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/private.h"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
34 #include "mac-learning.h"
35 #include "multipath.h"
42 #include "ofp-print.h"
43 #include "ofproto-sflow.h"
44 #include "poll-loop.h"
46 #include "unaligned.h"
48 #include "vlan-bitmap.h"
51 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
53 COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
54 COVERAGE_DEFINE(ofproto_dpif_expired);
55 COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
56 COVERAGE_DEFINE(ofproto_dpif_xlate);
57 COVERAGE_DEFINE(facet_changed_rule);
58 COVERAGE_DEFINE(facet_invalidated);
59 COVERAGE_DEFINE(facet_revalidate);
60 COVERAGE_DEFINE(facet_unexpected);
62 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
63 * flow translation. */
64 #define MAX_RESUBMIT_RECURSION 16
72 long long int used; /* Time last used; time created if not used. */
76 * - Do include packets and bytes from facets that have been deleted or
77 * whose own statistics have been folded into the rule.
79 * - Do include packets and bytes sent "by hand" that were accounted to
80 * the rule without any facet being involved (this is a rare corner
81 * case in rule_execute()).
83 * - Do not include packet or bytes that can be obtained from any facet's
84 * packet_count or byte_count member or that can be obtained from the
85 * datapath by, e.g., dpif_flow_get() for any facet.
87 uint64_t packet_count; /* Number of packets received. */
88 uint64_t byte_count; /* Number of bytes received. */
90 struct list facets; /* List of "struct facet"s. */
93 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
95 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
98 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *ofproto,
99 const struct flow *flow);
101 #define MAX_MIRRORS 32
102 typedef uint32_t mirror_mask_t;
103 #define MIRROR_MASK_C(X) UINT32_C(X)
104 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
106 struct ofproto_dpif *ofproto; /* Owning ofproto. */
107 size_t idx; /* In ofproto's "mirrors" array. */
108 void *aux; /* Key supplied by ofproto's client. */
109 char *name; /* Identifier for log messages. */
111 /* Selection criteria. */
112 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
113 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
114 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
116 /* Output (mutually exclusive). */
117 struct ofbundle *out; /* Output port or NULL. */
118 int out_vlan; /* Output VLAN or -1. */
121 static void mirror_destroy(struct ofmirror *);
123 /* A group of one or more OpenFlow ports. */
124 #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
126 struct ofproto_dpif *ofproto; /* Owning ofproto. */
127 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
128 void *aux; /* Key supplied by ofproto's client. */
129 char *name; /* Identifier for log messages. */
132 struct list ports; /* Contains "struct ofport"s. */
133 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
134 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
135 * NULL if all VLANs are trunked. */
136 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
137 struct bond *bond; /* Nonnull iff more than one port. */
140 bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */
142 /* Port mirroring info. */
143 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
144 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
145 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
148 static void bundle_remove(struct ofport *);
149 static void bundle_destroy(struct ofbundle *);
150 static void bundle_del_port(struct ofport_dpif *);
151 static void bundle_run(struct ofbundle *);
152 static void bundle_wait(struct ofbundle *);
154 struct action_xlate_ctx {
155 /* action_xlate_ctx_init() initializes these members. */
158 struct ofproto_dpif *ofproto;
160 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
161 * this flow when actions change header fields. */
164 /* The packet corresponding to 'flow', or a null pointer if we are
165 * revalidating without a packet to refer to. */
166 const struct ofpbuf *packet;
168 /* If nonnull, called just before executing a resubmit action.
170 * This is normally null so the client has to set it manually after
171 * calling action_xlate_ctx_init(). */
172 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
174 /* xlate_actions() initializes and uses these members. The client might want
175 * to look at them after it returns. */
177 struct ofpbuf *odp_actions; /* Datapath actions. */
178 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
179 bool may_set_up_flow; /* True ordinarily; false if the actions must
180 * be reassessed for every packet. */
181 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
183 /* xlate_actions() initializes and uses these members, but the client has no
184 * reason to look at them. */
186 int recurse; /* Recursion level, via xlate_table_action. */
187 int last_pop_priority; /* Offset in 'odp_actions' just past most
188 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
191 static void action_xlate_ctx_init(struct action_xlate_ctx *,
192 struct ofproto_dpif *, const struct flow *,
193 const struct ofpbuf *);
194 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
195 const union ofp_action *in, size_t n_in);
197 /* An exact-match instantiation of an OpenFlow flow. */
199 long long int used; /* Time last used; time created if not used. */
203 * - Do include packets and bytes sent "by hand", e.g. with
206 * - Do include packets and bytes that were obtained from the datapath
207 * when a flow was deleted (e.g. dpif_flow_del()) or when its
208 * statistics were reset (e.g. dpif_flow_put() with
209 * DPIF_FP_ZERO_STATS).
211 * - Do not include any packets or bytes that can currently be obtained
212 * from the datapath by, e.g., dpif_flow_get().
214 uint64_t packet_count; /* Number of packets received. */
215 uint64_t byte_count; /* Number of bytes received. */
217 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
218 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
220 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
221 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
222 long long int rs_used; /* Used time pushed to resubmit children. */
224 /* Number of bytes passed to account_cb. This may include bytes that can
225 * currently obtained from the datapath (thus, it can be greater than
227 uint64_t accounted_bytes;
229 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
230 struct list list_node; /* In owning rule's 'facets' list. */
231 struct rule_dpif *rule; /* Owning rule. */
232 struct flow flow; /* Exact-match flow. */
233 bool installed; /* Installed in datapath? */
234 bool may_install; /* True ordinarily; false if actions must
235 * be reassessed for every packet. */
236 size_t actions_len; /* Number of bytes in actions[]. */
237 struct nlattr *actions; /* Datapath actions. */
238 tag_type tags; /* Tags. */
239 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
242 static struct facet *facet_create(struct rule_dpif *, const struct flow *,
243 const struct ofpbuf *packet);
244 static void facet_remove(struct ofproto_dpif *, struct facet *);
245 static void facet_free(struct facet *);
247 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
248 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
249 const struct flow *);
250 static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
252 static void facet_execute(struct ofproto_dpif *, struct facet *,
253 struct ofpbuf *packet);
255 static int facet_put__(struct ofproto_dpif *, struct facet *,
256 const struct nlattr *actions, size_t actions_len,
257 struct dpif_flow_stats *);
258 static void facet_install(struct ofproto_dpif *, struct facet *,
260 static void facet_uninstall(struct ofproto_dpif *, struct facet *);
261 static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
263 static void facet_make_actions(struct ofproto_dpif *, struct facet *,
264 const struct ofpbuf *packet);
265 static void facet_update_time(struct ofproto_dpif *, struct facet *,
267 static void facet_update_stats(struct ofproto_dpif *, struct facet *,
268 const struct dpif_flow_stats *);
269 static void facet_push_stats(struct facet *);
270 static void facet_account(struct ofproto_dpif *, struct facet *,
271 uint64_t extra_bytes);
273 static bool facet_is_controller_flow(struct facet *);
275 static void flow_push_stats(const struct rule_dpif *,
276 struct flow *, uint64_t packets, uint64_t bytes,
283 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
284 struct list bundle_node; /* In struct ofbundle's "ports" list. */
285 struct cfm *cfm; /* Connectivity Fault Management, if any. */
286 tag_type tag; /* Tag associated with this port. */
289 static struct ofport_dpif *
290 ofport_dpif_cast(const struct ofport *ofport)
292 assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
293 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
296 static void port_run(struct ofport_dpif *);
297 static void port_wait(struct ofport_dpif *);
298 static int set_cfm(struct ofport *, const struct cfm *,
299 const uint16_t *remote_mps, size_t n_remote_mps);
301 struct ofproto_dpif {
310 struct netflow *netflow;
311 struct ofproto_sflow *sflow;
312 struct hmap bundles; /* Contains "struct ofbundle"s. */
313 struct mac_learning *ml;
314 struct ofmirror *mirrors[MAX_MIRRORS];
315 bool has_bonded_bundles;
318 struct timer next_expiration;
322 bool need_revalidate;
323 struct tag_set revalidate_set;
326 static void ofproto_dpif_unixctl_init(void);
328 static struct ofproto_dpif *
329 ofproto_dpif_cast(const struct ofproto *ofproto)
331 assert(ofproto->ofproto_class == &ofproto_dpif_class);
332 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
335 static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
337 static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
340 /* Packet processing. */
341 static void update_learning_table(struct ofproto_dpif *,
342 const struct flow *, int vlan,
344 static bool is_admissible(struct ofproto_dpif *, const struct flow *,
345 bool have_packet, tag_type *, int *vlanp,
346 struct ofbundle **in_bundlep);
347 static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
349 /* Flow expiration. */
350 static int expire(struct ofproto_dpif *);
353 static int send_packet(struct ofproto_dpif *,
354 uint32_t odp_port, uint16_t vlan_tci,
355 const struct ofpbuf *packet);
357 /* Global variables. */
358 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
360 /* Factory functions. */
363 enumerate_types(struct sset *types)
365 dp_enumerate_types(types);
369 enumerate_names(const char *type, struct sset *names)
371 return dp_enumerate_names(type, names);
375 del(const char *type, const char *name)
380 error = dpif_open(name, type, &dpif);
382 error = dpif_delete(dpif);
388 /* Basic life-cycle. */
390 static struct ofproto *
393 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
398 dealloc(struct ofproto *ofproto_)
400 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
405 construct(struct ofproto *ofproto_)
407 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
408 const char *name = ofproto->up.name;
412 error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
414 VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
418 ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
419 ofproto->n_matches = 0;
421 error = dpif_recv_set_mask(ofproto->dpif,
422 ((1u << DPIF_UC_MISS) |
423 (1u << DPIF_UC_ACTION) |
424 (1u << DPIF_UC_SAMPLE)));
426 VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
427 dpif_close(ofproto->dpif);
430 dpif_flow_flush(ofproto->dpif);
431 dpif_recv_purge(ofproto->dpif);
433 ofproto->netflow = NULL;
434 ofproto->sflow = NULL;
435 hmap_init(&ofproto->bundles);
436 ofproto->ml = mac_learning_create();
437 for (i = 0; i < MAX_MIRRORS; i++) {
438 ofproto->mirrors[i] = NULL;
440 ofproto->has_bonded_bundles = false;
442 timer_set_duration(&ofproto->next_expiration, 1000);
444 hmap_init(&ofproto->facets);
445 ofproto->need_revalidate = false;
446 tag_set_init(&ofproto->revalidate_set);
448 ofproto->up.tables = xmalloc(sizeof *ofproto->up.tables);
449 classifier_init(&ofproto->up.tables[0]);
450 ofproto->up.n_tables = 1;
452 ofproto_dpif_unixctl_init();
458 destruct(struct ofproto *ofproto_)
460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
463 for (i = 0; i < MAX_MIRRORS; i++) {
464 mirror_destroy(ofproto->mirrors[i]);
467 netflow_destroy(ofproto->netflow);
468 ofproto_sflow_destroy(ofproto->sflow);
469 hmap_destroy(&ofproto->bundles);
470 mac_learning_destroy(ofproto->ml);
472 hmap_destroy(&ofproto->facets);
474 dpif_close(ofproto->dpif);
478 run(struct ofproto *ofproto_)
480 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
481 struct ofport_dpif *ofport;
482 struct ofbundle *bundle;
485 dpif_run(ofproto->dpif);
487 for (i = 0; i < 50; i++) {
488 struct dpif_upcall packet;
491 error = dpif_recv(ofproto->dpif, &packet);
493 if (error == ENODEV) {
494 /* Datapath destroyed. */
500 handle_upcall(ofproto, &packet);
503 if (timer_expired(&ofproto->next_expiration)) {
504 int delay = expire(ofproto);
505 timer_set_duration(&ofproto->next_expiration, delay);
508 if (ofproto->netflow) {
509 netflow_run(ofproto->netflow);
511 if (ofproto->sflow) {
512 ofproto_sflow_run(ofproto->sflow);
515 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
518 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
522 /* Now revalidate if there's anything to do. */
523 if (ofproto->need_revalidate
524 || !tag_set_is_empty(&ofproto->revalidate_set)) {
525 struct tag_set revalidate_set = ofproto->revalidate_set;
526 bool revalidate_all = ofproto->need_revalidate;
527 struct facet *facet, *next;
529 /* Clear the revalidation flags. */
530 tag_set_init(&ofproto->revalidate_set);
531 ofproto->need_revalidate = false;
533 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
535 || tag_set_intersects(&revalidate_set, facet->tags)) {
536 facet_revalidate(ofproto, facet);
545 wait(struct ofproto *ofproto_)
547 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
548 struct ofport_dpif *ofport;
549 struct ofbundle *bundle;
551 dpif_wait(ofproto->dpif);
552 dpif_recv_wait(ofproto->dpif);
553 if (ofproto->sflow) {
554 ofproto_sflow_wait(ofproto->sflow);
556 if (!tag_set_is_empty(&ofproto->revalidate_set)) {
557 poll_immediate_wake();
559 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
562 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
565 if (ofproto->need_revalidate) {
566 /* Shouldn't happen, but if it does just go around again. */
567 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
568 poll_immediate_wake();
570 timer_wait(&ofproto->next_expiration);
575 flush(struct ofproto *ofproto_)
577 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
578 struct facet *facet, *next_facet;
580 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
581 /* Mark the facet as not installed so that facet_remove() doesn't
582 * bother trying to uninstall it. There is no point in uninstalling it
583 * individually since we are about to blow away all the facets with
584 * dpif_flow_flush(). */
585 facet->installed = false;
586 facet->dp_packet_count = 0;
587 facet->dp_byte_count = 0;
588 facet_remove(ofproto, facet);
590 dpif_flow_flush(ofproto->dpif);
594 get_features(struct ofproto *ofproto_ OVS_UNUSED,
595 bool *arp_match_ip, uint32_t *actions)
597 *arp_match_ip = true;
598 *actions = ((1u << OFPAT_OUTPUT) |
599 (1u << OFPAT_SET_VLAN_VID) |
600 (1u << OFPAT_SET_VLAN_PCP) |
601 (1u << OFPAT_STRIP_VLAN) |
602 (1u << OFPAT_SET_DL_SRC) |
603 (1u << OFPAT_SET_DL_DST) |
604 (1u << OFPAT_SET_NW_SRC) |
605 (1u << OFPAT_SET_NW_DST) |
606 (1u << OFPAT_SET_NW_TOS) |
607 (1u << OFPAT_SET_TP_SRC) |
608 (1u << OFPAT_SET_TP_DST) |
609 (1u << OFPAT_ENQUEUE));
613 get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
615 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
618 strcpy(ots->name, "classifier");
620 dpif_get_dp_stats(ofproto->dpif, &s);
621 put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
622 put_32aligned_be64(&ots->matched_count,
623 htonll(s.n_hit + ofproto->n_matches));
627 set_netflow(struct ofproto *ofproto_,
628 const struct netflow_options *netflow_options)
630 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
632 if (netflow_options) {
633 if (!ofproto->netflow) {
634 ofproto->netflow = netflow_create();
636 return netflow_set_options(ofproto->netflow, netflow_options);
638 netflow_destroy(ofproto->netflow);
639 ofproto->netflow = NULL;
644 static struct ofport *
647 struct ofport_dpif *port = xmalloc(sizeof *port);
652 port_dealloc(struct ofport *port_)
654 struct ofport_dpif *port = ofport_dpif_cast(port_);
659 port_construct(struct ofport *port_)
661 struct ofport_dpif *port = ofport_dpif_cast(port_);
662 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
664 port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
667 port->tag = tag_create_random();
669 if (ofproto->sflow) {
670 ofproto_sflow_add_port(ofproto->sflow, port->odp_port,
671 netdev_get_name(port->up.netdev));
678 port_destruct(struct ofport *port_)
680 struct ofport_dpif *port = ofport_dpif_cast(port_);
681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
683 bundle_remove(port_);
684 set_cfm(port_, NULL, NULL, 0);
685 if (ofproto->sflow) {
686 ofproto_sflow_del_port(ofproto->sflow, port->odp_port);
691 port_modified(struct ofport *port_)
693 struct ofport_dpif *port = ofport_dpif_cast(port_);
695 if (port->bundle && port->bundle->bond) {
696 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
701 port_reconfigured(struct ofport *port_, ovs_be32 old_config)
703 struct ofport_dpif *port = ofport_dpif_cast(port_);
704 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
705 ovs_be32 changed = old_config ^ port->up.opp.config;
707 if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
708 OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
709 ofproto->need_revalidate = true;
714 set_sflow(struct ofproto *ofproto_,
715 const struct ofproto_sflow_options *sflow_options)
717 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
718 struct ofproto_sflow *os = ofproto->sflow;
721 struct ofport_dpif *ofport;
723 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
724 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
725 ofproto_sflow_add_port(os, ofport->odp_port,
726 netdev_get_name(ofport->up.netdev));
729 ofproto_sflow_set_options(os, sflow_options);
731 ofproto_sflow_destroy(os);
732 ofproto->sflow = NULL;
738 set_cfm(struct ofport *ofport_, const struct cfm *cfm,
739 const uint16_t *remote_mps, size_t n_remote_mps)
741 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
748 ofport->cfm = cfm_create();
751 ofport->cfm->mpid = cfm->mpid;
752 ofport->cfm->interval = cfm->interval;
753 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
755 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
757 if (cfm_configure(ofport->cfm)) {
763 cfm_destroy(ofport->cfm);
769 get_cfm(struct ofport *ofport_, const struct cfm **cfmp)
771 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
778 /* Expires all MAC learning entries associated with 'port' and forces ofproto
779 * to revalidate every flow. */
781 bundle_flush_macs(struct ofbundle *bundle)
783 struct ofproto_dpif *ofproto = bundle->ofproto;
784 struct mac_learning *ml = ofproto->ml;
785 struct mac_entry *mac, *next_mac;
787 ofproto->need_revalidate = true;
788 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
789 if (mac->port.p == bundle) {
790 mac_learning_expire(ml, mac);
795 static struct ofbundle *
796 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
798 struct ofbundle *bundle;
800 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
802 if (bundle->aux == aux) {
809 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
810 * ones that are found to 'bundles'. */
812 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
813 void **auxes, size_t n_auxes,
814 struct hmapx *bundles)
819 for (i = 0; i < n_auxes; i++) {
820 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
822 hmapx_add(bundles, bundle);
828 bundle_del_port(struct ofport_dpif *port)
830 struct ofbundle *bundle = port->bundle;
832 list_remove(&port->bundle_node);
836 lacp_slave_unregister(bundle->lacp, port);
839 bond_slave_unregister(bundle->bond, port);
842 bundle->floodable = true;
843 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
844 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
845 bundle->floodable = false;
851 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
852 struct lacp_slave_settings *lacp)
854 struct ofport_dpif *port;
856 port = get_ofp_port(bundle->ofproto, ofp_port);
861 if (port->bundle != bundle) {
863 bundle_del_port(port);
866 port->bundle = bundle;
867 list_push_back(&bundle->ports, &port->bundle_node);
868 if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
869 bundle->floodable = false;
873 lacp_slave_register(bundle->lacp, port, lacp);
880 bundle_destroy(struct ofbundle *bundle)
882 struct ofproto_dpif *ofproto;
883 struct ofport_dpif *port, *next_port;
890 ofproto = bundle->ofproto;
891 for (i = 0; i < MAX_MIRRORS; i++) {
892 struct ofmirror *m = ofproto->mirrors[i];
894 if (m->out == bundle) {
896 } else if (hmapx_find_and_delete(&m->srcs, bundle)
897 || hmapx_find_and_delete(&m->dsts, bundle)) {
898 ofproto->need_revalidate = true;
903 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
904 bundle_del_port(port);
907 bundle_flush_macs(bundle);
908 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
910 free(bundle->trunks);
911 lacp_destroy(bundle->lacp);
912 bond_destroy(bundle->bond);
917 bundle_set(struct ofproto *ofproto_, void *aux,
918 const struct ofproto_bundle_settings *s)
920 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
921 bool need_flush = false;
922 const unsigned long *trunks;
923 struct ofport_dpif *port;
924 struct ofbundle *bundle;
929 bundle_destroy(bundle_lookup(ofproto, aux));
933 assert(s->n_slaves == 1 || s->bond != NULL);
934 assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
936 bundle = bundle_lookup(ofproto, aux);
938 bundle = xmalloc(sizeof *bundle);
940 bundle->ofproto = ofproto;
941 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
942 hash_pointer(aux, 0));
946 list_init(&bundle->ports);
948 bundle->trunks = NULL;
952 bundle->floodable = true;
954 bundle->src_mirrors = 0;
955 bundle->dst_mirrors = 0;
956 bundle->mirror_out = 0;
959 if (!bundle->name || strcmp(s->name, bundle->name)) {
961 bundle->name = xstrdup(s->name);
967 bundle->lacp = lacp_create();
969 lacp_configure(bundle->lacp, s->lacp);
971 lacp_destroy(bundle->lacp);
975 /* Update set of ports. */
977 for (i = 0; i < s->n_slaves; i++) {
978 if (!bundle_add_port(bundle, s->slaves[i],
979 s->lacp ? &s->lacp_slaves[i] : NULL)) {
983 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
984 struct ofport_dpif *next_port;
986 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
987 for (i = 0; i < s->n_slaves; i++) {
988 if (s->slaves[i] == odp_port_to_ofp_port(port->odp_port)) {
993 bundle_del_port(port);
997 assert(list_size(&bundle->ports) <= s->n_slaves);
999 if (list_is_empty(&bundle->ports)) {
1000 bundle_destroy(bundle);
1005 if (s->vlan != bundle->vlan) {
1006 bundle->vlan = s->vlan;
1010 /* Get trunked VLANs. */
1011 trunks = s->vlan == -1 ? NULL : s->trunks;
1012 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
1013 free(bundle->trunks);
1014 bundle->trunks = vlan_bitmap_clone(trunks);
1019 if (!list_is_short(&bundle->ports)) {
1020 bundle->ofproto->has_bonded_bundles = true;
1022 if (bond_reconfigure(bundle->bond, s->bond)) {
1023 ofproto->need_revalidate = true;
1026 bundle->bond = bond_create(s->bond);
1029 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1030 uint16_t stable_id = (bundle->lacp
1031 ? lacp_slave_get_port_id(bundle->lacp, port)
1033 bond_slave_register(bundle->bond, port, stable_id,
1037 bond_destroy(bundle->bond);
1038 bundle->bond = NULL;
1041 /* If we changed something that would affect MAC learning, un-learn
1042 * everything on this port and force flow revalidation. */
1044 bundle_flush_macs(bundle);
1051 bundle_remove(struct ofport *port_)
1053 struct ofport_dpif *port = ofport_dpif_cast(port_);
1054 struct ofbundle *bundle = port->bundle;
1057 bundle_del_port(port);
1058 if (list_is_empty(&bundle->ports)) {
1059 bundle_destroy(bundle);
1060 } else if (list_is_short(&bundle->ports)) {
1061 bond_destroy(bundle->bond);
1062 bundle->bond = NULL;
1068 send_pdu_cb(void *port_, const struct lacp_pdu *pdu)
1070 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
1071 struct ofport_dpif *port = port_;
1072 uint8_t ea[ETH_ADDR_LEN];
1075 error = netdev_get_etheraddr(port->up.netdev, ea);
1077 struct lacp_pdu *packet_pdu;
1078 struct ofpbuf packet;
1080 ofpbuf_init(&packet, 0);
1081 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
1082 sizeof *packet_pdu);
1084 error = netdev_send(port->up.netdev, &packet);
1086 VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
1087 "(%s)", port->bundle->name,
1088 netdev_get_name(port->up.netdev), strerror(error));
1090 ofpbuf_uninit(&packet);
1092 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
1093 "%s (%s)", port->bundle->name,
1094 netdev_get_name(port->up.netdev), strerror(error));
1099 bundle_send_learning_packets(struct ofbundle *bundle)
1101 struct ofproto_dpif *ofproto = bundle->ofproto;
1102 int error, n_packets, n_errors;
1103 struct mac_entry *e;
1105 error = n_packets = n_errors = 0;
1106 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
1107 if (e->port.p != bundle) {
1108 int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
1118 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1119 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
1120 "packets, last error was: %s",
1121 bundle->name, n_errors, n_packets, strerror(error));
1123 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
1124 bundle->name, n_packets);
1129 bundle_run(struct ofbundle *bundle)
1132 lacp_run(bundle->lacp, send_pdu_cb);
1135 struct ofport_dpif *port;
1137 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1138 bool may_enable = lacp_slave_may_enable(bundle->lacp, port);
1139 bond_slave_set_lacp_may_enable(bundle->bond, port, may_enable);
1142 bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
1143 lacp_negotiated(bundle->lacp));
1144 if (bond_should_send_learning_packets(bundle->bond)) {
1145 bundle_send_learning_packets(bundle);
1151 bundle_wait(struct ofbundle *bundle)
1154 lacp_wait(bundle->lacp);
1157 bond_wait(bundle->bond);
1164 mirror_scan(struct ofproto_dpif *ofproto)
1168 for (idx = 0; idx < MAX_MIRRORS; idx++) {
1169 if (!ofproto->mirrors[idx]) {
1176 static struct ofmirror *
1177 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
1181 for (i = 0; i < MAX_MIRRORS; i++) {
1182 struct ofmirror *mirror = ofproto->mirrors[i];
1183 if (mirror && mirror->aux == aux) {
1192 mirror_set(struct ofproto *ofproto_, void *aux,
1193 const struct ofproto_mirror_settings *s)
1195 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1196 mirror_mask_t mirror_bit;
1197 struct ofbundle *bundle;
1198 struct ofmirror *mirror;
1199 struct ofbundle *out;
1200 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
1201 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
1204 mirror = mirror_lookup(ofproto, aux);
1206 mirror_destroy(mirror);
1212 idx = mirror_scan(ofproto);
1214 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
1216 ofproto->up.name, MAX_MIRRORS, s->name);
1220 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
1221 mirror->ofproto = ofproto;
1223 mirror->out_vlan = -1;
1224 mirror->name = NULL;
1227 if (!mirror->name || strcmp(s->name, mirror->name)) {
1229 mirror->name = xstrdup(s->name);
1232 /* Get the new configuration. */
1233 if (s->out_bundle) {
1234 out = bundle_lookup(ofproto, s->out_bundle);
1236 mirror_destroy(mirror);
1242 out_vlan = s->out_vlan;
1244 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
1245 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
1247 /* If the configuration has not changed, do nothing. */
1248 if (hmapx_equals(&srcs, &mirror->srcs)
1249 && hmapx_equals(&dsts, &mirror->dsts)
1250 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
1251 && mirror->out == out
1252 && mirror->out_vlan == out_vlan)
1254 hmapx_destroy(&srcs);
1255 hmapx_destroy(&dsts);
1259 hmapx_swap(&srcs, &mirror->srcs);
1260 hmapx_destroy(&srcs);
1262 hmapx_swap(&dsts, &mirror->dsts);
1263 hmapx_destroy(&dsts);
1265 free(mirror->vlans);
1266 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
1269 mirror->out_vlan = out_vlan;
1271 /* Update bundles. */
1272 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1273 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
1274 if (hmapx_contains(&mirror->srcs, bundle)) {
1275 bundle->src_mirrors |= mirror_bit;
1277 bundle->src_mirrors &= ~mirror_bit;
1280 if (hmapx_contains(&mirror->dsts, bundle)) {
1281 bundle->dst_mirrors |= mirror_bit;
1283 bundle->dst_mirrors &= ~mirror_bit;
1286 if (mirror->out == bundle) {
1287 bundle->mirror_out |= mirror_bit;
1289 bundle->mirror_out &= ~mirror_bit;
1293 ofproto->need_revalidate = true;
1294 mac_learning_flush(ofproto->ml);
1300 mirror_destroy(struct ofmirror *mirror)
1302 struct ofproto_dpif *ofproto;
1303 mirror_mask_t mirror_bit;
1304 struct ofbundle *bundle;
1310 ofproto = mirror->ofproto;
1311 ofproto->need_revalidate = true;
1312 mac_learning_flush(ofproto->ml);
1314 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
1315 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1316 bundle->src_mirrors &= ~mirror_bit;
1317 bundle->dst_mirrors &= ~mirror_bit;
1318 bundle->mirror_out &= ~mirror_bit;
1321 hmapx_destroy(&mirror->srcs);
1322 hmapx_destroy(&mirror->dsts);
1323 free(mirror->vlans);
1325 ofproto->mirrors[mirror->idx] = NULL;
1331 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
1333 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1334 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
1335 ofproto->need_revalidate = true;
1336 mac_learning_flush(ofproto->ml);
1342 is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
1344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1345 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
1346 return bundle && bundle->mirror_out != 0;
1351 static struct ofport_dpif *
1352 get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
1354 return ofport_dpif_cast(ofproto_get_port(&ofproto->up, ofp_port));
1357 static struct ofport_dpif *
1358 get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
1360 return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
1364 ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
1365 struct dpif_port *dpif_port)
1367 ofproto_port->name = dpif_port->name;
1368 ofproto_port->type = dpif_port->type;
1369 ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
1373 port_run(struct ofport_dpif *ofport)
1376 cfm_run(ofport->cfm);
1378 if (cfm_should_send_ccm(ofport->cfm)) {
1379 struct ofpbuf packet;
1382 ofpbuf_init(&packet, 0);
1383 ccm = eth_compose(&packet, eth_addr_ccm, ofport->up.opp.hw_addr,
1384 ETH_TYPE_CFM, sizeof *ccm);
1385 cfm_compose_ccm(ofport->cfm, ccm);
1386 send_packet(ofproto_dpif_cast(ofport->up.ofproto),
1387 ofport->odp_port, 0, &packet);
1388 ofpbuf_uninit(&packet);
1394 port_wait(struct ofport_dpif *ofport)
1397 cfm_wait(ofport->cfm);
1402 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
1403 struct ofproto_port *ofproto_port)
1405 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1406 struct dpif_port dpif_port;
1409 error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
1411 ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
1417 port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
1419 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1423 error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
1425 *ofp_portp = odp_port_to_ofp_port(odp_port);
1431 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
1433 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1436 error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
1438 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
1440 /* The caller is going to close ofport->up.netdev. If this is a
1441 * bonded port, then the bond is using that netdev, so remove it
1442 * from the bond. The client will need to reconfigure everything
1443 * after deleting ports, so then the slave will get re-added. */
1444 bundle_remove(&ofport->up);
1450 struct port_dump_state {
1451 struct dpif_port_dump dump;
1456 port_dump_start(const struct ofproto *ofproto_, void **statep)
1458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1459 struct port_dump_state *state;
1461 *statep = state = xmalloc(sizeof *state);
1462 dpif_port_dump_start(&state->dump, ofproto->dpif);
1463 state->done = false;
1468 port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
1469 struct ofproto_port *port)
1471 struct port_dump_state *state = state_;
1472 struct dpif_port dpif_port;
1474 if (dpif_port_dump_next(&state->dump, &dpif_port)) {
1475 ofproto_port_from_dpif_port(port, &dpif_port);
1478 int error = dpif_port_dump_done(&state->dump);
1480 return error ? error : EOF;
1485 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
1487 struct port_dump_state *state = state_;
1490 dpif_port_dump_done(&state->dump);
1497 port_poll(const struct ofproto *ofproto_, char **devnamep)
1499 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1500 return dpif_port_poll(ofproto->dpif, devnamep);
1504 port_poll_wait(const struct ofproto *ofproto_)
1506 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1507 dpif_port_poll_wait(ofproto->dpif);
1511 port_is_lacp_current(const struct ofport *ofport_)
1513 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1514 return (ofport->bundle && ofport->bundle->lacp
1515 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
1519 /* Upcall handling. */
1521 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
1522 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
1523 * their individual configurations.
1525 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
1526 * Otherwise, ownership is transferred to this function. */
1528 send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
1529 const struct flow *flow, bool clone)
1531 struct ofputil_packet_in pin;
1533 pin.packet = upcall->packet;
1534 pin.in_port = flow->in_port;
1535 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
1536 pin.buffer_id = 0; /* not yet known */
1537 pin.send_len = upcall->userdata;
1538 connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
1539 clone ? NULL : upcall->packet);
1543 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
1544 const struct ofpbuf *packet)
1546 if (cfm_should_process_flow(flow)) {
1547 struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
1548 if (ofport && ofport->cfm) {
1549 cfm_process_heartbeat(ofport->cfm, packet);
1552 } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
1553 struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
1554 if (port && port->bundle && port->bundle->lacp) {
1555 const struct lacp_pdu *pdu = parse_lacp_packet(packet);
1557 lacp_process_pdu(port->bundle->lacp, port, pdu);
1566 handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1568 struct facet *facet;
1571 /* Obtain in_port and tun_id, at least. */
1572 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1574 /* Set header pointers in 'flow'. */
1575 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
1577 /* Handle 802.1ag and LACP. */
1578 if (process_special(ofproto, &flow, upcall->packet)) {
1579 ofpbuf_delete(upcall->packet);
1580 ofproto->n_matches++;
1584 /* Check with in-band control to see if this packet should be sent
1585 * to the local port regardless of the flow table. */
1586 if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
1587 send_packet(ofproto, OFPP_LOCAL, 0, upcall->packet);
1590 facet = facet_lookup_valid(ofproto, &flow);
1592 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow);
1594 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
1595 struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
1597 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
1598 COVERAGE_INC(ofproto_dpif_no_packet_in);
1599 /* XXX install 'drop' flow entry */
1600 ofpbuf_delete(upcall->packet);
1604 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
1608 send_packet_in(ofproto, upcall, &flow, false);
1612 facet = facet_create(rule, &flow, upcall->packet);
1613 } else if (!facet->may_install) {
1614 /* The facet is not installable, that is, we need to process every
1615 * packet, so process the current packet's actions into 'facet'. */
1616 facet_make_actions(ofproto, facet, upcall->packet);
1619 if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
1621 * Extra-special case for fail-open mode.
1623 * We are in fail-open mode and the packet matched the fail-open rule,
1624 * but we are connected to a controller too. We should send the packet
1625 * up to the controller in the hope that it will try to set up a flow
1626 * and thereby allow us to exit fail-open.
1628 * See the top-level comment in fail-open.c for more information.
1630 send_packet_in(ofproto, upcall, &flow, true);
1633 facet_execute(ofproto, facet, upcall->packet);
1634 facet_install(ofproto, facet, false);
1635 ofproto->n_matches++;
1639 handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
1643 switch (upcall->type) {
1644 case DPIF_UC_ACTION:
1645 COVERAGE_INC(ofproto_dpif_ctlr_action);
1646 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1647 send_packet_in(ofproto, upcall, &flow, false);
1650 case DPIF_UC_SAMPLE:
1651 if (ofproto->sflow) {
1652 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1653 ofproto_sflow_received(ofproto->sflow, upcall, &flow);
1655 ofpbuf_delete(upcall->packet);
1659 handle_miss_upcall(ofproto, upcall);
1662 case DPIF_N_UC_TYPES:
1664 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
1669 /* Flow expiration. */
1671 static int facet_max_idle(const struct ofproto_dpif *);
1672 static void update_stats(struct ofproto_dpif *);
1673 static void rule_expire(struct rule_dpif *);
1674 static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
1676 /* This function is called periodically by run(). Its job is to collect
1677 * updates for the flows that have been installed into the datapath, most
1678 * importantly when they last were used, and then use that information to
1679 * expire flows that have not been used recently.
1681 * Returns the number of milliseconds after which it should be called again. */
1683 expire(struct ofproto_dpif *ofproto)
1685 struct rule_dpif *rule, *next_rule;
1686 struct cls_cursor cursor;
1689 /* Update stats for each flow in the datapath. */
1690 update_stats(ofproto);
1692 /* Expire facets that have been idle too long. */
1693 dp_max_idle = facet_max_idle(ofproto);
1694 expire_facets(ofproto, dp_max_idle);
1696 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
1697 cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
1698 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1702 /* All outstanding data in existing flows has been accounted, so it's a
1703 * good time to do bond rebalancing. */
1704 if (ofproto->has_bonded_bundles) {
1705 struct ofbundle *bundle;
1707 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1709 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
1714 return MIN(dp_max_idle, 1000);
1717 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
1719 * This function also pushes statistics updates to rules which each facet
1720 * resubmits into. Generally these statistics will be accurate. However, if a
1721 * facet changes the rule it resubmits into at some time in between
1722 * update_stats() runs, it is possible that statistics accrued to the
1723 * old rule will be incorrectly attributed to the new rule. This could be
1724 * avoided by calling update_stats() whenever rules are created or
1725 * deleted. However, the performance impact of making so many calls to the
1726 * datapath do not justify the benefit of having perfectly accurate statistics.
1729 update_stats(struct ofproto_dpif *p)
1731 const struct dpif_flow_stats *stats;
1732 struct dpif_flow_dump dump;
1733 const struct nlattr *key;
1736 dpif_flow_dump_start(&dump, p->dpif);
1737 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
1738 struct facet *facet;
1741 if (odp_flow_key_to_flow(key, key_len, &flow)) {
1745 odp_flow_key_format(key, key_len, &s);
1746 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
1752 facet = facet_find(p, &flow);
1754 if (facet && facet->installed) {
1756 if (stats->n_packets >= facet->dp_packet_count) {
1757 uint64_t extra = stats->n_packets - facet->dp_packet_count;
1758 facet->packet_count += extra;
1760 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
1763 if (stats->n_bytes >= facet->dp_byte_count) {
1764 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
1766 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
1769 facet->dp_packet_count = stats->n_packets;
1770 facet->dp_byte_count = stats->n_bytes;
1772 facet_update_time(p, facet, stats->used);
1773 facet_account(p, facet, stats->n_bytes);
1774 facet_push_stats(facet);
1776 /* There's a flow in the datapath that we know nothing about.
1778 COVERAGE_INC(facet_unexpected);
1779 dpif_flow_del(p->dpif, key, key_len, NULL);
1782 dpif_flow_dump_done(&dump);
1785 /* Calculates and returns the number of milliseconds of idle time after which
1786 * facets should expire from the datapath and we should fold their statistics
1787 * into their parent rules in userspace. */
1789 facet_max_idle(const struct ofproto_dpif *ofproto)
1792 * Idle time histogram.
1794 * Most of the time a switch has a relatively small number of facets. When
1795 * this is the case we might as well keep statistics for all of them in
1796 * userspace and to cache them in the kernel datapath for performance as
1799 * As the number of facets increases, the memory required to maintain
1800 * statistics about them in userspace and in the kernel becomes
1801 * significant. However, with a large number of facets it is likely that
1802 * only a few of them are "heavy hitters" that consume a large amount of
1803 * bandwidth. At this point, only heavy hitters are worth caching in the
1804 * kernel and maintaining in userspaces; other facets we can discard.
1806 * The technique used to compute the idle time is to build a histogram with
1807 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
1808 * that is installed in the kernel gets dropped in the appropriate bucket.
1809 * After the histogram has been built, we compute the cutoff so that only
1810 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
1811 * cached. At least the most-recently-used bucket of facets is kept, so
1812 * actually an arbitrary number of facets can be kept in any given
1813 * expiration run (though the next run will delete most of those unless
1814 * they receive additional data).
1816 * This requires a second pass through the facets, in addition to the pass
1817 * made by update_stats(), because the former function never looks
1818 * at uninstallable facets.
1820 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
1821 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
1822 int buckets[N_BUCKETS] = { 0 };
1823 struct facet *facet;
1828 total = hmap_count(&ofproto->facets);
1829 if (total <= 1000) {
1830 return N_BUCKETS * BUCKET_WIDTH;
1833 /* Build histogram. */
1835 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
1836 long long int idle = now - facet->used;
1837 int bucket = (idle <= 0 ? 0
1838 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
1839 : (unsigned int) idle / BUCKET_WIDTH);
1843 /* Find the first bucket whose flows should be expired. */
1844 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
1845 if (buckets[bucket]) {
1848 subtotal += buckets[bucket++];
1849 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
1854 if (VLOG_IS_DBG_ENABLED()) {
1858 ds_put_cstr(&s, "keep");
1859 for (i = 0; i < N_BUCKETS; i++) {
1861 ds_put_cstr(&s, ", drop");
1864 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
1867 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
1871 return bucket * BUCKET_WIDTH;
1875 facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
1877 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
1878 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
1879 struct ofexpired expired;
1881 if (facet->installed) {
1882 struct dpif_flow_stats stats;
1884 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
1886 facet_update_stats(ofproto, facet, &stats);
1889 expired.flow = facet->flow;
1890 expired.packet_count = facet->packet_count;
1891 expired.byte_count = facet->byte_count;
1892 expired.used = facet->used;
1893 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
1898 expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
1900 long long int cutoff = time_msec() - dp_max_idle;
1901 struct facet *facet, *next_facet;
1903 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1904 facet_active_timeout(ofproto, facet);
1905 if (facet->used < cutoff) {
1906 facet_remove(ofproto, facet);
1911 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
1912 * then delete it entirely. */
1914 rule_expire(struct rule_dpif *rule)
1916 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
1917 struct facet *facet, *next_facet;
1921 /* Has 'rule' expired? */
1923 if (rule->up.hard_timeout
1924 && now > rule->up.created + rule->up.hard_timeout * 1000) {
1925 reason = OFPRR_HARD_TIMEOUT;
1926 } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
1927 && now > rule->used + rule->up.idle_timeout * 1000) {
1928 reason = OFPRR_IDLE_TIMEOUT;
1933 COVERAGE_INC(ofproto_dpif_expired);
1935 /* Update stats. (This is a no-op if the rule expired due to an idle
1936 * timeout, because that only happens when the rule has no facets left.) */
1937 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1938 facet_remove(ofproto, facet);
1941 /* Get rid of the rule. */
1942 ofproto_rule_expire(&rule->up, reason);
1947 /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
1948 * example 'packet' within that flow.
1950 * The caller must already have determined that no facet with an identical
1951 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
1952 * the ofproto's classifier table. */
1953 static struct facet *
1954 facet_create(struct rule_dpif *rule, const struct flow *flow,
1955 const struct ofpbuf *packet)
1957 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
1958 struct facet *facet;
1960 facet = xzalloc(sizeof *facet);
1961 facet->used = time_msec();
1962 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
1963 list_push_back(&rule->facets, &facet->list_node);
1965 facet->flow = *flow;
1966 netflow_flow_init(&facet->nf_flow);
1967 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
1969 facet_make_actions(ofproto, facet, packet);
1975 facet_free(struct facet *facet)
1977 free(facet->actions);
1981 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1982 * 'packet', which arrived on 'in_port'.
1984 * Takes ownership of 'packet'. */
1986 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
1987 const struct nlattr *odp_actions, size_t actions_len,
1988 struct ofpbuf *packet)
1990 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
1991 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
1992 /* As an optimization, avoid a round-trip from userspace to kernel to
1993 * userspace. This also avoids possibly filling up kernel packet
1994 * buffers along the way. */
1995 struct dpif_upcall upcall;
1997 upcall.type = DPIF_UC_ACTION;
1998 upcall.packet = packet;
2001 upcall.userdata = nl_attr_get_u64(odp_actions);
2002 upcall.sample_pool = 0;
2003 upcall.actions = NULL;
2004 upcall.actions_len = 0;
2006 send_packet_in(ofproto, &upcall, flow, false);
2012 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2013 ofpbuf_delete(packet);
2018 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2019 * statistics appropriately. 'packet' must have at least sizeof(struct
2020 * ofp_packet_in) bytes of headroom.
2022 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2023 * applying flow_extract() to 'packet' would yield the same flow as
2026 * 'facet' must have accurately composed ODP actions; that is, it must not be
2027 * in need of revalidation.
2029 * Takes ownership of 'packet'. */
2031 facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
2032 struct ofpbuf *packet)
2034 struct dpif_flow_stats stats;
2036 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2038 flow_extract_stats(&facet->flow, packet, &stats);
2039 stats.used = time_msec();
2040 if (execute_odp_actions(ofproto, &facet->flow,
2041 facet->actions, facet->actions_len, packet)) {
2042 facet_update_stats(ofproto, facet, &stats);
2046 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2048 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2049 * rule's statistics, via facet_uninstall().
2051 * - Removes 'facet' from its rule and from ofproto->facets.
2054 facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
2056 facet_uninstall(ofproto, facet);
2057 facet_flush_stats(ofproto, facet);
2058 hmap_remove(&ofproto->facets, &facet->hmap_node);
2059 list_remove(&facet->list_node);
2063 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2065 facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
2066 const struct ofpbuf *packet)
2068 const struct rule_dpif *rule = facet->rule;
2069 struct ofpbuf *odp_actions;
2070 struct action_xlate_ctx ctx;
2072 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2073 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2074 facet->tags = ctx.tags;
2075 facet->may_install = ctx.may_set_up_flow;
2076 facet->nf_flow.output_iface = ctx.nf_output_iface;
2078 if (facet->actions_len != odp_actions->size
2079 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2080 free(facet->actions);
2081 facet->actions_len = odp_actions->size;
2082 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2085 ofpbuf_delete(odp_actions);
2089 facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
2090 const struct nlattr *actions, size_t actions_len,
2091 struct dpif_flow_stats *stats)
2093 struct odputil_keybuf keybuf;
2094 enum dpif_flow_put_flags flags;
2097 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2099 flags |= DPIF_FP_ZERO_STATS;
2100 facet->dp_packet_count = 0;
2101 facet->dp_byte_count = 0;
2104 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2105 odp_flow_key_from_flow(&key, &facet->flow);
2107 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2108 actions, actions_len, stats);
2111 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2112 * 'zero_stats' is true, clears any existing statistics from the datapath for
2115 facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
2117 struct dpif_flow_stats stats;
2119 if (facet->may_install
2120 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2121 zero_stats ? &stats : NULL)) {
2122 facet->installed = true;
2127 facet_account(struct ofproto_dpif *ofproto,
2128 struct facet *facet, uint64_t extra_bytes)
2130 uint64_t total_bytes, n_bytes;
2131 struct ofbundle *in_bundle;
2132 const struct nlattr *a;
2137 total_bytes = facet->byte_count + extra_bytes;
2138 if (total_bytes <= facet->accounted_bytes) {
2141 n_bytes = total_bytes - facet->accounted_bytes;
2142 facet->accounted_bytes = total_bytes;
2144 /* Test that 'tags' is nonzero to ensure that only flows that include an
2145 * OFPP_NORMAL action are used for learning and bond slave rebalancing.
2146 * This works because OFPP_NORMAL always sets a nonzero tag value.
2148 * Feed information from the active flows back into the learning table to
2149 * ensure that table is always in sync with what is actually flowing
2150 * through the datapath. */
2152 || !is_admissible(ofproto, &facet->flow, false, &dummy,
2153 &vlan, &in_bundle)) {
2157 update_learning_table(ofproto, &facet->flow, vlan, in_bundle);
2159 if (!ofproto->has_bonded_bundles) {
2162 NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
2163 if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
2164 struct ofport_dpif *port;
2166 port = get_odp_port(ofproto, nl_attr_get_u32(a));
2167 if (port && port->bundle && port->bundle->bond) {
2168 bond_account(port->bundle->bond, &facet->flow, vlan, n_bytes);
2174 /* If 'rule' is installed in the datapath, uninstalls it. */
2176 facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
2178 if (facet->installed) {
2179 struct odputil_keybuf keybuf;
2180 struct dpif_flow_stats stats;
2183 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2184 odp_flow_key_from_flow(&key, &facet->flow);
2186 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2187 facet_update_stats(p, facet, &stats);
2189 facet->installed = false;
2190 facet->dp_packet_count = 0;
2191 facet->dp_byte_count = 0;
2193 assert(facet->dp_packet_count == 0);
2194 assert(facet->dp_byte_count == 0);
2198 /* Returns true if the only action for 'facet' is to send to the controller.
2199 * (We don't report NetFlow expiration messages for such facets because they
2200 * are just part of the control logic for the network, not real traffic). */
2202 facet_is_controller_flow(struct facet *facet)
2205 && facet->rule->up.n_actions == 1
2206 && action_outputs_to_port(&facet->rule->up.actions[0],
2207 htons(OFPP_CONTROLLER)));
2210 /* Folds all of 'facet''s statistics into its rule. Also updates the
2211 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2212 * 'facet''s statistics in the datapath should have been zeroed and folded into
2213 * its packet and byte counts before this function is called. */
2215 facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
2217 assert(!facet->dp_byte_count);
2218 assert(!facet->dp_packet_count);
2220 facet_push_stats(facet);
2221 facet_account(ofproto, facet, 0);
2223 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2224 struct ofexpired expired;
2225 expired.flow = facet->flow;
2226 expired.packet_count = facet->packet_count;
2227 expired.byte_count = facet->byte_count;
2228 expired.used = facet->used;
2229 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2232 facet->rule->packet_count += facet->packet_count;
2233 facet->rule->byte_count += facet->byte_count;
2235 /* Reset counters to prevent double counting if 'facet' ever gets
2237 facet->packet_count = 0;
2238 facet->byte_count = 0;
2239 facet->rs_packet_count = 0;
2240 facet->rs_byte_count = 0;
2241 facet->accounted_bytes = 0;
2243 netflow_flow_clear(&facet->nf_flow);
2246 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2247 * Returns it if found, otherwise a null pointer.
2249 * The returned facet might need revalidation; use facet_lookup_valid()
2250 * instead if that is important. */
2251 static struct facet *
2252 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
2254 struct facet *facet;
2256 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2258 if (flow_equal(flow, &facet->flow)) {
2266 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2267 * Returns it if found, otherwise a null pointer.
2269 * The returned facet is guaranteed to be valid. */
2270 static struct facet *
2271 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
2273 struct facet *facet = facet_find(ofproto, flow);
2275 /* The facet we found might not be valid, since we could be in need of
2276 * revalidation. If it is not valid, don't return it. */
2278 && ofproto->need_revalidate
2279 && !facet_revalidate(ofproto, facet)) {
2280 COVERAGE_INC(facet_invalidated);
2287 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2289 * - If the rule found is different from 'facet''s current rule, moves
2290 * 'facet' to the new rule and recompiles its actions.
2292 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2293 * where it is and recompiles its actions anyway.
2295 * - If there is none, destroys 'facet'.
2297 * Returns true if 'facet' still exists, false if it has been destroyed. */
2299 facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
2301 struct action_xlate_ctx ctx;
2302 struct ofpbuf *odp_actions;
2303 struct rule_dpif *new_rule;
2304 bool actions_changed;
2306 COVERAGE_INC(facet_revalidate);
2308 /* Determine the new rule. */
2309 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
2311 /* No new rule, so delete the facet. */
2312 facet_remove(ofproto, facet);
2316 /* Calculate new ODP actions.
2318 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2319 * emit a NetFlow expiration and, if so, we need to have the old state
2320 * around to properly compose it. */
2321 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2322 odp_actions = xlate_actions(&ctx,
2323 new_rule->up.actions, new_rule->up.n_actions);
2324 actions_changed = (facet->actions_len != odp_actions->size
2325 || memcmp(facet->actions, odp_actions->data,
2326 facet->actions_len));
2328 /* If the ODP actions changed or the installability changed, then we need
2329 * to talk to the datapath. */
2330 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2331 if (ctx.may_set_up_flow) {
2332 struct dpif_flow_stats stats;
2334 facet_put__(ofproto, facet,
2335 odp_actions->data, odp_actions->size, &stats);
2336 facet_update_stats(ofproto, facet, &stats);
2338 facet_uninstall(ofproto, facet);
2341 /* The datapath flow is gone or has zeroed stats, so push stats out of
2342 * 'facet' into 'rule'. */
2343 facet_flush_stats(ofproto, facet);
2346 /* Update 'facet' now that we've taken care of all the old state. */
2347 facet->tags = ctx.tags;
2348 facet->nf_flow.output_iface = ctx.nf_output_iface;
2349 facet->may_install = ctx.may_set_up_flow;
2350 if (actions_changed) {
2351 free(facet->actions);
2352 facet->actions_len = odp_actions->size;
2353 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2355 if (facet->rule != new_rule) {
2356 COVERAGE_INC(facet_changed_rule);
2357 list_remove(&facet->list_node);
2358 list_push_back(&new_rule->facets, &facet->list_node);
2359 facet->rule = new_rule;
2360 facet->used = new_rule->up.created;
2361 facet->rs_used = facet->used;
2364 ofpbuf_delete(odp_actions);
2369 /* Updates 'facet''s used time. Caller is responsible for calling
2370 * facet_push_stats() to update the flows which 'facet' resubmits into. */
2372 facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
2375 if (used > facet->used) {
2377 if (used > facet->rule->used) {
2378 facet->rule->used = used;
2380 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
2384 /* Folds the statistics from 'stats' into the counters in 'facet'.
2386 * Because of the meaning of a facet's counters, it only makes sense to do this
2387 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
2388 * packet that was sent by hand or if it represents statistics that have been
2389 * cleared out of the datapath. */
2391 facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
2392 const struct dpif_flow_stats *stats)
2394 if (stats->n_packets || stats->used > facet->used) {
2395 facet_update_time(ofproto, facet, stats->used);
2396 facet->packet_count += stats->n_packets;
2397 facet->byte_count += stats->n_bytes;
2398 facet_push_stats(facet);
2399 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
2404 facet_push_stats(struct facet *facet)
2406 uint64_t rs_packets, rs_bytes;
2408 assert(facet->packet_count >= facet->rs_packet_count);
2409 assert(facet->byte_count >= facet->rs_byte_count);
2410 assert(facet->used >= facet->rs_used);
2412 rs_packets = facet->packet_count - facet->rs_packet_count;
2413 rs_bytes = facet->byte_count - facet->rs_byte_count;
2415 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
2416 facet->rs_packet_count = facet->packet_count;
2417 facet->rs_byte_count = facet->byte_count;
2418 facet->rs_used = facet->used;
2420 flow_push_stats(facet->rule, &facet->flow,
2421 rs_packets, rs_bytes, facet->used);
2425 struct ofproto_push {
2426 struct action_xlate_ctx ctx;
2433 push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
2435 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
2438 rule->packet_count += push->packets;
2439 rule->byte_count += push->bytes;
2440 rule->used = MAX(push->used, rule->used);
2444 /* Pushes flow statistics to the rules which 'flow' resubmits into given
2445 * 'rule''s actions. */
2447 flow_push_stats(const struct rule_dpif *rule,
2448 struct flow *flow, uint64_t packets, uint64_t bytes,
2451 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2452 struct ofproto_push push;
2454 push.packets = packets;
2458 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
2459 push.ctx.resubmit_hook = push_resubmit;
2460 ofpbuf_delete(xlate_actions(&push.ctx,
2461 rule->up.actions, rule->up.n_actions));
2466 static struct rule_dpif *
2467 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
2469 return rule_dpif_cast(rule_from_cls_rule(
2470 classifier_lookup(&ofproto->up.tables[0],
2474 static struct rule *
2477 struct rule_dpif *rule = xmalloc(sizeof *rule);
2482 rule_dealloc(struct rule *rule_)
2484 struct rule_dpif *rule = rule_dpif_cast(rule_);
2489 rule_construct(struct rule *rule_)
2491 struct rule_dpif *rule = rule_dpif_cast(rule_);
2492 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2493 struct rule_dpif *old_rule;
2496 error = validate_actions(rule->up.actions, rule->up.n_actions,
2497 &rule->up.cr.flow, ofproto->max_ports);
2502 old_rule = rule_dpif_cast(rule_from_cls_rule(classifier_find_rule_exactly(
2503 &ofproto->up.tables[0],
2506 ofproto_rule_destroy(&old_rule->up);
2509 rule->used = rule->up.created;
2510 rule->packet_count = 0;
2511 rule->byte_count = 0;
2512 list_init(&rule->facets);
2513 classifier_insert(&ofproto->up.tables[0], &rule->up.cr);
2515 ofproto->need_revalidate = true;
2521 rule_destruct(struct rule *rule_)
2523 struct rule_dpif *rule = rule_dpif_cast(rule_);
2524 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2525 struct facet *facet, *next_facet;
2527 classifier_remove(&ofproto->up.tables[0], &rule->up.cr);
2528 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2529 facet_revalidate(ofproto, facet);
2531 ofproto->need_revalidate = true;
2535 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
2537 struct rule_dpif *rule = rule_dpif_cast(rule_);
2538 struct facet *facet;
2540 /* Start from historical data for 'rule' itself that are no longer tracked
2541 * in facets. This counts, for example, facets that have expired. */
2542 *packets = rule->packet_count;
2543 *bytes = rule->byte_count;
2545 /* Add any statistics that are tracked by facets. This includes
2546 * statistical data recently updated by ofproto_update_stats() as well as
2547 * stats for packets that were executed "by hand" via dpif_execute(). */
2548 LIST_FOR_EACH (facet, list_node, &rule->facets) {
2549 *packets += facet->packet_count;
2550 *bytes += facet->byte_count;
2555 rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
2557 struct rule_dpif *rule = rule_dpif_cast(rule_);
2558 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2559 struct action_xlate_ctx ctx;
2560 struct ofpbuf *odp_actions;
2561 struct facet *facet;
2564 /* First look for a related facet. If we find one, account it to that. */
2565 facet = facet_lookup_valid(ofproto, flow);
2566 if (facet && facet->rule == rule) {
2567 facet_execute(ofproto, facet, packet);
2571 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2572 * create a new facet for it and use that. */
2573 if (rule_dpif_lookup(ofproto, flow) == rule) {
2574 facet = facet_create(rule, flow, packet);
2575 facet_execute(ofproto, facet, packet);
2576 facet_install(ofproto, facet, true);
2580 /* We can't account anything to a facet. If we were to try, then that
2581 * facet would have a non-matching rule, busting our invariants. */
2582 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
2583 odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
2584 size = packet->size;
2585 if (execute_odp_actions(ofproto, flow, odp_actions->data,
2586 odp_actions->size, packet)) {
2587 rule->used = time_msec();
2588 rule->packet_count++;
2589 rule->byte_count += size;
2590 flow_push_stats(rule, flow, 1, size, rule->used);
2592 ofpbuf_delete(odp_actions);
2598 rule_modify_actions(struct rule *rule_,
2599 const union ofp_action *actions, size_t n_actions)
2601 struct rule_dpif *rule = rule_dpif_cast(rule_);
2602 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2605 error = validate_actions(actions, n_actions, &rule->up.cr.flow,
2606 ofproto->max_ports);
2608 ofproto->need_revalidate = true;
2613 /* Sends 'packet' out of port 'odp_port' within 'ofproto'. If 'vlan_tci' is
2614 * zero the packet will not have any 802.1Q hader; if it is nonzero, then the
2615 * packet will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
2617 * Returns 0 if successful, otherwise a positive errno value. */
2619 send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, uint16_t vlan_tci,
2620 const struct ofpbuf *packet)
2622 struct ofpbuf odp_actions;
2625 ofpbuf_init(&odp_actions, 32);
2626 if (vlan_tci != 0) {
2627 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2628 ntohs(vlan_tci & ~VLAN_CFI));
2630 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2631 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
2633 ofpbuf_uninit(&odp_actions);
2636 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
2637 ofproto->up.name, odp_port, strerror(error));
2642 /* OpenFlow to ODP action translation. */
2644 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2645 struct action_xlate_ctx *ctx);
2646 static bool xlate_normal(struct action_xlate_ctx *);
2649 add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
2651 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
2652 uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
2655 if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
2656 /* Forwarding disabled on port. */
2661 * We don't have an ofport record for this port, but it doesn't hurt to
2662 * allow forwarding to it anyhow. Maybe such a port will appear later
2663 * and we're pre-populating the flow table.
2667 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2668 ctx->nf_output_iface = ofp_port;
2672 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2674 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2675 struct rule_dpif *rule;
2676 uint16_t old_in_port;
2678 /* Look up a flow with 'in_port' as the input port. Then restore the
2679 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2680 * have surprising behavior). */
2681 old_in_port = ctx->flow.in_port;
2682 ctx->flow.in_port = in_port;
2683 rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow);
2684 ctx->flow.in_port = old_in_port;
2686 if (ctx->resubmit_hook) {
2687 ctx->resubmit_hook(ctx, rule);
2692 do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
2696 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2698 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2699 MAX_RESUBMIT_RECURSION);
2704 flood_packets(struct ofproto_dpif *ofproto,
2705 uint16_t ofp_in_port, ovs_be32 mask,
2706 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2708 struct ofport_dpif *ofport;
2710 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
2711 uint16_t ofp_port = ofport->up.ofp_port;
2712 if (ofp_port != ofp_in_port && !(ofport->up.opp.config & mask)) {
2713 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT,
2717 *nf_output_iface = NF_OUT_FLOOD;
2721 xlate_output_action__(struct action_xlate_ctx *ctx,
2722 uint16_t port, uint16_t max_len)
2724 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2726 ctx->nf_output_iface = NF_OUT_DROP;
2730 add_output_action(ctx, ctx->flow.in_port);
2733 xlate_table_action(ctx, ctx->flow.in_port);
2739 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD),
2740 &ctx->nf_output_iface, ctx->odp_actions);
2743 flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0),
2744 &ctx->nf_output_iface, ctx->odp_actions);
2746 case OFPP_CONTROLLER:
2747 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2750 add_output_action(ctx, OFPP_LOCAL);
2753 if (port != ctx->flow.in_port) {
2754 add_output_action(ctx, port);
2759 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2760 ctx->nf_output_iface = NF_OUT_FLOOD;
2761 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2762 ctx->nf_output_iface = prev_nf_output_iface;
2763 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2764 ctx->nf_output_iface != NF_OUT_FLOOD) {
2765 ctx->nf_output_iface = NF_OUT_MULTI;
2770 xlate_output_action(struct action_xlate_ctx *ctx,
2771 const struct ofp_action_output *oao)
2773 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2776 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2777 * optimization, because we're going to add another action that sets the
2778 * priority immediately after, or because there are no actions following the
2781 remove_pop_action(struct action_xlate_ctx *ctx)
2783 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2784 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2785 ctx->last_pop_priority = -1;
2790 add_pop_action(struct action_xlate_ctx *ctx)
2792 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2793 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2794 ctx->last_pop_priority = ctx->odp_actions->size;
2799 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2800 const struct ofp_action_enqueue *oae)
2802 uint16_t ofp_port, odp_port;
2806 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2809 /* Fall back to ordinary output action. */
2810 xlate_output_action__(ctx, ntohs(oae->port), 0);
2814 /* Figure out ODP output port. */
2815 ofp_port = ntohs(oae->port);
2816 if (ofp_port == OFPP_IN_PORT) {
2817 ofp_port = ctx->flow.in_port;
2819 odp_port = ofp_port_to_odp_port(ofp_port);
2821 /* Add ODP actions. */
2822 remove_pop_action(ctx);
2823 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2824 add_output_action(ctx, odp_port);
2825 add_pop_action(ctx);
2827 /* Update NetFlow output port. */
2828 if (ctx->nf_output_iface == NF_OUT_DROP) {
2829 ctx->nf_output_iface = odp_port;
2830 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2831 ctx->nf_output_iface = NF_OUT_MULTI;
2836 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2837 const struct nx_action_set_queue *nasq)
2842 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2845 /* Couldn't translate queue to a priority, so ignore. A warning
2846 * has already been logged. */
2850 remove_pop_action(ctx);
2851 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2855 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2857 ovs_be16 tci = ctx->flow.vlan_tci;
2858 if (!(tci & htons(VLAN_CFI))) {
2859 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2861 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2862 tci & ~htons(VLAN_CFI));
2866 struct xlate_reg_state {
2872 save_reg_state(const struct action_xlate_ctx *ctx,
2873 struct xlate_reg_state *state)
2875 state->vlan_tci = ctx->flow.vlan_tci;
2876 state->tun_id = ctx->flow.tun_id;
2880 update_reg_state(struct action_xlate_ctx *ctx,
2881 const struct xlate_reg_state *state)
2883 if (ctx->flow.vlan_tci != state->vlan_tci) {
2884 xlate_set_dl_tci(ctx);
2886 if (ctx->flow.tun_id != state->tun_id) {
2887 nl_msg_put_be64(ctx->odp_actions,
2888 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2893 xlate_autopath(struct action_xlate_ctx *ctx,
2894 const struct nx_action_autopath *naa)
2896 uint16_t ofp_port = ntohl(naa->id);
2897 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
2899 if (!port || !port->bundle) {
2900 ofp_port = OFPP_NONE;
2901 } else if (port->bundle->bond) {
2902 /* Autopath does not support VLAN hashing. */
2903 struct ofport_dpif *slave = bond_choose_output_slave(
2904 port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags);
2906 ofp_port = slave->up.ofp_port;
2909 autopath_execute(naa, &ctx->flow, ofp_port);
2913 xlate_nicira_action(struct action_xlate_ctx *ctx,
2914 const struct nx_action_header *nah)
2916 const struct nx_action_resubmit *nar;
2917 const struct nx_action_set_tunnel *nast;
2918 const struct nx_action_set_queue *nasq;
2919 const struct nx_action_multipath *nam;
2920 const struct nx_action_autopath *naa;
2921 enum nx_action_subtype subtype = ntohs(nah->subtype);
2922 struct xlate_reg_state state;
2925 assert(nah->vendor == htonl(NX_VENDOR_ID));
2927 case NXAST_RESUBMIT:
2928 nar = (const struct nx_action_resubmit *) nah;
2929 xlate_table_action(ctx, ntohs(nar->in_port));
2932 case NXAST_SET_TUNNEL:
2933 nast = (const struct nx_action_set_tunnel *) nah;
2934 tun_id = htonll(ntohl(nast->tun_id));
2935 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2936 ctx->flow.tun_id = tun_id;
2939 case NXAST_DROP_SPOOFED_ARP:
2940 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2941 nl_msg_put_flag(ctx->odp_actions,
2942 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2946 case NXAST_SET_QUEUE:
2947 nasq = (const struct nx_action_set_queue *) nah;
2948 xlate_set_queue_action(ctx, nasq);
2951 case NXAST_POP_QUEUE:
2952 add_pop_action(ctx);
2955 case NXAST_REG_MOVE:
2956 save_reg_state(ctx, &state);
2957 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2959 update_reg_state(ctx, &state);
2962 case NXAST_REG_LOAD:
2963 save_reg_state(ctx, &state);
2964 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2966 update_reg_state(ctx, &state);
2970 /* Nothing to do. */
2973 case NXAST_SET_TUNNEL64:
2974 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2975 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2976 ctx->flow.tun_id = tun_id;
2979 case NXAST_MULTIPATH:
2980 nam = (const struct nx_action_multipath *) nah;
2981 multipath_execute(nam, &ctx->flow);
2984 case NXAST_AUTOPATH:
2985 naa = (const struct nx_action_autopath *) nah;
2986 xlate_autopath(ctx, naa);
2989 /* If you add a new action here that modifies flow data, don't forget to
2990 * update the flow key in ctx->flow at the same time. */
2992 case NXAST_SNAT__OBSOLETE:
2994 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3000 do_xlate_actions(const union ofp_action *in, size_t n_in,
3001 struct action_xlate_ctx *ctx)
3003 const struct ofport_dpif *port;
3004 struct actions_iterator iter;
3005 const union ofp_action *ia;
3007 port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
3009 && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3010 port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3011 ? htonl(OFPPC_NO_RECV_STP)
3012 : htonl(OFPPC_NO_RECV))) {
3013 /* Drop this flow. */
3017 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3018 enum ofp_action_type type = ntohs(ia->type);
3019 const struct ofp_action_dl_addr *oada;
3023 xlate_output_action(ctx, &ia->output);
3026 case OFPAT_SET_VLAN_VID:
3027 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3028 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3029 xlate_set_dl_tci(ctx);
3032 case OFPAT_SET_VLAN_PCP:
3033 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3034 ctx->flow.vlan_tci |= htons(
3035 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3036 xlate_set_dl_tci(ctx);
3039 case OFPAT_STRIP_VLAN:
3040 ctx->flow.vlan_tci = htons(0);
3041 xlate_set_dl_tci(ctx);
3044 case OFPAT_SET_DL_SRC:
3045 oada = ((struct ofp_action_dl_addr *) ia);
3046 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3047 oada->dl_addr, ETH_ADDR_LEN);
3048 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3051 case OFPAT_SET_DL_DST:
3052 oada = ((struct ofp_action_dl_addr *) ia);
3053 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3054 oada->dl_addr, ETH_ADDR_LEN);
3055 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3058 case OFPAT_SET_NW_SRC:
3059 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3060 ia->nw_addr.nw_addr);
3061 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3064 case OFPAT_SET_NW_DST:
3065 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3066 ia->nw_addr.nw_addr);
3067 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3070 case OFPAT_SET_NW_TOS:
3071 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3073 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3076 case OFPAT_SET_TP_SRC:
3077 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3078 ia->tp_port.tp_port);
3079 ctx->flow.tp_src = ia->tp_port.tp_port;
3082 case OFPAT_SET_TP_DST:
3083 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3084 ia->tp_port.tp_port);
3085 ctx->flow.tp_dst = ia->tp_port.tp_port;
3089 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3093 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3097 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3104 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3105 struct ofproto_dpif *ofproto, const struct flow *flow,
3106 const struct ofpbuf *packet)
3108 ctx->ofproto = ofproto;
3110 ctx->packet = packet;
3111 ctx->resubmit_hook = NULL;
3114 static struct ofpbuf *
3115 xlate_actions(struct action_xlate_ctx *ctx,
3116 const union ofp_action *in, size_t n_in)
3118 COVERAGE_INC(ofproto_dpif_xlate);
3120 ctx->odp_actions = ofpbuf_new(512);
3122 ctx->may_set_up_flow = true;
3123 ctx->nf_output_iface = NF_OUT_DROP;
3125 ctx->last_pop_priority = -1;
3127 if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
3128 ctx->may_set_up_flow = false;
3130 do_xlate_actions(in, n_in, ctx);
3133 remove_pop_action(ctx);
3135 /* Check with in-band control to see if we're allowed to set up this
3137 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
3138 ctx->odp_actions->data,
3139 ctx->odp_actions->size)) {
3140 ctx->may_set_up_flow = false;
3143 return ctx->odp_actions;
3146 /* OFPP_NORMAL implementation. */
3149 struct ofport_dpif *port;
3154 struct dst builtin[32];
3156 size_t n, allocated;
3159 static void dst_set_init(struct dst_set *);
3160 static void dst_set_add(struct dst_set *, const struct dst *);
3161 static void dst_set_free(struct dst_set *);
3163 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
3166 set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
3167 const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
3169 dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE
3170 : in_bundle->vlan >= 0 ? in_bundle->vlan
3171 : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE
3172 : vlan_tci_to_vid(ctx->flow.vlan_tci));
3174 dst->port = (!out_bundle->bond
3175 ? ofbundle_get_a_port(out_bundle)
3176 : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
3177 dst->vlan, &ctx->tags));
3179 return dst->port != NULL;
3183 mirror_mask_ffs(mirror_mask_t mask)
3185 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
3190 dst_set_init(struct dst_set *set)
3192 set->dsts = set->builtin;
3194 set->allocated = ARRAY_SIZE(set->builtin);
3198 dst_set_add(struct dst_set *set, const struct dst *dst)
3200 if (set->n >= set->allocated) {
3201 size_t new_allocated;
3202 struct dst *new_dsts;
3204 new_allocated = set->allocated * 2;
3205 new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
3206 memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
3210 set->dsts = new_dsts;
3211 set->allocated = new_allocated;
3213 set->dsts[set->n++] = *dst;
3217 dst_set_free(struct dst_set *set)
3219 if (set->dsts != set->builtin) {
3225 dst_is_duplicate(const struct dst_set *set, const struct dst *test)
3228 for (i = 0; i < set->n; i++) {
3229 if (set->dsts[i].vlan == test->vlan
3230 && set->dsts[i].port == test->port) {
3238 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
3240 return bundle->vlan < 0 && vlan_bitmap_contains(bundle->trunks, vlan);
3244 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
3246 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
3249 /* Returns an arbitrary interface within 'bundle'. */
3250 static struct ofport_dpif *
3251 ofbundle_get_a_port(const struct ofbundle *bundle)
3253 return CONTAINER_OF(list_front(&bundle->ports),
3254 struct ofport_dpif, bundle_node);
3258 compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
3259 const struct ofbundle *in_bundle,
3260 const struct ofbundle *out_bundle, struct dst_set *set)
3264 if (out_bundle == OFBUNDLE_FLOOD) {
3265 struct ofbundle *bundle;
3267 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
3268 if (bundle != in_bundle
3269 && ofbundle_includes_vlan(bundle, vlan)
3270 && bundle->floodable
3271 && !bundle->mirror_out
3272 && set_dst(ctx, &dst, in_bundle, bundle)) {
3273 dst_set_add(set, &dst);
3276 ctx->nf_output_iface = NF_OUT_FLOOD;
3277 } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
3278 dst_set_add(set, &dst);
3279 ctx->nf_output_iface = dst.port->odp_port;
3284 vlan_is_mirrored(const struct ofmirror *m, int vlan)
3286 return vlan_bitmap_contains(m->vlans, vlan);
3290 compose_mirror_dsts(struct action_xlate_ctx *ctx,
3291 uint16_t vlan, const struct ofbundle *in_bundle,
3292 struct dst_set *set)
3294 struct ofproto_dpif *ofproto = ctx->ofproto;
3295 mirror_mask_t mirrors;
3299 mirrors = in_bundle->src_mirrors;
3300 for (i = 0; i < set->n; i++) {
3301 mirrors |= set->dsts[i].port->bundle->dst_mirrors;
3308 flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3309 if (flow_vlan == 0) {
3310 flow_vlan = OFP_VLAN_NONE;
3314 struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
3315 if (vlan_is_mirrored(m, vlan)) {
3319 if (set_dst(ctx, &dst, in_bundle, m->out)
3320 && !dst_is_duplicate(set, &dst)) {
3321 dst_set_add(set, &dst);
3324 struct ofbundle *bundle;
3326 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3327 if (ofbundle_includes_vlan(bundle, m->out_vlan)
3328 && set_dst(ctx, &dst, in_bundle, bundle))
3330 if (bundle->vlan < 0) {
3331 dst.vlan = m->out_vlan;
3333 if (dst_is_duplicate(set, &dst)) {
3337 /* Use the vlan tag on the original flow instead of
3338 * the one passed in the vlan parameter. This ensures
3339 * that we compare the vlan from before any implicit
3340 * tagging tags place. This is necessary because
3341 * dst->vlan is the final vlan, after removing implicit
3343 if (bundle == in_bundle && dst.vlan == flow_vlan) {
3344 /* Don't send out input port on same VLAN. */
3347 dst_set_add(set, &dst);
3352 mirrors &= mirrors - 1;
3357 compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
3358 const struct ofbundle *in_bundle,
3359 const struct ofbundle *out_bundle)
3361 uint16_t initial_vlan, cur_vlan;
3362 const struct dst *dst;
3366 compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
3367 compose_mirror_dsts(ctx, vlan, in_bundle, &set);
3369 /* Output all the packets we can without having to change the VLAN. */
3370 initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
3371 if (initial_vlan == 0) {
3372 initial_vlan = OFP_VLAN_NONE;
3374 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3375 if (dst->vlan != initial_vlan) {
3378 nl_msg_put_u32(ctx->odp_actions,
3379 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3382 /* Then output the rest. */
3383 cur_vlan = initial_vlan;
3384 for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
3385 if (dst->vlan == initial_vlan) {
3388 if (dst->vlan != cur_vlan) {
3389 if (dst->vlan == OFP_VLAN_NONE) {
3390 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
3393 tci = htons(dst->vlan & VLAN_VID_MASK);
3394 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
3395 nl_msg_put_be16(ctx->odp_actions,
3396 ODP_ACTION_ATTR_SET_DL_TCI, tci);
3398 cur_vlan = dst->vlan;
3400 nl_msg_put_u32(ctx->odp_actions,
3401 ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
3407 /* Returns the effective vlan of a packet, taking into account both the
3408 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
3409 * the packet is untagged and -1 indicates it has an invalid header and
3410 * should be dropped. */
3412 flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
3413 struct ofbundle *in_bundle, bool have_packet)
3415 int vlan = vlan_tci_to_vid(flow->vlan_tci);
3416 if (in_bundle->vlan >= 0) {
3419 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3420 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3421 "packet received on port %s configured with "
3422 "implicit VLAN %"PRIu16,
3423 ofproto->up.name, vlan,
3424 in_bundle->name, in_bundle->vlan);
3428 vlan = in_bundle->vlan;
3430 if (!ofbundle_includes_vlan(in_bundle, vlan)) {
3432 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3433 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
3434 "packet received on port %s not configured for "
3436 ofproto->up.name, vlan, in_bundle->name, vlan);
3445 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
3446 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
3447 * indicate this; newer upstream kernels use gratuitous ARP requests. */
3449 is_gratuitous_arp(const struct flow *flow)
3451 return (flow->dl_type == htons(ETH_TYPE_ARP)
3452 && eth_addr_is_broadcast(flow->dl_dst)
3453 && (flow->nw_proto == ARP_OP_REPLY
3454 || (flow->nw_proto == ARP_OP_REQUEST
3455 && flow->nw_src == flow->nw_dst)));
3459 update_learning_table(struct ofproto_dpif *ofproto,
3460 const struct flow *flow, int vlan,
3461 struct ofbundle *in_bundle)
3463 struct mac_entry *mac;
3465 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
3469 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
3470 if (is_gratuitous_arp(flow)) {
3471 /* We don't want to learn from gratuitous ARP packets that are
3472 * reflected back over bond slaves so we lock the learning table. */
3473 if (!in_bundle->bond) {
3474 mac_entry_set_grat_arp_lock(mac);
3475 } else if (mac_entry_is_grat_arp_locked(mac)) {
3480 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
3481 /* The log messages here could actually be useful in debugging,
3482 * so keep the rate limit relatively high. */
3483 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3484 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
3485 "on port %s in VLAN %d",
3486 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
3487 in_bundle->name, vlan);
3489 mac->port.p = in_bundle;
3490 tag_set_add(&ofproto->revalidate_set,
3491 mac_learning_changed(ofproto->ml, mac));
3495 /* Determines whether packets in 'flow' within 'br' should be forwarded or
3496 * dropped. Returns true if they may be forwarded, false if they should be
3499 * If 'have_packet' is true, it indicates that the caller is processing a
3500 * received packet. If 'have_packet' is false, then the caller is just
3501 * revalidating an existing flow because configuration has changed. Either
3502 * way, 'have_packet' only affects logging (there is no point in logging errors
3503 * during revalidation).
3505 * Sets '*in_portp' to the input port. This will be a null pointer if
3506 * flow->in_port does not designate a known input port (in which case
3507 * is_admissible() returns false).
3509 * When returning true, sets '*vlanp' to the effective VLAN of the input
3510 * packet, as returned by flow_get_vlan().
3512 * May also add tags to '*tags', although the current implementation only does
3513 * so in one special case.
3516 is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
3518 tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
3520 struct ofport_dpif *in_port;
3521 struct ofbundle *in_bundle;
3524 /* Find the port and bundle for the received packet. */
3525 in_port = get_ofp_port(ofproto, flow->in_port);
3526 *in_bundlep = in_bundle = in_port->bundle;
3527 if (!in_port || !in_bundle) {
3528 /* No interface? Something fishy... */
3530 /* Odd. A few possible reasons here:
3532 * - We deleted a port but there are still a few packets queued up
3535 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
3536 * we don't know about.
3538 * - Packet arrived on the local port but the local port is not
3541 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3543 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
3545 ofproto->up.name, flow->in_port);
3549 *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
3554 /* Drop frames for reserved multicast addresses. */
3555 if (eth_addr_is_reserved(flow->dl_dst)) {
3559 /* Drop frames on bundles reserved for mirroring. */
3560 if (in_bundle->mirror_out) {
3562 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3563 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
3564 "%s, which is reserved exclusively for mirroring",
3565 ofproto->up.name, in_bundle->name);
3570 if (in_bundle->bond) {
3571 struct mac_entry *mac;
3573 switch (bond_check_admissibility(in_bundle->bond, in_port,
3574 flow->dl_dst, tags)) {
3581 case BV_DROP_IF_MOVED:
3582 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
3583 if (mac && mac->port.p != in_bundle &&
3584 (!is_gratuitous_arp(flow)
3585 || mac_entry_is_grat_arp_locked(mac))) {
3595 /* If the composed actions may be applied to any packet in the given 'flow',
3596 * returns true. Otherwise, the actions should only be applied to 'packet', or
3597 * not at all, if 'packet' was NULL. */
3599 xlate_normal(struct action_xlate_ctx *ctx)
3601 struct ofbundle *in_bundle;
3602 struct ofbundle *out_bundle;
3603 struct mac_entry *mac;
3606 /* Check whether we should drop packets in this flow. */
3607 if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
3608 &ctx->tags, &vlan, &in_bundle)) {
3613 /* Learn source MAC (but don't try to learn from revalidation). */
3615 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
3618 /* Determine output bundle. */
3619 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
3622 out_bundle = mac->port.p;
3623 } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
3624 /* If we are revalidating but don't have a learning entry then eject
3625 * the flow. Installing a flow that floods packets opens up a window
3626 * of time where we could learn from a packet reflected on a bond and
3627 * blackhole packets before the learning table is updated to reflect
3628 * the correct port. */
3631 out_bundle = OFBUNDLE_FLOOD;
3634 /* Don't send packets out their input bundles. */
3635 if (in_bundle == out_bundle) {
3641 compose_actions(ctx, vlan, in_bundle, out_bundle);
3648 get_drop_frags(struct ofproto *ofproto_)
3650 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3653 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
3658 set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
3660 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3662 dpif_set_drop_frags(ofproto->dpif, drop_frags);
3666 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3667 const struct flow *flow,
3668 const union ofp_action *ofp_actions, size_t n_ofp_actions)
3670 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3673 error = validate_actions(ofp_actions, n_ofp_actions, flow,
3674 ofproto->max_ports);
3676 struct action_xlate_ctx ctx;
3677 struct ofpbuf *odp_actions;
3679 action_xlate_ctx_init(&ctx, ofproto, flow, packet);
3680 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3681 dpif_execute(ofproto->dpif, odp_actions->data, odp_actions->size,
3683 ofpbuf_delete(odp_actions);
3689 get_netflow_ids(const struct ofproto *ofproto_,
3690 uint8_t *engine_type, uint8_t *engine_id)
3692 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3694 dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
3697 static struct ofproto_dpif *
3698 ofproto_dpif_lookup(const char *name)
3700 struct ofproto *ofproto = ofproto_lookup(name);
3701 return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
3702 ? ofproto_dpif_cast(ofproto)
3707 ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
3708 const char *args, void *aux OVS_UNUSED)
3710 struct ds ds = DS_EMPTY_INITIALIZER;
3711 const struct ofproto_dpif *ofproto;
3712 const struct mac_entry *e;
3714 ofproto = ofproto_dpif_lookup(args);
3716 unixctl_command_reply(conn, 501, "no such bridge");
3720 ds_put_cstr(&ds, " port VLAN MAC Age\n");
3721 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3722 struct ofbundle *bundle = e->port.p;
3723 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
3724 ofbundle_get_a_port(bundle)->odp_port,
3725 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
3727 unixctl_command_reply(conn, 200, ds_cstr(&ds));
3731 struct ofproto_trace {
3732 struct action_xlate_ctx ctx;
3738 trace_format_rule(struct ds *result, int level, const struct rule *rule)
3740 ds_put_char_multiple(result, '\t', level);
3742 ds_put_cstr(result, "No match\n");
3746 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
3747 ntohll(rule->flow_cookie));
3748 cls_rule_format(&rule->cr, result);
3749 ds_put_char(result, '\n');
3751 ds_put_char_multiple(result, '\t', level);
3752 ds_put_cstr(result, "OpenFlow ");
3753 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
3754 rule->n_actions * sizeof *rule->actions);
3755 ds_put_char(result, '\n');
3759 trace_format_flow(struct ds *result, int level, const char *title,
3760 struct ofproto_trace *trace)
3762 ds_put_char_multiple(result, '\t', level);
3763 ds_put_format(result, "%s: ", title);
3764 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
3765 ds_put_cstr(result, "unchanged");
3767 flow_format(result, &trace->ctx.flow);
3768 trace->flow = trace->ctx.flow;
3770 ds_put_char(result, '\n');
3774 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
3776 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
3777 struct ds *result = trace->result;
3779 ds_put_char(result, '\n');
3780 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
3781 trace_format_rule(result, ctx->recurse + 1, &rule->up);
3785 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
3786 void *aux OVS_UNUSED)
3788 char *dpname, *in_port_s, *tun_id_s, *packet_s;
3789 char *args = xstrdup(args_);
3790 char *save_ptr = NULL;
3791 struct ofproto_dpif *ofproto;
3792 struct ofpbuf packet;
3793 struct rule_dpif *rule;
3800 ofpbuf_init(&packet, strlen(args) / 2);
3803 dpname = strtok_r(args, " ", &save_ptr);
3804 tun_id_s = strtok_r(NULL, " ", &save_ptr);
3805 in_port_s = strtok_r(NULL, " ", &save_ptr);
3806 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
3807 if (!dpname || !in_port_s || !packet_s) {
3808 unixctl_command_reply(conn, 501, "Bad command syntax");
3812 ofproto = ofproto_dpif_lookup(dpname);
3814 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
3819 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
3820 in_port = ofp_port_to_odp_port(atoi(in_port_s));
3822 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
3823 packet_s += strspn(packet_s, " ");
3824 if (*packet_s != '\0') {
3825 unixctl_command_reply(conn, 501, "Trailing garbage in command");
3828 if (packet.size < ETH_HEADER_LEN) {
3829 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
3833 ds_put_cstr(&result, "Packet: ");
3834 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
3835 ds_put_cstr(&result, s);
3838 flow_extract(&packet, tun_id, in_port, &flow);
3839 ds_put_cstr(&result, "Flow: ");
3840 flow_format(&result, &flow);
3841 ds_put_char(&result, '\n');
3843 rule = rule_dpif_lookup(ofproto, &flow);
3844 trace_format_rule(&result, 0, &rule->up);
3846 struct ofproto_trace trace;
3847 struct ofpbuf *odp_actions;
3849 trace.result = &result;
3851 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
3852 trace.ctx.resubmit_hook = trace_resubmit;
3853 odp_actions = xlate_actions(&trace.ctx,
3854 rule->up.actions, rule->up.n_actions);
3856 ds_put_char(&result, '\n');
3857 trace_format_flow(&result, 0, "Final flow", &trace);
3858 ds_put_cstr(&result, "Datapath actions: ");
3859 format_odp_actions(&result, odp_actions->data, odp_actions->size);
3860 ofpbuf_delete(odp_actions);
3863 unixctl_command_reply(conn, 200, ds_cstr(&result));
3866 ds_destroy(&result);
3867 ofpbuf_uninit(&packet);
3872 ofproto_dpif_unixctl_init(void)
3874 static bool registered;
3880 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
3881 unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
3884 const struct ofproto_class ofproto_dpif_class = {
3911 port_is_lacp_current,
3918 rule_modify_actions,
3931 is_mirror_output_bundle,