2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
29 #include "classifier.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "multipath.h"
44 #include "ofp-print.h"
46 #include "ofproto-sflow.h"
48 #include "openflow/nicira-ext.h"
49 #include "openflow/openflow.h"
50 #include "openvswitch/datapath-protocol.h"
54 #include "poll-loop.h"
57 #include "stream-ssl.h"
61 #include "unaligned.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(odp_overflow);
71 COVERAGE_DEFINE(ofproto_agg_request);
72 COVERAGE_DEFINE(ofproto_costly_flags);
73 COVERAGE_DEFINE(ofproto_ctlr_action);
74 COVERAGE_DEFINE(ofproto_del_rule);
75 COVERAGE_DEFINE(ofproto_error);
76 COVERAGE_DEFINE(ofproto_expiration);
77 COVERAGE_DEFINE(ofproto_expired);
78 COVERAGE_DEFINE(ofproto_flows_req);
79 COVERAGE_DEFINE(ofproto_flush);
80 COVERAGE_DEFINE(ofproto_invalidated);
81 COVERAGE_DEFINE(ofproto_no_packet_in);
82 COVERAGE_DEFINE(ofproto_ofconn_stuck);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_unexpected_rule);
90 COVERAGE_DEFINE(ofproto_uninstallable);
91 COVERAGE_DEFINE(ofproto_update_port);
93 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
94 * flow translation. */
95 #define MAX_RESUBMIT_RECURSION 16
100 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
101 struct netdev *netdev;
102 struct ofp_phy_port opp; /* In host byte order. */
104 struct cfm *cfm; /* Connectivity Fault Management, if any. */
107 static void ofport_free(struct ofport *);
108 static void ofport_run(struct ofproto *, struct ofport *);
109 static void ofport_wait(struct ofport *);
111 struct action_xlate_ctx {
112 /* action_xlate_ctx_init() initializes these members. */
115 struct ofproto *ofproto;
117 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
118 * this flow when actions change header fields. */
121 /* The packet corresponding to 'flow', or a null pointer if we are
122 * revalidating without a packet to refer to. */
123 const struct ofpbuf *packet;
125 /* If nonnull, called just before executing a resubmit action.
127 * This is normally null so the client has to set it manually after
128 * calling action_xlate_ctx_init(). */
129 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
131 /* If true, the speciality of 'flow' should be checked before executing
132 * its actions. If special_cb returns false on 'flow' rendered
133 * uninstallable and no actions will be executed. */
136 /* xlate_actions() initializes and uses these members. The client might want
137 * to look at them after it returns. */
139 struct ofpbuf *odp_actions; /* Datapath actions. */
140 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
141 bool may_set_up_flow; /* True ordinarily; false if the actions must
142 * be reassessed for every packet. */
143 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
145 /* xlate_actions() initializes and uses these members, but the client has no
146 * reason to look at them. */
148 int recurse; /* Recursion level, via xlate_table_action. */
149 int last_pop_priority; /* Offset in 'odp_actions' just past most
150 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
153 static void action_xlate_ctx_init(struct action_xlate_ctx *,
154 struct ofproto *, const struct flow *,
155 const struct ofpbuf *);
156 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
157 const union ofp_action *in, size_t n_in);
159 /* An OpenFlow flow. */
161 long long int used; /* Time last used; time created if not used. */
162 long long int created; /* Creation time. */
166 * - Do include packets and bytes from facets that have been deleted or
167 * whose own statistics have been folded into the rule.
169 * - Do include packets and bytes sent "by hand" that were accounted to
170 * the rule without any facet being involved (this is a rare corner
171 * case in rule_execute()).
173 * - Do not include packet or bytes that can be obtained from any facet's
174 * packet_count or byte_count member or that can be obtained from the
175 * datapath by, e.g., dpif_flow_get() for any facet.
177 uint64_t packet_count; /* Number of packets received. */
178 uint64_t byte_count; /* Number of bytes received. */
180 ovs_be64 flow_cookie; /* Controller-issued identifier. */
182 struct cls_rule cr; /* In owning ofproto's classifier. */
183 uint16_t idle_timeout; /* In seconds from time of last use. */
184 uint16_t hard_timeout; /* In seconds from time of creation. */
185 bool send_flow_removed; /* Send a flow removed message? */
186 int n_actions; /* Number of elements in actions[]. */
187 union ofp_action *actions; /* OpenFlow actions. */
188 struct list facets; /* List of "struct facet"s. */
191 static struct rule *rule_from_cls_rule(const struct cls_rule *);
192 static bool rule_is_hidden(const struct rule *);
194 static struct rule *rule_create(const struct cls_rule *,
195 const union ofp_action *, size_t n_actions,
196 uint16_t idle_timeout, uint16_t hard_timeout,
197 ovs_be64 flow_cookie, bool send_flow_removed);
198 static void rule_destroy(struct ofproto *, struct rule *);
199 static void rule_free(struct rule *);
201 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
202 static void rule_insert(struct ofproto *, struct rule *);
203 static void rule_remove(struct ofproto *, struct rule *);
205 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
206 static void rule_get_stats(const struct rule *, uint64_t *packets,
209 /* An exact-match instantiation of an OpenFlow flow. */
211 long long int used; /* Time last used; time created if not used. */
215 * - Do include packets and bytes sent "by hand", e.g. with
218 * - Do include packets and bytes that were obtained from the datapath
219 * when a flow was deleted (e.g. dpif_flow_del()) or when its
220 * statistics were reset (e.g. dpif_flow_put() with
221 * DPIF_FP_ZERO_STATS).
223 * - Do not include any packets or bytes that can currently be obtained
224 * from the datapath by, e.g., dpif_flow_get().
226 uint64_t packet_count; /* Number of packets received. */
227 uint64_t byte_count; /* Number of bytes received. */
229 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
230 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
232 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
233 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
234 long long int rs_used; /* Used time pushed to resubmit children. */
236 /* Number of bytes passed to account_cb. This may include bytes that can
237 * currently obtained from the datapath (thus, it can be greater than
239 uint64_t accounted_bytes;
241 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
242 struct list list_node; /* In owning rule's 'facets' list. */
243 struct rule *rule; /* Owning rule. */
244 struct flow flow; /* Exact-match flow. */
245 bool installed; /* Installed in datapath? */
246 bool may_install; /* True ordinarily; false if actions must
247 * be reassessed for every packet. */
248 size_t actions_len; /* Number of bytes in actions[]. */
249 struct nlattr *actions; /* Datapath actions. */
250 tag_type tags; /* Tags (set only by hooks). */
251 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
254 static struct facet *facet_create(struct ofproto *, struct rule *,
256 const struct ofpbuf *packet);
257 static void facet_remove(struct ofproto *, struct facet *);
258 static void facet_free(struct facet *);
260 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
261 static bool facet_revalidate(struct ofproto *, struct facet *);
263 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
264 static void facet_uninstall(struct ofproto *, struct facet *);
265 static void facet_flush_stats(struct ofproto *, struct facet *);
267 static void facet_make_actions(struct ofproto *, struct facet *,
268 const struct ofpbuf *packet);
269 static void facet_update_stats(struct ofproto *, struct facet *,
270 const struct dpif_flow_stats *);
271 static void facet_push_stats(struct ofproto *, struct facet *);
273 /* ofproto supports two kinds of OpenFlow connections:
275 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
276 * maintains persistent connections to these controllers and by default
277 * sends them asynchronous messages such as packet-ins.
279 * - "Service" connections, e.g. from ovs-ofctl. When these connections
280 * drop, it is the other side's responsibility to reconnect them if
281 * necessary. ofproto does not send them asynchronous messages by default.
283 * Currently, active (tcp, ssl, unix) connections are always "primary"
284 * connections and passive (ptcp, pssl, punix) connections are always "service"
285 * connections. There is no inherent reason for this, but it reflects the
289 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
290 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
293 /* A listener for incoming OpenFlow "service" connections. */
295 struct hmap_node node; /* In struct ofproto's "services" hmap. */
296 struct pvconn *pvconn; /* OpenFlow connection listener. */
298 /* These are not used by ofservice directly. They are settings for
299 * accepted "struct ofconn"s from the pvconn. */
300 int probe_interval; /* Max idle time before probing, in seconds. */
301 int rate_limit; /* Max packet-in rate in packets per second. */
302 int burst_limit; /* Limit on accumulating packet credits. */
305 static struct ofservice *ofservice_lookup(struct ofproto *,
307 static int ofservice_create(struct ofproto *,
308 const struct ofproto_controller *);
309 static void ofservice_reconfigure(struct ofservice *,
310 const struct ofproto_controller *);
311 static void ofservice_destroy(struct ofproto *, struct ofservice *);
313 /* An OpenFlow connection. */
315 struct ofproto *ofproto; /* The ofproto that owns this connection. */
316 struct list node; /* In struct ofproto's "all_conns" list. */
317 struct rconn *rconn; /* OpenFlow connection. */
318 enum ofconn_type type; /* Type. */
319 enum nx_flow_format flow_format; /* Currently selected flow format. */
321 /* OFPT_PACKET_IN related data. */
322 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
323 #define N_SCHEDULERS 2
324 struct pinsched *schedulers[N_SCHEDULERS];
325 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
326 int miss_send_len; /* Bytes to send of buffered packets. */
328 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
329 * requests, and the maximum number before we stop reading OpenFlow
331 #define OFCONN_REPLY_MAX 100
332 struct rconn_packet_counter *reply_counter;
334 /* type == OFCONN_PRIMARY only. */
335 enum nx_role role; /* Role. */
336 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
337 enum ofproto_band band; /* In-band or out-of-band? */
341 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
343 static void ofconn_destroy(struct ofconn *);
344 static void ofconn_run(struct ofconn *);
345 static void ofconn_wait(struct ofconn *);
346 static bool ofconn_receives_async_msgs(const struct ofconn *);
347 static char *ofconn_make_name(const struct ofproto *, const char *target);
348 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
350 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
351 struct rconn_packet_counter *counter);
353 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
354 const struct flow *, bool clone);
355 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
359 uint64_t datapath_id; /* Datapath ID. */
360 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
361 char *mfr_desc; /* Manufacturer. */
362 char *hw_desc; /* Hardware. */
363 char *sw_desc; /* Software version. */
364 char *serial_desc; /* Serial number. */
365 char *dp_desc; /* Datapath description. */
369 struct netdev_monitor *netdev_monitor;
370 struct hmap ports; /* Contains "struct ofport"s. */
371 struct shash port_by_name;
375 struct fail_open *fail_open;
376 struct netflow *netflow;
377 struct ofproto_sflow *sflow;
379 /* In-band control. */
380 struct in_band *in_band;
381 long long int next_in_band_update;
382 struct sockaddr_in *extra_in_band_remotes;
383 size_t n_extra_remotes;
387 struct classifier cls;
388 long long int next_expiration;
392 bool need_revalidate;
393 struct tag_set revalidate_set;
395 /* OpenFlow connections. */
396 struct hmap controllers; /* Controller "struct ofconn"s. */
397 struct list all_conns; /* Contains "struct ofconn"s. */
398 enum ofproto_fail_mode fail_mode;
400 /* OpenFlow listeners. */
401 struct hmap services; /* Contains "struct ofservice"s. */
402 struct pvconn **snoops;
405 /* Hooks for ovs-vswitchd. */
406 const struct ofhooks *ofhooks;
409 /* Used by default ofhooks. */
410 struct mac_learning *ml;
413 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
414 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
416 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
418 static const struct ofhooks default_ofhooks;
420 static uint64_t pick_datapath_id(const struct ofproto *);
421 static uint64_t pick_fallback_dpid(void);
423 static int ofproto_expire(struct ofproto *);
424 static void flow_push_stats(struct ofproto *, const struct rule *,
425 struct flow *, uint64_t packets, uint64_t bytes,
428 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
430 static void handle_openflow(struct ofconn *, struct ofpbuf *);
432 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
433 static void update_port(struct ofproto *, const char *devname);
434 static int init_ports(struct ofproto *);
435 static void reinit_ports(struct ofproto *);
437 static void ofproto_unixctl_init(void);
440 ofproto_create(const char *datapath, const char *datapath_type,
441 const struct ofhooks *ofhooks, void *aux,
442 struct ofproto **ofprotop)
450 ofproto_unixctl_init();
452 /* Connect to datapath and start listening for messages. */
453 error = dpif_open(datapath, datapath_type, &dpif);
455 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
458 error = dpif_recv_set_mask(dpif,
459 ((1u << DPIF_UC_MISS) |
460 (1u << DPIF_UC_ACTION) |
461 (1u << DPIF_UC_SAMPLE)));
463 VLOG_ERR("failed to listen on datapath %s: %s",
464 datapath, strerror(error));
468 dpif_flow_flush(dpif);
469 dpif_recv_purge(dpif);
471 /* Initialize settings. */
472 p = xzalloc(sizeof *p);
473 p->fallback_dpid = pick_fallback_dpid();
474 p->datapath_id = p->fallback_dpid;
475 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
476 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
477 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
478 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
479 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
481 /* Initialize datapath. */
483 p->netdev_monitor = netdev_monitor_create();
484 hmap_init(&p->ports);
485 shash_init(&p->port_by_name);
486 p->max_ports = dpif_get_max_ports(dpif);
488 /* Initialize submodules. */
493 /* Initialize in-band control. */
495 p->in_band_queue = -1;
497 /* Initialize flow table. */
498 classifier_init(&p->cls);
499 p->next_expiration = time_msec() + 1000;
501 /* Initialize facet table. */
502 hmap_init(&p->facets);
503 p->need_revalidate = false;
504 tag_set_init(&p->revalidate_set);
506 /* Initialize OpenFlow connections. */
507 list_init(&p->all_conns);
508 hmap_init(&p->controllers);
509 hmap_init(&p->services);
513 /* Initialize hooks. */
515 p->ofhooks = ofhooks;
519 p->ofhooks = &default_ofhooks;
521 p->ml = mac_learning_create();
524 /* Pick final datapath ID. */
525 p->datapath_id = pick_datapath_id(p);
526 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
528 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
535 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
537 uint64_t old_dpid = p->datapath_id;
538 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
539 if (p->datapath_id != old_dpid) {
540 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
542 /* Force all active connections to reconnect, since there is no way to
543 * notify a controller that the datapath ID has changed. */
544 ofproto_reconnect_controllers(p);
548 /* Creates a new controller in 'ofproto'. Some of the settings are initially
549 * drawn from 'c', but update_controller() needs to be called later to finish
550 * the new ofconn's configuration. */
552 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
554 char *name = ofconn_make_name(ofproto, c->target);
555 struct ofconn *ofconn;
557 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
558 ofconn->pktbuf = pktbuf_create();
559 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
560 rconn_connect(ofconn->rconn, c->target, name);
561 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
562 hash_string(c->target, 0));
567 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
568 * target (this is done by creating new ofconns and deleting old ones), but it
569 * can update the rest of an ofconn's settings. */
571 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
575 ofconn->band = c->band;
577 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
579 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
580 rconn_set_probe_interval(ofconn->rconn, probe_interval);
582 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
586 ofconn_get_target(const struct ofconn *ofconn)
588 return rconn_get_target(ofconn->rconn);
591 static struct ofconn *
592 find_controller_by_target(struct ofproto *ofproto, const char *target)
594 struct ofconn *ofconn;
596 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
597 hash_string(target, 0), &ofproto->controllers) {
598 if (!strcmp(ofconn_get_target(ofconn), target)) {
606 update_in_band_remotes(struct ofproto *ofproto)
608 const struct ofconn *ofconn;
609 struct sockaddr_in *addrs;
610 size_t max_addrs, n_addrs;
613 /* Allocate enough memory for as many remotes as we could possibly have. */
614 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
615 addrs = xmalloc(max_addrs * sizeof *addrs);
618 /* Add all the remotes. */
619 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
620 struct sockaddr_in *sin = &addrs[n_addrs];
622 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
626 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
627 if (sin->sin_addr.s_addr) {
628 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
632 for (i = 0; i < ofproto->n_extra_remotes; i++) {
633 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
636 /* Create or update or destroy in-band. */
638 if (!ofproto->in_band) {
639 in_band_create(ofproto, ofproto->dpif, &ofproto->in_band);
641 if (ofproto->in_band) {
642 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
644 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
645 ofproto->next_in_band_update = time_msec() + 1000;
647 in_band_destroy(ofproto->in_band);
648 ofproto->in_band = NULL;
656 update_fail_open(struct ofproto *p)
658 struct ofconn *ofconn;
660 if (!hmap_is_empty(&p->controllers)
661 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
662 struct rconn **rconns;
666 p->fail_open = fail_open_create(p);
670 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
671 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
672 rconns[n++] = ofconn->rconn;
675 fail_open_set_controllers(p->fail_open, rconns, n);
676 /* p->fail_open takes ownership of 'rconns'. */
678 fail_open_destroy(p->fail_open);
684 ofproto_set_controllers(struct ofproto *p,
685 const struct ofproto_controller *controllers,
686 size_t n_controllers)
688 struct shash new_controllers;
689 struct ofconn *ofconn, *next_ofconn;
690 struct ofservice *ofservice, *next_ofservice;
693 /* Create newly configured controllers and services.
694 * Create a name to ofproto_controller mapping in 'new_controllers'. */
695 shash_init(&new_controllers);
696 for (i = 0; i < n_controllers; i++) {
697 const struct ofproto_controller *c = &controllers[i];
699 if (!vconn_verify_name(c->target)) {
700 if (!find_controller_by_target(p, c->target)) {
701 add_controller(p, c);
703 } else if (!pvconn_verify_name(c->target)) {
704 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
708 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
709 dpif_name(p->dpif), c->target);
713 shash_add_once(&new_controllers, c->target, &controllers[i]);
716 /* Delete controllers that are no longer configured.
717 * Update configuration of all now-existing controllers. */
718 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
719 struct ofproto_controller *c;
721 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
723 ofconn_destroy(ofconn);
725 update_controller(ofconn, c);
729 /* Delete services that are no longer configured.
730 * Update configuration of all now-existing services. */
731 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
732 struct ofproto_controller *c;
734 c = shash_find_data(&new_controllers,
735 pvconn_get_name(ofservice->pvconn));
737 ofservice_destroy(p, ofservice);
739 ofservice_reconfigure(ofservice, c);
743 shash_destroy(&new_controllers);
745 update_in_band_remotes(p);
750 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
752 p->fail_mode = fail_mode;
756 /* Drops the connections between 'ofproto' and all of its controllers, forcing
757 * them to reconnect. */
759 ofproto_reconnect_controllers(struct ofproto *ofproto)
761 struct ofconn *ofconn;
763 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
764 rconn_reconnect(ofconn->rconn);
769 any_extras_changed(const struct ofproto *ofproto,
770 const struct sockaddr_in *extras, size_t n)
774 if (n != ofproto->n_extra_remotes) {
778 for (i = 0; i < n; i++) {
779 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
780 const struct sockaddr_in *new = &extras[i];
782 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
783 old->sin_port != new->sin_port) {
791 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
792 * in-band control should guarantee access, in the same way that in-band
793 * control guarantees access to OpenFlow controllers. */
795 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
796 const struct sockaddr_in *extras, size_t n)
798 if (!any_extras_changed(ofproto, extras, n)) {
802 free(ofproto->extra_in_band_remotes);
803 ofproto->n_extra_remotes = n;
804 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
806 update_in_band_remotes(ofproto);
809 /* Sets the OpenFlow queue used by flows set up by in-band control on
810 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
811 * flows will use the default queue. */
813 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
815 if (queue_id != ofproto->in_band_queue) {
816 ofproto->in_band_queue = queue_id;
817 update_in_band_remotes(ofproto);
822 ofproto_set_desc(struct ofproto *p,
823 const char *mfr_desc, const char *hw_desc,
824 const char *sw_desc, const char *serial_desc,
827 struct ofp_desc_stats *ods;
830 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
831 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
832 sizeof ods->mfr_desc);
835 p->mfr_desc = xstrdup(mfr_desc);
838 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
839 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
840 sizeof ods->hw_desc);
843 p->hw_desc = xstrdup(hw_desc);
846 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
847 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
848 sizeof ods->sw_desc);
851 p->sw_desc = xstrdup(sw_desc);
854 if (strlen(serial_desc) >= sizeof ods->serial_num) {
855 VLOG_WARN("truncating serial_desc, must be less than %zu "
857 sizeof ods->serial_num);
859 free(p->serial_desc);
860 p->serial_desc = xstrdup(serial_desc);
863 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
864 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
865 sizeof ods->dp_desc);
868 p->dp_desc = xstrdup(dp_desc);
873 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
874 const struct svec *svec)
876 struct pvconn **pvconns = *pvconnsp;
877 size_t n_pvconns = *n_pvconnsp;
881 for (i = 0; i < n_pvconns; i++) {
882 pvconn_close(pvconns[i]);
886 pvconns = xmalloc(svec->n * sizeof *pvconns);
888 for (i = 0; i < svec->n; i++) {
889 const char *name = svec->names[i];
890 struct pvconn *pvconn;
893 error = pvconn_open(name, &pvconn);
895 pvconns[n_pvconns++] = pvconn;
897 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
905 *n_pvconnsp = n_pvconns;
911 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
913 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
917 ofproto_set_netflow(struct ofproto *ofproto,
918 const struct netflow_options *nf_options)
920 if (nf_options && nf_options->collectors.n) {
921 if (!ofproto->netflow) {
922 ofproto->netflow = netflow_create();
924 return netflow_set_options(ofproto->netflow, nf_options);
926 netflow_destroy(ofproto->netflow);
927 ofproto->netflow = NULL;
933 ofproto_set_sflow(struct ofproto *ofproto,
934 const struct ofproto_sflow_options *oso)
936 struct ofproto_sflow *os = ofproto->sflow;
939 struct ofport *ofport;
941 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
942 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
943 ofproto_sflow_add_port(os, ofport->odp_port,
944 netdev_get_name(ofport->netdev));
947 ofproto_sflow_set_options(os, oso);
949 ofproto_sflow_destroy(os);
950 ofproto->sflow = NULL;
954 /* Connectivity Fault Management configuration. */
956 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
958 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
960 struct ofport *ofport = get_port(ofproto, port_no);
961 if (ofport && ofport->cfm){
962 cfm_destroy(ofport->cfm);
967 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
968 * basic configuration from the configuration members in 'cfm', and the set of
969 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
970 * Ignores the statistics members of 'cfm'.
972 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
974 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
975 const struct cfm *cfm,
976 const uint16_t *remote_mps, size_t n_remote_mps)
978 struct ofport *ofport;
980 ofport = get_port(ofproto, port_no);
982 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
983 dpif_name(ofproto->dpif), port_no);
988 ofport->cfm = cfm_create();
991 ofport->cfm->mpid = cfm->mpid;
992 ofport->cfm->interval = cfm->interval;
993 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
995 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
997 if (!cfm_configure(ofport->cfm)) {
998 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
999 dpif_name(ofproto->dpif), port_no,
1000 netdev_get_name(ofport->netdev));
1001 cfm_destroy(ofport->cfm);
1006 /* Returns the connectivity fault management object associated with 'port_no'
1007 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
1008 * 'port_no' or if that port does not have CFM configured. The caller must not
1009 * modify or destroy the returned object. */
1011 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
1013 struct ofport *ofport = get_port(ofproto, port_no);
1014 return ofport ? ofport->cfm : NULL;
1018 ofproto_get_datapath_id(const struct ofproto *ofproto)
1020 return ofproto->datapath_id;
1024 ofproto_has_primary_controller(const struct ofproto *ofproto)
1026 return !hmap_is_empty(&ofproto->controllers);
1029 enum ofproto_fail_mode
1030 ofproto_get_fail_mode(const struct ofproto *p)
1032 return p->fail_mode;
1036 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1040 for (i = 0; i < ofproto->n_snoops; i++) {
1041 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1046 ofproto_destroy(struct ofproto *p)
1048 struct ofservice *ofservice, *next_ofservice;
1049 struct ofconn *ofconn, *next_ofconn;
1050 struct ofport *ofport, *next_ofport;
1057 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1059 /* Destroy fail-open and in-band early, since they touch the classifier. */
1060 fail_open_destroy(p->fail_open);
1061 p->fail_open = NULL;
1063 in_band_destroy(p->in_band);
1065 free(p->extra_in_band_remotes);
1067 ofproto_flush_flows(p);
1068 classifier_destroy(&p->cls);
1069 hmap_destroy(&p->facets);
1071 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1072 ofconn_destroy(ofconn);
1074 hmap_destroy(&p->controllers);
1076 dpif_close(p->dpif);
1077 netdev_monitor_destroy(p->netdev_monitor);
1078 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1079 hmap_remove(&p->ports, &ofport->hmap_node);
1080 ofport_free(ofport);
1082 shash_destroy(&p->port_by_name);
1084 netflow_destroy(p->netflow);
1085 ofproto_sflow_destroy(p->sflow);
1087 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1088 ofservice_destroy(p, ofservice);
1090 hmap_destroy(&p->services);
1092 for (i = 0; i < p->n_snoops; i++) {
1093 pvconn_close(p->snoops[i]);
1097 mac_learning_destroy(p->ml);
1102 free(p->serial_desc);
1105 hmap_destroy(&p->ports);
1111 ofproto_run(struct ofproto *p)
1113 int error = ofproto_run1(p);
1115 error = ofproto_run2(p, false);
1121 process_port_change(struct ofproto *ofproto, int error, char *devname)
1123 if (error == ENOBUFS) {
1124 reinit_ports(ofproto);
1125 } else if (!error) {
1126 update_port(ofproto, devname);
1131 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1132 * means that 'ofconn' is more interesting for monitoring than a lower return
1135 snoop_preference(const struct ofconn *ofconn)
1137 switch (ofconn->role) {
1138 case NX_ROLE_MASTER:
1145 /* Shouldn't happen. */
1150 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1151 * Connects this vconn to a controller. */
1153 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1155 struct ofconn *ofconn, *best;
1157 /* Pick a controller for monitoring. */
1159 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1160 if (ofconn->type == OFCONN_PRIMARY
1161 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1167 rconn_add_monitor(best->rconn, vconn);
1169 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1175 ofproto_run1(struct ofproto *p)
1177 struct ofconn *ofconn, *next_ofconn;
1178 struct ofservice *ofservice;
1179 struct ofport *ofport;
1184 if (shash_is_empty(&p->port_by_name)) {
1188 for (i = 0; i < 50; i++) {
1189 struct dpif_upcall packet;
1191 error = dpif_recv(p->dpif, &packet);
1193 if (error == ENODEV) {
1194 /* Someone destroyed the datapath behind our back. The caller
1195 * better destroy us and give up, because we're just going to
1196 * spin from here on out. */
1197 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1198 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1199 dpif_name(p->dpif));
1205 handle_upcall(p, &packet);
1208 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1209 process_port_change(p, error, devname);
1211 while ((error = netdev_monitor_poll(p->netdev_monitor,
1212 &devname)) != EAGAIN) {
1213 process_port_change(p, error, devname);
1216 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1217 ofport_run(p, ofport);
1221 if (time_msec() >= p->next_in_band_update) {
1222 update_in_band_remotes(p);
1224 in_band_run(p->in_band);
1227 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1231 /* Fail-open maintenance. Do this after processing the ofconns since
1232 * fail-open checks the status of the controller rconn. */
1234 fail_open_run(p->fail_open);
1237 HMAP_FOR_EACH (ofservice, node, &p->services) {
1238 struct vconn *vconn;
1241 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1243 struct rconn *rconn;
1246 rconn = rconn_create(ofservice->probe_interval, 0);
1247 name = ofconn_make_name(p, vconn_get_name(vconn));
1248 rconn_connect_unreliably(rconn, vconn, name);
1251 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1252 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1253 ofservice->burst_limit);
1254 } else if (retval != EAGAIN) {
1255 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1259 for (i = 0; i < p->n_snoops; i++) {
1260 struct vconn *vconn;
1263 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1265 add_snooper(p, vconn);
1266 } else if (retval != EAGAIN) {
1267 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1271 if (time_msec() >= p->next_expiration) {
1272 int delay = ofproto_expire(p);
1273 p->next_expiration = time_msec() + delay;
1274 COVERAGE_INC(ofproto_expiration);
1278 netflow_run(p->netflow);
1281 ofproto_sflow_run(p->sflow);
1288 ofproto_run2(struct ofproto *p, bool revalidate_all)
1290 /* Figure out what we need to revalidate now, if anything. */
1291 struct tag_set revalidate_set = p->revalidate_set;
1292 if (p->need_revalidate) {
1293 revalidate_all = true;
1296 /* Clear the revalidation flags. */
1297 tag_set_init(&p->revalidate_set);
1298 p->need_revalidate = false;
1300 /* Now revalidate if there's anything to do. */
1301 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1302 struct facet *facet, *next;
1304 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1306 || tag_set_intersects(&revalidate_set, facet->tags)) {
1307 facet_revalidate(p, facet);
1316 ofproto_wait(struct ofproto *p)
1318 struct ofservice *ofservice;
1319 struct ofconn *ofconn;
1320 struct ofport *ofport;
1323 dpif_recv_wait(p->dpif);
1324 dpif_port_poll_wait(p->dpif);
1325 netdev_monitor_poll_wait(p->netdev_monitor);
1326 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1327 ofport_wait(ofport);
1329 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1330 ofconn_wait(ofconn);
1333 poll_timer_wait_until(p->next_in_band_update);
1334 in_band_wait(p->in_band);
1337 fail_open_wait(p->fail_open);
1340 ofproto_sflow_wait(p->sflow);
1342 if (!tag_set_is_empty(&p->revalidate_set)) {
1343 poll_immediate_wake();
1345 if (p->need_revalidate) {
1346 /* Shouldn't happen, but if it does just go around again. */
1347 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1348 poll_immediate_wake();
1349 } else if (p->next_expiration != LLONG_MAX) {
1350 poll_timer_wait_until(p->next_expiration);
1352 HMAP_FOR_EACH (ofservice, node, &p->services) {
1353 pvconn_wait(ofservice->pvconn);
1355 for (i = 0; i < p->n_snoops; i++) {
1356 pvconn_wait(p->snoops[i]);
1361 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1363 tag_set_add(&ofproto->revalidate_set, tag);
1367 ofproto_get_revalidate_set(struct ofproto *ofproto)
1369 return &ofproto->revalidate_set;
1373 ofproto_is_alive(const struct ofproto *p)
1375 return !hmap_is_empty(&p->controllers);
1379 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1382 const struct ofconn *ofconn;
1386 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1387 const struct rconn *rconn = ofconn->rconn;
1388 time_t now = time_now();
1389 time_t last_connection = rconn_get_last_connection(rconn);
1390 time_t last_disconnect = rconn_get_last_disconnect(rconn);
1391 const int last_error = rconn_get_last_error(rconn);
1392 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1394 shash_add(info, rconn_get_target(rconn), cinfo);
1396 cinfo->is_connected = rconn_is_connected(rconn);
1397 cinfo->role = ofconn->role;
1402 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1403 cinfo->pairs.values[cinfo->pairs.n++] =
1404 xstrdup(ovs_retval_to_string(last_error));
1407 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1408 cinfo->pairs.values[cinfo->pairs.n++] =
1409 xstrdup(rconn_get_state(rconn));
1411 if (last_connection != TIME_MIN) {
1412 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
1413 cinfo->pairs.values[cinfo->pairs.n++]
1414 = xasprintf("%ld", (long int) (now - last_connection));
1417 if (last_disconnect != TIME_MIN) {
1418 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
1419 cinfo->pairs.values[cinfo->pairs.n++]
1420 = xasprintf("%ld", (long int) (now - last_disconnect));
1426 ofproto_free_ofproto_controller_info(struct shash *info)
1428 struct shash_node *node;
1430 SHASH_FOR_EACH (node, info) {
1431 struct ofproto_controller_info *cinfo = node->data;
1432 while (cinfo->pairs.n) {
1433 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1437 shash_destroy(info);
1440 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1442 * This is almost the same as calling dpif_port_del() directly on the
1443 * datapath, but it also makes 'ofproto' close its open netdev for the port
1444 * (if any). This makes it possible to create a new netdev of a different
1445 * type under the same name, which otherwise the netdev library would refuse
1446 * to do because of the conflict. (The netdev would eventually get closed on
1447 * the next trip through ofproto_run(), but this interface is more direct.)
1449 * Returns 0 if successful, otherwise a positive errno. */
1451 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1453 struct ofport *ofport = get_port(ofproto, odp_port);
1454 const char *name = ofport ? ofport->opp.name : "<unknown>";
1457 error = dpif_port_del(ofproto->dpif, odp_port);
1459 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1460 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1461 } else if (ofport) {
1462 /* 'name' is ofport->opp.name and update_port() is going to destroy
1463 * 'ofport'. Just in case update_port() refers to 'name' after it
1464 * destroys 'ofport', make a copy of it around the update_port()
1466 char *devname = xstrdup(name);
1467 update_port(ofproto, devname);
1473 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1474 * true if 'odp_port' exists and should be included, false otherwise. */
1476 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1478 struct ofport *ofport = get_port(ofproto, odp_port);
1479 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1482 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1483 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1484 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1486 * Returns 0 if successful, otherwise a positive errno value. */
1488 ofproto_send_packet(struct ofproto *ofproto,
1489 uint32_t port_no, uint16_t vlan_tci,
1490 const struct ofpbuf *packet)
1492 struct ofpbuf odp_actions;
1495 ofpbuf_init(&odp_actions, 32);
1496 if (vlan_tci != 0) {
1497 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1498 ntohs(vlan_tci & ~VLAN_CFI));
1500 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1501 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1503 ofpbuf_uninit(&odp_actions);
1506 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1507 dpif_name(ofproto->dpif), port_no, strerror(error));
1512 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1513 * performs the 'n_actions' actions in 'actions'. The new flow will not
1516 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1517 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1518 * controllers; otherwise, it will be hidden.
1520 * The caller retains ownership of 'cls_rule' and 'actions'. */
1522 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1523 const union ofp_action *actions, size_t n_actions)
1526 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1527 rule_insert(p, rule);
1531 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1535 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1538 rule_remove(ofproto, rule);
1543 ofproto_flush_flows(struct ofproto *ofproto)
1545 struct facet *facet, *next_facet;
1546 struct rule *rule, *next_rule;
1547 struct cls_cursor cursor;
1549 COVERAGE_INC(ofproto_flush);
1551 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1552 /* Mark the facet as not installed so that facet_remove() doesn't
1553 * bother trying to uninstall it. There is no point in uninstalling it
1554 * individually since we are about to blow away all the facets with
1555 * dpif_flow_flush(). */
1556 facet->installed = false;
1557 facet->dp_packet_count = 0;
1558 facet->dp_byte_count = 0;
1559 facet_remove(ofproto, facet);
1562 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1563 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1564 rule_remove(ofproto, rule);
1567 dpif_flow_flush(ofproto->dpif);
1568 if (ofproto->in_band) {
1569 in_band_flushed(ofproto->in_band);
1571 if (ofproto->fail_open) {
1572 fail_open_flushed(ofproto->fail_open);
1577 reinit_ports(struct ofproto *p)
1579 struct dpif_port_dump dump;
1580 struct shash_node *node;
1581 struct shash devnames;
1582 struct ofport *ofport;
1583 struct dpif_port dpif_port;
1585 COVERAGE_INC(ofproto_reinit_ports);
1587 shash_init(&devnames);
1588 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1589 shash_add_once (&devnames, ofport->opp.name, NULL);
1591 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1592 shash_add_once (&devnames, dpif_port.name, NULL);
1595 SHASH_FOR_EACH (node, &devnames) {
1596 update_port(p, node->name);
1598 shash_destroy(&devnames);
1601 static struct ofport *
1602 make_ofport(const struct dpif_port *dpif_port)
1604 struct netdev_options netdev_options;
1605 enum netdev_flags flags;
1606 struct ofport *ofport;
1607 struct netdev *netdev;
1610 memset(&netdev_options, 0, sizeof netdev_options);
1611 netdev_options.name = dpif_port->name;
1612 netdev_options.type = dpif_port->type;
1613 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1615 error = netdev_open(&netdev_options, &netdev);
1617 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1618 "cannot be opened (%s)",
1619 dpif_port->name, dpif_port->port_no,
1620 dpif_port->name, strerror(error));
1624 ofport = xzalloc(sizeof *ofport);
1625 ofport->netdev = netdev;
1626 ofport->odp_port = dpif_port->port_no;
1627 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1628 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1629 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1631 netdev_get_flags(netdev, &flags);
1632 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1634 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1636 netdev_get_features(netdev,
1637 &ofport->opp.curr, &ofport->opp.advertised,
1638 &ofport->opp.supported, &ofport->opp.peer);
1643 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1645 if (get_port(p, dpif_port->port_no)) {
1646 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1647 dpif_port->port_no);
1649 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1650 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1659 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1661 const struct ofp_phy_port *a = &a_->opp;
1662 const struct ofp_phy_port *b = &b_->opp;
1664 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1665 return (a->port_no == b->port_no
1666 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1667 && !strcmp(a->name, b->name)
1668 && a->state == b->state
1669 && a->config == b->config
1670 && a->curr == b->curr
1671 && a->advertised == b->advertised
1672 && a->supported == b->supported
1673 && a->peer == b->peer);
1677 send_port_status(struct ofproto *p, const struct ofport *ofport,
1680 /* XXX Should limit the number of queued port status change messages. */
1681 struct ofconn *ofconn;
1682 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1683 struct ofp_port_status *ops;
1686 /* Primary controllers, even slaves, should always get port status
1687 updates. Otherwise obey ofconn_receives_async_msgs(). */
1688 if (ofconn->type != OFCONN_PRIMARY
1689 && !ofconn_receives_async_msgs(ofconn)) {
1693 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1694 ops->reason = reason;
1695 ops->desc = ofport->opp;
1696 hton_ofp_phy_port(&ops->desc);
1697 queue_tx(b, ofconn, NULL);
1702 ofport_install(struct ofproto *p, struct ofport *ofport)
1704 const char *netdev_name = ofport->opp.name;
1706 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1707 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1708 shash_add(&p->port_by_name, netdev_name, ofport);
1710 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1715 ofport_remove(struct ofproto *p, struct ofport *ofport)
1717 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1718 hmap_remove(&p->ports, &ofport->hmap_node);
1719 shash_delete(&p->port_by_name,
1720 shash_find(&p->port_by_name, ofport->opp.name));
1722 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1727 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1730 cfm_run(ofport->cfm);
1732 if (cfm_should_send_ccm(ofport->cfm)) {
1733 struct ofpbuf packet;
1736 ofpbuf_init(&packet, 0);
1737 ccm = compose_packet(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1738 ETH_TYPE_CFM, sizeof *ccm);
1739 cfm_compose_ccm(ofport->cfm, ccm);
1740 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1741 ofpbuf_uninit(&packet);
1747 ofport_wait(struct ofport *ofport)
1750 cfm_wait(ofport->cfm);
1755 ofport_free(struct ofport *ofport)
1758 cfm_destroy(ofport->cfm);
1759 netdev_close(ofport->netdev);
1764 static struct ofport *
1765 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1767 struct ofport *port;
1769 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1770 hash_int(odp_port, 0), &ofproto->ports) {
1771 if (port->odp_port == odp_port) {
1779 update_port(struct ofproto *p, const char *devname)
1781 struct dpif_port dpif_port;
1782 struct ofport *old_ofport;
1783 struct ofport *new_ofport;
1786 COVERAGE_INC(ofproto_update_port);
1788 /* Query the datapath for port information. */
1789 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1791 /* Find the old ofport. */
1792 old_ofport = shash_find_data(&p->port_by_name, devname);
1795 /* There's no port named 'devname' but there might be a port with
1796 * the same port number. This could happen if a port is deleted
1797 * and then a new one added in its place very quickly, or if a port
1798 * is renamed. In the former case we want to send an OFPPR_DELETE
1799 * and an OFPPR_ADD, and in the latter case we want to send a
1800 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1801 * the old port's ifindex against the new port, or perhaps less
1802 * reliably but more portably by comparing the old port's MAC
1803 * against the new port's MAC. However, this code isn't that smart
1804 * and always sends an OFPPR_MODIFY (XXX). */
1805 old_ofport = get_port(p, dpif_port.port_no);
1807 } else if (error != ENOENT && error != ENODEV) {
1808 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1809 "%s", strerror(error));
1813 /* Create a new ofport. */
1814 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1816 /* Eliminate a few pathological cases. */
1817 if (!old_ofport && !new_ofport) {
1819 } else if (old_ofport && new_ofport) {
1820 /* Most of the 'config' bits are OpenFlow soft state, but
1821 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1822 * OpenFlow bits from old_ofport. (make_ofport() only sets
1823 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1824 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1826 if (ofport_equal(old_ofport, new_ofport)) {
1827 /* False alarm--no change. */
1828 ofport_free(new_ofport);
1833 /* Now deal with the normal cases. */
1835 ofport_remove(p, old_ofport);
1838 ofport_install(p, new_ofport);
1840 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1841 (!old_ofport ? OFPPR_ADD
1842 : !new_ofport ? OFPPR_DELETE
1844 ofport_free(old_ofport);
1847 dpif_port_destroy(&dpif_port);
1851 init_ports(struct ofproto *p)
1853 struct dpif_port_dump dump;
1854 struct dpif_port dpif_port;
1856 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1857 if (!ofport_conflicts(p, &dpif_port)) {
1858 struct ofport *ofport = make_ofport(&dpif_port);
1860 ofport_install(p, ofport);
1868 static struct ofconn *
1869 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1871 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1872 ofconn->ofproto = p;
1873 list_push_back(&p->all_conns, &ofconn->node);
1874 ofconn->rconn = rconn;
1875 ofconn->type = type;
1876 ofconn->flow_format = NXFF_OPENFLOW10;
1877 ofconn->role = NX_ROLE_OTHER;
1878 ofconn->packet_in_counter = rconn_packet_counter_create ();
1879 ofconn->pktbuf = NULL;
1880 ofconn->miss_send_len = 0;
1881 ofconn->reply_counter = rconn_packet_counter_create ();
1886 ofconn_destroy(struct ofconn *ofconn)
1888 if (ofconn->type == OFCONN_PRIMARY) {
1889 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1892 list_remove(&ofconn->node);
1893 rconn_destroy(ofconn->rconn);
1894 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1895 rconn_packet_counter_destroy(ofconn->reply_counter);
1896 pktbuf_destroy(ofconn->pktbuf);
1901 ofconn_run(struct ofconn *ofconn)
1903 struct ofproto *p = ofconn->ofproto;
1907 for (i = 0; i < N_SCHEDULERS; i++) {
1908 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1911 rconn_run(ofconn->rconn);
1913 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1914 /* Limit the number of iterations to prevent other tasks from
1916 for (iteration = 0; iteration < 50; iteration++) {
1917 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1922 fail_open_maybe_recover(p->fail_open);
1924 handle_openflow(ofconn, of_msg);
1925 ofpbuf_delete(of_msg);
1929 if (!rconn_is_alive(ofconn->rconn)) {
1930 ofconn_destroy(ofconn);
1935 ofconn_wait(struct ofconn *ofconn)
1939 for (i = 0; i < N_SCHEDULERS; i++) {
1940 pinsched_wait(ofconn->schedulers[i]);
1942 rconn_run_wait(ofconn->rconn);
1943 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1944 rconn_recv_wait(ofconn->rconn);
1946 COVERAGE_INC(ofproto_ofconn_stuck);
1950 /* Returns true if 'ofconn' should receive asynchronous messages. */
1952 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1954 if (ofconn->type == OFCONN_PRIMARY) {
1955 /* Primary controllers always get asynchronous messages unless they
1956 * have configured themselves as "slaves". */
1957 return ofconn->role != NX_ROLE_SLAVE;
1959 /* Service connections don't get asynchronous messages unless they have
1960 * explicitly asked for them by setting a nonzero miss send length. */
1961 return ofconn->miss_send_len > 0;
1965 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1966 * and 'target', suitable for use in log messages for identifying the
1969 * The name is dynamically allocated. The caller should free it (with free())
1970 * when it is no longer needed. */
1972 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1974 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1978 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1982 for (i = 0; i < N_SCHEDULERS; i++) {
1983 struct pinsched **s = &ofconn->schedulers[i];
1987 *s = pinsched_create(rate, burst);
1989 pinsched_set_limits(*s, rate, burst);
1992 pinsched_destroy(*s);
1999 ofservice_reconfigure(struct ofservice *ofservice,
2000 const struct ofproto_controller *c)
2002 ofservice->probe_interval = c->probe_interval;
2003 ofservice->rate_limit = c->rate_limit;
2004 ofservice->burst_limit = c->burst_limit;
2007 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
2008 * positive errno value. */
2010 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
2012 struct ofservice *ofservice;
2013 struct pvconn *pvconn;
2016 error = pvconn_open(c->target, &pvconn);
2021 ofservice = xzalloc(sizeof *ofservice);
2022 hmap_insert(&ofproto->services, &ofservice->node,
2023 hash_string(c->target, 0));
2024 ofservice->pvconn = pvconn;
2026 ofservice_reconfigure(ofservice, c);
2032 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
2034 hmap_remove(&ofproto->services, &ofservice->node);
2035 pvconn_close(ofservice->pvconn);
2039 /* Finds and returns the ofservice within 'ofproto' that has the given
2040 * 'target', or a null pointer if none exists. */
2041 static struct ofservice *
2042 ofservice_lookup(struct ofproto *ofproto, const char *target)
2044 struct ofservice *ofservice;
2046 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
2047 &ofproto->services) {
2048 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2055 /* Returns true if 'rule' should be hidden from the controller.
2057 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2058 * (e.g. by in-band control) and are intentionally hidden from the
2061 rule_is_hidden(const struct rule *rule)
2063 return rule->cr.priority > UINT16_MAX;
2066 /* Creates and returns a new rule initialized as specified.
2068 * The caller is responsible for inserting the rule into the classifier (with
2069 * rule_insert()). */
2070 static struct rule *
2071 rule_create(const struct cls_rule *cls_rule,
2072 const union ofp_action *actions, size_t n_actions,
2073 uint16_t idle_timeout, uint16_t hard_timeout,
2074 ovs_be64 flow_cookie, bool send_flow_removed)
2076 struct rule *rule = xzalloc(sizeof *rule);
2077 rule->cr = *cls_rule;
2078 rule->idle_timeout = idle_timeout;
2079 rule->hard_timeout = hard_timeout;
2080 rule->flow_cookie = flow_cookie;
2081 rule->used = rule->created = time_msec();
2082 rule->send_flow_removed = send_flow_removed;
2083 list_init(&rule->facets);
2084 if (n_actions > 0) {
2085 rule->n_actions = n_actions;
2086 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2092 static struct rule *
2093 rule_from_cls_rule(const struct cls_rule *cls_rule)
2095 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2099 rule_free(struct rule *rule)
2101 free(rule->actions);
2105 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2106 * destroying any that no longer has a rule (which is probably all of them).
2108 * The caller must have already removed 'rule' from the classifier. */
2110 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2112 struct facet *facet, *next_facet;
2113 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2114 facet_revalidate(ofproto, facet);
2119 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2120 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2123 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2125 const union ofp_action *oa;
2126 struct actions_iterator i;
2128 if (out_port == htons(OFPP_NONE)) {
2131 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2132 oa = actions_next(&i)) {
2133 if (action_outputs_to_port(oa, out_port)) {
2140 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2141 * 'packet', which arrived on 'in_port'.
2143 * Takes ownership of 'packet'. */
2145 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2146 const struct nlattr *odp_actions, size_t actions_len,
2147 struct ofpbuf *packet)
2149 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2150 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2151 /* As an optimization, avoid a round-trip from userspace to kernel to
2152 * userspace. This also avoids possibly filling up kernel packet
2153 * buffers along the way. */
2154 struct dpif_upcall upcall;
2156 upcall.type = DPIF_UC_ACTION;
2157 upcall.packet = packet;
2160 upcall.userdata = nl_attr_get_u64(odp_actions);
2161 upcall.sample_pool = 0;
2162 upcall.actions = NULL;
2163 upcall.actions_len = 0;
2165 send_packet_in(ofproto, &upcall, flow, false);
2171 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2172 ofpbuf_delete(packet);
2177 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2178 * statistics appropriately. 'packet' must have at least sizeof(struct
2179 * ofp_packet_in) bytes of headroom.
2181 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2182 * applying flow_extract() to 'packet' would yield the same flow as
2185 * 'facet' must have accurately composed ODP actions; that is, it must not be
2186 * in need of revalidation.
2188 * Takes ownership of 'packet'. */
2190 facet_execute(struct ofproto *ofproto, struct facet *facet,
2191 struct ofpbuf *packet)
2193 struct dpif_flow_stats stats;
2195 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2197 flow_extract_stats(&facet->flow, packet, &stats);
2198 stats.used = time_msec();
2199 if (execute_odp_actions(ofproto, &facet->flow,
2200 facet->actions, facet->actions_len, packet)) {
2201 facet_update_stats(ofproto, facet, &stats);
2205 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2206 * statistics (or the statistics for one of its facets) appropriately.
2207 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2209 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2210 * with statistics for 'packet' either way.
2212 * Takes ownership of 'packet'. */
2214 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2215 struct ofpbuf *packet)
2217 struct action_xlate_ctx ctx;
2218 struct ofpbuf *odp_actions;
2219 struct facet *facet;
2223 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2225 flow_extract(packet, 0, in_port, &flow);
2227 /* First look for a related facet. If we find one, account it to that. */
2228 facet = facet_lookup_valid(ofproto, &flow);
2229 if (facet && facet->rule == rule) {
2230 facet_execute(ofproto, facet, packet);
2234 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2235 * create a new facet for it and use that. */
2236 if (rule_lookup(ofproto, &flow) == rule) {
2237 facet = facet_create(ofproto, rule, &flow, packet);
2238 facet_execute(ofproto, facet, packet);
2239 facet_install(ofproto, facet, true);
2243 /* We can't account anything to a facet. If we were to try, then that
2244 * facet would have a non-matching rule, busting our invariants. */
2245 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2246 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2247 size = packet->size;
2248 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2249 odp_actions->size, packet)) {
2250 rule->used = time_msec();
2251 rule->packet_count++;
2252 rule->byte_count += size;
2253 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
2255 ofpbuf_delete(odp_actions);
2258 /* Inserts 'rule' into 'p''s flow table. */
2260 rule_insert(struct ofproto *p, struct rule *rule)
2262 struct rule *displaced_rule;
2264 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2265 if (displaced_rule) {
2266 rule_destroy(p, displaced_rule);
2268 p->need_revalidate = true;
2271 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2272 * 'flow' and an example 'packet' within that flow.
2274 * The caller must already have determined that no facet with an identical
2275 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2276 * 'ofproto''s classifier table. */
2277 static struct facet *
2278 facet_create(struct ofproto *ofproto, struct rule *rule,
2279 const struct flow *flow, const struct ofpbuf *packet)
2281 struct facet *facet;
2283 facet = xzalloc(sizeof *facet);
2284 facet->used = time_msec();
2285 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2286 list_push_back(&rule->facets, &facet->list_node);
2288 facet->flow = *flow;
2289 netflow_flow_init(&facet->nf_flow);
2290 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2292 facet_make_actions(ofproto, facet, packet);
2298 facet_free(struct facet *facet)
2300 free(facet->actions);
2304 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2306 * - Removes 'rule' from the classifier.
2308 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2309 * destroys them), via rule_destroy().
2312 rule_remove(struct ofproto *ofproto, struct rule *rule)
2314 COVERAGE_INC(ofproto_del_rule);
2315 ofproto->need_revalidate = true;
2316 classifier_remove(&ofproto->cls, &rule->cr);
2317 rule_destroy(ofproto, rule);
2320 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2322 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2323 * rule's statistics, via facet_uninstall().
2325 * - Removes 'facet' from its rule and from ofproto->facets.
2328 facet_remove(struct ofproto *ofproto, struct facet *facet)
2330 facet_uninstall(ofproto, facet);
2331 facet_flush_stats(ofproto, facet);
2332 hmap_remove(&ofproto->facets, &facet->hmap_node);
2333 list_remove(&facet->list_node);
2337 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2339 facet_make_actions(struct ofproto *p, struct facet *facet,
2340 const struct ofpbuf *packet)
2342 const struct rule *rule = facet->rule;
2343 struct ofpbuf *odp_actions;
2344 struct action_xlate_ctx ctx;
2346 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2347 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2348 facet->tags = ctx.tags;
2349 facet->may_install = ctx.may_set_up_flow;
2350 facet->nf_flow.output_iface = ctx.nf_output_iface;
2352 if (facet->actions_len != odp_actions->size
2353 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2354 free(facet->actions);
2355 facet->actions_len = odp_actions->size;
2356 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2359 ofpbuf_delete(odp_actions);
2363 facet_put__(struct ofproto *ofproto, struct facet *facet,
2364 const struct nlattr *actions, size_t actions_len,
2365 struct dpif_flow_stats *stats)
2367 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2368 enum dpif_flow_put_flags flags;
2371 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2373 flags |= DPIF_FP_ZERO_STATS;
2374 facet->dp_packet_count = 0;
2375 facet->dp_byte_count = 0;
2378 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2379 odp_flow_key_from_flow(&key, &facet->flow);
2380 assert(key.base == keybuf);
2382 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2383 actions, actions_len, stats);
2386 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2387 * 'zero_stats' is true, clears any existing statistics from the datapath for
2390 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2392 struct dpif_flow_stats stats;
2394 if (facet->may_install
2395 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2396 zero_stats ? &stats : NULL)) {
2397 facet->installed = true;
2401 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2402 * to the accounting hook function in the ofhooks structure. */
2404 facet_account(struct ofproto *ofproto,
2405 struct facet *facet, uint64_t extra_bytes)
2407 uint64_t total_bytes = facet->byte_count + extra_bytes;
2409 if (ofproto->ofhooks->account_flow_cb
2410 && total_bytes > facet->accounted_bytes)
2412 ofproto->ofhooks->account_flow_cb(
2413 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2414 total_bytes - facet->accounted_bytes, ofproto->aux);
2415 facet->accounted_bytes = total_bytes;
2419 /* If 'rule' is installed in the datapath, uninstalls it. */
2421 facet_uninstall(struct ofproto *p, struct facet *facet)
2423 if (facet->installed) {
2424 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2425 struct dpif_flow_stats stats;
2428 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2429 odp_flow_key_from_flow(&key, &facet->flow);
2430 assert(key.base == keybuf);
2432 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2433 facet_update_stats(p, facet, &stats);
2435 facet->installed = false;
2436 facet->dp_packet_count = 0;
2437 facet->dp_byte_count = 0;
2439 assert(facet->dp_packet_count == 0);
2440 assert(facet->dp_byte_count == 0);
2444 /* Returns true if the only action for 'facet' is to send to the controller.
2445 * (We don't report NetFlow expiration messages for such facets because they
2446 * are just part of the control logic for the network, not real traffic). */
2448 facet_is_controller_flow(struct facet *facet)
2451 && facet->rule->n_actions == 1
2452 && action_outputs_to_port(&facet->rule->actions[0],
2453 htons(OFPP_CONTROLLER)));
2456 /* Folds all of 'facet''s statistics into its rule. Also updates the
2457 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2458 * 'facet''s statistics in the datapath should have been zeroed and folded into
2459 * its packet and byte counts before this function is called. */
2461 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2463 assert(!facet->dp_byte_count);
2464 assert(!facet->dp_packet_count);
2466 facet_push_stats(ofproto, facet);
2467 facet_account(ofproto, facet, 0);
2469 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2470 struct ofexpired expired;
2471 expired.flow = facet->flow;
2472 expired.packet_count = facet->packet_count;
2473 expired.byte_count = facet->byte_count;
2474 expired.used = facet->used;
2475 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2478 facet->rule->packet_count += facet->packet_count;
2479 facet->rule->byte_count += facet->byte_count;
2481 /* Reset counters to prevent double counting if 'facet' ever gets
2483 facet->packet_count = 0;
2484 facet->byte_count = 0;
2485 facet->rs_packet_count = 0;
2486 facet->rs_byte_count = 0;
2487 facet->accounted_bytes = 0;
2489 netflow_flow_clear(&facet->nf_flow);
2492 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2493 * Returns it if found, otherwise a null pointer.
2495 * The returned facet might need revalidation; use facet_lookup_valid()
2496 * instead if that is important. */
2497 static struct facet *
2498 facet_find(struct ofproto *ofproto, const struct flow *flow)
2500 struct facet *facet;
2502 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2504 if (flow_equal(flow, &facet->flow)) {
2512 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2513 * Returns it if found, otherwise a null pointer.
2515 * The returned facet is guaranteed to be valid. */
2516 static struct facet *
2517 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2519 struct facet *facet = facet_find(ofproto, flow);
2521 /* The facet we found might not be valid, since we could be in need of
2522 * revalidation. If it is not valid, don't return it. */
2524 && ofproto->need_revalidate
2525 && !facet_revalidate(ofproto, facet)) {
2526 COVERAGE_INC(ofproto_invalidated);
2533 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2535 * - If the rule found is different from 'facet''s current rule, moves
2536 * 'facet' to the new rule and recompiles its actions.
2538 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2539 * where it is and recompiles its actions anyway.
2541 * - If there is none, destroys 'facet'.
2543 * Returns true if 'facet' still exists, false if it has been destroyed. */
2545 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2547 struct action_xlate_ctx ctx;
2548 struct ofpbuf *odp_actions;
2549 struct rule *new_rule;
2550 bool actions_changed;
2552 COVERAGE_INC(facet_revalidate);
2554 /* Determine the new rule. */
2555 new_rule = rule_lookup(ofproto, &facet->flow);
2557 /* No new rule, so delete the facet. */
2558 facet_remove(ofproto, facet);
2562 /* Calculate new ODP actions.
2564 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2565 * emit a NetFlow expiration and, if so, we need to have the old state
2566 * around to properly compose it. */
2567 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2568 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2569 actions_changed = (facet->actions_len != odp_actions->size
2570 || memcmp(facet->actions, odp_actions->data,
2571 facet->actions_len));
2573 /* If the ODP actions changed or the installability changed, then we need
2574 * to talk to the datapath. */
2575 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2576 if (ctx.may_set_up_flow) {
2577 struct dpif_flow_stats stats;
2579 facet_put__(ofproto, facet,
2580 odp_actions->data, odp_actions->size, &stats);
2581 facet_update_stats(ofproto, facet, &stats);
2583 facet_uninstall(ofproto, facet);
2586 /* The datapath flow is gone or has zeroed stats, so push stats out of
2587 * 'facet' into 'rule'. */
2588 facet_flush_stats(ofproto, facet);
2591 /* Update 'facet' now that we've taken care of all the old state. */
2592 facet->tags = ctx.tags;
2593 facet->nf_flow.output_iface = ctx.nf_output_iface;
2594 facet->may_install = ctx.may_set_up_flow;
2595 if (actions_changed) {
2596 free(facet->actions);
2597 facet->actions_len = odp_actions->size;
2598 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2600 if (facet->rule != new_rule) {
2601 COVERAGE_INC(facet_changed_rule);
2602 list_remove(&facet->list_node);
2603 list_push_back(&new_rule->facets, &facet->list_node);
2604 facet->rule = new_rule;
2605 facet->used = new_rule->created;
2606 facet->rs_used = facet->used;
2609 ofpbuf_delete(odp_actions);
2615 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2616 struct rconn_packet_counter *counter)
2618 update_openflow_length(msg);
2619 if (rconn_send(ofconn->rconn, msg, counter)) {
2625 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2628 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2630 COVERAGE_INC(ofproto_error);
2631 queue_tx(buf, ofconn, ofconn->reply_counter);
2636 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2638 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2643 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2645 struct ofp_switch_features *osf;
2647 struct ofport *port;
2649 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2650 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2651 osf->n_buffers = htonl(pktbuf_capacity());
2653 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2654 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2655 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2656 (1u << OFPAT_SET_VLAN_VID) |
2657 (1u << OFPAT_SET_VLAN_PCP) |
2658 (1u << OFPAT_STRIP_VLAN) |
2659 (1u << OFPAT_SET_DL_SRC) |
2660 (1u << OFPAT_SET_DL_DST) |
2661 (1u << OFPAT_SET_NW_SRC) |
2662 (1u << OFPAT_SET_NW_DST) |
2663 (1u << OFPAT_SET_NW_TOS) |
2664 (1u << OFPAT_SET_TP_SRC) |
2665 (1u << OFPAT_SET_TP_DST) |
2666 (1u << OFPAT_ENQUEUE));
2668 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2669 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2672 queue_tx(buf, ofconn, ofconn->reply_counter);
2677 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2680 struct ofp_switch_config *osc;
2684 /* Figure out flags. */
2685 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2686 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2689 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2690 osc->flags = htons(flags);
2691 osc->miss_send_len = htons(ofconn->miss_send_len);
2692 queue_tx(buf, ofconn, ofconn->reply_counter);
2698 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2700 uint16_t flags = ntohs(osc->flags);
2702 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2703 switch (flags & OFPC_FRAG_MASK) {
2704 case OFPC_FRAG_NORMAL:
2705 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2707 case OFPC_FRAG_DROP:
2708 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2711 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2717 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2722 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2723 struct action_xlate_ctx *ctx);
2726 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2728 const struct ofport *ofport = get_port(ctx->ofproto, port);
2731 if (ofport->opp.config & OFPPC_NO_FWD) {
2732 /* Forwarding disabled on port. */
2737 * We don't have an ofport record for this port, but it doesn't hurt to
2738 * allow forwarding to it anyhow. Maybe such a port will appear later
2739 * and we're pre-populating the flow table.
2743 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2744 ctx->nf_output_iface = port;
2747 static struct rule *
2748 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2750 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2754 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2756 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2757 uint16_t old_in_port;
2760 /* Look up a flow with 'in_port' as the input port. Then restore the
2761 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2762 * have surprising behavior). */
2763 old_in_port = ctx->flow.in_port;
2764 ctx->flow.in_port = in_port;
2765 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2766 ctx->flow.in_port = old_in_port;
2768 if (ctx->resubmit_hook) {
2769 ctx->resubmit_hook(ctx, rule);
2774 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2778 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2780 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2781 MAX_RESUBMIT_RECURSION);
2786 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2787 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2789 struct ofport *ofport;
2791 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2792 uint16_t odp_port = ofport->odp_port;
2793 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2794 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2797 *nf_output_iface = NF_OUT_FLOOD;
2801 xlate_output_action__(struct action_xlate_ctx *ctx,
2802 uint16_t port, uint16_t max_len)
2805 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2807 ctx->nf_output_iface = NF_OUT_DROP;
2811 add_output_action(ctx, ctx->flow.in_port);
2814 xlate_table_action(ctx, ctx->flow.in_port);
2817 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2818 ctx->odp_actions, &ctx->tags,
2819 &ctx->nf_output_iface,
2820 ctx->ofproto->aux)) {
2821 COVERAGE_INC(ofproto_uninstallable);
2822 ctx->may_set_up_flow = false;
2826 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2827 &ctx->nf_output_iface, ctx->odp_actions);
2830 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2831 &ctx->nf_output_iface, ctx->odp_actions);
2833 case OFPP_CONTROLLER:
2834 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2837 add_output_action(ctx, ODPP_LOCAL);
2840 odp_port = ofp_port_to_odp_port(port);
2841 if (odp_port != ctx->flow.in_port) {
2842 add_output_action(ctx, odp_port);
2847 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2848 ctx->nf_output_iface = NF_OUT_FLOOD;
2849 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2850 ctx->nf_output_iface = prev_nf_output_iface;
2851 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2852 ctx->nf_output_iface != NF_OUT_FLOOD) {
2853 ctx->nf_output_iface = NF_OUT_MULTI;
2858 xlate_output_action(struct action_xlate_ctx *ctx,
2859 const struct ofp_action_output *oao)
2861 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2864 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2865 * optimization, because we're going to add another action that sets the
2866 * priority immediately after, or because there are no actions following the
2869 remove_pop_action(struct action_xlate_ctx *ctx)
2871 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2872 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2873 ctx->last_pop_priority = -1;
2878 add_pop_action(struct action_xlate_ctx *ctx)
2880 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2881 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2882 ctx->last_pop_priority = ctx->odp_actions->size;
2887 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2888 const struct ofp_action_enqueue *oae)
2890 uint16_t ofp_port, odp_port;
2894 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2897 /* Fall back to ordinary output action. */
2898 xlate_output_action__(ctx, ntohs(oae->port), 0);
2902 /* Figure out ODP output port. */
2903 ofp_port = ntohs(oae->port);
2904 if (ofp_port != OFPP_IN_PORT) {
2905 odp_port = ofp_port_to_odp_port(ofp_port);
2907 odp_port = ctx->flow.in_port;
2910 /* Add ODP actions. */
2911 remove_pop_action(ctx);
2912 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2913 add_output_action(ctx, odp_port);
2914 add_pop_action(ctx);
2916 /* Update NetFlow output port. */
2917 if (ctx->nf_output_iface == NF_OUT_DROP) {
2918 ctx->nf_output_iface = odp_port;
2919 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2920 ctx->nf_output_iface = NF_OUT_MULTI;
2925 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2926 const struct nx_action_set_queue *nasq)
2931 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2934 /* Couldn't translate queue to a priority, so ignore. A warning
2935 * has already been logged. */
2939 remove_pop_action(ctx);
2940 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2944 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2946 ovs_be16 tci = ctx->flow.vlan_tci;
2947 if (!(tci & htons(VLAN_CFI))) {
2948 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2950 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2951 tci & ~htons(VLAN_CFI));
2955 struct xlate_reg_state {
2961 save_reg_state(const struct action_xlate_ctx *ctx,
2962 struct xlate_reg_state *state)
2964 state->vlan_tci = ctx->flow.vlan_tci;
2965 state->tun_id = ctx->flow.tun_id;
2969 update_reg_state(struct action_xlate_ctx *ctx,
2970 const struct xlate_reg_state *state)
2972 if (ctx->flow.vlan_tci != state->vlan_tci) {
2973 xlate_set_dl_tci(ctx);
2975 if (ctx->flow.tun_id != state->tun_id) {
2976 nl_msg_put_be64(ctx->odp_actions,
2977 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2982 xlate_nicira_action(struct action_xlate_ctx *ctx,
2983 const struct nx_action_header *nah)
2985 const struct nx_action_resubmit *nar;
2986 const struct nx_action_set_tunnel *nast;
2987 const struct nx_action_set_queue *nasq;
2988 const struct nx_action_multipath *nam;
2989 enum nx_action_subtype subtype = ntohs(nah->subtype);
2990 struct xlate_reg_state state;
2993 assert(nah->vendor == htonl(NX_VENDOR_ID));
2995 case NXAST_RESUBMIT:
2996 nar = (const struct nx_action_resubmit *) nah;
2997 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
3000 case NXAST_SET_TUNNEL:
3001 nast = (const struct nx_action_set_tunnel *) nah;
3002 tun_id = htonll(ntohl(nast->tun_id));
3003 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3004 ctx->flow.tun_id = tun_id;
3007 case NXAST_DROP_SPOOFED_ARP:
3008 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
3009 nl_msg_put_flag(ctx->odp_actions,
3010 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
3014 case NXAST_SET_QUEUE:
3015 nasq = (const struct nx_action_set_queue *) nah;
3016 xlate_set_queue_action(ctx, nasq);
3019 case NXAST_POP_QUEUE:
3020 add_pop_action(ctx);
3023 case NXAST_REG_MOVE:
3024 save_reg_state(ctx, &state);
3025 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
3027 update_reg_state(ctx, &state);
3030 case NXAST_REG_LOAD:
3031 save_reg_state(ctx, &state);
3032 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
3034 update_reg_state(ctx, &state);
3038 /* Nothing to do. */
3041 case NXAST_SET_TUNNEL64:
3042 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
3043 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3044 ctx->flow.tun_id = tun_id;
3047 case NXAST_MULTIPATH:
3048 nam = (const struct nx_action_multipath *) nah;
3049 multipath_execute(nam, &ctx->flow);
3052 /* If you add a new action here that modifies flow data, don't forget to
3053 * update the flow key in ctx->flow at the same time. */
3055 case NXAST_SNAT__OBSOLETE:
3057 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3063 do_xlate_actions(const union ofp_action *in, size_t n_in,
3064 struct action_xlate_ctx *ctx)
3066 struct actions_iterator iter;
3067 const union ofp_action *ia;
3068 const struct ofport *port;
3070 port = get_port(ctx->ofproto, ctx->flow.in_port);
3071 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3072 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3073 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3074 /* Drop this flow. */
3078 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3079 enum ofp_action_type type = ntohs(ia->type);
3080 const struct ofp_action_dl_addr *oada;
3084 xlate_output_action(ctx, &ia->output);
3087 case OFPAT_SET_VLAN_VID:
3088 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3089 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3090 xlate_set_dl_tci(ctx);
3093 case OFPAT_SET_VLAN_PCP:
3094 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3095 ctx->flow.vlan_tci |= htons(
3096 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3097 xlate_set_dl_tci(ctx);
3100 case OFPAT_STRIP_VLAN:
3101 ctx->flow.vlan_tci = htons(0);
3102 xlate_set_dl_tci(ctx);
3105 case OFPAT_SET_DL_SRC:
3106 oada = ((struct ofp_action_dl_addr *) ia);
3107 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3108 oada->dl_addr, ETH_ADDR_LEN);
3109 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3112 case OFPAT_SET_DL_DST:
3113 oada = ((struct ofp_action_dl_addr *) ia);
3114 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3115 oada->dl_addr, ETH_ADDR_LEN);
3116 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3119 case OFPAT_SET_NW_SRC:
3120 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3121 ia->nw_addr.nw_addr);
3122 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3125 case OFPAT_SET_NW_DST:
3126 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3127 ia->nw_addr.nw_addr);
3128 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3131 case OFPAT_SET_NW_TOS:
3132 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3134 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3137 case OFPAT_SET_TP_SRC:
3138 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3139 ia->tp_port.tp_port);
3140 ctx->flow.tp_src = ia->tp_port.tp_port;
3143 case OFPAT_SET_TP_DST:
3144 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3145 ia->tp_port.tp_port);
3146 ctx->flow.tp_dst = ia->tp_port.tp_port;
3150 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3154 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3158 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3165 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3166 struct ofproto *ofproto, const struct flow *flow,
3167 const struct ofpbuf *packet)
3169 ctx->ofproto = ofproto;
3171 ctx->packet = packet;
3172 ctx->resubmit_hook = NULL;
3173 ctx->check_special = true;
3177 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
3178 const struct ofpbuf *packet)
3180 struct ofport *ofport;
3182 ofport = get_port(ofproto, flow->in_port);
3183 if (ofport && ofport->cfm) {
3184 cfm_process_heartbeat(ofport->cfm, packet);
3188 static struct ofpbuf *
3189 xlate_actions(struct action_xlate_ctx *ctx,
3190 const union ofp_action *in, size_t n_in)
3192 COVERAGE_INC(ofproto_ofp2odp);
3194 ctx->odp_actions = ofpbuf_new(512);
3196 ctx->may_set_up_flow = true;
3197 ctx->nf_output_iface = NF_OUT_DROP;
3199 ctx->last_pop_priority = -1;
3201 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
3203 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
3205 ctx->may_set_up_flow = false;
3206 } else if (ctx->check_special
3207 && ctx->ofproto->ofhooks->special_cb
3208 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3209 ctx->ofproto->aux)) {
3210 ctx->may_set_up_flow = false;
3212 do_xlate_actions(in, n_in, ctx);
3215 remove_pop_action(ctx);
3217 /* Check with in-band control to see if we're allowed to set up this
3219 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3220 ctx->odp_actions->data, ctx->odp_actions->size)) {
3221 ctx->may_set_up_flow = false;
3224 return ctx->odp_actions;
3227 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3228 * error message code (composed with ofp_mkerr()) for the caller to propagate
3229 * upward. Otherwise, returns 0.
3231 * The log message mentions 'msg_type'. */
3233 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3235 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3236 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3237 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3240 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3247 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3249 struct ofproto *p = ofconn->ofproto;
3250 struct ofp_packet_out *opo;
3251 struct ofpbuf payload, *buffer;
3252 union ofp_action *ofp_actions;
3253 struct action_xlate_ctx ctx;
3254 struct ofpbuf *odp_actions;
3255 struct ofpbuf request;
3257 size_t n_ofp_actions;
3261 COVERAGE_INC(ofproto_packet_out);
3263 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3268 /* Get ofp_packet_out. */
3269 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3270 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3273 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3274 &ofp_actions, &n_ofp_actions);
3280 if (opo->buffer_id != htonl(UINT32_MAX)) {
3281 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3283 if (error || !buffer) {
3292 /* Extract flow, check actions. */
3293 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3295 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3301 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3302 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3303 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3304 ofpbuf_delete(odp_actions);
3307 ofpbuf_delete(buffer);
3312 update_port_config(struct ofproto *p, struct ofport *port,
3313 uint32_t config, uint32_t mask)
3315 mask &= config ^ port->opp.config;
3316 if (mask & OFPPC_PORT_DOWN) {
3317 if (config & OFPPC_PORT_DOWN) {
3318 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3320 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3323 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3324 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3325 if (mask & REVALIDATE_BITS) {
3326 COVERAGE_INC(ofproto_costly_flags);
3327 port->opp.config ^= mask & REVALIDATE_BITS;
3328 p->need_revalidate = true;
3330 #undef REVALIDATE_BITS
3331 if (mask & OFPPC_NO_PACKET_IN) {
3332 port->opp.config ^= OFPPC_NO_PACKET_IN;
3337 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3339 struct ofproto *p = ofconn->ofproto;
3340 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3341 struct ofport *port;
3344 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3349 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3351 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3352 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3353 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3355 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3356 if (opm->advertise) {
3357 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3363 static struct ofpbuf *
3364 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3366 struct ofp_stats_reply *osr;
3369 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3370 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3372 osr->flags = htons(0);
3376 static struct ofpbuf *
3377 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3379 const struct ofp_stats_request *osr
3380 = (const struct ofp_stats_request *) request;
3381 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3385 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3386 struct ofpbuf **msgp)
3388 struct ofpbuf *msg = *msgp;
3389 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3390 if (nbytes + msg->size > UINT16_MAX) {
3391 struct ofp_stats_reply *reply = msg->data;
3392 reply->flags = htons(OFPSF_REPLY_MORE);
3393 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3394 queue_tx(msg, ofconn, ofconn->reply_counter);
3396 return ofpbuf_put_uninit(*msgp, nbytes);
3399 static struct ofpbuf *
3400 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3402 struct nicira_stats_msg *nsm;
3405 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3406 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3407 nsm->type = htons(OFPST_VENDOR);
3408 nsm->flags = htons(0);
3409 nsm->vendor = htonl(NX_VENDOR_ID);
3410 nsm->subtype = subtype;
3414 static struct ofpbuf *
3415 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3417 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3421 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3422 struct ofpbuf **msgp)
3424 struct ofpbuf *msg = *msgp;
3425 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3426 if (nbytes + msg->size > UINT16_MAX) {
3427 struct nicira_stats_msg *reply = msg->data;
3428 reply->flags = htons(OFPSF_REPLY_MORE);
3429 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3430 queue_tx(msg, ofconn, ofconn->reply_counter);
3432 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3436 handle_desc_stats_request(struct ofconn *ofconn,
3437 const struct ofp_header *request)
3439 struct ofproto *p = ofconn->ofproto;
3440 struct ofp_desc_stats *ods;
3443 msg = start_ofp_stats_reply(request, sizeof *ods);
3444 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3445 memset(ods, 0, sizeof *ods);
3446 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3447 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3448 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3449 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3450 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3451 queue_tx(msg, ofconn, ofconn->reply_counter);
3457 handle_table_stats_request(struct ofconn *ofconn,
3458 const struct ofp_header *request)
3460 struct ofproto *p = ofconn->ofproto;
3461 struct ofp_table_stats *ots;
3464 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3466 /* Classifier table. */
3467 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3468 memset(ots, 0, sizeof *ots);
3469 strcpy(ots->name, "classifier");
3470 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3471 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3472 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3473 ots->active_count = htonl(classifier_count(&p->cls));
3474 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3475 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3477 queue_tx(msg, ofconn, ofconn->reply_counter);
3482 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3483 struct ofpbuf **msgp)
3485 struct netdev_stats stats;
3486 struct ofp_port_stats *ops;
3488 /* Intentionally ignore return value, since errors will set
3489 * 'stats' to all-1s, which is correct for OpenFlow, and
3490 * netdev_get_stats() will log errors. */
3491 netdev_get_stats(port->netdev, &stats);
3493 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3494 ops->port_no = htons(port->opp.port_no);
3495 memset(ops->pad, 0, sizeof ops->pad);
3496 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3497 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3498 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3499 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3500 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3501 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3502 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3503 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3504 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3505 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3506 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3507 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3511 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3513 struct ofproto *p = ofconn->ofproto;
3514 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3515 struct ofp_port_stats *ops;
3517 struct ofport *port;
3519 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3520 if (psr->port_no != htons(OFPP_NONE)) {
3521 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3523 append_port_stat(port, ofconn, &msg);
3526 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3527 append_port_stat(port, ofconn, &msg);
3531 queue_tx(msg, ofconn, ofconn->reply_counter);
3536 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
3538 long long int msecs = time_msec() - start;
3539 *sec = msecs / 1000;
3540 *nsec = (msecs % 1000) * (1000 * 1000);
3544 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
3548 calc_flow_duration__(start, &sec, &nsec);
3549 *sec_be = htonl(sec);
3550 *nsec_be = htonl(nsec);
3554 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3555 ovs_be16 out_port, struct ofpbuf **replyp)
3557 struct ofp_flow_stats *ofs;
3558 uint64_t packet_count, byte_count;
3560 size_t act_len, len;
3562 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3566 act_len = sizeof *rule->actions * rule->n_actions;
3567 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3569 rule_get_stats(rule, &packet_count, &byte_count);
3571 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3572 ofs->length = htons(len);
3575 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3576 rule->flow_cookie, &cookie);
3577 put_32aligned_be64(&ofs->cookie, cookie);
3578 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3579 ofs->priority = htons(rule->cr.priority);
3580 ofs->idle_timeout = htons(rule->idle_timeout);
3581 ofs->hard_timeout = htons(rule->hard_timeout);
3582 memset(ofs->pad2, 0, sizeof ofs->pad2);
3583 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3584 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3585 if (rule->n_actions > 0) {
3586 memcpy(ofs->actions, rule->actions, act_len);
3591 is_valid_table(uint8_t table_id)
3593 if (table_id == 0 || table_id == 0xff) {
3596 /* It would probably be better to reply with an error but there doesn't
3597 * seem to be any appropriate value, so that might just be
3599 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3606 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3608 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3609 struct ofpbuf *reply;
3611 COVERAGE_INC(ofproto_flows_req);
3612 reply = start_ofp_stats_reply(oh, 1024);
3613 if (is_valid_table(fsr->table_id)) {
3614 struct cls_cursor cursor;
3615 struct cls_rule target;
3618 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3620 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3621 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3622 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3625 queue_tx(reply, ofconn, ofconn->reply_counter);
3631 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3632 ovs_be16 out_port, struct ofpbuf **replyp)
3634 struct nx_flow_stats *nfs;
3635 uint64_t packet_count, byte_count;
3636 size_t act_len, start_len;
3637 struct ofpbuf *reply;
3639 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3643 rule_get_stats(rule, &packet_count, &byte_count);
3645 act_len = sizeof *rule->actions * rule->n_actions;
3647 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3648 start_len = (*replyp)->size;
3651 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3654 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3655 nfs->cookie = rule->flow_cookie;
3656 nfs->priority = htons(rule->cr.priority);
3657 nfs->idle_timeout = htons(rule->idle_timeout);
3658 nfs->hard_timeout = htons(rule->hard_timeout);
3659 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3660 memset(nfs->pad2, 0, sizeof nfs->pad2);
3661 nfs->packet_count = htonll(packet_count);
3662 nfs->byte_count = htonll(byte_count);
3663 if (rule->n_actions > 0) {
3664 ofpbuf_put(reply, rule->actions, act_len);
3666 nfs->length = htons(reply->size - start_len);
3670 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3672 struct nx_flow_stats_request *nfsr;
3673 struct cls_rule target;
3674 struct ofpbuf *reply;
3678 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3680 /* Dissect the message. */
3681 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3682 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3687 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3690 COVERAGE_INC(ofproto_flows_req);
3691 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3692 if (is_valid_table(nfsr->table_id)) {
3693 struct cls_cursor cursor;
3696 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3697 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3698 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3701 queue_tx(reply, ofconn, ofconn->reply_counter);
3707 flow_stats_ds(struct rule *rule, struct ds *results)
3709 uint64_t packet_count, byte_count;
3710 size_t act_len = sizeof *rule->actions * rule->n_actions;
3712 rule_get_stats(rule, &packet_count, &byte_count);
3714 ds_put_format(results, "duration=%llds, ",
3715 (time_msec() - rule->created) / 1000);
3716 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3717 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3718 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3719 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3720 cls_rule_format(&rule->cr, results);
3721 ds_put_char(results, ',');
3723 ofp_print_actions(results, &rule->actions->header, act_len);
3725 ds_put_cstr(results, "drop");
3727 ds_put_cstr(results, "\n");
3730 /* Adds a pretty-printed description of all flows to 'results', including
3731 * hidden flows (e.g., set up by in-band control). */
3733 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3735 struct cls_cursor cursor;
3738 cls_cursor_init(&cursor, &p->cls, NULL);
3739 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3740 flow_stats_ds(rule, results);
3745 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3746 ovs_be16 out_port, uint8_t table_id,
3747 struct ofp_aggregate_stats_reply *oasr)
3749 uint64_t total_packets = 0;
3750 uint64_t total_bytes = 0;
3753 COVERAGE_INC(ofproto_agg_request);
3755 if (is_valid_table(table_id)) {
3756 struct cls_cursor cursor;
3759 cls_cursor_init(&cursor, &ofproto->cls, target);
3760 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3761 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3762 uint64_t packet_count;
3763 uint64_t byte_count;
3765 rule_get_stats(rule, &packet_count, &byte_count);
3767 total_packets += packet_count;
3768 total_bytes += byte_count;
3774 oasr->flow_count = htonl(n_flows);
3775 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3776 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3777 memset(oasr->pad, 0, sizeof oasr->pad);
3781 handle_aggregate_stats_request(struct ofconn *ofconn,
3782 const struct ofp_header *oh)
3784 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3785 struct ofp_aggregate_stats_reply *reply;
3786 struct cls_rule target;
3789 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3792 msg = start_ofp_stats_reply(oh, sizeof *reply);
3793 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3794 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3795 request->table_id, reply);
3796 queue_tx(msg, ofconn, ofconn->reply_counter);
3801 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3803 struct nx_aggregate_stats_request *request;
3804 struct ofp_aggregate_stats_reply *reply;
3805 struct cls_rule target;
3810 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3812 /* Dissect the message. */
3813 request = ofpbuf_pull(&b, sizeof *request);
3814 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3819 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3823 COVERAGE_INC(ofproto_flows_req);
3824 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3825 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3826 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3827 request->table_id, reply);
3828 queue_tx(buf, ofconn, ofconn->reply_counter);
3833 struct queue_stats_cbdata {
3834 struct ofconn *ofconn;
3835 struct ofport *ofport;
3840 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3841 const struct netdev_queue_stats *stats)
3843 struct ofp_queue_stats *reply;
3845 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3846 reply->port_no = htons(cbdata->ofport->opp.port_no);
3847 memset(reply->pad, 0, sizeof reply->pad);
3848 reply->queue_id = htonl(queue_id);
3849 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3850 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3851 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3855 handle_queue_stats_dump_cb(uint32_t queue_id,
3856 struct netdev_queue_stats *stats,
3859 struct queue_stats_cbdata *cbdata = cbdata_;
3861 put_queue_stats(cbdata, queue_id, stats);
3865 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3866 struct queue_stats_cbdata *cbdata)
3868 cbdata->ofport = port;
3869 if (queue_id == OFPQ_ALL) {
3870 netdev_dump_queue_stats(port->netdev,
3871 handle_queue_stats_dump_cb, cbdata);
3873 struct netdev_queue_stats stats;
3875 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3876 put_queue_stats(cbdata, queue_id, &stats);
3882 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3884 struct ofproto *ofproto = ofconn->ofproto;
3885 const struct ofp_queue_stats_request *qsr;
3886 struct queue_stats_cbdata cbdata;
3887 struct ofport *port;
3888 unsigned int port_no;
3891 qsr = ofputil_stats_body(oh);
3893 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3896 COVERAGE_INC(ofproto_queue_req);
3898 cbdata.ofconn = ofconn;
3899 cbdata.msg = start_ofp_stats_reply(oh, 128);
3901 port_no = ntohs(qsr->port_no);
3902 queue_id = ntohl(qsr->queue_id);
3903 if (port_no == OFPP_ALL) {
3904 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3905 handle_queue_stats_for_port(port, queue_id, &cbdata);
3907 } else if (port_no < ofproto->max_ports) {
3908 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3910 handle_queue_stats_for_port(port, queue_id, &cbdata);
3913 ofpbuf_delete(cbdata.msg);
3914 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3916 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3921 /* Updates 'facet''s used time. Caller is responsible for calling
3922 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3924 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3927 if (used > facet->used) {
3929 if (used > facet->rule->used) {
3930 facet->rule->used = used;
3932 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3936 /* Folds the statistics from 'stats' into the counters in 'facet'.
3938 * Because of the meaning of a facet's counters, it only makes sense to do this
3939 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3940 * packet that was sent by hand or if it represents statistics that have been
3941 * cleared out of the datapath. */
3943 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3944 const struct dpif_flow_stats *stats)
3946 if (stats->n_packets || stats->used > facet->used) {
3947 facet_update_time(ofproto, facet, stats->used);
3948 facet->packet_count += stats->n_packets;
3949 facet->byte_count += stats->n_bytes;
3950 facet_push_stats(ofproto, facet);
3951 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3956 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3958 uint64_t rs_packets, rs_bytes;
3960 assert(facet->packet_count >= facet->rs_packet_count);
3961 assert(facet->byte_count >= facet->rs_byte_count);
3962 assert(facet->used >= facet->rs_used);
3964 rs_packets = facet->packet_count - facet->rs_packet_count;
3965 rs_bytes = facet->byte_count - facet->rs_byte_count;
3967 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3968 facet->rs_packet_count = facet->packet_count;
3969 facet->rs_byte_count = facet->byte_count;
3970 facet->rs_used = facet->used;
3972 flow_push_stats(ofproto, facet->rule, &facet->flow,
3973 rs_packets, rs_bytes, facet->used);
3977 struct ofproto_push {
3978 struct action_xlate_ctx ctx;
3985 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3987 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3990 rule->packet_count += push->packets;
3991 rule->byte_count += push->bytes;
3992 rule->used = MAX(push->used, rule->used);
3996 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3997 * 'rule''s actions. */
3999 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
4000 struct flow *flow, uint64_t packets, uint64_t bytes,
4003 struct ofproto_push push;
4005 push.packets = packets;
4009 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
4010 push.ctx.resubmit_hook = push_resubmit;
4011 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
4014 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4015 * in which no matching flow already exists in the flow table.
4017 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4018 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
4019 * OpenFlow error code as encoded by ofp_mkerr() on failure.
4021 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4024 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
4026 struct ofproto *p = ofconn->ofproto;
4027 struct ofpbuf *packet;
4032 if (fm->flags & OFPFF_CHECK_OVERLAP
4033 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
4034 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
4038 if (fm->buffer_id != UINT32_MAX) {
4039 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
4043 in_port = UINT16_MAX;
4046 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
4047 fm->idle_timeout, fm->hard_timeout, fm->cookie,
4048 fm->flags & OFPFF_SEND_FLOW_REM);
4049 rule_insert(p, rule);
4051 rule_execute(p, rule, in_port, packet);
4056 static struct rule *
4057 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
4059 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
4063 send_buffered_packet(struct ofconn *ofconn,
4064 struct rule *rule, uint32_t buffer_id)
4066 struct ofpbuf *packet;
4070 if (buffer_id == UINT32_MAX) {
4074 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
4079 rule_execute(ofconn->ofproto, rule, in_port, packet);
4084 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4086 struct modify_flows_cbdata {
4087 struct ofproto *ofproto;
4088 const struct flow_mod *fm;
4092 static int modify_flow(struct ofproto *, const struct flow_mod *,
4095 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4096 * encoded by ofp_mkerr() on failure.
4098 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4101 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
4103 struct ofproto *p = ofconn->ofproto;
4104 struct rule *match = NULL;
4105 struct cls_cursor cursor;
4108 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4109 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4110 if (!rule_is_hidden(rule)) {
4112 modify_flow(p, fm, rule);
4117 /* This credits the packet to whichever flow happened to match last.
4118 * That's weird. Maybe we should do a lookup for the flow that
4119 * actually matches the packet? Who knows. */
4120 send_buffered_packet(ofconn, match, fm->buffer_id);
4123 return add_flow(ofconn, fm);
4127 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4128 * code as encoded by ofp_mkerr() on failure.
4130 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4133 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4135 struct ofproto *p = ofconn->ofproto;
4136 struct rule *rule = find_flow_strict(p, fm);
4137 if (rule && !rule_is_hidden(rule)) {
4138 modify_flow(p, fm, rule);
4139 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4141 return add_flow(ofconn, fm);
4145 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4146 * been identified as a flow in 'p''s flow table to be modified, by changing
4147 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4148 * ofp_action[] structures). */
4150 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4152 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4154 rule->flow_cookie = fm->cookie;
4156 /* If the actions are the same, do nothing. */
4157 if (fm->n_actions == rule->n_actions
4159 || !memcmp(fm->actions, rule->actions, actions_len))) {
4163 /* Replace actions. */
4164 free(rule->actions);
4165 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4166 rule->n_actions = fm->n_actions;
4168 p->need_revalidate = true;
4173 /* OFPFC_DELETE implementation. */
4175 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4177 /* Implements OFPFC_DELETE. */
4179 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4181 struct rule *rule, *next_rule;
4182 struct cls_cursor cursor;
4184 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4185 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4186 delete_flow(p, rule, htons(fm->out_port));
4190 /* Implements OFPFC_DELETE_STRICT. */
4192 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4194 struct rule *rule = find_flow_strict(p, fm);
4196 delete_flow(p, rule, htons(fm->out_port));
4200 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4201 * been identified as a flow to delete from 'p''s flow table, by deleting the
4202 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4205 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4206 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4207 * specified 'out_port'. */
4209 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4211 if (rule_is_hidden(rule)) {
4215 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4219 rule_send_removed(p, rule, OFPRR_DELETE);
4220 rule_remove(p, rule);
4224 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4226 struct ofproto *p = ofconn->ofproto;
4230 error = reject_slave_controller(ofconn, "flow_mod");
4235 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4240 /* We do not support the emergency flow cache. It will hopefully get
4241 * dropped from OpenFlow in the near future. */
4242 if (fm.flags & OFPFF_EMERG) {
4243 /* There isn't a good fit for an error code, so just state that the
4244 * flow table is full. */
4245 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4248 error = validate_actions(fm.actions, fm.n_actions,
4249 &fm.cr.flow, p->max_ports);
4254 switch (fm.command) {
4256 return add_flow(ofconn, &fm);
4259 return modify_flows_loose(ofconn, &fm);
4261 case OFPFC_MODIFY_STRICT:
4262 return modify_flow_strict(ofconn, &fm);
4265 delete_flows_loose(p, &fm);
4268 case OFPFC_DELETE_STRICT:
4269 delete_flow_strict(p, &fm);
4273 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4278 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4280 const struct nxt_tun_id_cookie *msg
4281 = (const struct nxt_tun_id_cookie *) oh;
4283 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4288 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4290 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4291 struct nx_role_request *reply;
4295 if (ofconn->type != OFCONN_PRIMARY) {
4296 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4298 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4301 role = ntohl(nrr->role);
4302 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4303 && role != NX_ROLE_SLAVE) {
4304 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4306 /* There's no good error code for this. */
4307 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4310 if (role == NX_ROLE_MASTER) {
4311 struct ofconn *other;
4313 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4314 if (other->role == NX_ROLE_MASTER) {
4315 other->role = NX_ROLE_SLAVE;
4319 ofconn->role = role;
4321 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4322 reply->role = htonl(role);
4323 queue_tx(buf, ofconn, ofconn->reply_counter);
4329 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4331 const struct nxt_set_flow_format *msg
4332 = (const struct nxt_set_flow_format *) oh;
4335 format = ntohl(msg->format);
4336 if (format == NXFF_OPENFLOW10
4337 || format == NXFF_TUN_ID_FROM_COOKIE
4338 || format == NXFF_NXM) {
4339 ofconn->flow_format = format;
4342 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4347 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4349 struct ofp_header *ob;
4352 /* Currently, everything executes synchronously, so we can just
4353 * immediately send the barrier reply. */
4354 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4355 queue_tx(buf, ofconn, ofconn->reply_counter);
4360 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4362 const struct ofp_header *oh = msg->data;
4363 const struct ofputil_msg_type *type;
4366 error = ofputil_decode_msg_type(oh, &type);
4371 switch (ofputil_msg_type_code(type)) {
4372 /* OpenFlow requests. */
4373 case OFPUTIL_OFPT_ECHO_REQUEST:
4374 return handle_echo_request(ofconn, oh);
4376 case OFPUTIL_OFPT_FEATURES_REQUEST:
4377 return handle_features_request(ofconn, oh);
4379 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4380 return handle_get_config_request(ofconn, oh);
4382 case OFPUTIL_OFPT_SET_CONFIG:
4383 return handle_set_config(ofconn, msg->data);
4385 case OFPUTIL_OFPT_PACKET_OUT:
4386 return handle_packet_out(ofconn, oh);
4388 case OFPUTIL_OFPT_PORT_MOD:
4389 return handle_port_mod(ofconn, oh);
4391 case OFPUTIL_OFPT_FLOW_MOD:
4392 return handle_flow_mod(ofconn, oh);
4394 case OFPUTIL_OFPT_BARRIER_REQUEST:
4395 return handle_barrier_request(ofconn, oh);
4397 /* OpenFlow replies. */
4398 case OFPUTIL_OFPT_ECHO_REPLY:
4401 /* Nicira extension requests. */
4402 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4403 return handle_tun_id_from_cookie(ofconn, oh);
4405 case OFPUTIL_NXT_ROLE_REQUEST:
4406 return handle_role_request(ofconn, oh);
4408 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4409 return handle_nxt_set_flow_format(ofconn, oh);
4411 case OFPUTIL_NXT_FLOW_MOD:
4412 return handle_flow_mod(ofconn, oh);
4414 /* OpenFlow statistics requests. */
4415 case OFPUTIL_OFPST_DESC_REQUEST:
4416 return handle_desc_stats_request(ofconn, oh);
4418 case OFPUTIL_OFPST_FLOW_REQUEST:
4419 return handle_flow_stats_request(ofconn, oh);
4421 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4422 return handle_aggregate_stats_request(ofconn, oh);
4424 case OFPUTIL_OFPST_TABLE_REQUEST:
4425 return handle_table_stats_request(ofconn, oh);
4427 case OFPUTIL_OFPST_PORT_REQUEST:
4428 return handle_port_stats_request(ofconn, oh);
4430 case OFPUTIL_OFPST_QUEUE_REQUEST:
4431 return handle_queue_stats_request(ofconn, oh);
4433 /* Nicira extension statistics requests. */
4434 case OFPUTIL_NXST_FLOW_REQUEST:
4435 return handle_nxst_flow(ofconn, oh);
4437 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4438 return handle_nxst_aggregate(ofconn, oh);
4440 case OFPUTIL_INVALID:
4441 case OFPUTIL_OFPT_HELLO:
4442 case OFPUTIL_OFPT_ERROR:
4443 case OFPUTIL_OFPT_FEATURES_REPLY:
4444 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4445 case OFPUTIL_OFPT_PACKET_IN:
4446 case OFPUTIL_OFPT_FLOW_REMOVED:
4447 case OFPUTIL_OFPT_PORT_STATUS:
4448 case OFPUTIL_OFPT_BARRIER_REPLY:
4449 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4450 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4451 case OFPUTIL_OFPST_DESC_REPLY:
4452 case OFPUTIL_OFPST_FLOW_REPLY:
4453 case OFPUTIL_OFPST_QUEUE_REPLY:
4454 case OFPUTIL_OFPST_PORT_REPLY:
4455 case OFPUTIL_OFPST_TABLE_REPLY:
4456 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4457 case OFPUTIL_NXT_ROLE_REPLY:
4458 case OFPUTIL_NXT_FLOW_REMOVED:
4459 case OFPUTIL_NXST_FLOW_REPLY:
4460 case OFPUTIL_NXST_AGGREGATE_REPLY:
4462 if (VLOG_IS_WARN_ENABLED()) {
4463 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4464 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4467 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4468 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4470 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4476 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4478 int error = handle_openflow__(ofconn, ofp_msg);
4480 send_error_oh(ofconn, ofp_msg->data, error);
4482 COVERAGE_INC(ofproto_recv_openflow);
4486 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4488 struct facet *facet;
4491 /* Obtain in_port and tun_id, at least. */
4492 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4494 /* Set header pointers in 'flow'. */
4495 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4497 if (cfm_should_process_flow(&flow)) {
4498 ofproto_process_cfm(p, &flow, upcall->packet);
4499 ofpbuf_delete(upcall->packet);
4501 } else if (p->ofhooks->special_cb
4502 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4503 ofpbuf_delete(upcall->packet);
4507 /* Check with in-band control to see if this packet should be sent
4508 * to the local port regardless of the flow table. */
4509 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4510 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
4513 facet = facet_lookup_valid(p, &flow);
4515 struct rule *rule = rule_lookup(p, &flow);
4517 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4518 struct ofport *port = get_port(p, flow.in_port);
4520 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4521 COVERAGE_INC(ofproto_no_packet_in);
4522 /* XXX install 'drop' flow entry */
4523 ofpbuf_delete(upcall->packet);
4527 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4531 COVERAGE_INC(ofproto_packet_in);
4532 send_packet_in(p, upcall, &flow, false);
4536 facet = facet_create(p, rule, &flow, upcall->packet);
4537 } else if (!facet->may_install) {
4538 /* The facet is not installable, that is, we need to process every
4539 * packet, so process the current packet's actions into 'facet'. */
4540 facet_make_actions(p, facet, upcall->packet);
4543 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4545 * Extra-special case for fail-open mode.
4547 * We are in fail-open mode and the packet matched the fail-open rule,
4548 * but we are connected to a controller too. We should send the packet
4549 * up to the controller in the hope that it will try to set up a flow
4550 * and thereby allow us to exit fail-open.
4552 * See the top-level comment in fail-open.c for more information.
4554 send_packet_in(p, upcall, &flow, true);
4557 facet_execute(p, facet, upcall->packet);
4558 facet_install(p, facet, false);
4562 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4566 switch (upcall->type) {
4567 case DPIF_UC_ACTION:
4568 COVERAGE_INC(ofproto_ctlr_action);
4569 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4570 send_packet_in(p, upcall, &flow, false);
4573 case DPIF_UC_SAMPLE:
4575 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4576 ofproto_sflow_received(p->sflow, upcall, &flow);
4578 ofpbuf_delete(upcall->packet);
4582 handle_miss_upcall(p, upcall);
4585 case DPIF_N_UC_TYPES:
4587 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4592 /* Flow expiration. */
4594 static int ofproto_dp_max_idle(const struct ofproto *);
4595 static void ofproto_update_stats(struct ofproto *);
4596 static void rule_expire(struct ofproto *, struct rule *);
4597 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4599 /* This function is called periodically by ofproto_run(). Its job is to
4600 * collect updates for the flows that have been installed into the datapath,
4601 * most importantly when they last were used, and then use that information to
4602 * expire flows that have not been used recently.
4604 * Returns the number of milliseconds after which it should be called again. */
4606 ofproto_expire(struct ofproto *ofproto)
4608 struct rule *rule, *next_rule;
4609 struct cls_cursor cursor;
4612 /* Update stats for each flow in the datapath. */
4613 ofproto_update_stats(ofproto);
4615 /* Expire facets that have been idle too long. */
4616 dp_max_idle = ofproto_dp_max_idle(ofproto);
4617 ofproto_expire_facets(ofproto, dp_max_idle);
4619 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4620 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4621 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4622 rule_expire(ofproto, rule);
4625 /* Let the hook know that we're at a stable point: all outstanding data
4626 * in existing flows has been accounted to the account_cb. Thus, the
4627 * hook can now reasonably do operations that depend on having accurate
4628 * flow volume accounting (currently, that's just bond rebalancing). */
4629 if (ofproto->ofhooks->account_checkpoint_cb) {
4630 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4633 return MIN(dp_max_idle, 1000);
4636 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4638 * This function also pushes statistics updates to rules which each facet
4639 * resubmits into. Generally these statistics will be accurate. However, if a
4640 * facet changes the rule it resubmits into at some time in between
4641 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4642 * old rule will be incorrectly attributed to the new rule. This could be
4643 * avoided by calling ofproto_update_stats() whenever rules are created or
4644 * deleted. However, the performance impact of making so many calls to the
4645 * datapath do not justify the benefit of having perfectly accurate statistics.
4648 ofproto_update_stats(struct ofproto *p)
4650 const struct dpif_flow_stats *stats;
4651 struct dpif_flow_dump dump;
4652 const struct nlattr *key;
4655 dpif_flow_dump_start(&dump, p->dpif);
4656 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4657 struct facet *facet;
4660 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4664 odp_flow_key_format(key, key_len, &s);
4665 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4671 facet = facet_find(p, &flow);
4673 if (facet && facet->installed) {
4675 if (stats->n_packets >= facet->dp_packet_count) {
4676 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4678 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4681 if (stats->n_bytes >= facet->dp_byte_count) {
4682 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4684 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4687 facet->dp_packet_count = stats->n_packets;
4688 facet->dp_byte_count = stats->n_bytes;
4690 facet_update_time(p, facet, stats->used);
4691 facet_account(p, facet, stats->n_bytes);
4692 facet_push_stats(p, facet);
4694 /* There's a flow in the datapath that we know nothing about.
4696 COVERAGE_INC(ofproto_unexpected_rule);
4697 dpif_flow_del(p->dpif, key, key_len, NULL);
4700 dpif_flow_dump_done(&dump);
4703 /* Calculates and returns the number of milliseconds of idle time after which
4704 * facets should expire from the datapath and we should fold their statistics
4705 * into their parent rules in userspace. */
4707 ofproto_dp_max_idle(const struct ofproto *ofproto)
4710 * Idle time histogram.
4712 * Most of the time a switch has a relatively small number of facets. When
4713 * this is the case we might as well keep statistics for all of them in
4714 * userspace and to cache them in the kernel datapath for performance as
4717 * As the number of facets increases, the memory required to maintain
4718 * statistics about them in userspace and in the kernel becomes
4719 * significant. However, with a large number of facets it is likely that
4720 * only a few of them are "heavy hitters" that consume a large amount of
4721 * bandwidth. At this point, only heavy hitters are worth caching in the
4722 * kernel and maintaining in userspaces; other facets we can discard.
4724 * The technique used to compute the idle time is to build a histogram with
4725 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4726 * that is installed in the kernel gets dropped in the appropriate bucket.
4727 * After the histogram has been built, we compute the cutoff so that only
4728 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4729 * cached. At least the most-recently-used bucket of facets is kept, so
4730 * actually an arbitrary number of facets can be kept in any given
4731 * expiration run (though the next run will delete most of those unless
4732 * they receive additional data).
4734 * This requires a second pass through the facets, in addition to the pass
4735 * made by ofproto_update_stats(), because the former function never looks
4736 * at uninstallable facets.
4738 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4739 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4740 int buckets[N_BUCKETS] = { 0 };
4741 struct facet *facet;
4746 total = hmap_count(&ofproto->facets);
4747 if (total <= 1000) {
4748 return N_BUCKETS * BUCKET_WIDTH;
4751 /* Build histogram. */
4753 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4754 long long int idle = now - facet->used;
4755 int bucket = (idle <= 0 ? 0
4756 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4757 : (unsigned int) idle / BUCKET_WIDTH);
4761 /* Find the first bucket whose flows should be expired. */
4762 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4763 if (buckets[bucket]) {
4766 subtotal += buckets[bucket++];
4767 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4772 if (VLOG_IS_DBG_ENABLED()) {
4776 ds_put_cstr(&s, "keep");
4777 for (i = 0; i < N_BUCKETS; i++) {
4779 ds_put_cstr(&s, ", drop");
4782 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4785 VLOG_INFO("%s: %s (msec:count)",
4786 dpif_name(ofproto->dpif), ds_cstr(&s));
4790 return bucket * BUCKET_WIDTH;
4794 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4796 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4797 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4798 struct ofexpired expired;
4800 if (facet->installed) {
4801 struct dpif_flow_stats stats;
4803 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4805 facet_update_stats(ofproto, facet, &stats);
4808 expired.flow = facet->flow;
4809 expired.packet_count = facet->packet_count;
4810 expired.byte_count = facet->byte_count;
4811 expired.used = facet->used;
4812 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4817 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4819 long long int cutoff = time_msec() - dp_max_idle;
4820 struct facet *facet, *next_facet;
4822 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4823 facet_active_timeout(ofproto, facet);
4824 if (facet->used < cutoff) {
4825 facet_remove(ofproto, facet);
4830 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4831 * then delete it entirely. */
4833 rule_expire(struct ofproto *ofproto, struct rule *rule)
4835 struct facet *facet, *next_facet;
4839 /* Has 'rule' expired? */
4841 if (rule->hard_timeout
4842 && now > rule->created + rule->hard_timeout * 1000) {
4843 reason = OFPRR_HARD_TIMEOUT;
4844 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4845 && now >rule->used + rule->idle_timeout * 1000) {
4846 reason = OFPRR_IDLE_TIMEOUT;
4851 COVERAGE_INC(ofproto_expired);
4853 /* Update stats. (This is a no-op if the rule expired due to an idle
4854 * timeout, because that only happens when the rule has no facets left.) */
4855 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4856 facet_remove(ofproto, facet);
4859 /* Get rid of the rule. */
4860 if (!rule_is_hidden(rule)) {
4861 rule_send_removed(ofproto, rule, reason);
4863 rule_remove(ofproto, rule);
4867 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4869 struct ofputil_flow_removed fr;
4870 struct ofconn *ofconn;
4872 if (!rule->send_flow_removed) {
4877 fr.cookie = rule->flow_cookie;
4879 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4880 fr.idle_timeout = rule->idle_timeout;
4881 fr.packet_count = rule->packet_count;
4882 fr.byte_count = rule->byte_count;
4884 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4885 if (!rconn_is_connected(ofconn->rconn)
4886 || !ofconn_receives_async_msgs(ofconn)) {
4890 /* Account flow expirations under ofconn->reply_counter, the counter
4891 * for replies to OpenFlow requests. That works because preventing
4892 * OpenFlow requests from being processed also prevents new flows from
4893 * being added (and expiring). (It also prevents processing OpenFlow
4894 * requests that would not add new flows, so it is imperfect.) */
4895 queue_tx(ofputil_encode_flow_removed(&fr, ofconn->flow_format),
4896 ofconn, ofconn->reply_counter);
4900 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4901 * The returned statistics include statistics for all of 'rule''s facets. */
4903 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4906 struct facet *facet;
4908 /* Start from historical data for 'rule' itself that are no longer tracked
4909 * in facets. This counts, for example, facets that have expired. */
4910 p = rule->packet_count;
4911 b = rule->byte_count;
4913 /* Add any statistics that are tracked by facets. This includes
4914 * statistical data recently updated by ofproto_update_stats() as well as
4915 * stats for packets that were executed "by hand" via dpif_execute(). */
4916 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4917 p += facet->packet_count;
4918 b += facet->byte_count;
4925 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4927 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4929 struct ofconn *ofconn = ofconn_;
4931 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4932 ofconn->packet_in_counter, 100);
4935 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4936 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4937 * scheduler for sending.
4939 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4940 * Otherwise, ownership is transferred to this function. */
4942 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4943 const struct flow *flow, bool clone)
4945 struct ofproto *ofproto = ofconn->ofproto;
4946 struct ofputil_packet_in pin;
4949 /* Figure out the easy parts. */
4950 pin.packet = upcall->packet;
4951 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4952 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4954 /* Get OpenFlow buffer_id. */
4955 if (upcall->type == DPIF_UC_ACTION) {
4956 pin.buffer_id = UINT32_MAX;
4957 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4958 pin.buffer_id = pktbuf_get_null();
4959 } else if (!ofconn->pktbuf) {
4960 pin.buffer_id = UINT32_MAX;
4962 pin.buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet,
4966 /* Figure out how much of the packet to send. */
4967 pin.send_len = upcall->packet->size;
4968 if (pin.buffer_id != UINT32_MAX) {
4969 pin.send_len = MIN(pin.send_len, ofconn->miss_send_len);
4971 if (upcall->type == DPIF_UC_ACTION) {
4972 pin.send_len = MIN(pin.send_len, upcall->userdata);
4975 /* Make OFPT_PACKET_IN and hand over to packet scheduler. It might
4976 * immediately call into do_send_packet_in() or it might buffer it for a
4977 * while (until a later call to pinsched_run()). */
4978 msg = ofputil_encode_packet_in(&pin, clone ? NULL : upcall->packet);
4979 pinsched_send(ofconn->schedulers[upcall->type == DPIF_UC_MISS ? 0 : 1],
4980 flow->in_port, msg, do_send_packet_in, ofconn);
4983 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4984 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4985 * their individual configurations.
4987 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4988 * Otherwise, ownership is transferred to this function. */
4990 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4991 const struct flow *flow, bool clone)
4993 struct ofconn *ofconn, *prev;
4996 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4997 if (ofconn_receives_async_msgs(ofconn)) {
4999 schedule_packet_in(prev, upcall, flow, true);
5005 schedule_packet_in(prev, upcall, flow, clone);
5006 } else if (!clone) {
5007 ofpbuf_delete(upcall->packet);
5012 pick_datapath_id(const struct ofproto *ofproto)
5014 const struct ofport *port;
5016 port = get_port(ofproto, ODPP_LOCAL);
5018 uint8_t ea[ETH_ADDR_LEN];
5021 error = netdev_get_etheraddr(port->netdev, ea);
5023 return eth_addr_to_uint64(ea);
5025 VLOG_WARN("could not get MAC address for %s (%s)",
5026 netdev_get_name(port->netdev), strerror(error));
5028 return ofproto->fallback_dpid;
5032 pick_fallback_dpid(void)
5034 uint8_t ea[ETH_ADDR_LEN];
5035 eth_addr_nicira_random(ea);
5036 return eth_addr_to_uint64(ea);
5040 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
5041 void *aux OVS_UNUSED)
5043 const struct shash_node *node;
5047 SHASH_FOR_EACH (node, &all_ofprotos) {
5048 ds_put_format(&results, "%s\n", node->name);
5050 unixctl_command_reply(conn, 200, ds_cstr(&results));
5051 ds_destroy(&results);
5054 struct ofproto_trace {
5055 struct action_xlate_ctx ctx;
5061 trace_format_rule(struct ds *result, int level, const struct rule *rule)
5063 ds_put_char_multiple(result, '\t', level);
5065 ds_put_cstr(result, "No match\n");
5069 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
5070 ntohll(rule->flow_cookie));
5071 cls_rule_format(&rule->cr, result);
5072 ds_put_char(result, '\n');
5074 ds_put_char_multiple(result, '\t', level);
5075 ds_put_cstr(result, "OpenFlow ");
5076 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
5077 rule->n_actions * sizeof *rule->actions);
5078 ds_put_char(result, '\n');
5082 trace_format_flow(struct ds *result, int level, const char *title,
5083 struct ofproto_trace *trace)
5085 ds_put_char_multiple(result, '\t', level);
5086 ds_put_format(result, "%s: ", title);
5087 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5088 ds_put_cstr(result, "unchanged");
5090 flow_format(result, &trace->ctx.flow);
5091 trace->flow = trace->ctx.flow;
5093 ds_put_char(result, '\n');
5097 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
5099 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5100 struct ds *result = trace->result;
5102 ds_put_char(result, '\n');
5103 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5104 trace_format_rule(result, ctx->recurse + 1, rule);
5108 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5109 void *aux OVS_UNUSED)
5111 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5112 char *args = xstrdup(args_);
5113 char *save_ptr = NULL;
5114 struct ofproto *ofproto;
5115 struct ofpbuf packet;
5123 ofpbuf_init(&packet, strlen(args) / 2);
5126 dpname = strtok_r(args, " ", &save_ptr);
5127 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5128 in_port_s = strtok_r(NULL, " ", &save_ptr);
5129 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5130 if (!dpname || !in_port_s || !packet_s) {
5131 unixctl_command_reply(conn, 501, "Bad command syntax");
5135 ofproto = shash_find_data(&all_ofprotos, dpname);
5137 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5142 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5143 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5145 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5146 packet_s += strspn(packet_s, " ");
5147 if (*packet_s != '\0') {
5148 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5151 if (packet.size < ETH_HEADER_LEN) {
5152 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5156 ds_put_cstr(&result, "Packet: ");
5157 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5158 ds_put_cstr(&result, s);
5161 flow_extract(&packet, tun_id, in_port, &flow);
5162 ds_put_cstr(&result, "Flow: ");
5163 flow_format(&result, &flow);
5164 ds_put_char(&result, '\n');
5166 rule = rule_lookup(ofproto, &flow);
5167 trace_format_rule(&result, 0, rule);
5169 struct ofproto_trace trace;
5170 struct ofpbuf *odp_actions;
5172 trace.result = &result;
5174 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5175 trace.ctx.resubmit_hook = trace_resubmit;
5176 odp_actions = xlate_actions(&trace.ctx,
5177 rule->actions, rule->n_actions);
5179 ds_put_char(&result, '\n');
5180 trace_format_flow(&result, 0, "Final flow", &trace);
5181 ds_put_cstr(&result, "Datapath actions: ");
5182 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5183 ofpbuf_delete(odp_actions);
5186 unixctl_command_reply(conn, 200, ds_cstr(&result));
5189 ds_destroy(&result);
5190 ofpbuf_uninit(&packet);
5195 ofproto_unixctl_init(void)
5197 static bool registered;
5203 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5204 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5208 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5209 struct ofpbuf *odp_actions, tag_type *tags,
5210 uint16_t *nf_output_iface, void *ofproto_)
5212 struct ofproto *ofproto = ofproto_;
5213 struct mac_entry *dst_mac;
5215 /* Drop frames for reserved multicast addresses. */
5216 if (eth_addr_is_reserved(flow->dl_dst)) {
5220 /* Learn source MAC (but don't try to learn from revalidation). */
5222 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
5223 struct mac_entry *src_mac;
5225 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
5226 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
5227 /* The log messages here could actually be useful in debugging,
5228 * so keep the rate limit relatively high. */
5229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5230 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5231 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5233 ofproto_revalidate(ofproto,
5234 mac_learning_changed(ofproto->ml, src_mac));
5235 src_mac->port.i = flow->in_port;
5239 /* Determine output port. */
5240 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
5242 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5243 nf_output_iface, odp_actions);
5245 int out_port = dst_mac->port.i;
5246 if (out_port != flow->in_port) {
5247 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5248 *nf_output_iface = out_port;
5257 static const struct ofhooks default_ofhooks = {
5258 default_normal_ofhook_cb,