling action_xlate_ctx_init(). */
173 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
175 /* xlate_actions() initializes and uses these members. The client might want
176 * to look at them after it returns. */
178 struct ofpbuf *odp_actions; /* Datapath actions. */
179 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
180 bool may_set_up_flow; /* True ordinarily; false if the actions must
181 * be reassessed for every packet. */
182 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
184 /* xlate_actions() initializes and uses these members, but the client has no
185 * reason to look at them. */
187 int recurse; /* Recursion level, via xlate_table_action. */
188 uint32_t priority; /* Current flow priority. 0 if none. */
189 struct flow base_flow; /* Flow at the last commit. */
190 uint32_t base_priority; /* Priority at the last commit. */
193 static void action_xlate_ctx_init(struct action_xlate_ctx *,
194 struct ofproto_dpif *, const struct flow *,
195 const struct ofpbuf *);
196 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
197 const union ofp_action *in, size_t n_in);
199 /* An exact-match instantiation of an OpenFlow flow. */
201 long long int used; /* Time last used; time created if not used. */
205 * - Do include packets and bytes sent "by hand", e.g. with
208 * - Do include packets and bytes that were obtained from the datapath
209 * when a flow was deleted (e.g. dpif_flow_del()) or when its
210 * statistics were reset (e.g. dpif_flow_put() with
211 * DPIF_FP_ZERO_STATS).
213 * - Do not include any packets or bytes that can currently be obtained
214 * from the datapath by, e.g., dpif_flow_get().
216 uint64_t packet_count; /* Number of packets received. */
217 uint64_t byte_count; /* Number of bytes received. */
219 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
220 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
222 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
223 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
224 long long int rs_used; /* Used time pushed to resubmit children. */
226 /* Number of bytes passed to account_cb. This may include bytes that can
227 * currently obtained from the datapath (thus, it can be greater than
229 uint64_t accounted_bytes;
231 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
232 struct list list_node; /* In owning rule's 'facets' list. */
233 struct rule_dpif *rule; /* Owning rule. */
234 struct flow flow; /* Exact-match flow. */
235 bool installed; /* Installed in datapath? */
236 bool may_install; /* True ordinarily; false if actions must
237 * be reassessed for every packet. */
238 size_t actions_len; /* Number of bytes in actions[]. */
239 struct nlattr *actions; /* Datapath actions. */
240 tag_type tags; /* Tags. */
241 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
244 static struct facet *facet_create(struct rule_dpif *, const struct flow *,
245 const struct ofpbuf *packet);
246 static void facet_remove(struct ofproto_dpif *, struct facet *);
247 static void facet_free(struct facet *);
249 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
250 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
251 const struct flow *);
252 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
254 static void facet_execute(struct ofproto_dpif *, struct facet *,
255 struct ofpbuf *packet);
257 static int facet_put__(struct ofproto_dpif *, struct facet *,
258 const struct nlattr *actions, size_t actions_len,
259 struct dpif_flow_stats *);
260 static void facet_install(struct ofproto_dpif *, struct facet *,
262 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
263 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
265 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
266 const struct ofpbuf *packet);
267 static void facet_update_time(struct ofproto_dpif *, struct facet *,
269 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
270 const struct dpif_flow_stats *);
271 static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *);
272 static void facet_push_stats(struct facet *);
273 static void facet_account(struct ofproto_dpif *, struct facet *,
274 uint64_t extra_bytes);
276 static bool facet_is_controller_flow(struct facet *);
278 static void flow_push_stats(const struct rule_dpif *,
279 struct flow *, uint64_t packets, uint64_t bytes,
286 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
287 struct list bundle_node; /* In struct ofbundle's "ports" list. */
288 struct cfm *cfm; /* Connectivity Fault Management, if any. */
289 tag_type tag; /* Tag associated with this port. */
290 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
291 bool may_enable; /* May be enabled in bonds. */
294 static struct ofport_dpif *
295 ofport_dpif_cast(const struct ofport *ofport)
297 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
298 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
301 static void port_run(struct ofport_dpif *);
302 static void port_wait(struct ofport_dpif *);
303 static int set_cfm(struct ofport *, const struct cfm_settings *);
305 struct dpif_completion {
306 struct list list_node;
307 struct ofoperation *op;
310 struct ofproto_dpif {
319 struct netflow *netflow;
320 struct dpif_sflow *sflow;
321 struct hmap bundles; /* Contains "struct ofbundle"s. */
322 struct mac_learning *ml;
323 struct ofmirror *mirrors[MAX_MIRRORS];
324 bool has_bonded_bundles;
327 struct timer next_expiration;
331 bool need_revalidate;
332 struct tag_set revalidate_set;
334 /* Support for debugging async flow mods. */
335 struct list completions;
337 bool has_bundle_action; /* True when the first bundle action appears. */
340 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
341 * for debugging the asynchronous flow_mod implementation.) */
344 static void ofproto_dpif_unixctl_init(void);
346 static struct ofproto_dpif *
347 ofproto_dpif_cast(const struct ofproto *ofproto)
349 assert(ofproto->ofproto_class == &ofproto_dpif_class);
350 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
353 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
355 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
358 /* Packet processing. */
359 static void update_learning_table(struct ofproto_dpif *,
360 const struct flow *, int vlan,
362 static bool is_admissible(struct ofproto_dpif *, const struct flow *,
363 bool have_packet, tag_type *, int *vlanp,
364 struct ofbundle **in_bundlep);
365 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
367 /* Flow expiration. */
368 static int expire(struct ofproto_dpif *);
371 static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
372 const struct ofpbuf *packet);
374 /* Global variables. */
375 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
377 /* Factory functions. */
380 enumerate_types(struct sset *types)
382 dp_enumerate_types(types);
386 enumerate_names(const char *type, struct sset *names)
388 return dp_enumerate_names(type, names);
392 del(const char *type, const char *name)
397 error = dpif_open(name, type, &dpif);
399 error = dpif_delete(dpif);
405 /* Basic life-cycle. */
407 static struct ofproto *
410 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
415 dealloc(struct ofproto *ofproto_)
417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
422 construct(struct ofproto *ofproto_, int *n_tablesp)
424 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
425 const char *name = ofproto->up.name;
429 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
431 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
435 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
436 ofproto->n_matches = 0;
438 error = dpif_recv_set_mask(ofproto->dpif,
439 ((1u << DPIF_UC_MISS) |
440 (1u << DPIF_UC_ACTION) |
441 (1u << DPIF_UC_SAMPLE)));
443 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
444 dpif_close(ofproto->dpif);
447 dpif_flow_flush(ofproto->dpif);
448 dpif_recv_purge(ofproto->dpif);
450 ofproto->netflow = NULL;
451 ofproto->sflow = NULL;
452 hmap_init(&ofproto->bundles);
453 ofproto->ml = mac_learning_create();
454 for (i = 0; i < MAX_MIRRORS; i++) {
455 ofproto->mirrors[i] = NULL;
457 ofproto->has_bonded_bundles = false;
459 timer_set_duration(&ofproto->next_expiration, 1000);
461 hmap_init(&ofproto->facets);
462 ofproto->need_revalidate = false;
463 tag_set_init(&ofproto->revalidate_set);
465 list_init(&ofproto->completions);
467 ofproto_dpif_unixctl_init();
469 ofproto->has_bundle_action = false;
476 complete_operations(struct ofproto_dpif *ofproto)
478 struct dpif_completion *c, *next;
480 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
481 ofoperation_complete(c->op, 0);
482 list_remove(&c->list_node);
488 destruct(struct ofproto *ofproto_)
490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
491 struct rule_dpif *rule, *next_rule;
492 struct classifier *table;
495 complete_operations(ofproto);
497 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
498 struct cls_cursor cursor;
500 cls_cursor_init(&cursor, table, NULL);
501 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
502 ofproto_rule_destroy(&rule->up);
506 for (i = 0; i < MAX_MIRRORS; i++) {
507 mirror_destroy(ofproto->mirrors[i]);
510 netflow_destroy(ofproto->netflow);
511 dpif_sflow_destroy(ofproto->sflow);
512 hmap_destroy(&ofproto->bundles);
513 mac_learning_destroy(ofproto->ml);
515 hmap_destroy(&ofproto->facets);
517 dpif_close(ofproto->dpif);
521 run(struct ofproto *ofproto_)
523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
524 struct ofport_dpif *ofport;
525 struct ofbundle *bundle;
529 complete_operations(ofproto);
531 dpif_run(ofproto->dpif);
533 for (i = 0; i < 50; i++) {
534 struct dpif_upcall packet;
537 error = dpif_recv(ofproto->dpif, &packet);
539 if (error == ENODEV) {
540 /* Datapath destroyed. */
546 handle_upcall(ofproto, &packet);
549 if (timer_expired(&ofproto->next_expiration)) {
550 int delay = expire(ofproto);
551 timer_set_duration(&ofproto->next_expiration, delay);
554 if (ofproto->netflow) {
555 netflow_run(ofproto->netflow);
557 if (ofproto->sflow) {
558 dpif_sflow_run(ofproto->sflow);
561 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
564 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
568 /* Now revalidate if there's anything to do. */
569 if (ofproto->need_revalidate
570 || !tag_set_is_empty(&ofproto->revalidate_set)) {
571 struct tag_set revalidate_set = ofproto->revalidate_set;
572 bool revalidate_all = ofproto->need_revalidate;
573 struct facet *facet, *next;
575 /* Clear the revalidation flags. */
576 tag_set_init(&ofproto->revalidate_set);
577 ofproto->need_revalidate = false;
579 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
581 || tag_set_intersects(&revalidate_set, facet->tags)) {
582 facet_revalidate(ofproto, facet);
591 wait(struct ofproto *ofproto_)
593 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
594 struct ofport_dpif *ofport;
595 struct ofbundle *bundle;
597 if (!clogged && !list_is_empty(&ofproto->completions)) {
598 poll_immediate_wake();
601 dpif_wait(ofproto->dpif);
602 dpif_recv_wait(ofproto->dpif);
603 if (ofproto->sflow) {
604 dpif_sflow_wait(ofproto->sflow);
606 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
607 poll_immediate_wake();
609 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
612 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
615 if (ofproto->need_revalidate) {
616 /* Shouldn't happen, but if it does just go around again. */
617 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
618 poll_immediate_wake();
620 timer_wait(&ofproto->next_expiration);
625 flush(struct ofproto *ofproto_)
627 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
628 struct facet *facet, *next_facet;
630 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
631 /* Mark the facet as not installed so that facet_remove() doesn't
632 * bother trying to uninstall it. There is no point in uninstalling it
633 * individually since we are about to blow away all the facets with
634 * dpif_flow_flush(). */
635 facet->installed = false;
636 facet->dp_packet_count = 0;
637 facet->dp_byte_count = 0;
638 facet_remove(ofproto, facet);
640 dpif_flow_flush(ofproto->dpif);
644 get_features(struct ofproto *ofproto_ OVS_UNUSED,
645 bool *arp_match_ip, uint32_t *actions)
647 *arp_match_ip = true;
648 *actions = ((1u << OFPAT_OUTPUT) |
649 (1u << OFPAT_SET_VLAN_VID) |
650 (1u << OFPAT_SET_VLAN_PCP) |
651 (1u << OFPAT_STRIP_VLAN) |
652 (1u << OFPAT_SET_DL_SRC) |
653 (1u << OFPAT_SET_DL_DST) |
654 (1u << OFPAT_SET_NW_SRC) |
655 (1u << OFPAT_SET_NW_DST) |
656 (1u << OFPAT_SET_NW_TOS) |
657 (1u << OFPAT_SET_TP_SRC) |
658 (1u << OFPAT_SET_TP_DST) |
659 (1u << OFPAT_ENQUEUE));
663 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
665 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
668 strcpy(ots->name, "classifier");
670 dpif_get_dp_stats(ofproto->dpif, &s);
671 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
672 put_32aligned_be64(&ots->matched_count,
673 htonll(s.n_hit + ofproto->n_matches));
677 set_netflow(struct ofproto *ofproto_,
678 const struct netflow_options *netflow_options)
680 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
682 if (netflow_options) {
683 if (!ofproto->netflow) {
684 ofproto->netflow = netflow_create();
686 return netflow_set_options(ofproto->netflow, netflow_options);
688 netflow_destroy(ofproto->netflow);
689 ofproto->netflow = NULL;
694 static struct ofport *
697 struct ofport_dpif *port = xmalloc(sizeof *port);
702 port_dealloc(struct ofport *port_)
704 struct ofport_dpif *port = ofport_dpif_cast(port_);
709 port_construct(struct ofport *port_)
711 struct ofport_dpif *port = ofport_dpif_cast(port_);
712 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
714 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
717 port->tag = tag_create_random();
718 port->may_enable = true;
720 if (ofproto->sflow) {
721 dpif_sflow_add_port(ofproto->sflow, port->odp_port,
722 netdev_get_name(port->up.netdev));
729 port_destruct(struct ofport *port_)
731 struct ofport_dpif *port = ofport_dpif_cast(port_);
732 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
734 bundle_remove(port_);
735 set_cfm(port_, NULL);
736 if (ofproto->sflow) {
737 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
742 port_modified(struct ofport *port_)
744 struct ofport_dpif *port = ofport_dpif_cast(port_);
746 if (port->bundle && port->bundle->bond) {
747 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
752 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
754 struct ofport_dpif *port = ofport_dpif_cast(port_);
755 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
756 ovs_be32 changed = old_config ^ port->up.opp.config;
758 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
759 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
760 ofproto->need_revalidate = true;
765 set_sflow(struct ofproto *ofproto_,
766 const struct ofproto_sflow_options *sflow_options)
768 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
769 struct dpif_sflow *ds = ofproto->sflow;
772 struct ofport_dpif *ofport;
774 ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif);
775 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
776 dpif_sflow_add_port(ds, ofport->odp_port,
777 netdev_get_name(ofport->up.netdev));
780 dpif_sflow_set_options(ds, sflow_options);
782 dpif_sflow_destroy(ds);
783 ofproto->sflow = NULL;
789 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
791 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
798 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
801 if (cfm_configure(ofport->cfm, s)) {
807 cfm_destroy(ofport->cfm);
813 get_cfm_fault(const struct ofport *ofport_)
815 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
817 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
822 /* Expires all MAC learning entries associated with 'port' and forces ofproto
823 * to revalidate every flow. */
825 bundle_flush_macs(struct ofbundle *bundle)
827 struct ofproto_dpif *ofproto = bundle->ofproto;
828 struct mac_learning *ml = ofproto->ml;
829 struct mac_entry *mac, *next_mac;
831 ofproto->need_revalidate = true;
832 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
833 if (mac->port.p == bundle) {
834 mac_learning_expire(ml, mac);
839 static struct ofbundle *
840 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
842 struct ofbundle *bundle;
844 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
846 if (bundle->aux == aux) {
853 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
854 * ones that are found to 'bundles'. */
856 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
857 void **auxes, size_t n_auxes,
858 struct hmapx *bundles)
863 for (i = 0; i < n_auxes; i++) {
864 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
866 hmapx_add(bundles, bundle);
872 bundle_del_port(struct ofport_dpif *port)
874 struct ofbundle *bundle = port->bundle;
876 bundle->ofproto->need_revalidate = true;
878 list_remove(&port->bundle_node);
882 lacp_slave_unregister(bundle->lacp, port);
885 bond_slave_unregister(bundle->bond, port);
888 bundle->floodable = true;
889 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
890 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
891 bundle->floodable = false;
897 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
898 struct lacp_slave_settings *lacp,
899 uint32_t bond_stable_id)
901 struct ofport_dpif *port;
903 port = get_ofp_port(bundle->ofproto, ofp_port);
908 if (port->bundle != bundle) {
909 bundle->ofproto->need_revalidate = true;
911 bundle_del_port(port);
914 port->bundle = bundle;
915 list_push_back(&bundle->ports, &port->bundle_node);
916 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
917 bundle->floodable = false;
921 lacp_slave_register(bundle->lacp, port, lacp);
924 port->bond_stable_id = bond_stable_id;
930 bundle_destroy(struct ofbundle *bundle)
932 struct ofproto_dpif *ofproto;
933 struct ofport_dpif *port, *next_port;
940 ofproto = bundle->ofproto;
941 for (i = 0; i < MAX_MIRRORS; i++) {
942 struct ofmirror *m = ofproto->mirrors[i];
944 if (m->out == bundle) {
946 } else if (hmapx_find_and_delete(&m->srcs, bundle)
947 || hmapx_find_and_delete(&m->dsts, bundle)) {
948 ofproto->need_revalidate = true;
953 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
954 bundle_del_port(port);
957 bundle_flush_macs(bundle);
958 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
960 free(bundle->trunks);
961 lacp_destroy(bundle->lacp);
962 bond_destroy(bundle->bond);
967 bundle_set(struct ofproto *ofproto_, void *aux,
968 const struct ofproto_bundle_settings *s)
970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
971 bool need_flush = false;
972 const unsigned long *trunks;
973 struct ofport_dpif *port;
974 struct ofbundle *bundle;
979 bundle_destroy(bundle_lookup(ofproto, aux));
983 assert(s->n_slaves == 1 || s->bond != NULL);
984 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
986 bundle = bundle_lookup(ofproto, aux);
988 bundle = xmalloc(sizeof *bundle);
990 bundle->ofproto = ofproto;
991 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
992 hash_pointer(aux, 0));
996 list_init(&bundle->ports);
998 bundle->trunks = NULL;
1000 bundle->bond = NULL;
1002 bundle->floodable = true;
1004 bundle->src_mirrors = 0;
1005 bundle->dst_mirrors = 0;
1006 bundle->mirror_out = 0;
1009 if (!bundle->name || strcmp(s->name, bundle->name)) {
1011 bundle->name = xstrdup(s->name);
1016 if (!bundle->lacp) {
1017 bundle->lacp = lacp_create();
1019 lacp_configure(bundle->lacp, s->lacp);
1021 lacp_destroy(bundle->lacp);
1022 bundle->lacp = NULL;
1025 /* Update set of ports. */
1027 for (i = 0; i < s->n_slaves; i++) {
1028 if (!bundle_add_port(bundle, s->slaves[i],
1029 s->lacp ? &s->lacp_slaves[i] : NULL,
1030 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
1034 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
1035 struct ofport_dpif *next_port;
1037 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
1038 for (i = 0; i < s->n_slaves; i++) {
1039 if (s->slaves[i] == port->up.ofp_port) {
1044 bundle_del_port(port);
1048 assert(list_size(&bundle->ports) <= s->n_slaves);
1050 if (list_is_empty(&bundle->ports)) {
1051 bundle_destroy(bundle);
1056 if (s->vlan != bundle->vlan) {
1057 bundle->vlan = s->vlan;
1061 /* Get trunked VLANs. */
1062 trunks = s->vlan == -1 ? NULL : s->trunks;
1063 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1064 free(bundle->trunks);
1065 bundle->trunks = vlan_bitmap_clone(trunks);
1070 if (!list_is_short(&bundle->ports)) {
1071 bundle->ofproto->has_bonded_bundles = true;
1073 if (bond_reconfigure(bundle->bond, s->bond)) {
1074 ofproto->need_revalidate = true;
1077 bundle->bond = bond_create(s->bond);
1078 ofproto->need_revalidate = true;
1081 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1082 bond_slave_register(bundle->bond, port, port->bond_stable_id,
1086 bond_destroy(bundle->bond);
1087 bundle->bond = NULL;
1090 /* If we changed something that would affect MAC learning, un-learn
1091 * everything on this port and force flow revalidation. */
1093 bundle_flush_macs(bundle);
1100 bundle_remove(struct ofport *port_)
1102 struct ofport_dpif *port = ofport_dpif_cast(port_);
1103 struct ofbundle *bundle = port->bundle;
1106 bundle_del_port(port);
1107 if (list_is_empty(&bundle->ports)) {
1108 bundle_destroy(bundle);
1109 } else if (list_is_short(&bundle->ports)) {
1110 bond_destroy(bundle->bond);
1111 bundle->bond = NULL;
1117 send_pdu_cb(void *port_, const struct lacp_pdu *pdu)
1119 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1120 struct ofport_dpif *port = port_;
1121 uint8_t ea[ETH_ADDR_LEN];
1124 error = netdev_get_etheraddr(port->up.netdev, ea);
1126 struct lacp_pdu *packet_pdu;
1127 struct ofpbuf packet;
1129 ofpbuf_init(&packet, 0);
1130 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1131 sizeof *packet_pdu);
1133 error = netdev_send(port->up.netdev, &packet);
1135 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1136 "(%s)", port->bundle->name,
1137 netdev_get_name(port->up.netdev), strerror(error));
1139 ofpbuf_uninit(&packet);
1141 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1142 "%s (%s)", port->bundle->name,
1143 netdev_get_name(port->up.netdev), strerror(error));
1148 bundle_send_learning_packets(struct ofbundle *bundle)
1150 struct ofproto_dpif *ofproto = bundle->ofproto;
1151 int error, n_packets, n_errors;
1152 struct mac_entry *e;
1154 error = n_packets = n_errors = 0;
1155 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1156 if (e->port.p != bundle) {
1157 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1167 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1168 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1169 "packets, last error was: %s",
1170 bundle->name, n_errors, n_packets, strerror(error));
1172 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1173 bundle->name, n_packets);
1178 bundle_run(struct ofbundle *bundle)
1181 lacp_run(bundle->lacp, send_pdu_cb);
1184 struct ofport_dpif *port;
1186 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1187 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
1190 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1191 lacp_negotiated(bundle->lacp));
1192 if (bond_should_send_learning_packets(bundle->bond)) {
1193 bundle_send_learning_packets(bundle);
1199 bundle_wait(struct ofbundle *bundle)
1202 lacp_wait(bundle->lacp);
1205 bond_wait(bundle->bond);
1212 mirror_scan(struct ofproto_dpif *ofproto)
1216 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1217 if (!ofproto->mirrors[idx]) {
1224 static struct ofmirror *
1225 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1229 for (i = 0; i < MAX_MIRRORS; i++) {
1230 struct ofmirror *mirror = ofproto->mirrors[i];
1231 if (mirror && mirror->aux == aux) {
1240 mirror_set(struct ofproto *ofproto_, void *aux,
1241 const struct ofproto_mirror_settings *s)
1243 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1244 mirror_mask_t mirror_bit;
1245 struct ofbundle *bundle;
1246 struct ofmirror *mirror;
1247 struct ofbundle *out;
1248 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1249 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1252 mirror = mirror_lookup(ofproto, aux);
1254 mirror_destroy(mirror);
1260 idx = mirror_scan(ofproto);
1262 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1264 ofproto->up.name, MAX_MIRRORS, s->name);
1268 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1269 mirror->ofproto = ofproto;
1272 mirror->out_vlan = -1;
1273 mirror->name = NULL;
1276 if (!mirror->name || strcmp(s->name, mirror->name)) {
1278 mirror->name = xstrdup(s->name);
1281 /* Get the new configuration. */
1282 if (s->out_bundle) {
1283 out = bundle_lookup(ofproto, s->out_bundle);
1285 mirror_destroy(mirror);
1291 out_vlan = s->out_vlan;
1293 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1294 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1296 /* If the configuration has not changed, do nothing. */
1297 if (hmapx_equals(&srcs, &mirror->srcs)
1298 && hmapx_equals(&dsts, &mirror->dsts)
1299 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1300 && mirror->out == out
1301 && mirror->out_vlan == out_vlan)
1303 hmapx_destroy(&srcs);
1304 hmapx_destroy(&dsts);
1308 hmapx_swap(&srcs, &mirror->srcs);
1309 hmapx_destroy(&srcs);
1311 hmapx_swap(&dsts, &mirror->dsts);
1312 hmapx_destroy(&dsts);
1314 free(mirror->vlans);
1315 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1318 mirror->out_vlan = out_vlan;
1320 /* Update bundles. */
1321 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1322 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1323 if (hmapx_contains(&mirror->srcs, bundle)) {
1324 bundle->src_mirrors |= mirror_bit;
1326 bundle->src_mirrors &= ~mirror_bit;
1329 if (hmapx_contains(&mirror->dsts, bundle)) {
1330 bundle->dst_mirrors |= mirror_bit;
1332 bundle->dst_mirrors &= ~mirror_bit;
1335 if (mirror->out == bundle) {
1336 bundle->mirror_out |= mirror_bit;
1338 bundle->mirror_out &= ~mirror_bit;
1342 ofproto->need_revalidate = true;
1343 mac_learning_flush(ofproto->ml);
1349 mirror_destroy(struct ofmirror *mirror)
1351 struct ofproto_dpif *ofproto;
1352 mirror_mask_t mirror_bit;
1353 struct ofbundle *bundle;
1359 ofproto = mirror->ofproto;
1360 ofproto->need_revalidate = true;
1361 mac_learning_flush(ofproto->ml);
1363 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1364 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1365 bundle->src_mirrors &= ~mirror_bit;
1366 bundle->dst_mirrors &= ~mirror_bit;
1367 bundle->mirror_out &= ~mirror_bit;
1370 hmapx_destroy(&mirror->srcs);
1371 hmapx_destroy(&mirror->dsts);
1372 free(mirror->vlans);
1374 ofproto->mirrors[mirror->idx] = NULL;
1380 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1382 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1383 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1384 ofproto->need_revalidate = true;
1385 mac_learning_flush(ofproto->ml);
1391 is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
1393 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1394 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1395 return bundle && bundle->mirror_out != 0;
1400 static struct ofport_dpif *
1401 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1403 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
1404 return ofport ? ofport_dpif_cast(ofport) : NULL;
1407 static struct ofport_dpif *
1408 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1410 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1414 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1415 struct dpif_port *dpif_port)
1417 ofproto_port->name = dpif_port->name;
1418 ofproto_port->type = dpif_port->type;
1419 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1423 port_run(struct ofport_dpif *ofport)
1425 bool enable = netdev_get_carrier(ofport->up.netdev);
1428 cfm_run(ofport->cfm);
1430 if (cfm_should_send_ccm(ofport->cfm)) {
1431 struct ofpbuf packet;
1433 ofpbuf_init(&packet, 0);
1434 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr);
1435 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1436 ofport->odp_port, &packet);
1437 ofpbuf_uninit(&packet);
1440 enable = enable && !cfm_get_fault(ofport->cfm);
1443 if (ofport->bundle) {
1444 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
1447 if (ofport->may_enable != enable) {
1448 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1450 if (ofproto->has_bundle_action) {
1451 ofproto->need_revalidate = true;
1455 ofport->may_enable = enable;
1459 port_wait(struct ofport_dpif *ofport)
1462 cfm_wait(ofport->cfm);
1467 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1468 struct ofproto_port *ofproto_port)
1470 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1471 struct dpif_port dpif_port;
1474 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1476 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1482 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1484 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1488 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1490 *ofp_portp = odp_port_to_ofp_port(odp_port);
1496 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1501 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1503 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1505 /* The caller is going to close ofport->up.netdev. If this is a
1506 * bonded port, then the bond is using that netdev, so remove it
1507 * from the bond. The client will need to reconfigure everything
1508 * after deleting ports, so then the slave will get re-added. */
1509 bundle_remove(&ofport->up);
1515 struct port_dump_state {
1516 struct dpif_port_dump dump;
1521 port_dump_start(const struct ofproto *ofproto_, void **statep)
1523 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1524 struct port_dump_state *state;
1526 *statep = state = xmalloc(sizeof *state);
1527 dpif_port_dump_start(&state->dump, ofproto->dpif);
1528 state->done = false;
1533 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1534 struct ofproto_port *port)
1536 struct port_dump_state *state = state_;
1537 struct dpif_port dpif_port;
1539 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1540 ofproto_port_from_dpif_port(port, &dpif_port);
1543 int error = dpif_port_dump_done(&state->dump);
1545 return error ? error : EOF;
1550 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1552 struct port_dump_state *state = state_;
1555 dpif_port_dump_done(&state->dump);
1562 port_poll(const struct ofproto *ofproto_, char **devnamep)
1564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1565 return dpif_port_poll(ofproto->dpif, devnamep);
1569 port_poll_wait(const struct ofproto *ofproto_)
1571 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1572 dpif_port_poll_wait(ofproto->dpif);
1576 port_is_lacp_current(const struct ofport *ofport_)
1578 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1579 return (ofport->bundle && ofport->bundle->lacp
1580 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
1584 /* Upcall handling. */
1586 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
1587 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
1588 * their individual configurations.
1590 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
1591 * Otherwise, ownership is transferred to this function. */
1593 send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
1594 const struct flow *flow, bool clone)
1596 struct ofputil_packet_in pin;
1598 pin.packet = upcall->packet;
1599 pin.in_port = flow->in_port;
1600 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
1601 pin.buffer_id = 0; /* not yet known */
1602 pin.send_len = upcall->userdata;
1603 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
1604 clone ? NULL : upcall->packet);
1608 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
1609 const struct ofpbuf *packet)
1611 if (cfm_should_process_flow(flow)) {
1612 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
1613 if (packet && ofport && ofport->cfm) {
1614 cfm_process_heartbeat(ofport->cfm, packet);
1617 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
1618 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
1619 if (packet && port && port->bundle && port->bundle->lacp) {
1620 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
1622 lacp_process_pdu(port->bundle->lacp, port, pdu);
1631 handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1633 struct facet *facet;
1636 /* Obtain in_port and tun_id, at least. */
1637 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1639 /* Set header pointers in 'flow'. */
1640 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
1642 /* Handle 802.1ag and LACP. */
1643 if (process_special(ofproto, &flow, upcall->packet)) {
1644 ofpbuf_delete(upcall->packet);
1645 ofproto->n_matches++;
1649 /* Check with in-band control to see if this packet should be sent
1650 * to the local port regardless of the flow table. */
1651 if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
1652 send_packet(ofproto, ODPP_LOCAL, upcall->packet);
1655 facet = facet_lookup_valid(ofproto, &flow);
1657 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow);
1659 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
1660 struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
1662 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
1663 COVERAGE_INC(ofproto_dpif_no_packet_in);
1664 /* XXX install 'drop' flow entry */
1665 ofpbuf_delete(upcall->packet);
1669 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
1673 send_packet_in(ofproto, upcall, &flow, false);
1677 facet = facet_create(rule, &flow, upcall->packet);
1678 } else if (!facet->may_install) {
1679 /* The facet is not installable, that is, we need to process every
1680 * packet, so process the current packet's actions into 'facet'. */
1681 facet_make_actions(ofproto, facet, upcall->packet);
1684 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
1686 * Extra-special case for fail-open mode.
1688 * We are in fail-open mode and the packet matched the fail-open rule,
1689 * but we are connected to a controller too. We should send the packet
1690 * up to the controller in the hope that it will try to set up a flow
1691 * and thereby allow us to exit fail-open.
1693 * See the top-level comment in fail-open.c for more information.
1695 send_packet_in(ofproto, upcall, &flow, true);
1698 facet_execute(ofproto, facet, upcall->packet);
1699 facet_install(ofproto, facet, false);
1700 ofproto->n_matches++;
1704 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1708 switch (upcall->type) {
1709 case DPIF_UC_ACTION:
1710 COVERAGE_INC(ofproto_dpif_ctlr_action);
1711 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1712 send_packet_in(ofproto, upcall, &flow, false);
1715 case DPIF_UC_SAMPLE:
1716 if (ofproto->sflow) {
1717 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1718 dpif_sflow_received(ofproto->sflow, upcall, &flow);
1720 ofpbuf_delete(upcall->packet);
1724 handle_miss_upcall(ofproto, upcall);
1727 case DPIF_N_UC_TYPES:
1729 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
1734 /* Flow expiration. */
1736 static int facet_max_idle(const struct ofproto_dpif *);
1737 static void update_stats(struct ofproto_dpif *);
1738 static void rule_expire(struct rule_dpif *);
1739 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
1741 /* This function is called periodically by run(). Its job is to collect
1742 * updates for the flows that have been installed into the datapath, most
1743 * importantly when they last were used, and then use that information to
1744 * expire flows that have not been used recently.
1746 * Returns the number of milliseconds after which it should be called again. */
1748 expire(struct ofproto_dpif *ofproto)
1750 struct rule_dpif *rule, *next_rule;
1751 struct classifier *table;
1754 /* Update stats for each flow in the datapath. */
1755 update_stats(ofproto);
1757 /* Expire facets that have been idle too long. */
1758 dp_max_idle = facet_max_idle(ofproto);
1759 expire_facets(ofproto, dp_max_idle);
1761 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
1762 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1763 struct cls_cursor cursor;
1765 cls_cursor_init(&cursor, table, NULL);
1766 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1771 /* All outstanding data in existing flows has been accounted, so it's a
1772 * good time to do bond rebalancing. */
1773 if (ofproto->has_bonded_bundles) {
1774 struct ofbundle *bundle;
1776 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1778 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
1783 return MIN(dp_max_idle, 1000);
1786 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
1788 * This function also pushes statistics updates to rules which each facet
1789 * resubmits into. Generally these statistics will be accurate. However, if a
1790 * facet changes the rule it resubmits into at some time in between
1791 * update_stats() runs, it is possible that statistics accrued to the
1792 * old rule will be incorrectly attributed to the new rule. This could be
1793 * avoided by calling update_stats() whenever rules are created or
1794 * deleted. However, the performance impact of making so many calls to the
1795 * datapath do not justify the benefit of having perfectly accurate statistics.
1798 update_stats(struct ofproto_dpif *p)
1800 const struct dpif_flow_stats *stats;
1801 struct dpif_flow_dump dump;
1802 const struct nlattr *key;
1805 dpif_flow_dump_start(&dump, p->dpif);
1806 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
1807 struct facet *facet;
1810 if (odp_flow_key_to_flow(key, key_len, &flow)) {
1814 odp_flow_key_format(key, key_len, &s);
1815 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
1821 facet = facet_find(p, &flow);
1823 if (facet && facet->installed) {
1825 if (stats->n_packets >= facet->dp_packet_count) {
1826 uint64_t extra = stats->n_packets - facet->dp_packet_count;
1827 facet->packet_count += extra;
1829 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
1832 if (stats->n_bytes >= facet->dp_byte_count) {
1833 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
1835 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
1838 facet->dp_packet_count = stats->n_packets;
1839 facet->dp_byte_count = stats->n_bytes;
1841 facet_update_time(p, facet, stats->used);
1842 facet_account(p, facet, stats->n_bytes);
1843 facet_push_stats(facet);
1845 /* There's a flow in the datapath that we know nothing about.
1847 COVERAGE_INC(facet_unexpected);
1848 dpif_flow_del(p->dpif, key, key_len, NULL);
1851 dpif_flow_dump_done(&dump);
1854 /* Calculates and returns the number of milliseconds of idle time after which
1855 * facets should expire from the datapath and we should fold their statistics
1856 * into their parent rules in userspace. */
1858 facet_max_idle(const struct ofproto_dpif *ofproto)
1861 * Idle time histogram.
1863 * Most of the time a switch has a relatively small number of facets. When
1864 * this is the case we might as well keep statistics for all of them in
1865 * userspace and to cache them in the kernel datapath for performance as
1868 * As the number of facets increases, the memory required to maintain
1869 * statistics about them in userspace and in the kernel becomes
1870 * significant. However, with a large number of facets it is likely that
1871 * only a few of them are "heavy hitters" that consume a large amount of
1872 * bandwidth. At this point, only heavy hitters are worth caching in the
1873 * kernel and maintaining in userspaces; other facets we can discard.
1875 * The technique used to compute the idle time is to build a histogram with
1876 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
1877 * that is installed in the kernel gets dropped in the appropriate bucket.
1878 * After the histogram has been built, we compute the cutoff so that only
1879 * the most-recently-used 1% of facets (but at least
1880 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
1881 * the most-recently-used bucket of facets is kept, so actually an
1882 * arbitrary number of facets can be kept in any given expiration run
1883 * (though the next run will delete most of those unless they receive
1886 * This requires a second pass through the facets, in addition to the pass
1887 * made by update_stats(), because the former function never looks
1888 * at uninstallable facets.
1890 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
1891 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
1892 int buckets[N_BUCKETS] = { 0 };
1893 int total, subtotal, bucket;
1894 struct facet *facet;
1898 total = hmap_count(&ofproto->facets);
1899 if (total <= ofproto->up.flow_eviction_threshold) {
1900 return N_BUCKETS * BUCKET_WIDTH;
1903 /* Build histogram. */
1905 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1906 long long int idle = now - facet->used;
1907 int bucket = (idle <= 0 ? 0
1908 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
1909 : (unsigned int) idle / BUCKET_WIDTH);
1913 /* Find the first bucket whose flows should be expired. */
1914 subtotal = bucket = 0;
1916 subtotal += buckets[bucket++];
1917 } while (bucket < N_BUCKETS &&
1918 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
1920 if (VLOG_IS_DBG_ENABLED()) {
1924 ds_put_cstr(&s, "keep");
1925 for (i = 0; i < N_BUCKETS; i++) {
1927 ds_put_cstr(&s, ", drop");
1930 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
1933 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
1937 return bucket * BUCKET_WIDTH;
1941 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
1943 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
1944 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
1945 struct ofexpired expired;
1947 if (facet->installed) {
1948 struct dpif_flow_stats stats;
1950 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
1952 facet_update_stats(ofproto, facet, &stats);
1955 expired.flow = facet->flow;
1956 expired.packet_count = facet->packet_count;
1957 expired.byte_count = facet->byte_count;
1958 expired.used = facet->used;
1959 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1964 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
1966 long long int cutoff = time_msec() - dp_max_idle;
1967 struct facet *facet, *next_facet;
1969 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1970 facet_active_timeout(ofproto, facet);
1971 if (facet->used < cutoff) {
1972 facet_remove(ofproto, facet);
1977 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
1978 * then delete it entirely. */
1980 rule_expire(struct rule_dpif *rule)
1982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
1983 struct facet *facet, *next_facet;
1987 /* Has 'rule' expired? */
1989 if (rule->up.hard_timeout
1990 && now > rule->up.created + rule->up.hard_timeout * 1000) {
1991 reason = OFPRR_HARD_TIMEOUT;
1992 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
1993 && now > rule->used + rule->up.idle_timeout * 1000) {
1994 reason = OFPRR_IDLE_TIMEOUT;
1999 COVERAGE_INC(ofproto_dpif_expired);
2001 /* Update stats. (This is a no-op if the rule expired due to an idle
2002 * timeout, because that only happens when the rule has no facets left.) */
2003 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2004 facet_remove(ofproto, facet);
2007 /* Get rid of the rule. */
2008 ofproto_rule_expire(&rule->up, reason);
2013 /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
2014 * example 'packet' within that flow.
2016 * The caller must already have determined that no facet with an identical
2017 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2018 * the ofproto's classifier table. */
2019 static struct facet *
2020 facet_create(struct rule_dpif *rule, const struct flow *flow,
2021 const struct ofpbuf *packet)
2023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2024 struct facet *facet;
2026 facet = xzalloc(sizeof *facet);
2027 facet->used = time_msec();
2028 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2029 list_push_back(&rule->facets, &facet->list_node);
2031 facet->flow = *flow;
2032 netflow_flow_init(&facet->nf_flow);
2033 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2035 facet_make_actions(ofproto, facet, packet);
2041 facet_free(struct facet *facet)
2043 free(facet->actions);
2047 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2048 * 'packet', which arrived on 'in_port'.
2050 * Takes ownership of 'packet'. */
2052 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
2053 const struct nlattr *odp_actions, size_t actions_len,
2054 struct ofpbuf *packet)
2056 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2057 && odp_actions->nla_type == ODP_ACTION_ATTR_USERSPACE) {
2058 /* As an optimization, avoid a round-trip from userspace to kernel to
2059 * userspace. This also avoids possibly filling up kernel packet
2060 * buffers along the way. */
2061 struct dpif_upcall upcall;
2063 upcall.type = DPIF_UC_ACTION;
2064 upcall.packet = packet;
2067 upcall.userdata = nl_attr_get_u64(odp_actions);
2068 upcall.sample_pool = 0;
2069 upcall.actions = NULL;
2070 upcall.actions_len = 0;
2072 send_packet_in(ofproto, &upcall, flow, false);
2076 struct odputil_keybuf keybuf;
2080 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2081 odp_flow_key_from_flow(&key, flow);
2083 error = dpif_execute(ofproto->dpif, key.data, key.size,
2084 odp_actions, actions_len, packet);
2086 ofpbuf_delete(packet);
2091 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2092 * statistics appropriately. 'packet' must have at least sizeof(struct
2093 * ofp_packet_in) bytes of headroom.
2095 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2096 * applying flow_extract() to 'packet' would yield the same flow as
2099 * 'facet' must have accurately composed ODP actions; that is, it must not be
2100 * in need of revalidation.
2102 * Takes ownership of 'packet'. */
2104 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2105 struct ofpbuf *packet)
2107 struct dpif_flow_stats stats;
2109 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2111 flow_extract_stats(&facet->flow, packet, &stats);
2112 stats.used = time_msec();
2113 if (execute_odp_actions(ofproto, &facet->flow,
2114 facet->actions, facet->actions_len, packet)) {
2115 facet_update_stats(ofproto, facet, &stats);
2119 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2121 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2122 * rule's statistics, via facet_uninstall().
2124 * - Removes 'facet' from its rule and from ofproto->facets.
2127 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2129 facet_uninstall(ofproto, facet);
2130 facet_flush_stats(ofproto, facet);
2131 hmap_remove(&ofproto->facets, &facet->hmap_node);
2132 list_remove(&facet->list_node);
2136 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2138 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2139 const struct ofpbuf *packet)
2141 const struct rule_dpif *rule = facet->rule;
2142 struct ofpbuf *odp_actions;
2143 struct action_xlate_ctx ctx;
2145 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2146 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2147 facet->tags = ctx.tags;
2148 facet->may_install = ctx.may_set_up_flow;
2149 facet->nf_flow.output_iface = ctx.nf_output_iface;
2151 if (facet->actions_len != odp_actions->size
2152 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2153 free(facet->actions);
2154 facet->actions_len = odp_actions->size;
2155 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2158 ofpbuf_delete(odp_actions);
2161 /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len'
2162 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
2163 * in the datapath will be zeroed and 'stats' will be updated with traffic new
2164 * since 'facet' was last updated.
2166 * Returns 0 if successful, otherwise a positive errno value.*/
2168 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2169 const struct nlattr *actions, size_t actions_len,
2170 struct dpif_flow_stats *stats)
2172 struct odputil_keybuf keybuf;
2173 enum dpif_flow_put_flags flags;
2177 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2179 flags |= DPIF_FP_ZERO_STATS;
2182 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2183 odp_flow_key_from_flow(&key, &facet->flow);
2185 ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2186 actions, actions_len, stats);
2189 facet_reset_dp_stats(facet, stats);
2195 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2196 * 'zero_stats' is true, clears any existing statistics from the datapath for
2199 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2201 struct dpif_flow_stats stats;
2203 if (facet->may_install
2204 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2205 zero_stats ? &stats : NULL)) {
2206 facet->installed = true;
2211 vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci)
2213 return vlan_tci != htons(0) ? vlan_tci_to_vid(vlan_tci) : OFP_VLAN_NONE;
2217 facet_account(struct ofproto_dpif *ofproto,
2218 struct facet *facet, uint64_t extra_bytes)
2220 uint64_t total_bytes, n_bytes;
2221 struct ofbundle *in_bundle;
2222 const struct nlattr *a;
2228 total_bytes = facet->byte_count + extra_bytes;
2229 if (total_bytes <= facet->accounted_bytes) {
2232 n_bytes = total_bytes - facet->accounted_bytes;
2233 facet->accounted_bytes = total_bytes;
2235 /* Test that 'tags' is nonzero to ensure that only flows that include an
2236 * OFPP_NORMAL action are used for learning and bond slave rebalancing.
2237 * This works because OFPP_NORMAL always sets a nonzero tag value.
2239 * Feed information from the active flows back into the learning table to
2240 * ensure that table is always in sync with what is actually flowing
2241 * through the datapath. */
2243 || !is_admissible(ofproto, &facet->flow, false, &dummy,
2244 &vlan, &in_bundle)) {
2248 update_learning_table(ofproto, &facet->flow, vlan, in_bundle);
2250 if (!ofproto->has_bonded_bundles) {
2254 /* This loop feeds byte counters to bond_account() for rebalancing to use
2255 * as a basis. We also need to track the actual VLAN on which the packet
2256 * is going to be sent to ensure that it matches the one passed to
2257 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
2259 vlan_tci = facet->flow.vlan_tci;
2260 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2261 struct ofport_dpif *port;
2263 switch (nl_attr_type(a)) {
2264 case ODP_ACTION_ATTR_OUTPUT:
2265 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2266 if (port && port->bundle && port->bundle->bond) {
2267 bond_account(port->bundle->bond, &facet->flow,
2268 vlan_tci_to_openflow_vlan(vlan_tci), n_bytes);
2272 case ODP_ACTION_ATTR_STRIP_VLAN:
2273 vlan_tci = htons(0);
2276 case ODP_ACTION_ATTR_SET_DL_TCI:
2277 vlan_tci = nl_attr_get_be16(a);
2283 /* If 'rule' is installed in the datapath, uninstalls it. */
2285 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2287 if (facet->installed) {
2288 struct odputil_keybuf keybuf;
2289 struct dpif_flow_stats stats;
2293 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2294 odp_flow_key_from_flow(&key, &facet->flow);
2296 error = dpif_flow_del(p->dpif, key.data, key.size, &stats);
2297 facet_reset_dp_stats(facet, &stats);
2299 facet_update_stats(p, facet, &stats);
2301 facet->installed = false;
2303 assert(facet->dp_packet_count == 0);
2304 assert(facet->dp_byte_count == 0);
2308 /* Returns true if the only action for 'facet' is to send to the controller.
2309 * (We don't report NetFlow expiration messages for such facets because they
2310 * are just part of the control logic for the network, not real traffic). */
2312 facet_is_controller_flow(struct facet *facet)
2315 && facet->rule->up.n_actions == 1
2316 && action_outputs_to_port(&facet->rule->up.actions[0],
2317 htons(OFPP_CONTROLLER)));
2320 /* Resets 'facet''s datapath statistics counters. This should be called when
2321 * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null,
2322 * it should contain the statistics returned by dpif when 'facet' was reset in
2323 * the datapath. 'stats' will be modified to only included statistics new
2324 * since 'facet' was last updated. */
2326 facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats)
2328 if (stats && facet->dp_packet_count <= stats->n_packets
2329 && facet->dp_byte_count <= stats->n_bytes) {
2330 stats->n_packets -= facet->dp_packet_count;
2331 stats->n_bytes -= facet->dp_byte_count;
2334 facet->dp_packet_count = 0;
2335 facet->dp_byte_count = 0;
2338 /* Folds all of 'facet''s statistics into its rule. Also updates the
2339 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2340 * 'facet''s statistics in the datapath should have been zeroed and folded into
2341 * its packet and byte counts before this function is called. */
2343 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2345 assert(!facet->dp_byte_count);
2346 assert(!facet->dp_packet_count);
2348 facet_push_stats(facet);
2349 facet_account(ofproto, facet, 0);
2351 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2352 struct ofexpired expired;
2353 expired.flow = facet->flow;
2354 expired.packet_count = facet->packet_count;
2355 expired.byte_count = facet->byte_count;
2356 expired.used = facet->used;
2357 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2360 facet->rule->packet_count += facet->packet_count;
2361 facet->rule->byte_count += facet->byte_count;
2363 /* Reset counters to prevent double counting if 'facet' ever gets
2365 facet->packet_count = 0;
2366 facet->byte_count = 0;
2367 facet->rs_packet_count = 0;
2368 facet->rs_byte_count = 0;
2369 facet->accounted_bytes = 0;
2371 netflow_flow_clear(&facet->nf_flow);
2374 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2375 * Returns it if found, otherwise a null pointer.
2377 * The returned facet might need revalidation; use facet_lookup_valid()
2378 * instead if that is important. */
2379 static struct facet *
2380 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
2382 struct facet *facet;
2384 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2386 if (flow_equal(flow, &facet->flow)) {
2394 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2395 * Returns it if found, otherwise a null pointer.
2397 * The returned facet is guaranteed to be valid. */
2398 static struct facet *
2399 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
2401 struct facet *facet = facet_find(ofproto, flow);
2403 /* The facet we found might not be valid, since we could be in need of
2404 * revalidation. If it is not valid, don't return it. */
2406 && ofproto->need_revalidate
2407 && !facet_revalidate(ofproto, facet)) {
2408 COVERAGE_INC(facet_invalidated);
2415 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2417 * - If the rule found is different from 'facet''s current rule, moves
2418 * 'facet' to the new rule and recompiles its actions.
2420 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2421 * where it is and recompiles its actions anyway.
2423 * - If there is none, destroys 'facet'.
2425 * Returns true if 'facet' still exists, false if it has been destroyed. */
2427 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
2429 struct action_xlate_ctx ctx;
2430 struct ofpbuf *odp_actions;
2431 struct rule_dpif *new_rule;
2432 bool actions_changed;
2434 COVERAGE_INC(facet_revalidate);
2436 /* Determine the new rule. */
2437 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
2439 /* No new rule, so delete the facet. */
2440 facet_remove(ofproto, facet);
2444 /* Calculate new ODP actions.
2446 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2447 * emit a NetFlow expiration and, if so, we need to have the old state
2448 * around to properly compose it. */
2449 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2450 odp_actions = xlate_actions(&ctx,
2451 new_rule->up.actions, new_rule->up.n_actions);
2452 actions_changed = (facet->actions_len != odp_actions->size
2453 || memcmp(facet->actions, odp_actions->data,
2454 facet->actions_len));
2456 /* If the ODP actions changed or the installability changed, then we need
2457 * to talk to the datapath. */
2458 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2459 if (ctx.may_set_up_flow) {
2460 struct dpif_flow_stats stats;
2462 facet_put__(ofproto, facet,
2463 odp_actions->data, odp_actions->size, &stats);
2464 facet_update_stats(ofproto, facet, &stats);
2466 facet_uninstall(ofproto, facet);
2469 /* The datapath flow is gone or has zeroed stats, so push stats out of
2470 * 'facet' into 'rule'. */
2471 facet_flush_stats(ofproto, facet);
2474 /* Update 'facet' now that we've taken care of all the old state. */
2475 facet->tags = ctx.tags;
2476 facet->nf_flow.output_iface = ctx.nf_output_iface;
2477 facet->may_install = ctx.may_set_up_flow;
2478 if (actions_changed) {
2479 free(facet->actions);
2480 facet->actions_len = odp_actions->size;
2481 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2483 if (facet->rule != new_rule) {
2484 COVERAGE_INC(facet_changed_rule);
2485 list_remove(&facet->list_node);
2486 list_push_back(&new_rule->facets, &facet->list_node);
2487 facet->rule = new_rule;
2488 facet->used = new_rule->up.created;
2489 facet->rs_used = facet->used;
2492 ofpbuf_delete(odp_actions);
2497 /* Updates 'facet''s used time. Caller is responsible for calling
2498 * facet_push_stats() to update the flows which 'facet' resubmits into. */
2500 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
2503 if (used > facet->used) {
2505 if (used > facet->rule->used) {
2506 facet->rule->used = used;
2508 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
2512 /* Folds the statistics from 'stats' into the counters in 'facet'.
2514 * Because of the meaning of a facet's counters, it only makes sense to do this
2515 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
2516 * packet that was sent by hand or if it represents statistics that have been
2517 * cleared out of the datapath. */
2519 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
2520 const struct dpif_flow_stats *stats)
2522 if (stats->n_packets || stats->used > facet->used) {
2523 facet_update_time(ofproto, facet, stats->used);
2524 facet->packet_count += stats->n_packets;
2525 facet->byte_count += stats->n_bytes;
2526 facet_push_stats(facet);
2527 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
2532 facet_push_stats(struct facet *facet)
2534 uint64_t rs_packets, rs_bytes;
2536 assert(facet->packet_count >= facet->rs_packet_count);
2537 assert(facet->byte_count >= facet->rs_byte_count);
2538 assert(facet->used >= facet->rs_used);
2540 rs_packets = facet->packet_count - facet->rs_packet_count;
2541 rs_bytes = facet->byte_count - facet->rs_byte_count;
2543 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
2544 facet->rs_packet_count = facet->packet_count;
2545 facet->rs_byte_count = facet->byte_count;
2546 facet->rs_used = facet->used;
2548 flow_push_stats(facet->rule, &facet->flow,
2549 rs_packets, rs_bytes, facet->used);
2553 struct ofproto_push {
2554 struct action_xlate_ctx ctx;
2561 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
2563 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
2566 rule->packet_count += push->packets;
2567 rule->byte_count += push->bytes;
2568 rule->used = MAX(push->used, rule->used);
2572 /* Pushes flow statistics to the rules which 'flow' resubmits into given
2573 * 'rule''s actions. */
2575 flow_push_stats(const struct rule_dpif *rule,
2576 struct flow *flow, uint64_t packets, uint64_t bytes,
2579 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2580 struct ofproto_push push;
2582 push.packets = packets;
2586 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
2587 push.ctx.resubmit_hook = push_resubmit;
2588 ofpbuf_delete(xlate_actions(&push.ctx,
2589 rule->up.actions, rule->up.n_actions));
2594 static struct rule_dpif *
2595 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
2597 return rule_dpif_cast(rule_from_cls_rule(
2598 classifier_lookup(&ofproto->up.tables[0],
2603 complete_operation(struct rule_dpif *rule)
2605 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2607 ofproto->need_revalidate = true;
2609 struct dpif_completion *c = xmalloc(sizeof *c);
2610 c->op = rule->up.pending;
2611 list_push_back(&ofproto->completions, &c->list_node);
2613 ofoperation_complete(rule->up.pending, 0);
2617 static struct rule *
2620 struct rule_dpif *rule = xmalloc(sizeof *rule);
2625 rule_dealloc(struct rule *rule_)
2627 struct rule_dpif *rule = rule_dpif_cast(rule_);
2632 rule_construct(struct rule *rule_)
2634 struct rule_dpif *rule = rule_dpif_cast(rule_);
2635 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2636 struct rule_dpif *victim;
2639 error = validate_actions(rule->up.actions, rule->up.n_actions,
2640 &rule->up.cr.flow, ofproto->max_ports);
2645 rule->used = rule->up.created;
2646 rule->packet_count = 0;
2647 rule->byte_count = 0;
2649 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
2650 if (victim && !list_is_empty(&victim->facets)) {
2651 struct facet *facet;
2653 rule->facets = victim->facets;
2654 list_moved(&rule->facets);
2655 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2659 /* Must avoid list_moved() in this case. */
2660 list_init(&rule->facets);
2663 complete_operation(rule);
2668 rule_destruct(struct rule *rule_)
2670 struct rule_dpif *rule = rule_dpif_cast(rule_);
2671 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2672 struct facet *facet, *next_facet;
2674 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2675 facet_revalidate(ofproto, facet);
2678 complete_operation(rule);
2682 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
2684 struct rule_dpif *rule = rule_dpif_cast(rule_);
2685 struct facet *facet;
2687 /* Start from historical data for 'rule' itself that are no longer tracked
2688 * in facets. This counts, for example, facets that have expired. */
2689 *packets = rule->packet_count;
2690 *bytes = rule->byte_count;
2692 /* Add any statistics that are tracked by facets. This includes
2693 * statistical data recently updated by ofproto_update_stats() as well as
2694 * stats for packets that were executed "by hand" via dpif_execute(). */
2695 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2696 *packets += facet->packet_count;
2697 *bytes += facet->byte_count;
2702 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
2704 struct rule_dpif *rule = rule_dpif_cast(rule_);
2705 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2706 struct action_xlate_ctx ctx;
2707 struct ofpbuf *odp_actions;
2708 struct facet *facet;
2711 /* First look for a related facet. If we find one, account it to that. */
2712 facet = facet_lookup_valid(ofproto, flow);
2713 if (facet && facet->rule == rule) {
2714 facet_execute(ofproto, facet, packet);
2718 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2719 * create a new facet for it and use that. */
2720 if (rule_dpif_lookup(ofproto, flow) == rule) {
2721 facet = facet_create(rule, flow, packet);
2722 facet_execute(ofproto, facet, packet);
2723 facet_install(ofproto, facet, true);
2727 /* We can't account anything to a facet. If we were to try, then that
2728 * facet would have a non-matching rule, busting our invariants. */
2729 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
2730 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2731 size = packet->size;
2732 if (execute_odp_actions(ofproto, flow, odp_actions->data,
2733 odp_actions->size, packet)) {
2734 rule->used = time_msec();
2735 rule->packet_count++;
2736 rule->byte_count += size;
2737 flow_push_stats(rule, flow, 1, size, rule->used);
2739 ofpbuf_delete(odp_actions);
2745 rule_modify_actions(struct rule *rule_)
2747 struct rule_dpif *rule = rule_dpif_cast(rule_);
2748 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2751 error = validate_actions(rule->up.actions, rule->up.n_actions,
2752 &rule->up.cr.flow, ofproto->max_ports);
2754 ofoperation_complete(rule->up.pending, error);
2758 complete_operation(rule);
2761 /* Sends 'packet' out of port 'odp_port' within 'p'.
2762 * Returns 0 if successful, otherwise a positive errno value. */
2764 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
2765 const struct ofpbuf *packet)
2767 struct ofpbuf key, odp_actions;
2768 struct odputil_keybuf keybuf;
2772 flow_extract((struct ofpbuf *) packet, 0, 0, &flow);
2773 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2774 odp_flow_key_from_flow(&key, &flow);
2776 ofpbuf_init(&odp_actions, 32);
2777 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2778 error = dpif_execute(ofproto->dpif,
2780 odp_actions.data, odp_actions.size,
2782 ofpbuf_uninit(&odp_actions);
2785 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
2786 ofproto->up.name, odp_port, strerror(error));
2791 /* OpenFlow to ODP action translation. */
2793 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2794 struct action_xlate_ctx *ctx);
2795 static void xlate_normal(struct action_xlate_ctx *);
2798 commit_odp_actions(struct action_xlate_ctx *ctx)
2800 const struct flow *flow = &ctx->flow;
2801 struct flow *base = &ctx->base_flow;
2802 struct ofpbuf *odp_actions = ctx->odp_actions;
2804 if (base->tun_id != flow->tun_id) {
2805 nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id);
2806 base->tun_id = flow->tun_id;
2809 if (base->nw_src != flow->nw_src) {
2810 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src);
2811 base->nw_src = flow->nw_src;
2814 if (base->nw_dst != flow->nw_dst) {
2815 nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst);
2816 base->nw_dst = flow->nw_dst;
2819 if (base->nw_tos != flow->nw_tos) {
2820 nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos);
2821 base->nw_tos = flow->nw_tos;
2824 if (base->vlan_tci != flow->vlan_tci) {
2825 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
2826 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2828 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2829 flow->vlan_tci & ~htons(VLAN_CFI));
2831 base->vlan_tci = flow->vlan_tci;
2834 if (base->tp_src != flow->tp_src) {
2835 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src);
2836 base->tp_src = flow->tp_src;
2839 if (base->tp_dst != flow->tp_dst) {
2840 nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst);
2841 base->tp_dst = flow->tp_dst;
2844 if (!eth_addr_equals(base->dl_src, flow->dl_src)) {
2845 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
2846 flow->dl_src, ETH_ADDR_LEN);
2847 memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN);
2850 if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) {
2851 nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
2852 flow->dl_dst, ETH_ADDR_LEN);
2853 memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN);
2856 if (ctx->base_priority != ctx->priority) {
2857 if (ctx->priority) {
2858 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY,
2861 nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2863 ctx->base_priority = ctx->priority;
2868 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
2870 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
2871 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
2874 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
2875 /* Forwarding disabled on port. */
2880 * We don't have an ofport record for this port, but it doesn't hurt to
2881 * allow forwarding to it anyhow. Maybe such a port will appear later
2882 * and we're pre-populating the flow table.
2886 commit_odp_actions(ctx);
2887 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2888 ctx->nf_output_iface = ofp_port;
2892 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2894 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2895 struct rule_dpif *rule;
2896 uint16_t old_in_port;
2898 /* Look up a flow with 'in_port' as the input port. Then restore the
2899 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2900 * have surprising behavior). */
2901 old_in_port = ctx->flow.in_port;
2902 ctx->flow.in_port = in_port;
2903 rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow);
2904 ctx->flow.in_port = old_in_port;
2906 if (ctx->resubmit_hook) {
2907 ctx->resubmit_hook(ctx, rule);
2912 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
2916 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2918 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2919 MAX_RESUBMIT_RECURSION);
2924 flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask)
2926 struct ofport_dpif *ofport;
2928 commit_odp_actions(ctx);
2929 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
2930 uint16_t ofp_port = ofport->up.ofp_port;
2931 if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) {
2932 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT,
2937 ctx->nf_output_iface = NF_OUT_FLOOD;
2941 xlate_output_action__(struct action_xlate_ctx *ctx,
2942 uint16_t port, uint16_t max_len)
2944 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2946 ctx->nf_output_iface = NF_OUT_DROP;
2950 add_output_action(ctx, ctx->flow.in_port);
2953 xlate_table_action(ctx, ctx->flow.in_port);
2959 flood_packets(ctx, htonl(OFPPC_NO_FLOOD));
2962 flood_packets(ctx, htonl(0));
2964 case OFPP_CONTROLLER:
2965 commit_odp_actions(ctx);
2966 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_USERSPACE, max_len);
2969 add_output_action(ctx, OFPP_LOCAL);
2974 if (port != ctx->flow.in_port) {
2975 add_output_action(ctx, port);
2980 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2981 ctx->nf_output_iface = NF_OUT_FLOOD;
2982 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2983 ctx->nf_output_iface = prev_nf_output_iface;
2984 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2985 ctx->nf_output_iface != NF_OUT_FLOOD) {
2986 ctx->nf_output_iface = NF_OUT_MULTI;
2991 xlate_output_action(struct action_xlate_ctx *ctx,
2992 const struct ofp_action_output *oao)
2994 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2998 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2999 const struct ofp_action_enqueue *oae)
3001 uint16_t ofp_port, odp_port;
3002 uint32_t ctx_priority, priority;
3005 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
3008 /* Fall back to ordinary output action. */
3009 xlate_output_action__(ctx, ntohs(oae->port), 0);
3013 /* Figure out ODP output port. */
3014 ofp_port = ntohs(oae->port);
3015 if (ofp_port == OFPP_IN_PORT) {
3016 ofp_port = ctx->flow.in_port;
3018 odp_port = ofp_port_to_odp_port(ofp_port);
3020 /* Add ODP actions. */
3021 ctx_priority = ctx->priority;
3022 ctx->priority = priority;
3023 add_output_action(ctx, odp_port);
3024 ctx->priority = ctx_priority;
3026 /* Update NetFlow output port. */
3027 if (ctx->nf_output_iface == NF_OUT_DROP) {
3028 ctx->nf_output_iface = odp_port;
3029 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3030 ctx->nf_output_iface = NF_OUT_MULTI;
3035 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3036 const struct nx_action_set_queue *nasq)
3041 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3044 /* Couldn't translate queue to a priority, so ignore. A warning
3045 * has already been logged. */
3049 ctx->priority = priority;
3052 struct xlate_reg_state {
3058 xlate_autopath(struct action_xlate_ctx *ctx,
3059 const struct nx_action_autopath *naa)
3061 uint16_t ofp_port = ntohl(naa->id);
3062 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
3064 if (!port || !port->bundle) {
3065 ofp_port = OFPP_NONE;
3066 } else if (port->bundle->bond) {
3067 /* Autopath does not support VLAN hashing. */
3068 struct ofport_dpif *slave = bond_choose_output_slave(
3069 port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags);
3071 ofp_port = slave->up.ofp_port;
3074 autopath_execute(naa, &ctx->flow, ofp_port);
3078 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
3080 struct ofproto_dpif *ofproto = ofproto_;
3081 struct ofport_dpif *port;
3091 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3094 port = get_ofp_port(ofproto, ofp_port);
3095 return port ? port->may_enable : false;
3100 do_xlate_actions(const union ofp_action *in, size_t n_in,
3101 struct action_xlate_ctx *ctx)
3103 const struct ofport_dpif *port;
3104 const union ofp_action *ia;
3107 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
3109 && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3110 port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3111 ? htonl(OFPPC_NO_RECV_STP)
3112 : htonl(OFPPC_NO_RECV))) {
3113 /* Drop this flow. */
3117 OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) {
3118 const struct ofp_action_dl_addr *oada;
3119 const struct nx_action_resubmit *nar;
3120 const struct nx_action_set_tunnel *nast;
3121 const struct nx_action_set_queue *nasq;
3122 const struct nx_action_multipath *nam;
3123 const struct nx_action_autopath *naa;
3124 const struct nx_action_bundle *nab;
3125 enum ofputil_action_code code;
3128 code = ofputil_decode_action_unsafe(ia);
3130 case OFPUTIL_OFPAT_OUTPUT:
3131 xlate_output_action(ctx, &ia->output);
3134 case OFPUTIL_OFPAT_SET_VLAN_VID:
3135 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3136 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3139 case OFPUTIL_OFPAT_SET_VLAN_PCP:
3140 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3141 ctx->flow.vlan_tci |= htons(
3142 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3145 case OFPUTIL_OFPAT_STRIP_VLAN:
3146 ctx->flow.vlan_tci = htons(0);
3149 case OFPUTIL_OFPAT_SET_DL_SRC:
3150 oada = ((struct ofp_action_dl_addr *) ia);
3151 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3154 case OFPUTIL_OFPAT_SET_DL_DST:
3155 oada = ((struct ofp_action_dl_addr *) ia);
3156 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3159 case OFPUTIL_OFPAT_SET_NW_SRC:
3160 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3163 case OFPUTIL_OFPAT_SET_NW_DST:
3164 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3167 case OFPUTIL_OFPAT_SET_NW_TOS:
3168 ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK;
3171 case OFPUTIL_OFPAT_SET_TP_SRC:
3172 ctx->flow.tp_src = ia->tp_port.tp_port;
3175 case OFPUTIL_OFPAT_SET_TP_DST:
3176 ctx->flow.tp_dst = ia->tp_port.tp_port;
3179 case OFPUTIL_OFPAT_ENQUEUE:
3180 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3183 case OFPUTIL_NXAST_RESUBMIT:
3184 nar = (const struct nx_action_resubmit *) ia;
3185 xlate_table_action(ctx, ntohs(nar->in_port));
3188 case OFPUTIL_NXAST_SET_TUNNEL:
3189 nast = (const struct nx_action_set_tunnel *) ia;
3190 tun_id = htonll(ntohl(nast->tun_id));
3191 ctx->flow.tun_id = tun_id;
3194 case OFPUTIL_NXAST_SET_QUEUE:
3195 nasq = (const struct nx_action_set_queue *) ia;
3196 xlate_set_queue_action(ctx, nasq);
3199 case OFPUTIL_NXAST_POP_QUEUE:
3203 case OFPUTIL_NXAST_REG_MOVE:
3204 nxm_execute_reg_move((const struct nx_action_reg_move *) ia,
3208 case OFPUTIL_NXAST_REG_LOAD:
3209 nxm_execute_reg_load((const struct nx_action_reg_load *) ia,
3213 case OFPUTIL_NXAST_NOTE:
3214 /* Nothing to do. */
3217 case OFPUTIL_NXAST_SET_TUNNEL64:
3218 tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id;
3219 ctx->flow.tun_id = tun_id;
3222 case OFPUTIL_NXAST_MULTIPATH:
3223 nam = (const struct nx_action_multipath *) ia;
3224 multipath_execute(nam, &ctx->flow);
3227 case OFPUTIL_NXAST_AUTOPATH:
3228 naa = (const struct nx_action_autopath *) ia;
3229 xlate_autopath(ctx, naa);
3232 case OFPUTIL_NXAST_BUNDLE:
3233 ctx->ofproto->has_bundle_action = true;
3234 nab = (const struct nx_action_bundle *) ia;
3235 xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow,
3240 case OFPUTIL_NXAST_BUNDLE_LOAD:
3241 ctx->ofproto->has_bundle_action = true;
3242 nab = (const struct nx_action_bundle *) ia;
3243 bundle_execute_load(nab, &ctx->flow, slave_enabled_cb,
3251 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3252 struct ofproto_dpif *ofproto, const struct flow *flow,
3253 const struct ofpbuf *packet)
3255 ctx->ofproto = ofproto;
3257 ctx->packet = packet;
3258 ctx->resubmit_hook = NULL;
3261 static struct ofpbuf *
3262 xlate_actions(struct action_xlate_ctx *ctx,
3263 const union ofp_action *in, size_t n_in)
3265 COVERAGE_INC(ofproto_dpif_xlate);
3267 ctx->odp_actions = ofpbuf_new(512);
3269 ctx->may_set_up_flow = true;
3270 ctx->nf_output_iface = NF_OUT_DROP;
3273 ctx->base_priority = 0;
3274 ctx->base_flow = ctx->flow;
3276 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
3277 ctx->may_set_up_flow = false;
3279 do_xlate_actions(in, n_in, ctx);
3282 /* Check with in-band control to see if we're allowed to set up this
3284 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
3285 ctx->odp_actions->data,
3286 ctx->odp_actions->size)) {
3287 ctx->may_set_up_flow = false;
3290 return ctx->odp_actions;
3293 /* OFPP_NORMAL implementation. */
3296 struct ofport_dpif *port;
3301 struct dst builtin[32];
3303 size_t n, allocated;
3306 static void dst_set_init(struct dst_set *);
3307 static void dst_set_add(struct dst_set *, const struct dst *);
3308 static void dst_set_free(struct dst_set *);
3310 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
3313 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
3314 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
3316 dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE
3317 : in_bundle->vlan >= 0 ? in_bundle->vlan
3318 : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE
3319 : vlan_tci_to_vid(ctx->flow.vlan_tci));
3321 dst->port = (!out_bundle->bond
3322 ? ofbundle_get_a_port(out_bundle)
3323 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
3324 dst->vlan, &ctx->tags));
3326 return dst->port != NULL;
3330 mirror_mask_ffs(mirror_mask_t mask)
3332 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
3337 dst_set_init(struct dst_set *set)
3339 set->dsts = set->builtin;
3341 set->allocated = ARRAY_SIZE(set->builtin);
3345 dst_set_add(struct dst_set *set, const struct dst *dst)
3347 if (set->n >= set->allocated) {
3348 size_t new_allocated;
3349 struct dst *new_dsts;
3351 new_allocated = set->allocated * 2;
3352 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
3353 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
3357 set->dsts = new_dsts;
3358 set->allocated = new_allocated;
3360 set->dsts[set->n++] = *dst;
3364 dst_set_free(struct dst_set *set)
3366 if (set->dsts != set->builtin) {
3372 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
3375 for (i = 0; i < set->n; i++) {
3376 if (set->dsts[i].vlan == test->vlan
3377 && set->dsts[i].port == test->port) {
3385 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
3387 return (bundle->vlan < 0
3388 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
3392 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
3394 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
3397 /* Returns an arbitrary interface within 'bundle'. */
3398 static struct ofport_dpif *
3399 ofbundle_get_a_port(const struct ofbundle *bundle)
3401 return CONTAINER_OF(list_front(&bundle->ports),
3402 struct ofport_dpif, bundle_node);
3406 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
3407 const struct ofbundle *in_bundle,
3408 const struct ofbundle *out_bundle, struct dst_set *set)
3412 if (out_bundle == OFBUNDLE_FLOOD) {
3413 struct ofbundle *bundle;
3415 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
3416 if (bundle != in_bundle
3417 && ofbundle_includes_vlan(bundle, vlan)
3418 && bundle->floodable
3419 && !bundle->mirror_out
3420 && set_dst(ctx, &dst, in_bundle, bundle)) {
3421 dst_set_add(set, &dst);
3424 ctx->nf_output_iface = NF_OUT_FLOOD;
3425 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
3426 dst_set_add(set, &dst);
3427 ctx->nf_output_iface = dst.port->odp_port;
3432 vlan_is_mirrored(const struct ofmirror *m, int vlan)
3434 return !m->vlans || bitmap_is_set(m->vlans, vlan);
3437 /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored
3438 * to a VLAN. In general most packets may be mirrored but we want to drop
3439 * protocols that may confuse switches. */
3441 eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN])
3443 /* If you change this function's behavior, please update corresponding
3444 * documentation in vswitch.xml at the same time. */
3445 if (dst[0] != 0x01) {
3446 /* All the currently banned MACs happen to start with 01 currently, so
3447 * this is a quick way to eliminate most of the good ones. */
3449 if (eth_addr_is_reserved(dst)) {
3450 /* Drop STP, IEEE pause frames, and other reserved protocols
3451 * (01-80-c2-00-00-0x). */
3455 if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) {
3457 if ((dst[3] & 0xfe) == 0xcc &&
3458 (dst[4] & 0xfe) == 0xcc &&
3459 (dst[5] & 0xfe) == 0xcc) {
3460 /* Drop the following protocols plus others following the same
3463 CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc)
3464 Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd)
3465 STP Uplink Fast (01-00-0c-cd-cd-cd) */
3469 if (!(dst[3] | dst[4] | dst[5])) {
3470 /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */
3479 compose_mirror_dsts(struct action_xlate_ctx *ctx,
3480 uint16_t vlan, const struct ofbundle *in_bundle,
3481 struct dst_set *set)
3483 struct ofproto_dpif *ofproto = ctx->ofproto;
3484 mirror_mask_t mirrors;
3488 mirrors = in_bundle->src_mirrors;
3489 for (i = 0; i < set->n; i++) {
3490 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
3497 flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3498 if (flow_vlan == 0) {
3499 flow_vlan = OFP_VLAN_NONE;
3503 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
3504 if (vlan_is_mirrored(m, vlan)) {
3508 if (set_dst(ctx, &dst, in_bundle, m->out)
3509 && !dst_is_duplicate(set, &dst)) {
3510 dst_set_add(set, &dst);
3512 } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) {
3513 struct ofbundle *bundle;
3515 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3516 if (ofbundle_includes_vlan(bundle, m->out_vlan)
3517 && set_dst(ctx, &dst, in_bundle, bundle))
3519 if (bundle->vlan < 0) {
3520 dst.vlan = m->out_vlan;
3522 if (dst_is_duplicate(set, &dst)) {
3526 /* Use the vlan tag on the original flow instead of
3527 * the one passed in the vlan parameter. This ensures
3528 * that we compare the vlan from before any implicit
3529 * tagging tags place. This is necessary because
3530 * dst->vlan is the final vlan, after removing implicit
3532 if (bundle == in_bundle && dst.vlan == flow_vlan) {
3533 /* Don't send out input port on same VLAN. */
3536 dst_set_add(set, &dst);
3541 mirrors &= mirrors - 1;
3546 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
3547 const struct ofbundle *in_bundle,
3548 const struct ofbundle *out_bundle)
3550 uint16_t initial_vlan, cur_vlan;
3551 const struct dst *dst;
3555 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
3556 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
3558 /* Output all the packets we can without having to change the VLAN. */
3559 initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3560 if (initial_vlan == 0) {
3561 initial_vlan = OFP_VLAN_NONE;
3563 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3564 if (dst->vlan != initial_vlan) {
3567 nl_msg_put_u32(ctx->odp_actions,
3568 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3571 /* Then output the rest. */
3572 cur_vlan = initial_vlan;
3573 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3574 if (dst->vlan == initial_vlan) {
3577 if (dst->vlan != cur_vlan) {
3578 if (dst->vlan == OFP_VLAN_NONE) {
3579 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
3582 tci = htons(dst->vlan & VLAN_VID_MASK);
3583 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
3584 nl_msg_put_be16(ctx->odp_actions,
3585 ODP_ACTION_ATTR_SET_DL_TCI, tci);
3587 cur_vlan = dst->vlan;
3589 nl_msg_put_u32(ctx->odp_actions,
3590 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3596 /* Returns the effective vlan of a packet, taking into account both the
3597 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
3598 * the packet is untagged and -1 indicates it has an invalid header and
3599 * should be dropped. */
3601 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
3602 struct ofbundle *in_bundle, bool have_packet)
3604 int vlan = vlan_tci_to_vid(flow->vlan_tci);
3605 if (in_bundle->vlan >= 0) {
3608 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3609 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3610 "packet received on port %s configured with "
3611 "implicit VLAN %"PRIu16,
3612 ofproto->up.name, vlan,
3613 in_bundle->name, in_bundle->vlan);
3617 vlan = in_bundle->vlan;
3619 if (!ofbundle_includes_vlan(in_bundle, vlan)) {
3621 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3622 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3623 "packet received on port %s not configured for "
3625 ofproto->up.name, vlan, in_bundle->name, vlan);
3634 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
3635 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
3636 * indicate this; newer upstream kernels use gratuitous ARP requests. */
3638 is_gratuitous_arp(const struct flow *flow)
3640 return (flow->dl_type == htons(ETH_TYPE_ARP)
3641 && eth_addr_is_broadcast(flow->dl_dst)
3642 && (flow->nw_proto == ARP_OP_REPLY
3643 || (flow->nw_proto == ARP_OP_REQUEST
3644 && flow->nw_src == flow->nw_dst)));
3648 update_learning_table(struct ofproto_dpif *ofproto,
3649 const struct flow *flow, int vlan,
3650 struct ofbundle *in_bundle)
3652 struct mac_entry *mac;
3654 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
3658 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
3659 if (is_gratuitous_arp(flow)) {
3660 /* We don't want to learn from gratuitous ARP packets that are
3661 * reflected back over bond slaves so we lock the learning table. */
3662 if (!in_bundle->bond) {
3663 mac_entry_set_grat_arp_lock(mac);
3664 } else if (mac_entry_is_grat_arp_locked(mac)) {
3669 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
3670 /* The log messages here could actually be useful in debugging,
3671 * so keep the rate limit relatively high. */
3672 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3673 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
3674 "on port %s in VLAN %d",
3675 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
3676 in_bundle->name, vlan);
3678 mac->port.p = in_bundle;
3679 tag_set_add(&ofproto->revalidate_set,
3680 mac_learning_changed(ofproto->ml, mac));
3684 /* Determines whether packets in 'flow' within 'br' should be forwarded or
3685 * dropped. Returns true if they may be forwarded, false if they should be
3688 * If 'have_packet' is true, it indicates that the caller is processing a
3689 * received packet. If 'have_packet' is false, then the caller is just
3690 * revalidating an existing flow because configuration has changed. Either
3691 * way, 'have_packet' only affects logging (there is no point in logging errors
3692 * during revalidation).
3694 * Sets '*in_portp' to the input port. This will be a null pointer if
3695 * flow->in_port does not designate a known input port (in which case
3696 * is_admissible() returns false).
3698 * When returning true, sets '*vlanp' to the effective VLAN of the input
3699 * packet, as returned by flow_get_vlan().
3701 * May also add tags to '*tags', although the current implementation only does
3702 * so in one special case.
3705 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
3707 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
3709 struct ofport_dpif *in_port;
3710 struct ofbundle *in_bundle;
3713 /* Find the port and bundle for the received packet. */
3714 in_port = get_ofp_port(ofproto, flow->in_port);
3715 *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
3716 if (!in_port || !in_bundle) {
3717 /* No interface? Something fishy... */
3719 /* Odd. A few possible reasons here:
3721 * - We deleted a port but there are still a few packets queued up
3724 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
3725 * we don't know about.
3727 * - Packet arrived on the local port but the local port is not
3730 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3732 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3734 ofproto->up.name, flow->in_port);
3738 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
3743 /* Drop frames for reserved multicast addresses. */
3744 if (eth_addr_is_reserved(flow->dl_dst)) {
3748 /* Drop frames on bundles reserved for mirroring. */
3749 if (in_bundle->mirror_out) {
3751 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3752 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3753 "%s, which is reserved exclusively for mirroring",
3754 ofproto->up.name, in_bundle->name);
3759 if (in_bundle->bond) {
3760 struct mac_entry *mac;
3762 switch (bond_check_admissibility(in_bundle->bond, in_port,
3763 flow->dl_dst, tags)) {
3770 case BV_DROP_IF_MOVED:
3771 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
3772 if (mac && mac->port.p != in_bundle &&
3773 (!is_gratuitous_arp(flow)
3774 || mac_entry_is_grat_arp_locked(mac))) {
3785 xlate_normal(struct action_xlate_ctx *ctx)
3787 struct ofbundle *in_bundle;
3788 struct ofbundle *out_bundle;
3789 struct mac_entry *mac;
3792 /* Check whether we should drop packets in this flow. */
3793 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
3794 &ctx->tags, &vlan, &in_bundle)) {
3799 /* Learn source MAC (but don't try to learn from revalidation). */
3801 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
3804 /* Determine output bundle. */
3805 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
3808 out_bundle = mac->port.p;
3809 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
3810 /* If we are revalidating but don't have a learning entry then eject
3811 * the flow. Installing a flow that floods packets opens up a window
3812 * of time where we could learn from a packet reflected on a bond and
3813 * blackhole packets before the learning table is updated to reflect
3814 * the correct port. */
3815 ctx->may_set_up_flow = false;
3818 out_bundle = OFBUNDLE_FLOOD;
3821 /* Don't send packets out their input bundles. */
3822 if (in_bundle == out_bundle) {
3828 compose_actions(ctx, vlan, in_bundle, out_bundle);
3833 get_drop_frags(struct ofproto *ofproto_)
3835 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3838 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
3843 set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
3845 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3847 dpif_set_drop_frags(ofproto->dpif, drop_frags);
3851 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3852 const struct flow *flow,
3853 const union ofp_action *ofp_actions, size_t n_ofp_actions)
3855 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3858 error = validate_actions(ofp_actions, n_ofp_actions, flow,
3859 ofproto->max_ports);
3861 struct odputil_keybuf keybuf;
3862 struct action_xlate_ctx ctx;
3863 struct ofpbuf *odp_actions;
3866 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3867 odp_flow_key_from_flow(&key, flow);
3869 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3870 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3871 dpif_execute(ofproto->dpif, key.data, key.size,
3872 odp_actions->data, odp_actions->size, packet);
3873 ofpbuf_delete(odp_actions);
3879 get_netflow_ids(const struct ofproto *ofproto_,
3880 uint8_t *engine_type, uint8_t *engine_id)
3882 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3884 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3887 static struct ofproto_dpif *
3888 ofproto_dpif_lookup(const char *name)
3890 struct ofproto *ofproto = ofproto_lookup(name);
3891 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
3892 ? ofproto_dpif_cast(ofproto)
3897 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
3898 const char *args, void *aux OVS_UNUSED)
3900 struct ds ds = DS_EMPTY_INITIALIZER;
3901 const struct ofproto_dpif *ofproto;
3902 const struct mac_entry *e;
3904 ofproto = ofproto_dpif_lookup(args);
3906 unixctl_command_reply(conn, 501, "no such bridge");
3910 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3911 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3912 struct ofbundle *bundle = e->port.p;
3913 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
3914 ofbundle_get_a_port(bundle)->odp_port,
3915 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
3917 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3921 struct ofproto_trace {
3922 struct action_xlate_ctx ctx;
3928 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3930 ds_put_char_multiple(result, '\t', level);
3932 ds_put_cstr(result, "No match\n");
3936 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
3937 ntohll(rule->up.flow_cookie));
3938 cls_rule_format(&rule->up.cr, result);
3939 ds_put_char(result, '\n');
3941 ds_put_char_multiple(result, '\t', level);
3942 ds_put_cstr(result, "OpenFlow ");
3943 ofp_print_actions(result, rule->up.actions, rule->up.n_actions);
3944 ds_put_char(result, '\n');
3948 trace_format_flow(struct ds *result, int level, const char *title,
3949 struct ofproto_trace *trace)
3951 ds_put_char_multiple(result, '\t', level);
3952 ds_put_format(result, "%s: ", title);
3953 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
3954 ds_put_cstr(result, "unchanged");
3956 flow_format(result, &trace->ctx.flow);
3957 trace->flow = trace->ctx.flow;
3959 ds_put_char(result, '\n');
3963 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3965 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
3966 struct ds *result = trace->result;
3968 ds_put_char(result, '\n');
3969 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
3970 trace_format_rule(result, ctx->recurse + 1, rule);
3974 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
3975 void *aux OVS_UNUSED)
3977 char *dpname, *arg1, *arg2, *arg3;
3978 char *args = xstrdup(args_);
3979 char *save_ptr = NULL;
3980 struct ofproto_dpif *ofproto;
3981 struct ofpbuf odp_key;
3982 struct ofpbuf *packet;
3983 struct rule_dpif *rule;
3989 ofpbuf_init(&odp_key, 0);
3992 dpname = strtok_r(args, " ", &save_ptr);
3993 arg1 = strtok_r(NULL, " ", &save_ptr);
3994 arg2 = strtok_r(NULL, " ", &save_ptr);
3995 arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
3996 if (dpname && arg1 && !arg2 && !arg3) {
3997 /* ofproto/trace dpname flow */
4000 /* Convert string to ODP key. */
4001 ofpbuf_init(&odp_key, 0);
4002 error = odp_flow_key_from_string(arg1, &odp_key);
4004 unixctl_command_reply(conn, 501, "Bad flow syntax");
4008 /* Convert odp_key to flow. */
4009 error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow);
4011 unixctl_command_reply(conn, 501, "Invalid flow");
4014 } else if (dpname && arg1 && arg2 && arg3) {
4015 /* ofproto/trace dpname tun_id in_port packet */
4019 tun_id = htonll(strtoull(arg1, NULL, 0));
4020 in_port = ofp_port_to_odp_port(atoi(arg2));
4022 packet = ofpbuf_new(strlen(args) / 2);
4023 arg3 = ofpbuf_put_hex(packet, arg3, NULL);
4024 arg3 += strspn(arg3, " ");
4025 if (*arg3 != '\0') {
4026 unixctl_command_reply(conn, 501, "Trailing garbage in command");
4029 if (packet->size < ETH_HEADER_LEN) {
4030 unixctl_command_reply(conn, 501,
4031 "Packet data too short for Ethernet");
4035 ds_put_cstr(&result, "Packet: ");
4036 s = ofp_packet_to_string(packet->data, packet->size, packet->size);
4037 ds_put_cstr(&result, s);
4040 flow_extract(packet, tun_id, in_port, &flow);
4042 unixctl_command_reply(conn, 501, "Bad command syntax");
4046 ofproto = ofproto_dpif_lookup(dpname);
4048 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
4053 ds_put_cstr(&result, "Flow: ");
4054 flow_format(&result, &flow);
4055 ds_put_char(&result, '\n');
4057 rule = rule_dpif_lookup(ofproto, &flow);
4058 trace_format_rule(&result, 0, rule);
4060 struct ofproto_trace trace;
4061 struct ofpbuf *odp_actions;
4063 trace.result = &result;
4065 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet);
4066 trace.ctx.resubmit_hook = trace_resubmit;
4067 odp_actions = xlate_actions(&trace.ctx,
4068 rule->up.actions, rule->up.n_actions);
4070 ds_put_char(&result, '\n');
4071 trace_format_flow(&result, 0, "Final flow", &trace);
4072 ds_put_cstr(&result, "Datapath actions: ");
4073 format_odp_actions(&result, odp_actions->data, odp_actions->size);
4074 ofpbuf_delete(odp_actions);
4076 if (!trace.ctx.may_set_up_flow) {
4078 ds_put_cstr(&result, "\nThis flow is not cachable.");
4080 ds_put_cstr(&result, "\nThe datapath actions are incomplete--"
4081 "for complete actions, please supply a packet.");
4086 unixctl_command_reply(conn, 200, ds_cstr(&result));
4089 ds_destroy(&result);
4090 ofpbuf_delete(packet);
4091 ofpbuf_uninit(&odp_key);
4096 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED,
4097 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4100 unixctl_command_reply(conn, 200, NULL);
4104 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED,
4105 const char *args_ OVS_UNUSED, void *aux OVS_UNUSED)
4108 unixctl_command_reply(conn, 200, NULL);
4112 ofproto_dpif_unixctl_init(void)
4114 static bool registered;
4120 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
4121 unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
4123 unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL);
4124 unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL);
4127 const struct ofproto_class ofproto_dpif_class = {
4154 port_is_lacp_current,
4155 NULL, /* rule_choose_table */
4162 rule_modify_actions,
4175 is_mirror_output_bundle,