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 *);
347 static bool ofconn_receives_async_msgs(const struct ofconn *);
348 static char *ofconn_make_name(const struct ofproto *, const char *target);
349 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
351 static struct ofproto *ofconn_get_ofproto(struct ofconn *);
353 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
354 struct rconn_packet_counter *counter);
356 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
357 const struct flow *, bool clone);
358 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
362 uint64_t datapath_id; /* Datapath ID. */
363 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
364 char *mfr_desc; /* Manufacturer. */
365 char *hw_desc; /* Hardware. */
366 char *sw_desc; /* Software version. */
367 char *serial_desc; /* Serial number. */
368 char *dp_desc; /* Datapath description. */
372 struct netdev_monitor *netdev_monitor;
373 struct hmap ports; /* Contains "struct ofport"s. */
374 struct shash port_by_name;
378 struct fail_open *fail_open;
379 struct netflow *netflow;
380 struct ofproto_sflow *sflow;
382 /* In-band control. */
383 struct in_band *in_band;
384 long long int next_in_band_update;
385 struct sockaddr_in *extra_in_band_remotes;
386 size_t n_extra_remotes;
390 struct classifier cls;
391 long long int next_expiration;
395 bool need_revalidate;
396 struct tag_set revalidate_set;
398 /* OpenFlow connections. */
399 struct hmap controllers; /* Controller "struct ofconn"s. */
400 struct list all_conns; /* Contains "struct ofconn"s. */
401 enum ofproto_fail_mode fail_mode;
403 /* OpenFlow listeners. */
404 struct hmap services; /* Contains "struct ofservice"s. */
405 struct pvconn **snoops;
408 /* Hooks for ovs-vswitchd. */
409 const struct ofhooks *ofhooks;
412 /* Used by default ofhooks. */
413 struct mac_learning *ml;
416 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
417 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
419 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
421 static const struct ofhooks default_ofhooks;
423 static uint64_t pick_datapath_id(const struct ofproto *);
424 static uint64_t pick_fallback_dpid(void);
426 static int ofproto_expire(struct ofproto *);
427 static void flow_push_stats(struct ofproto *, const struct rule *,
428 struct flow *, uint64_t packets, uint64_t bytes,
431 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
433 static void handle_openflow(struct ofconn *, struct ofpbuf *);
435 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
436 static void update_port(struct ofproto *, const char *devname);
437 static int init_ports(struct ofproto *);
438 static void reinit_ports(struct ofproto *);
440 static void ofproto_unixctl_init(void);
443 ofproto_create(const char *datapath, const char *datapath_type,
444 const struct ofhooks *ofhooks, void *aux,
445 struct ofproto **ofprotop)
453 ofproto_unixctl_init();
455 /* Connect to datapath and start listening for messages. */
456 error = dpif_open(datapath, datapath_type, &dpif);
458 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
461 error = dpif_recv_set_mask(dpif,
462 ((1u << DPIF_UC_MISS) |
463 (1u << DPIF_UC_ACTION) |
464 (1u << DPIF_UC_SAMPLE)));
466 VLOG_ERR("failed to listen on datapath %s: %s",
467 datapath, strerror(error));
471 dpif_flow_flush(dpif);
472 dpif_recv_purge(dpif);
474 /* Initialize settings. */
475 p = xzalloc(sizeof *p);
476 p->fallback_dpid = pick_fallback_dpid();
477 p->datapath_id = p->fallback_dpid;
478 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
479 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
480 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
481 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
482 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
484 /* Initialize datapath. */
486 p->netdev_monitor = netdev_monitor_create();
487 hmap_init(&p->ports);
488 shash_init(&p->port_by_name);
489 p->max_ports = dpif_get_max_ports(dpif);
491 /* Initialize submodules. */
496 /* Initialize in-band control. */
498 p->in_band_queue = -1;
500 /* Initialize flow table. */
501 classifier_init(&p->cls);
502 p->next_expiration = time_msec() + 1000;
504 /* Initialize facet table. */
505 hmap_init(&p->facets);
506 p->need_revalidate = false;
507 tag_set_init(&p->revalidate_set);
509 /* Initialize OpenFlow connections. */
510 list_init(&p->all_conns);
511 hmap_init(&p->controllers);
512 hmap_init(&p->services);
516 /* Initialize hooks. */
518 p->ofhooks = ofhooks;
522 p->ofhooks = &default_ofhooks;
524 p->ml = mac_learning_create();
527 /* Pick final datapath ID. */
528 p->datapath_id = pick_datapath_id(p);
529 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
531 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
538 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
540 uint64_t old_dpid = p->datapath_id;
541 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
542 if (p->datapath_id != old_dpid) {
543 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
545 /* Force all active connections to reconnect, since there is no way to
546 * notify a controller that the datapath ID has changed. */
547 ofproto_reconnect_controllers(p);
551 /* Creates a new controller in 'ofproto'. Some of the settings are initially
552 * drawn from 'c', but update_controller() needs to be called later to finish
553 * the new ofconn's configuration. */
555 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
557 char *name = ofconn_make_name(ofproto, c->target);
558 struct ofconn *ofconn;
560 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
561 ofconn->pktbuf = pktbuf_create();
562 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
563 rconn_connect(ofconn->rconn, c->target, name);
564 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
565 hash_string(c->target, 0));
570 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
571 * target (this is done by creating new ofconns and deleting old ones), but it
572 * can update the rest of an ofconn's settings. */
574 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
578 ofconn->band = c->band;
580 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
582 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
583 rconn_set_probe_interval(ofconn->rconn, probe_interval);
585 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
589 ofconn_get_target(const struct ofconn *ofconn)
591 return rconn_get_target(ofconn->rconn);
594 static struct ofconn *
595 find_controller_by_target(struct ofproto *ofproto, const char *target)
597 struct ofconn *ofconn;
599 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
600 hash_string(target, 0), &ofproto->controllers) {
601 if (!strcmp(ofconn_get_target(ofconn), target)) {
609 update_in_band_remotes(struct ofproto *ofproto)
611 const struct ofconn *ofconn;
612 struct sockaddr_in *addrs;
613 size_t max_addrs, n_addrs;
616 /* Allocate enough memory for as many remotes as we could possibly have. */
617 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
618 addrs = xmalloc(max_addrs * sizeof *addrs);
621 /* Add all the remotes. */
622 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
623 struct sockaddr_in *sin = &addrs[n_addrs];
625 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
629 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
630 if (sin->sin_addr.s_addr) {
631 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
635 for (i = 0; i < ofproto->n_extra_remotes; i++) {
636 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
639 /* Create or update or destroy in-band. */
641 if (!ofproto->in_band) {
642 in_band_create(ofproto, ofproto->dpif, &ofproto->in_band);
644 if (ofproto->in_band) {
645 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
647 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
648 ofproto->next_in_band_update = time_msec() + 1000;
650 in_band_destroy(ofproto->in_band);
651 ofproto->in_band = NULL;
659 update_fail_open(struct ofproto *p)
661 struct ofconn *ofconn;
663 if (!hmap_is_empty(&p->controllers)
664 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
665 struct rconn **rconns;
669 p->fail_open = fail_open_create(p);
673 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
674 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
675 rconns[n++] = ofconn->rconn;
678 fail_open_set_controllers(p->fail_open, rconns, n);
679 /* p->fail_open takes ownership of 'rconns'. */
681 fail_open_destroy(p->fail_open);
687 ofproto_set_controllers(struct ofproto *p,
688 const struct ofproto_controller *controllers,
689 size_t n_controllers)
691 struct shash new_controllers;
692 struct ofconn *ofconn, *next_ofconn;
693 struct ofservice *ofservice, *next_ofservice;
696 /* Create newly configured controllers and services.
697 * Create a name to ofproto_controller mapping in 'new_controllers'. */
698 shash_init(&new_controllers);
699 for (i = 0; i < n_controllers; i++) {
700 const struct ofproto_controller *c = &controllers[i];
702 if (!vconn_verify_name(c->target)) {
703 if (!find_controller_by_target(p, c->target)) {
704 add_controller(p, c);
706 } else if (!pvconn_verify_name(c->target)) {
707 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
711 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
712 dpif_name(p->dpif), c->target);
716 shash_add_once(&new_controllers, c->target, &controllers[i]);
719 /* Delete controllers that are no longer configured.
720 * Update configuration of all now-existing controllers. */
721 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
722 struct ofproto_controller *c;
724 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
726 ofconn_destroy(ofconn);
728 update_controller(ofconn, c);
732 /* Delete services that are no longer configured.
733 * Update configuration of all now-existing services. */
734 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
735 struct ofproto_controller *c;
737 c = shash_find_data(&new_controllers,
738 pvconn_get_name(ofservice->pvconn));
740 ofservice_destroy(p, ofservice);
742 ofservice_reconfigure(ofservice, c);
746 shash_destroy(&new_controllers);
748 update_in_band_remotes(p);
753 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
755 p->fail_mode = fail_mode;
759 /* Drops the connections between 'ofproto' and all of its controllers, forcing
760 * them to reconnect. */
762 ofproto_reconnect_controllers(struct ofproto *ofproto)
764 struct ofconn *ofconn;
766 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
767 rconn_reconnect(ofconn->rconn);
772 any_extras_changed(const struct ofproto *ofproto,
773 const struct sockaddr_in *extras, size_t n)
777 if (n != ofproto->n_extra_remotes) {
781 for (i = 0; i < n; i++) {
782 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
783 const struct sockaddr_in *new = &extras[i];
785 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
786 old->sin_port != new->sin_port) {
794 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
795 * in-band control should guarantee access, in the same way that in-band
796 * control guarantees access to OpenFlow controllers. */
798 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
799 const struct sockaddr_in *extras, size_t n)
801 if (!any_extras_changed(ofproto, extras, n)) {
805 free(ofproto->extra_in_band_remotes);
806 ofproto->n_extra_remotes = n;
807 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
809 update_in_band_remotes(ofproto);
812 /* Sets the OpenFlow queue used by flows set up by in-band control on
813 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
814 * flows will use the default queue. */
816 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
818 if (queue_id != ofproto->in_band_queue) {
819 ofproto->in_band_queue = queue_id;
820 update_in_band_remotes(ofproto);
825 ofproto_set_desc(struct ofproto *p,
826 const char *mfr_desc, const char *hw_desc,
827 const char *sw_desc, const char *serial_desc,
830 struct ofp_desc_stats *ods;
833 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
834 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
835 sizeof ods->mfr_desc);
838 p->mfr_desc = xstrdup(mfr_desc);
841 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
842 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
843 sizeof ods->hw_desc);
846 p->hw_desc = xstrdup(hw_desc);
849 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
850 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
851 sizeof ods->sw_desc);
854 p->sw_desc = xstrdup(sw_desc);
857 if (strlen(serial_desc) >= sizeof ods->serial_num) {
858 VLOG_WARN("truncating serial_desc, must be less than %zu "
860 sizeof ods->serial_num);
862 free(p->serial_desc);
863 p->serial_desc = xstrdup(serial_desc);
866 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
867 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
868 sizeof ods->dp_desc);
871 p->dp_desc = xstrdup(dp_desc);
876 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
877 const struct svec *svec)
879 struct pvconn **pvconns = *pvconnsp;
880 size_t n_pvconns = *n_pvconnsp;
884 for (i = 0; i < n_pvconns; i++) {
885 pvconn_close(pvconns[i]);
889 pvconns = xmalloc(svec->n * sizeof *pvconns);
891 for (i = 0; i < svec->n; i++) {
892 const char *name = svec->names[i];
893 struct pvconn *pvconn;
896 error = pvconn_open(name, &pvconn);
898 pvconns[n_pvconns++] = pvconn;
900 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
908 *n_pvconnsp = n_pvconns;
914 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
916 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
920 ofproto_set_netflow(struct ofproto *ofproto,
921 const struct netflow_options *nf_options)
923 if (nf_options && nf_options->collectors.n) {
924 if (!ofproto->netflow) {
925 ofproto->netflow = netflow_create();
927 return netflow_set_options(ofproto->netflow, nf_options);
929 netflow_destroy(ofproto->netflow);
930 ofproto->netflow = NULL;
936 ofproto_set_sflow(struct ofproto *ofproto,
937 const struct ofproto_sflow_options *oso)
939 struct ofproto_sflow *os = ofproto->sflow;
942 struct ofport *ofport;
944 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
945 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
946 ofproto_sflow_add_port(os, ofport->odp_port,
947 netdev_get_name(ofport->netdev));
950 ofproto_sflow_set_options(os, oso);
952 ofproto_sflow_destroy(os);
953 ofproto->sflow = NULL;
957 /* Connectivity Fault Management configuration. */
959 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
961 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
963 struct ofport *ofport = get_port(ofproto, port_no);
964 if (ofport && ofport->cfm){
965 cfm_destroy(ofport->cfm);
970 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
971 * basic configuration from the configuration members in 'cfm', and the set of
972 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
973 * Ignores the statistics members of 'cfm'.
975 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
977 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
978 const struct cfm *cfm,
979 const uint16_t *remote_mps, size_t n_remote_mps)
981 struct ofport *ofport;
983 ofport = get_port(ofproto, port_no);
985 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
986 dpif_name(ofproto->dpif), port_no);
991 ofport->cfm = cfm_create();
994 ofport->cfm->mpid = cfm->mpid;
995 ofport->cfm->interval = cfm->interval;
996 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
998 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
1000 if (!cfm_configure(ofport->cfm)) {
1001 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
1002 dpif_name(ofproto->dpif), port_no,
1003 netdev_get_name(ofport->netdev));
1004 cfm_destroy(ofport->cfm);
1009 /* Returns the connectivity fault management object associated with 'port_no'
1010 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
1011 * 'port_no' or if that port does not have CFM configured. The caller must not
1012 * modify or destroy the returned object. */
1014 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
1016 struct ofport *ofport = get_port(ofproto, port_no);
1017 return ofport ? ofport->cfm : NULL;
1021 ofproto_get_datapath_id(const struct ofproto *ofproto)
1023 return ofproto->datapath_id;
1027 ofproto_has_primary_controller(const struct ofproto *ofproto)
1029 return !hmap_is_empty(&ofproto->controllers);
1032 enum ofproto_fail_mode
1033 ofproto_get_fail_mode(const struct ofproto *p)
1035 return p->fail_mode;
1039 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1043 for (i = 0; i < ofproto->n_snoops; i++) {
1044 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1049 ofproto_destroy(struct ofproto *p)
1051 struct ofservice *ofservice, *next_ofservice;
1052 struct ofconn *ofconn, *next_ofconn;
1053 struct ofport *ofport, *next_ofport;
1060 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1062 /* Destroy fail-open and in-band early, since they touch the classifier. */
1063 fail_open_destroy(p->fail_open);
1064 p->fail_open = NULL;
1066 in_band_destroy(p->in_band);
1068 free(p->extra_in_band_remotes);
1070 ofproto_flush_flows(p);
1071 classifier_destroy(&p->cls);
1072 hmap_destroy(&p->facets);
1074 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1075 ofconn_destroy(ofconn);
1077 hmap_destroy(&p->controllers);
1079 dpif_close(p->dpif);
1080 netdev_monitor_destroy(p->netdev_monitor);
1081 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1082 hmap_remove(&p->ports, &ofport->hmap_node);
1083 ofport_free(ofport);
1085 shash_destroy(&p->port_by_name);
1087 netflow_destroy(p->netflow);
1088 ofproto_sflow_destroy(p->sflow);
1090 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1091 ofservice_destroy(p, ofservice);
1093 hmap_destroy(&p->services);
1095 for (i = 0; i < p->n_snoops; i++) {
1096 pvconn_close(p->snoops[i]);
1100 mac_learning_destroy(p->ml);
1105 free(p->serial_desc);
1108 hmap_destroy(&p->ports);
1114 ofproto_run(struct ofproto *p)
1116 int error = ofproto_run1(p);
1118 error = ofproto_run2(p, false);
1124 process_port_change(struct ofproto *ofproto, int error, char *devname)
1126 if (error == ENOBUFS) {
1127 reinit_ports(ofproto);
1128 } else if (!error) {
1129 update_port(ofproto, devname);
1134 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1135 * means that 'ofconn' is more interesting for monitoring than a lower return
1138 snoop_preference(const struct ofconn *ofconn)
1140 switch (ofconn->role) {
1141 case NX_ROLE_MASTER:
1148 /* Shouldn't happen. */
1153 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1154 * Connects this vconn to a controller. */
1156 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1158 struct ofconn *ofconn, *best;
1160 /* Pick a controller for monitoring. */
1162 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1163 if (ofconn->type == OFCONN_PRIMARY
1164 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1170 rconn_add_monitor(best->rconn, vconn);
1172 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1178 ofproto_run1(struct ofproto *p)
1180 struct ofconn *ofconn, *next_ofconn;
1181 struct ofservice *ofservice;
1182 struct ofport *ofport;
1187 if (shash_is_empty(&p->port_by_name)) {
1191 for (i = 0; i < 50; i++) {
1192 struct dpif_upcall packet;
1194 error = dpif_recv(p->dpif, &packet);
1196 if (error == ENODEV) {
1197 /* Someone destroyed the datapath behind our back. The caller
1198 * better destroy us and give up, because we're just going to
1199 * spin from here on out. */
1200 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1201 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1202 dpif_name(p->dpif));
1208 handle_upcall(p, &packet);
1211 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1212 process_port_change(p, error, devname);
1214 while ((error = netdev_monitor_poll(p->netdev_monitor,
1215 &devname)) != EAGAIN) {
1216 process_port_change(p, error, devname);
1219 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1220 ofport_run(p, ofport);
1224 if (time_msec() >= p->next_in_band_update) {
1225 update_in_band_remotes(p);
1227 in_band_run(p->in_band);
1230 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1234 /* Fail-open maintenance. Do this after processing the ofconns since
1235 * fail-open checks the status of the controller rconn. */
1237 fail_open_run(p->fail_open);
1240 HMAP_FOR_EACH (ofservice, node, &p->services) {
1241 struct vconn *vconn;
1244 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1246 struct rconn *rconn;
1249 rconn = rconn_create(ofservice->probe_interval, 0);
1250 name = ofconn_make_name(p, vconn_get_name(vconn));
1251 rconn_connect_unreliably(rconn, vconn, name);
1254 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1255 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1256 ofservice->burst_limit);
1257 } else if (retval != EAGAIN) {
1258 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1262 for (i = 0; i < p->n_snoops; i++) {
1263 struct vconn *vconn;
1266 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1268 add_snooper(p, vconn);
1269 } else if (retval != EAGAIN) {
1270 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1274 if (time_msec() >= p->next_expiration) {
1275 int delay = ofproto_expire(p);
1276 p->next_expiration = time_msec() + delay;
1277 COVERAGE_INC(ofproto_expiration);
1281 netflow_run(p->netflow);
1284 ofproto_sflow_run(p->sflow);
1291 ofproto_run2(struct ofproto *p, bool revalidate_all)
1293 /* Figure out what we need to revalidate now, if anything. */
1294 struct tag_set revalidate_set = p->revalidate_set;
1295 if (p->need_revalidate) {
1296 revalidate_all = true;
1299 /* Clear the revalidation flags. */
1300 tag_set_init(&p->revalidate_set);
1301 p->need_revalidate = false;
1303 /* Now revalidate if there's anything to do. */
1304 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1305 struct facet *facet, *next;
1307 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1309 || tag_set_intersects(&revalidate_set, facet->tags)) {
1310 facet_revalidate(p, facet);
1319 ofproto_wait(struct ofproto *p)
1321 struct ofservice *ofservice;
1322 struct ofconn *ofconn;
1323 struct ofport *ofport;
1326 dpif_recv_wait(p->dpif);
1327 dpif_port_poll_wait(p->dpif);
1328 netdev_monitor_poll_wait(p->netdev_monitor);
1329 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1330 ofport_wait(ofport);
1332 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1333 ofconn_wait(ofconn);
1336 poll_timer_wait_until(p->next_in_band_update);
1337 in_band_wait(p->in_band);
1340 fail_open_wait(p->fail_open);
1343 ofproto_sflow_wait(p->sflow);
1345 if (!tag_set_is_empty(&p->revalidate_set)) {
1346 poll_immediate_wake();
1348 if (p->need_revalidate) {
1349 /* Shouldn't happen, but if it does just go around again. */
1350 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1351 poll_immediate_wake();
1352 } else if (p->next_expiration != LLONG_MAX) {
1353 poll_timer_wait_until(p->next_expiration);
1355 HMAP_FOR_EACH (ofservice, node, &p->services) {
1356 pvconn_wait(ofservice->pvconn);
1358 for (i = 0; i < p->n_snoops; i++) {
1359 pvconn_wait(p->snoops[i]);
1364 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1366 tag_set_add(&ofproto->revalidate_set, tag);
1370 ofproto_get_revalidate_set(struct ofproto *ofproto)
1372 return &ofproto->revalidate_set;
1376 ofproto_is_alive(const struct ofproto *p)
1378 return !hmap_is_empty(&p->controllers);
1382 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1385 const struct ofconn *ofconn;
1389 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1390 const struct rconn *rconn = ofconn->rconn;
1391 time_t now = time_now();
1392 time_t last_connection = rconn_get_last_connection(rconn);
1393 time_t last_disconnect = rconn_get_last_disconnect(rconn);
1394 const int last_error = rconn_get_last_error(rconn);
1395 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1397 shash_add(info, rconn_get_target(rconn), cinfo);
1399 cinfo->is_connected = rconn_is_connected(rconn);
1400 cinfo->role = ofconn->role;
1405 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1406 cinfo->pairs.values[cinfo->pairs.n++] =
1407 xstrdup(ovs_retval_to_string(last_error));
1410 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1411 cinfo->pairs.values[cinfo->pairs.n++] =
1412 xstrdup(rconn_get_state(rconn));
1414 if (last_connection != TIME_MIN) {
1415 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
1416 cinfo->pairs.values[cinfo->pairs.n++]
1417 = xasprintf("%ld", (long int) (now - last_connection));
1420 if (last_disconnect != TIME_MIN) {
1421 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
1422 cinfo->pairs.values[cinfo->pairs.n++]
1423 = xasprintf("%ld", (long int) (now - last_disconnect));
1429 ofproto_free_ofproto_controller_info(struct shash *info)
1431 struct shash_node *node;
1433 SHASH_FOR_EACH (node, info) {
1434 struct ofproto_controller_info *cinfo = node->data;
1435 while (cinfo->pairs.n) {
1436 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1440 shash_destroy(info);
1443 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1445 * This is almost the same as calling dpif_port_del() directly on the
1446 * datapath, but it also makes 'ofproto' close its open netdev for the port
1447 * (if any). This makes it possible to create a new netdev of a different
1448 * type under the same name, which otherwise the netdev library would refuse
1449 * to do because of the conflict. (The netdev would eventually get closed on
1450 * the next trip through ofproto_run(), but this interface is more direct.)
1452 * Returns 0 if successful, otherwise a positive errno. */
1454 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1456 struct ofport *ofport = get_port(ofproto, odp_port);
1457 const char *name = ofport ? ofport->opp.name : "<unknown>";
1460 error = dpif_port_del(ofproto->dpif, odp_port);
1462 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1463 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1464 } else if (ofport) {
1465 /* 'name' is ofport->opp.name and update_port() is going to destroy
1466 * 'ofport'. Just in case update_port() refers to 'name' after it
1467 * destroys 'ofport', make a copy of it around the update_port()
1469 char *devname = xstrdup(name);
1470 update_port(ofproto, devname);
1476 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1477 * true if 'odp_port' exists and should be included, false otherwise. */
1479 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1481 struct ofport *ofport = get_port(ofproto, odp_port);
1482 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1485 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1486 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1487 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1489 * Returns 0 if successful, otherwise a positive errno value. */
1491 ofproto_send_packet(struct ofproto *ofproto,
1492 uint32_t port_no, uint16_t vlan_tci,
1493 const struct ofpbuf *packet)
1495 struct ofpbuf odp_actions;
1498 ofpbuf_init(&odp_actions, 32);
1499 if (vlan_tci != 0) {
1500 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1501 ntohs(vlan_tci & ~VLAN_CFI));
1503 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1504 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1506 ofpbuf_uninit(&odp_actions);
1509 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1510 dpif_name(ofproto->dpif), port_no, strerror(error));
1515 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1516 * performs the 'n_actions' actions in 'actions'. The new flow will not
1519 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1520 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1521 * controllers; otherwise, it will be hidden.
1523 * The caller retains ownership of 'cls_rule' and 'actions'. */
1525 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1526 const union ofp_action *actions, size_t n_actions)
1529 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1530 rule_insert(p, rule);
1534 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1538 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1541 rule_remove(ofproto, rule);
1546 ofproto_flush_flows(struct ofproto *ofproto)
1548 struct facet *facet, *next_facet;
1549 struct rule *rule, *next_rule;
1550 struct cls_cursor cursor;
1552 COVERAGE_INC(ofproto_flush);
1554 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1555 /* Mark the facet as not installed so that facet_remove() doesn't
1556 * bother trying to uninstall it. There is no point in uninstalling it
1557 * individually since we are about to blow away all the facets with
1558 * dpif_flow_flush(). */
1559 facet->installed = false;
1560 facet->dp_packet_count = 0;
1561 facet->dp_byte_count = 0;
1562 facet_remove(ofproto, facet);
1565 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1566 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1567 rule_remove(ofproto, rule);
1570 dpif_flow_flush(ofproto->dpif);
1571 if (ofproto->in_band) {
1572 in_band_flushed(ofproto->in_band);
1574 if (ofproto->fail_open) {
1575 fail_open_flushed(ofproto->fail_open);
1580 reinit_ports(struct ofproto *p)
1582 struct dpif_port_dump dump;
1583 struct shash_node *node;
1584 struct shash devnames;
1585 struct ofport *ofport;
1586 struct dpif_port dpif_port;
1588 COVERAGE_INC(ofproto_reinit_ports);
1590 shash_init(&devnames);
1591 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1592 shash_add_once (&devnames, ofport->opp.name, NULL);
1594 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1595 shash_add_once (&devnames, dpif_port.name, NULL);
1598 SHASH_FOR_EACH (node, &devnames) {
1599 update_port(p, node->name);
1601 shash_destroy(&devnames);
1604 static struct ofport *
1605 make_ofport(const struct dpif_port *dpif_port)
1607 struct netdev_options netdev_options;
1608 enum netdev_flags flags;
1609 struct ofport *ofport;
1610 struct netdev *netdev;
1613 memset(&netdev_options, 0, sizeof netdev_options);
1614 netdev_options.name = dpif_port->name;
1615 netdev_options.type = dpif_port->type;
1616 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1618 error = netdev_open(&netdev_options, &netdev);
1620 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1621 "cannot be opened (%s)",
1622 dpif_port->name, dpif_port->port_no,
1623 dpif_port->name, strerror(error));
1627 ofport = xzalloc(sizeof *ofport);
1628 ofport->netdev = netdev;
1629 ofport->odp_port = dpif_port->port_no;
1630 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1631 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1632 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1634 netdev_get_flags(netdev, &flags);
1635 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1637 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1639 netdev_get_features(netdev,
1640 &ofport->opp.curr, &ofport->opp.advertised,
1641 &ofport->opp.supported, &ofport->opp.peer);
1646 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1648 if (get_port(p, dpif_port->port_no)) {
1649 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1650 dpif_port->port_no);
1652 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1653 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1662 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1664 const struct ofp_phy_port *a = &a_->opp;
1665 const struct ofp_phy_port *b = &b_->opp;
1667 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1668 return (a->port_no == b->port_no
1669 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1670 && !strcmp(a->name, b->name)
1671 && a->state == b->state
1672 && a->config == b->config
1673 && a->curr == b->curr
1674 && a->advertised == b->advertised
1675 && a->supported == b->supported
1676 && a->peer == b->peer);
1680 send_port_status(struct ofproto *p, const struct ofport *ofport,
1683 /* XXX Should limit the number of queued port status change messages. */
1684 struct ofconn *ofconn;
1685 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1686 struct ofp_port_status *ops;
1689 /* Primary controllers, even slaves, should always get port status
1690 updates. Otherwise obey ofconn_receives_async_msgs(). */
1691 if (ofconn->type != OFCONN_PRIMARY
1692 && !ofconn_receives_async_msgs(ofconn)) {
1696 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1697 ops->reason = reason;
1698 ops->desc = ofport->opp;
1699 hton_ofp_phy_port(&ops->desc);
1700 queue_tx(b, ofconn, NULL);
1705 ofport_install(struct ofproto *p, struct ofport *ofport)
1707 const char *netdev_name = ofport->opp.name;
1709 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1710 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1711 shash_add(&p->port_by_name, netdev_name, ofport);
1713 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1718 ofport_remove(struct ofproto *p, struct ofport *ofport)
1720 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1721 hmap_remove(&p->ports, &ofport->hmap_node);
1722 shash_delete(&p->port_by_name,
1723 shash_find(&p->port_by_name, ofport->opp.name));
1725 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1730 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1733 cfm_run(ofport->cfm);
1735 if (cfm_should_send_ccm(ofport->cfm)) {
1736 struct ofpbuf packet;
1739 ofpbuf_init(&packet, 0);
1740 ccm = compose_packet(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1741 ETH_TYPE_CFM, sizeof *ccm);
1742 cfm_compose_ccm(ofport->cfm, ccm);
1743 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1744 ofpbuf_uninit(&packet);
1750 ofport_wait(struct ofport *ofport)
1753 cfm_wait(ofport->cfm);
1758 ofport_free(struct ofport *ofport)
1761 cfm_destroy(ofport->cfm);
1762 netdev_close(ofport->netdev);
1767 static struct ofport *
1768 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1770 struct ofport *port;
1772 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1773 hash_int(odp_port, 0), &ofproto->ports) {
1774 if (port->odp_port == odp_port) {
1782 update_port(struct ofproto *p, const char *devname)
1784 struct dpif_port dpif_port;
1785 struct ofport *old_ofport;
1786 struct ofport *new_ofport;
1789 COVERAGE_INC(ofproto_update_port);
1791 /* Query the datapath for port information. */
1792 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1794 /* Find the old ofport. */
1795 old_ofport = shash_find_data(&p->port_by_name, devname);
1798 /* There's no port named 'devname' but there might be a port with
1799 * the same port number. This could happen if a port is deleted
1800 * and then a new one added in its place very quickly, or if a port
1801 * is renamed. In the former case we want to send an OFPPR_DELETE
1802 * and an OFPPR_ADD, and in the latter case we want to send a
1803 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1804 * the old port's ifindex against the new port, or perhaps less
1805 * reliably but more portably by comparing the old port's MAC
1806 * against the new port's MAC. However, this code isn't that smart
1807 * and always sends an OFPPR_MODIFY (XXX). */
1808 old_ofport = get_port(p, dpif_port.port_no);
1810 } else if (error != ENOENT && error != ENODEV) {
1811 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1812 "%s", strerror(error));
1816 /* Create a new ofport. */
1817 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1819 /* Eliminate a few pathological cases. */
1820 if (!old_ofport && !new_ofport) {
1822 } else if (old_ofport && new_ofport) {
1823 /* Most of the 'config' bits are OpenFlow soft state, but
1824 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1825 * OpenFlow bits from old_ofport. (make_ofport() only sets
1826 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1827 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1829 if (ofport_equal(old_ofport, new_ofport)) {
1830 /* False alarm--no change. */
1831 ofport_free(new_ofport);
1836 /* Now deal with the normal cases. */
1838 ofport_remove(p, old_ofport);
1841 ofport_install(p, new_ofport);
1843 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1844 (!old_ofport ? OFPPR_ADD
1845 : !new_ofport ? OFPPR_DELETE
1847 ofport_free(old_ofport);
1850 dpif_port_destroy(&dpif_port);
1854 init_ports(struct ofproto *p)
1856 struct dpif_port_dump dump;
1857 struct dpif_port dpif_port;
1859 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1860 if (!ofport_conflicts(p, &dpif_port)) {
1861 struct ofport *ofport = make_ofport(&dpif_port);
1863 ofport_install(p, ofport);
1871 static struct ofconn *
1872 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1874 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1875 ofconn->ofproto = p;
1876 list_push_back(&p->all_conns, &ofconn->node);
1877 ofconn->rconn = rconn;
1878 ofconn->type = type;
1879 ofconn->flow_format = NXFF_OPENFLOW10;
1880 ofconn->role = NX_ROLE_OTHER;
1881 ofconn->packet_in_counter = rconn_packet_counter_create ();
1882 ofconn->pktbuf = NULL;
1883 ofconn->miss_send_len = 0;
1884 ofconn->reply_counter = rconn_packet_counter_create ();
1889 ofconn_destroy(struct ofconn *ofconn)
1891 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1893 if (ofconn->type == OFCONN_PRIMARY) {
1894 hmap_remove(&ofproto->controllers, &ofconn->hmap_node);
1897 list_remove(&ofconn->node);
1898 rconn_destroy(ofconn->rconn);
1899 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1900 rconn_packet_counter_destroy(ofconn->reply_counter);
1901 pktbuf_destroy(ofconn->pktbuf);
1906 ofconn_run(struct ofconn *ofconn)
1908 struct ofproto *p = ofconn_get_ofproto(ofconn);
1912 for (i = 0; i < N_SCHEDULERS; i++) {
1913 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1916 rconn_run(ofconn->rconn);
1918 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1919 /* Limit the number of iterations to prevent other tasks from
1921 for (iteration = 0; iteration < 50; iteration++) {
1922 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1927 fail_open_maybe_recover(p->fail_open);
1929 handle_openflow(ofconn, of_msg);
1930 ofpbuf_delete(of_msg);
1934 if (!rconn_is_alive(ofconn->rconn)) {
1935 ofconn_destroy(ofconn);
1940 ofconn_wait(struct ofconn *ofconn)
1944 for (i = 0; i < N_SCHEDULERS; i++) {
1945 pinsched_wait(ofconn->schedulers[i]);
1947 rconn_run_wait(ofconn->rconn);
1948 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1949 rconn_recv_wait(ofconn->rconn);
1951 COVERAGE_INC(ofproto_ofconn_stuck);
1955 /* Returns true if 'ofconn' should receive asynchronous messages. */
1957 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1959 if (ofconn->type == OFCONN_PRIMARY) {
1960 /* Primary controllers always get asynchronous messages unless they
1961 * have configured themselves as "slaves". */
1962 return ofconn->role != NX_ROLE_SLAVE;
1964 /* Service connections don't get asynchronous messages unless they have
1965 * explicitly asked for them by setting a nonzero miss send length. */
1966 return ofconn->miss_send_len > 0;
1970 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1971 * and 'target', suitable for use in log messages for identifying the
1974 * The name is dynamically allocated. The caller should free it (with free())
1975 * when it is no longer needed. */
1977 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1979 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1983 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1987 for (i = 0; i < N_SCHEDULERS; i++) {
1988 struct pinsched **s = &ofconn->schedulers[i];
1992 *s = pinsched_create(rate, burst);
1994 pinsched_set_limits(*s, rate, burst);
1997 pinsched_destroy(*s);
2003 static struct ofproto *
2004 ofconn_get_ofproto(struct ofconn *ofconn)
2006 return ofconn->ofproto;
2010 ofservice_reconfigure(struct ofservice *ofservice,
2011 const struct ofproto_controller *c)
2013 ofservice->probe_interval = c->probe_interval;
2014 ofservice->rate_limit = c->rate_limit;
2015 ofservice->burst_limit = c->burst_limit;
2018 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
2019 * positive errno value. */
2021 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
2023 struct ofservice *ofservice;
2024 struct pvconn *pvconn;
2027 error = pvconn_open(c->target, &pvconn);
2032 ofservice = xzalloc(sizeof *ofservice);
2033 hmap_insert(&ofproto->services, &ofservice->node,
2034 hash_string(c->target, 0));
2035 ofservice->pvconn = pvconn;
2037 ofservice_reconfigure(ofservice, c);
2043 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
2045 hmap_remove(&ofproto->services, &ofservice->node);
2046 pvconn_close(ofservice->pvconn);
2050 /* Finds and returns the ofservice within 'ofproto' that has the given
2051 * 'target', or a null pointer if none exists. */
2052 static struct ofservice *
2053 ofservice_lookup(struct ofproto *ofproto, const char *target)
2055 struct ofservice *ofservice;
2057 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
2058 &ofproto->services) {
2059 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2066 /* Returns true if 'rule' should be hidden from the controller.
2068 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2069 * (e.g. by in-band control) and are intentionally hidden from the
2072 rule_is_hidden(const struct rule *rule)
2074 return rule->cr.priority > UINT16_MAX;
2077 /* Creates and returns a new rule initialized as specified.
2079 * The caller is responsible for inserting the rule into the classifier (with
2080 * rule_insert()). */
2081 static struct rule *
2082 rule_create(const struct cls_rule *cls_rule,
2083 const union ofp_action *actions, size_t n_actions,
2084 uint16_t idle_timeout, uint16_t hard_timeout,
2085 ovs_be64 flow_cookie, bool send_flow_removed)
2087 struct rule *rule = xzalloc(sizeof *rule);
2088 rule->cr = *cls_rule;
2089 rule->idle_timeout = idle_timeout;
2090 rule->hard_timeout = hard_timeout;
2091 rule->flow_cookie = flow_cookie;
2092 rule->used = rule->created = time_msec();
2093 rule->send_flow_removed = send_flow_removed;
2094 list_init(&rule->facets);
2095 if (n_actions > 0) {
2096 rule->n_actions = n_actions;
2097 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2103 static struct rule *
2104 rule_from_cls_rule(const struct cls_rule *cls_rule)
2106 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2110 rule_free(struct rule *rule)
2112 free(rule->actions);
2116 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2117 * destroying any that no longer has a rule (which is probably all of them).
2119 * The caller must have already removed 'rule' from the classifier. */
2121 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2123 struct facet *facet, *next_facet;
2124 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2125 facet_revalidate(ofproto, facet);
2130 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2131 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2134 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2136 const union ofp_action *oa;
2137 struct actions_iterator i;
2139 if (out_port == htons(OFPP_NONE)) {
2142 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2143 oa = actions_next(&i)) {
2144 if (action_outputs_to_port(oa, out_port)) {
2151 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2152 * 'packet', which arrived on 'in_port'.
2154 * Takes ownership of 'packet'. */
2156 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2157 const struct nlattr *odp_actions, size_t actions_len,
2158 struct ofpbuf *packet)
2160 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2161 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2162 /* As an optimization, avoid a round-trip from userspace to kernel to
2163 * userspace. This also avoids possibly filling up kernel packet
2164 * buffers along the way. */
2165 struct dpif_upcall upcall;
2167 upcall.type = DPIF_UC_ACTION;
2168 upcall.packet = packet;
2171 upcall.userdata = nl_attr_get_u64(odp_actions);
2172 upcall.sample_pool = 0;
2173 upcall.actions = NULL;
2174 upcall.actions_len = 0;
2176 send_packet_in(ofproto, &upcall, flow, false);
2182 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2183 ofpbuf_delete(packet);
2188 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2189 * statistics appropriately. 'packet' must have at least sizeof(struct
2190 * ofp_packet_in) bytes of headroom.
2192 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2193 * applying flow_extract() to 'packet' would yield the same flow as
2196 * 'facet' must have accurately composed ODP actions; that is, it must not be
2197 * in need of revalidation.
2199 * Takes ownership of 'packet'. */
2201 facet_execute(struct ofproto *ofproto, struct facet *facet,
2202 struct ofpbuf *packet)
2204 struct dpif_flow_stats stats;
2206 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2208 flow_extract_stats(&facet->flow, packet, &stats);
2209 stats.used = time_msec();
2210 if (execute_odp_actions(ofproto, &facet->flow,
2211 facet->actions, facet->actions_len, packet)) {
2212 facet_update_stats(ofproto, facet, &stats);
2216 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2217 * statistics (or the statistics for one of its facets) appropriately.
2218 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2220 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2221 * with statistics for 'packet' either way.
2223 * Takes ownership of 'packet'. */
2225 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2226 struct ofpbuf *packet)
2228 struct action_xlate_ctx ctx;
2229 struct ofpbuf *odp_actions;
2230 struct facet *facet;
2234 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2236 flow_extract(packet, 0, in_port, &flow);
2238 /* First look for a related facet. If we find one, account it to that. */
2239 facet = facet_lookup_valid(ofproto, &flow);
2240 if (facet && facet->rule == rule) {
2241 facet_execute(ofproto, facet, packet);
2245 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2246 * create a new facet for it and use that. */
2247 if (rule_lookup(ofproto, &flow) == rule) {
2248 facet = facet_create(ofproto, rule, &flow, packet);
2249 facet_execute(ofproto, facet, packet);
2250 facet_install(ofproto, facet, true);
2254 /* We can't account anything to a facet. If we were to try, then that
2255 * facet would have a non-matching rule, busting our invariants. */
2256 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2257 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2258 size = packet->size;
2259 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2260 odp_actions->size, packet)) {
2261 rule->used = time_msec();
2262 rule->packet_count++;
2263 rule->byte_count += size;
2264 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
2266 ofpbuf_delete(odp_actions);
2269 /* Inserts 'rule' into 'p''s flow table. */
2271 rule_insert(struct ofproto *p, struct rule *rule)
2273 struct rule *displaced_rule;
2275 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2276 if (displaced_rule) {
2277 rule_destroy(p, displaced_rule);
2279 p->need_revalidate = true;
2282 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2283 * 'flow' and an example 'packet' within that flow.
2285 * The caller must already have determined that no facet with an identical
2286 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2287 * 'ofproto''s classifier table. */
2288 static struct facet *
2289 facet_create(struct ofproto *ofproto, struct rule *rule,
2290 const struct flow *flow, const struct ofpbuf *packet)
2292 struct facet *facet;
2294 facet = xzalloc(sizeof *facet);
2295 facet->used = time_msec();
2296 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2297 list_push_back(&rule->facets, &facet->list_node);
2299 facet->flow = *flow;
2300 netflow_flow_init(&facet->nf_flow);
2301 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2303 facet_make_actions(ofproto, facet, packet);
2309 facet_free(struct facet *facet)
2311 free(facet->actions);
2315 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2317 * - Removes 'rule' from the classifier.
2319 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2320 * destroys them), via rule_destroy().
2323 rule_remove(struct ofproto *ofproto, struct rule *rule)
2325 COVERAGE_INC(ofproto_del_rule);
2326 ofproto->need_revalidate = true;
2327 classifier_remove(&ofproto->cls, &rule->cr);
2328 rule_destroy(ofproto, rule);
2331 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2333 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2334 * rule's statistics, via facet_uninstall().
2336 * - Removes 'facet' from its rule and from ofproto->facets.
2339 facet_remove(struct ofproto *ofproto, struct facet *facet)
2341 facet_uninstall(ofproto, facet);
2342 facet_flush_stats(ofproto, facet);
2343 hmap_remove(&ofproto->facets, &facet->hmap_node);
2344 list_remove(&facet->list_node);
2348 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2350 facet_make_actions(struct ofproto *p, struct facet *facet,
2351 const struct ofpbuf *packet)
2353 const struct rule *rule = facet->rule;
2354 struct ofpbuf *odp_actions;
2355 struct action_xlate_ctx ctx;
2357 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2358 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2359 facet->tags = ctx.tags;
2360 facet->may_install = ctx.may_set_up_flow;
2361 facet->nf_flow.output_iface = ctx.nf_output_iface;
2363 if (facet->actions_len != odp_actions->size
2364 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2365 free(facet->actions);
2366 facet->actions_len = odp_actions->size;
2367 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2370 ofpbuf_delete(odp_actions);
2374 facet_put__(struct ofproto *ofproto, struct facet *facet,
2375 const struct nlattr *actions, size_t actions_len,
2376 struct dpif_flow_stats *stats)
2378 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2379 enum dpif_flow_put_flags flags;
2382 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2384 flags |= DPIF_FP_ZERO_STATS;
2385 facet->dp_packet_count = 0;
2386 facet->dp_byte_count = 0;
2389 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2390 odp_flow_key_from_flow(&key, &facet->flow);
2391 assert(key.base == keybuf);
2393 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2394 actions, actions_len, stats);
2397 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2398 * 'zero_stats' is true, clears any existing statistics from the datapath for
2401 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2403 struct dpif_flow_stats stats;
2405 if (facet->may_install
2406 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2407 zero_stats ? &stats : NULL)) {
2408 facet->installed = true;
2412 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2413 * to the accounting hook function in the ofhooks structure. */
2415 facet_account(struct ofproto *ofproto,
2416 struct facet *facet, uint64_t extra_bytes)
2418 uint64_t total_bytes = facet->byte_count + extra_bytes;
2420 if (ofproto->ofhooks->account_flow_cb
2421 && total_bytes > facet->accounted_bytes)
2423 ofproto->ofhooks->account_flow_cb(
2424 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2425 total_bytes - facet->accounted_bytes, ofproto->aux);
2426 facet->accounted_bytes = total_bytes;
2430 /* If 'rule' is installed in the datapath, uninstalls it. */
2432 facet_uninstall(struct ofproto *p, struct facet *facet)
2434 if (facet->installed) {
2435 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2436 struct dpif_flow_stats stats;
2439 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2440 odp_flow_key_from_flow(&key, &facet->flow);
2441 assert(key.base == keybuf);
2443 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2444 facet_update_stats(p, facet, &stats);
2446 facet->installed = false;
2447 facet->dp_packet_count = 0;
2448 facet->dp_byte_count = 0;
2450 assert(facet->dp_packet_count == 0);
2451 assert(facet->dp_byte_count == 0);
2455 /* Returns true if the only action for 'facet' is to send to the controller.
2456 * (We don't report NetFlow expiration messages for such facets because they
2457 * are just part of the control logic for the network, not real traffic). */
2459 facet_is_controller_flow(struct facet *facet)
2462 && facet->rule->n_actions == 1
2463 && action_outputs_to_port(&facet->rule->actions[0],
2464 htons(OFPP_CONTROLLER)));
2467 /* Folds all of 'facet''s statistics into its rule. Also updates the
2468 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2469 * 'facet''s statistics in the datapath should have been zeroed and folded into
2470 * its packet and byte counts before this function is called. */
2472 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2474 assert(!facet->dp_byte_count);
2475 assert(!facet->dp_packet_count);
2477 facet_push_stats(ofproto, facet);
2478 facet_account(ofproto, facet, 0);
2480 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2481 struct ofexpired expired;
2482 expired.flow = facet->flow;
2483 expired.packet_count = facet->packet_count;
2484 expired.byte_count = facet->byte_count;
2485 expired.used = facet->used;
2486 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2489 facet->rule->packet_count += facet->packet_count;
2490 facet->rule->byte_count += facet->byte_count;
2492 /* Reset counters to prevent double counting if 'facet' ever gets
2494 facet->packet_count = 0;
2495 facet->byte_count = 0;
2496 facet->rs_packet_count = 0;
2497 facet->rs_byte_count = 0;
2498 facet->accounted_bytes = 0;
2500 netflow_flow_clear(&facet->nf_flow);
2503 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2504 * Returns it if found, otherwise a null pointer.
2506 * The returned facet might need revalidation; use facet_lookup_valid()
2507 * instead if that is important. */
2508 static struct facet *
2509 facet_find(struct ofproto *ofproto, const struct flow *flow)
2511 struct facet *facet;
2513 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2515 if (flow_equal(flow, &facet->flow)) {
2523 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2524 * Returns it if found, otherwise a null pointer.
2526 * The returned facet is guaranteed to be valid. */
2527 static struct facet *
2528 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2530 struct facet *facet = facet_find(ofproto, flow);
2532 /* The facet we found might not be valid, since we could be in need of
2533 * revalidation. If it is not valid, don't return it. */
2535 && ofproto->need_revalidate
2536 && !facet_revalidate(ofproto, facet)) {
2537 COVERAGE_INC(ofproto_invalidated);
2544 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2546 * - If the rule found is different from 'facet''s current rule, moves
2547 * 'facet' to the new rule and recompiles its actions.
2549 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2550 * where it is and recompiles its actions anyway.
2552 * - If there is none, destroys 'facet'.
2554 * Returns true if 'facet' still exists, false if it has been destroyed. */
2556 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2558 struct action_xlate_ctx ctx;
2559 struct ofpbuf *odp_actions;
2560 struct rule *new_rule;
2561 bool actions_changed;
2563 COVERAGE_INC(facet_revalidate);
2565 /* Determine the new rule. */
2566 new_rule = rule_lookup(ofproto, &facet->flow);
2568 /* No new rule, so delete the facet. */
2569 facet_remove(ofproto, facet);
2573 /* Calculate new ODP actions.
2575 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2576 * emit a NetFlow expiration and, if so, we need to have the old state
2577 * around to properly compose it. */
2578 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2579 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2580 actions_changed = (facet->actions_len != odp_actions->size
2581 || memcmp(facet->actions, odp_actions->data,
2582 facet->actions_len));
2584 /* If the ODP actions changed or the installability changed, then we need
2585 * to talk to the datapath. */
2586 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2587 if (ctx.may_set_up_flow) {
2588 struct dpif_flow_stats stats;
2590 facet_put__(ofproto, facet,
2591 odp_actions->data, odp_actions->size, &stats);
2592 facet_update_stats(ofproto, facet, &stats);
2594 facet_uninstall(ofproto, facet);
2597 /* The datapath flow is gone or has zeroed stats, so push stats out of
2598 * 'facet' into 'rule'. */
2599 facet_flush_stats(ofproto, facet);
2602 /* Update 'facet' now that we've taken care of all the old state. */
2603 facet->tags = ctx.tags;
2604 facet->nf_flow.output_iface = ctx.nf_output_iface;
2605 facet->may_install = ctx.may_set_up_flow;
2606 if (actions_changed) {
2607 free(facet->actions);
2608 facet->actions_len = odp_actions->size;
2609 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2611 if (facet->rule != new_rule) {
2612 COVERAGE_INC(facet_changed_rule);
2613 list_remove(&facet->list_node);
2614 list_push_back(&new_rule->facets, &facet->list_node);
2615 facet->rule = new_rule;
2616 facet->used = new_rule->created;
2617 facet->rs_used = facet->used;
2620 ofpbuf_delete(odp_actions);
2626 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2627 struct rconn_packet_counter *counter)
2629 update_openflow_length(msg);
2630 if (rconn_send(ofconn->rconn, msg, counter)) {
2636 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2639 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2641 COVERAGE_INC(ofproto_error);
2642 queue_tx(buf, ofconn, ofconn->reply_counter);
2647 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2649 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2654 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2656 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2657 struct ofp_switch_features *osf;
2659 struct ofport *port;
2661 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2662 osf->datapath_id = htonll(ofproto->datapath_id);
2663 osf->n_buffers = htonl(pktbuf_capacity());
2665 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2666 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2667 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2668 (1u << OFPAT_SET_VLAN_VID) |
2669 (1u << OFPAT_SET_VLAN_PCP) |
2670 (1u << OFPAT_STRIP_VLAN) |
2671 (1u << OFPAT_SET_DL_SRC) |
2672 (1u << OFPAT_SET_DL_DST) |
2673 (1u << OFPAT_SET_NW_SRC) |
2674 (1u << OFPAT_SET_NW_DST) |
2675 (1u << OFPAT_SET_NW_TOS) |
2676 (1u << OFPAT_SET_TP_SRC) |
2677 (1u << OFPAT_SET_TP_DST) |
2678 (1u << OFPAT_ENQUEUE));
2680 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2681 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2684 queue_tx(buf, ofconn, ofconn->reply_counter);
2689 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2691 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2693 struct ofp_switch_config *osc;
2697 /* Figure out flags. */
2698 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2699 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2702 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2703 osc->flags = htons(flags);
2704 osc->miss_send_len = htons(ofconn->miss_send_len);
2705 queue_tx(buf, ofconn, ofconn->reply_counter);
2711 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2713 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2714 uint16_t flags = ntohs(osc->flags);
2716 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2717 switch (flags & OFPC_FRAG_MASK) {
2718 case OFPC_FRAG_NORMAL:
2719 dpif_set_drop_frags(ofproto->dpif, false);
2721 case OFPC_FRAG_DROP:
2722 dpif_set_drop_frags(ofproto->dpif, true);
2725 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2731 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2736 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2737 struct action_xlate_ctx *ctx);
2740 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2742 const struct ofport *ofport = get_port(ctx->ofproto, port);
2745 if (ofport->opp.config & OFPPC_NO_FWD) {
2746 /* Forwarding disabled on port. */
2751 * We don't have an ofport record for this port, but it doesn't hurt to
2752 * allow forwarding to it anyhow. Maybe such a port will appear later
2753 * and we're pre-populating the flow table.
2757 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2758 ctx->nf_output_iface = port;
2761 static struct rule *
2762 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2764 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2768 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2770 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2771 uint16_t old_in_port;
2774 /* Look up a flow with 'in_port' as the input port. Then restore the
2775 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2776 * have surprising behavior). */
2777 old_in_port = ctx->flow.in_port;
2778 ctx->flow.in_port = in_port;
2779 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2780 ctx->flow.in_port = old_in_port;
2782 if (ctx->resubmit_hook) {
2783 ctx->resubmit_hook(ctx, rule);
2788 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2792 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2794 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2795 MAX_RESUBMIT_RECURSION);
2800 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2801 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2803 struct ofport *ofport;
2805 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2806 uint16_t odp_port = ofport->odp_port;
2807 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2808 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2811 *nf_output_iface = NF_OUT_FLOOD;
2815 xlate_output_action__(struct action_xlate_ctx *ctx,
2816 uint16_t port, uint16_t max_len)
2819 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2821 ctx->nf_output_iface = NF_OUT_DROP;
2825 add_output_action(ctx, ctx->flow.in_port);
2828 xlate_table_action(ctx, ctx->flow.in_port);
2831 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2832 ctx->odp_actions, &ctx->tags,
2833 &ctx->nf_output_iface,
2834 ctx->ofproto->aux)) {
2835 COVERAGE_INC(ofproto_uninstallable);
2836 ctx->may_set_up_flow = false;
2840 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2841 &ctx->nf_output_iface, ctx->odp_actions);
2844 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2845 &ctx->nf_output_iface, ctx->odp_actions);
2847 case OFPP_CONTROLLER:
2848 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2851 add_output_action(ctx, ODPP_LOCAL);
2854 odp_port = ofp_port_to_odp_port(port);
2855 if (odp_port != ctx->flow.in_port) {
2856 add_output_action(ctx, odp_port);
2861 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2862 ctx->nf_output_iface = NF_OUT_FLOOD;
2863 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2864 ctx->nf_output_iface = prev_nf_output_iface;
2865 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2866 ctx->nf_output_iface != NF_OUT_FLOOD) {
2867 ctx->nf_output_iface = NF_OUT_MULTI;
2872 xlate_output_action(struct action_xlate_ctx *ctx,
2873 const struct ofp_action_output *oao)
2875 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2878 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2879 * optimization, because we're going to add another action that sets the
2880 * priority immediately after, or because there are no actions following the
2883 remove_pop_action(struct action_xlate_ctx *ctx)
2885 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2886 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2887 ctx->last_pop_priority = -1;
2892 add_pop_action(struct action_xlate_ctx *ctx)
2894 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2895 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2896 ctx->last_pop_priority = ctx->odp_actions->size;
2901 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2902 const struct ofp_action_enqueue *oae)
2904 uint16_t ofp_port, odp_port;
2908 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2911 /* Fall back to ordinary output action. */
2912 xlate_output_action__(ctx, ntohs(oae->port), 0);
2916 /* Figure out ODP output port. */
2917 ofp_port = ntohs(oae->port);
2918 if (ofp_port != OFPP_IN_PORT) {
2919 odp_port = ofp_port_to_odp_port(ofp_port);
2921 odp_port = ctx->flow.in_port;
2924 /* Add ODP actions. */
2925 remove_pop_action(ctx);
2926 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2927 add_output_action(ctx, odp_port);
2928 add_pop_action(ctx);
2930 /* Update NetFlow output port. */
2931 if (ctx->nf_output_iface == NF_OUT_DROP) {
2932 ctx->nf_output_iface = odp_port;
2933 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2934 ctx->nf_output_iface = NF_OUT_MULTI;
2939 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2940 const struct nx_action_set_queue *nasq)
2945 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2948 /* Couldn't translate queue to a priority, so ignore. A warning
2949 * has already been logged. */
2953 remove_pop_action(ctx);
2954 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2958 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2960 ovs_be16 tci = ctx->flow.vlan_tci;
2961 if (!(tci & htons(VLAN_CFI))) {
2962 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2964 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2965 tci & ~htons(VLAN_CFI));
2969 struct xlate_reg_state {
2975 save_reg_state(const struct action_xlate_ctx *ctx,
2976 struct xlate_reg_state *state)
2978 state->vlan_tci = ctx->flow.vlan_tci;
2979 state->tun_id = ctx->flow.tun_id;
2983 update_reg_state(struct action_xlate_ctx *ctx,
2984 const struct xlate_reg_state *state)
2986 if (ctx->flow.vlan_tci != state->vlan_tci) {
2987 xlate_set_dl_tci(ctx);
2989 if (ctx->flow.tun_id != state->tun_id) {
2990 nl_msg_put_be64(ctx->odp_actions,
2991 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2996 xlate_nicira_action(struct action_xlate_ctx *ctx,
2997 const struct nx_action_header *nah)
2999 const struct nx_action_resubmit *nar;
3000 const struct nx_action_set_tunnel *nast;
3001 const struct nx_action_set_queue *nasq;
3002 const struct nx_action_multipath *nam;
3003 enum nx_action_subtype subtype = ntohs(nah->subtype);
3004 struct xlate_reg_state state;
3007 assert(nah->vendor == htonl(NX_VENDOR_ID));
3009 case NXAST_RESUBMIT:
3010 nar = (const struct nx_action_resubmit *) nah;
3011 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
3014 case NXAST_SET_TUNNEL:
3015 nast = (const struct nx_action_set_tunnel *) nah;
3016 tun_id = htonll(ntohl(nast->tun_id));
3017 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3018 ctx->flow.tun_id = tun_id;
3021 case NXAST_DROP_SPOOFED_ARP:
3022 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
3023 nl_msg_put_flag(ctx->odp_actions,
3024 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
3028 case NXAST_SET_QUEUE:
3029 nasq = (const struct nx_action_set_queue *) nah;
3030 xlate_set_queue_action(ctx, nasq);
3033 case NXAST_POP_QUEUE:
3034 add_pop_action(ctx);
3037 case NXAST_REG_MOVE:
3038 save_reg_state(ctx, &state);
3039 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
3041 update_reg_state(ctx, &state);
3044 case NXAST_REG_LOAD:
3045 save_reg_state(ctx, &state);
3046 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
3048 update_reg_state(ctx, &state);
3052 /* Nothing to do. */
3055 case NXAST_SET_TUNNEL64:
3056 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
3057 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3058 ctx->flow.tun_id = tun_id;
3061 case NXAST_MULTIPATH:
3062 nam = (const struct nx_action_multipath *) nah;
3063 multipath_execute(nam, &ctx->flow);
3066 /* If you add a new action here that modifies flow data, don't forget to
3067 * update the flow key in ctx->flow at the same time. */
3069 case NXAST_SNAT__OBSOLETE:
3071 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3077 do_xlate_actions(const union ofp_action *in, size_t n_in,
3078 struct action_xlate_ctx *ctx)
3080 struct actions_iterator iter;
3081 const union ofp_action *ia;
3082 const struct ofport *port;
3084 port = get_port(ctx->ofproto, ctx->flow.in_port);
3085 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3086 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3087 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3088 /* Drop this flow. */
3092 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3093 enum ofp_action_type type = ntohs(ia->type);
3094 const struct ofp_action_dl_addr *oada;
3098 xlate_output_action(ctx, &ia->output);
3101 case OFPAT_SET_VLAN_VID:
3102 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3103 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3104 xlate_set_dl_tci(ctx);
3107 case OFPAT_SET_VLAN_PCP:
3108 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3109 ctx->flow.vlan_tci |= htons(
3110 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3111 xlate_set_dl_tci(ctx);
3114 case OFPAT_STRIP_VLAN:
3115 ctx->flow.vlan_tci = htons(0);
3116 xlate_set_dl_tci(ctx);
3119 case OFPAT_SET_DL_SRC:
3120 oada = ((struct ofp_action_dl_addr *) ia);
3121 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3122 oada->dl_addr, ETH_ADDR_LEN);
3123 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3126 case OFPAT_SET_DL_DST:
3127 oada = ((struct ofp_action_dl_addr *) ia);
3128 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3129 oada->dl_addr, ETH_ADDR_LEN);
3130 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3133 case OFPAT_SET_NW_SRC:
3134 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3135 ia->nw_addr.nw_addr);
3136 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3139 case OFPAT_SET_NW_DST:
3140 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3141 ia->nw_addr.nw_addr);
3142 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3145 case OFPAT_SET_NW_TOS:
3146 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3148 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3151 case OFPAT_SET_TP_SRC:
3152 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3153 ia->tp_port.tp_port);
3154 ctx->flow.tp_src = ia->tp_port.tp_port;
3157 case OFPAT_SET_TP_DST:
3158 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3159 ia->tp_port.tp_port);
3160 ctx->flow.tp_dst = ia->tp_port.tp_port;
3164 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3168 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3172 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3179 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3180 struct ofproto *ofproto, const struct flow *flow,
3181 const struct ofpbuf *packet)
3183 ctx->ofproto = ofproto;
3185 ctx->packet = packet;
3186 ctx->resubmit_hook = NULL;
3187 ctx->check_special = true;
3191 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
3192 const struct ofpbuf *packet)
3194 struct ofport *ofport;
3196 ofport = get_port(ofproto, flow->in_port);
3197 if (ofport && ofport->cfm) {
3198 cfm_process_heartbeat(ofport->cfm, packet);
3202 static struct ofpbuf *
3203 xlate_actions(struct action_xlate_ctx *ctx,
3204 const union ofp_action *in, size_t n_in)
3206 COVERAGE_INC(ofproto_ofp2odp);
3208 ctx->odp_actions = ofpbuf_new(512);
3210 ctx->may_set_up_flow = true;
3211 ctx->nf_output_iface = NF_OUT_DROP;
3213 ctx->last_pop_priority = -1;
3215 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
3217 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
3219 ctx->may_set_up_flow = false;
3220 } else if (ctx->check_special
3221 && ctx->ofproto->ofhooks->special_cb
3222 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3223 ctx->ofproto->aux)) {
3224 ctx->may_set_up_flow = false;
3226 do_xlate_actions(in, n_in, ctx);
3229 remove_pop_action(ctx);
3231 /* Check with in-band control to see if we're allowed to set up this
3233 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3234 ctx->odp_actions->data, ctx->odp_actions->size)) {
3235 ctx->may_set_up_flow = false;
3238 return ctx->odp_actions;
3241 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3242 * error message code (composed with ofp_mkerr()) for the caller to propagate
3243 * upward. Otherwise, returns 0.
3245 * The log message mentions 'msg_type'. */
3247 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3249 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3250 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3251 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3254 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3261 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3263 struct ofproto *p = ofconn_get_ofproto(ofconn);
3264 struct ofp_packet_out *opo;
3265 struct ofpbuf payload, *buffer;
3266 union ofp_action *ofp_actions;
3267 struct action_xlate_ctx ctx;
3268 struct ofpbuf *odp_actions;
3269 struct ofpbuf request;
3271 size_t n_ofp_actions;
3275 COVERAGE_INC(ofproto_packet_out);
3277 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3282 /* Get ofp_packet_out. */
3283 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3284 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3287 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3288 &ofp_actions, &n_ofp_actions);
3294 if (opo->buffer_id != htonl(UINT32_MAX)) {
3295 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3297 if (error || !buffer) {
3306 /* Extract flow, check actions. */
3307 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3309 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3315 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3316 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3317 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3318 ofpbuf_delete(odp_actions);
3321 ofpbuf_delete(buffer);
3326 update_port_config(struct ofproto *p, struct ofport *port,
3327 uint32_t config, uint32_t mask)
3329 mask &= config ^ port->opp.config;
3330 if (mask & OFPPC_PORT_DOWN) {
3331 if (config & OFPPC_PORT_DOWN) {
3332 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3334 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3337 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3338 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3339 if (mask & REVALIDATE_BITS) {
3340 COVERAGE_INC(ofproto_costly_flags);
3341 port->opp.config ^= mask & REVALIDATE_BITS;
3342 p->need_revalidate = true;
3344 #undef REVALIDATE_BITS
3345 if (mask & OFPPC_NO_PACKET_IN) {
3346 port->opp.config ^= OFPPC_NO_PACKET_IN;
3351 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3353 struct ofproto *p = ofconn_get_ofproto(ofconn);
3354 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3355 struct ofport *port;
3358 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3363 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3365 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3366 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3367 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3369 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3370 if (opm->advertise) {
3371 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3377 static struct ofpbuf *
3378 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3380 struct ofp_stats_reply *osr;
3383 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3384 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3386 osr->flags = htons(0);
3390 static struct ofpbuf *
3391 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3393 const struct ofp_stats_request *osr
3394 = (const struct ofp_stats_request *) request;
3395 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3399 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3400 struct ofpbuf **msgp)
3402 struct ofpbuf *msg = *msgp;
3403 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3404 if (nbytes + msg->size > UINT16_MAX) {
3405 struct ofp_stats_reply *reply = msg->data;
3406 reply->flags = htons(OFPSF_REPLY_MORE);
3407 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3408 queue_tx(msg, ofconn, ofconn->reply_counter);
3410 return ofpbuf_put_uninit(*msgp, nbytes);
3413 static struct ofpbuf *
3414 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3416 struct nicira_stats_msg *nsm;
3419 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3420 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3421 nsm->type = htons(OFPST_VENDOR);
3422 nsm->flags = htons(0);
3423 nsm->vendor = htonl(NX_VENDOR_ID);
3424 nsm->subtype = subtype;
3428 static struct ofpbuf *
3429 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3431 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3435 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3436 struct ofpbuf **msgp)
3438 struct ofpbuf *msg = *msgp;
3439 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3440 if (nbytes + msg->size > UINT16_MAX) {
3441 struct nicira_stats_msg *reply = msg->data;
3442 reply->flags = htons(OFPSF_REPLY_MORE);
3443 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3444 queue_tx(msg, ofconn, ofconn->reply_counter);
3446 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3450 handle_desc_stats_request(struct ofconn *ofconn,
3451 const struct ofp_header *request)
3453 struct ofproto *p = ofconn_get_ofproto(ofconn);
3454 struct ofp_desc_stats *ods;
3457 msg = start_ofp_stats_reply(request, sizeof *ods);
3458 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3459 memset(ods, 0, sizeof *ods);
3460 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3461 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3462 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3463 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3464 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3465 queue_tx(msg, ofconn, ofconn->reply_counter);
3471 handle_table_stats_request(struct ofconn *ofconn,
3472 const struct ofp_header *request)
3474 struct ofproto *p = ofconn_get_ofproto(ofconn);
3475 struct ofp_table_stats *ots;
3478 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3480 /* Classifier table. */
3481 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3482 memset(ots, 0, sizeof *ots);
3483 strcpy(ots->name, "classifier");
3484 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3485 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3486 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3487 ots->active_count = htonl(classifier_count(&p->cls));
3488 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3489 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3491 queue_tx(msg, ofconn, ofconn->reply_counter);
3496 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3497 struct ofpbuf **msgp)
3499 struct netdev_stats stats;
3500 struct ofp_port_stats *ops;
3502 /* Intentionally ignore return value, since errors will set
3503 * 'stats' to all-1s, which is correct for OpenFlow, and
3504 * netdev_get_stats() will log errors. */
3505 netdev_get_stats(port->netdev, &stats);
3507 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3508 ops->port_no = htons(port->opp.port_no);
3509 memset(ops->pad, 0, sizeof ops->pad);
3510 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3511 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3512 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3513 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3514 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3515 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3516 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3517 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3518 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3519 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3520 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3521 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3525 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3527 struct ofproto *p = ofconn_get_ofproto(ofconn);
3528 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3529 struct ofp_port_stats *ops;
3531 struct ofport *port;
3533 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3534 if (psr->port_no != htons(OFPP_NONE)) {
3535 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3537 append_port_stat(port, ofconn, &msg);
3540 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3541 append_port_stat(port, ofconn, &msg);
3545 queue_tx(msg, ofconn, ofconn->reply_counter);
3550 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
3552 long long int msecs = time_msec() - start;
3553 *sec = msecs / 1000;
3554 *nsec = (msecs % 1000) * (1000 * 1000);
3558 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
3562 calc_flow_duration__(start, &sec, &nsec);
3563 *sec_be = htonl(sec);
3564 *nsec_be = htonl(nsec);
3568 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3569 ovs_be16 out_port, struct ofpbuf **replyp)
3571 struct ofp_flow_stats *ofs;
3572 uint64_t packet_count, byte_count;
3574 size_t act_len, len;
3576 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3580 act_len = sizeof *rule->actions * rule->n_actions;
3581 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3583 rule_get_stats(rule, &packet_count, &byte_count);
3585 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3586 ofs->length = htons(len);
3589 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3590 rule->flow_cookie, &cookie);
3591 put_32aligned_be64(&ofs->cookie, cookie);
3592 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3593 ofs->priority = htons(rule->cr.priority);
3594 ofs->idle_timeout = htons(rule->idle_timeout);
3595 ofs->hard_timeout = htons(rule->hard_timeout);
3596 memset(ofs->pad2, 0, sizeof ofs->pad2);
3597 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3598 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3599 if (rule->n_actions > 0) {
3600 memcpy(ofs->actions, rule->actions, act_len);
3605 is_valid_table(uint8_t table_id)
3607 if (table_id == 0 || table_id == 0xff) {
3610 /* It would probably be better to reply with an error but there doesn't
3611 * seem to be any appropriate value, so that might just be
3613 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3620 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3622 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3623 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3624 struct ofpbuf *reply;
3626 COVERAGE_INC(ofproto_flows_req);
3627 reply = start_ofp_stats_reply(oh, 1024);
3628 if (is_valid_table(fsr->table_id)) {
3629 struct cls_cursor cursor;
3630 struct cls_rule target;
3633 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3635 cls_cursor_init(&cursor, &ofproto->cls, &target);
3636 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3637 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3640 queue_tx(reply, ofconn, ofconn->reply_counter);
3646 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3647 ovs_be16 out_port, struct ofpbuf **replyp)
3649 struct nx_flow_stats *nfs;
3650 uint64_t packet_count, byte_count;
3651 size_t act_len, start_len;
3652 struct ofpbuf *reply;
3654 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3658 rule_get_stats(rule, &packet_count, &byte_count);
3660 act_len = sizeof *rule->actions * rule->n_actions;
3662 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3663 start_len = (*replyp)->size;
3666 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3669 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3670 nfs->cookie = rule->flow_cookie;
3671 nfs->priority = htons(rule->cr.priority);
3672 nfs->idle_timeout = htons(rule->idle_timeout);
3673 nfs->hard_timeout = htons(rule->hard_timeout);
3674 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3675 memset(nfs->pad2, 0, sizeof nfs->pad2);
3676 nfs->packet_count = htonll(packet_count);
3677 nfs->byte_count = htonll(byte_count);
3678 if (rule->n_actions > 0) {
3679 ofpbuf_put(reply, rule->actions, act_len);
3681 nfs->length = htons(reply->size - start_len);
3685 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3687 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3688 struct nx_flow_stats_request *nfsr;
3689 struct cls_rule target;
3690 struct ofpbuf *reply;
3694 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3696 /* Dissect the message. */
3697 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3698 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3703 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3706 COVERAGE_INC(ofproto_flows_req);
3707 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3708 if (is_valid_table(nfsr->table_id)) {
3709 struct cls_cursor cursor;
3712 cls_cursor_init(&cursor, &ofproto->cls, &target);
3713 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3714 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3717 queue_tx(reply, ofconn, ofconn->reply_counter);
3723 flow_stats_ds(struct rule *rule, struct ds *results)
3725 uint64_t packet_count, byte_count;
3726 size_t act_len = sizeof *rule->actions * rule->n_actions;
3728 rule_get_stats(rule, &packet_count, &byte_count);
3730 ds_put_format(results, "duration=%llds, ",
3731 (time_msec() - rule->created) / 1000);
3732 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3733 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3734 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3735 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3736 cls_rule_format(&rule->cr, results);
3737 ds_put_char(results, ',');
3739 ofp_print_actions(results, &rule->actions->header, act_len);
3741 ds_put_cstr(results, "drop");
3743 ds_put_cstr(results, "\n");
3746 /* Adds a pretty-printed description of all flows to 'results', including
3747 * hidden flows (e.g., set up by in-band control). */
3749 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3751 struct cls_cursor cursor;
3754 cls_cursor_init(&cursor, &p->cls, NULL);
3755 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3756 flow_stats_ds(rule, results);
3761 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3762 ovs_be16 out_port, uint8_t table_id,
3763 struct ofp_aggregate_stats_reply *oasr)
3765 uint64_t total_packets = 0;
3766 uint64_t total_bytes = 0;
3769 COVERAGE_INC(ofproto_agg_request);
3771 if (is_valid_table(table_id)) {
3772 struct cls_cursor cursor;
3775 cls_cursor_init(&cursor, &ofproto->cls, target);
3776 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3777 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3778 uint64_t packet_count;
3779 uint64_t byte_count;
3781 rule_get_stats(rule, &packet_count, &byte_count);
3783 total_packets += packet_count;
3784 total_bytes += byte_count;
3790 oasr->flow_count = htonl(n_flows);
3791 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3792 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3793 memset(oasr->pad, 0, sizeof oasr->pad);
3797 handle_aggregate_stats_request(struct ofconn *ofconn,
3798 const struct ofp_header *oh)
3800 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3801 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3802 struct ofp_aggregate_stats_reply *reply;
3803 struct cls_rule target;
3806 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3809 msg = start_ofp_stats_reply(oh, sizeof *reply);
3810 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3811 query_aggregate_stats(ofproto, &target, request->out_port,
3812 request->table_id, reply);
3813 queue_tx(msg, ofconn, ofconn->reply_counter);
3818 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3820 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3821 struct nx_aggregate_stats_request *request;
3822 struct ofp_aggregate_stats_reply *reply;
3823 struct cls_rule target;
3828 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3830 /* Dissect the message. */
3831 request = ofpbuf_pull(&b, sizeof *request);
3832 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3837 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3841 COVERAGE_INC(ofproto_flows_req);
3842 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3843 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3844 query_aggregate_stats(ofproto, &target, request->out_port,
3845 request->table_id, reply);
3846 queue_tx(buf, ofconn, ofconn->reply_counter);
3851 struct queue_stats_cbdata {
3852 struct ofconn *ofconn;
3853 struct ofport *ofport;
3858 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3859 const struct netdev_queue_stats *stats)
3861 struct ofp_queue_stats *reply;
3863 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3864 reply->port_no = htons(cbdata->ofport->opp.port_no);
3865 memset(reply->pad, 0, sizeof reply->pad);
3866 reply->queue_id = htonl(queue_id);
3867 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3868 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3869 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3873 handle_queue_stats_dump_cb(uint32_t queue_id,
3874 struct netdev_queue_stats *stats,
3877 struct queue_stats_cbdata *cbdata = cbdata_;
3879 put_queue_stats(cbdata, queue_id, stats);
3883 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3884 struct queue_stats_cbdata *cbdata)
3886 cbdata->ofport = port;
3887 if (queue_id == OFPQ_ALL) {
3888 netdev_dump_queue_stats(port->netdev,
3889 handle_queue_stats_dump_cb, cbdata);
3891 struct netdev_queue_stats stats;
3893 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3894 put_queue_stats(cbdata, queue_id, &stats);
3900 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3902 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3903 const struct ofp_queue_stats_request *qsr;
3904 struct queue_stats_cbdata cbdata;
3905 struct ofport *port;
3906 unsigned int port_no;
3909 qsr = ofputil_stats_body(oh);
3911 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3914 COVERAGE_INC(ofproto_queue_req);
3916 cbdata.ofconn = ofconn;
3917 cbdata.msg = start_ofp_stats_reply(oh, 128);
3919 port_no = ntohs(qsr->port_no);
3920 queue_id = ntohl(qsr->queue_id);
3921 if (port_no == OFPP_ALL) {
3922 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3923 handle_queue_stats_for_port(port, queue_id, &cbdata);
3925 } else if (port_no < ofproto->max_ports) {
3926 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3928 handle_queue_stats_for_port(port, queue_id, &cbdata);
3931 ofpbuf_delete(cbdata.msg);
3932 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3934 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3939 /* Updates 'facet''s used time. Caller is responsible for calling
3940 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3942 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3945 if (used > facet->used) {
3947 if (used > facet->rule->used) {
3948 facet->rule->used = used;
3950 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3954 /* Folds the statistics from 'stats' into the counters in 'facet'.
3956 * Because of the meaning of a facet's counters, it only makes sense to do this
3957 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3958 * packet that was sent by hand or if it represents statistics that have been
3959 * cleared out of the datapath. */
3961 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3962 const struct dpif_flow_stats *stats)
3964 if (stats->n_packets || stats->used > facet->used) {
3965 facet_update_time(ofproto, facet, stats->used);
3966 facet->packet_count += stats->n_packets;
3967 facet->byte_count += stats->n_bytes;
3968 facet_push_stats(ofproto, facet);
3969 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3974 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3976 uint64_t rs_packets, rs_bytes;
3978 assert(facet->packet_count >= facet->rs_packet_count);
3979 assert(facet->byte_count >= facet->rs_byte_count);
3980 assert(facet->used >= facet->rs_used);
3982 rs_packets = facet->packet_count - facet->rs_packet_count;
3983 rs_bytes = facet->byte_count - facet->rs_byte_count;
3985 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3986 facet->rs_packet_count = facet->packet_count;
3987 facet->rs_byte_count = facet->byte_count;
3988 facet->rs_used = facet->used;
3990 flow_push_stats(ofproto, facet->rule, &facet->flow,
3991 rs_packets, rs_bytes, facet->used);
3995 struct ofproto_push {
3996 struct action_xlate_ctx ctx;
4003 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4005 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
4008 rule->packet_count += push->packets;
4009 rule->byte_count += push->bytes;
4010 rule->used = MAX(push->used, rule->used);
4014 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4015 * 'rule''s actions. */
4017 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
4018 struct flow *flow, uint64_t packets, uint64_t bytes,
4021 struct ofproto_push push;
4023 push.packets = packets;
4027 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
4028 push.ctx.resubmit_hook = push_resubmit;
4029 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
4032 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4033 * in which no matching flow already exists in the flow table.
4035 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4036 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
4037 * OpenFlow error code as encoded by ofp_mkerr() on failure.
4039 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4042 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
4044 struct ofproto *p = ofconn_get_ofproto(ofconn);
4045 struct ofpbuf *packet;
4050 if (fm->flags & OFPFF_CHECK_OVERLAP
4051 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
4052 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
4056 if (fm->buffer_id != UINT32_MAX) {
4057 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
4061 in_port = UINT16_MAX;
4064 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
4065 fm->idle_timeout, fm->hard_timeout, fm->cookie,
4066 fm->flags & OFPFF_SEND_FLOW_REM);
4067 rule_insert(p, rule);
4069 rule_execute(p, rule, in_port, packet);
4074 static struct rule *
4075 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
4077 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
4081 send_buffered_packet(struct ofconn *ofconn,
4082 struct rule *rule, uint32_t buffer_id)
4084 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4085 struct ofpbuf *packet;
4089 if (buffer_id == UINT32_MAX) {
4093 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
4098 rule_execute(ofproto, rule, in_port, packet);
4103 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4105 struct modify_flows_cbdata {
4106 struct ofproto *ofproto;
4107 const struct flow_mod *fm;
4111 static int modify_flow(struct ofproto *, const struct flow_mod *,
4114 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4115 * encoded by ofp_mkerr() on failure.
4117 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4120 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
4122 struct ofproto *p = ofconn_get_ofproto(ofconn);
4123 struct rule *match = NULL;
4124 struct cls_cursor cursor;
4127 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4128 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4129 if (!rule_is_hidden(rule)) {
4131 modify_flow(p, fm, rule);
4136 /* This credits the packet to whichever flow happened to match last.
4137 * That's weird. Maybe we should do a lookup for the flow that
4138 * actually matches the packet? Who knows. */
4139 send_buffered_packet(ofconn, match, fm->buffer_id);
4142 return add_flow(ofconn, fm);
4146 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4147 * code as encoded by ofp_mkerr() on failure.
4149 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4152 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4154 struct ofproto *p = ofconn_get_ofproto(ofconn);
4155 struct rule *rule = find_flow_strict(p, fm);
4156 if (rule && !rule_is_hidden(rule)) {
4157 modify_flow(p, fm, rule);
4158 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4160 return add_flow(ofconn, fm);
4164 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4165 * been identified as a flow in 'p''s flow table to be modified, by changing
4166 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4167 * ofp_action[] structures). */
4169 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4171 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4173 rule->flow_cookie = fm->cookie;
4175 /* If the actions are the same, do nothing. */
4176 if (fm->n_actions == rule->n_actions
4178 || !memcmp(fm->actions, rule->actions, actions_len))) {
4182 /* Replace actions. */
4183 free(rule->actions);
4184 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4185 rule->n_actions = fm->n_actions;
4187 p->need_revalidate = true;
4192 /* OFPFC_DELETE implementation. */
4194 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4196 /* Implements OFPFC_DELETE. */
4198 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4200 struct rule *rule, *next_rule;
4201 struct cls_cursor cursor;
4203 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4204 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4205 delete_flow(p, rule, htons(fm->out_port));
4209 /* Implements OFPFC_DELETE_STRICT. */
4211 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4213 struct rule *rule = find_flow_strict(p, fm);
4215 delete_flow(p, rule, htons(fm->out_port));
4219 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4220 * been identified as a flow to delete from 'p''s flow table, by deleting the
4221 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4224 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4225 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4226 * specified 'out_port'. */
4228 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4230 if (rule_is_hidden(rule)) {
4234 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4238 rule_send_removed(p, rule, OFPRR_DELETE);
4239 rule_remove(p, rule);
4243 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4245 struct ofproto *p = ofconn_get_ofproto(ofconn);
4249 error = reject_slave_controller(ofconn, "flow_mod");
4254 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4259 /* We do not support the emergency flow cache. It will hopefully get
4260 * dropped from OpenFlow in the near future. */
4261 if (fm.flags & OFPFF_EMERG) {
4262 /* There isn't a good fit for an error code, so just state that the
4263 * flow table is full. */
4264 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4267 error = validate_actions(fm.actions, fm.n_actions,
4268 &fm.cr.flow, p->max_ports);
4273 switch (fm.command) {
4275 return add_flow(ofconn, &fm);
4278 return modify_flows_loose(ofconn, &fm);
4280 case OFPFC_MODIFY_STRICT:
4281 return modify_flow_strict(ofconn, &fm);
4284 delete_flows_loose(p, &fm);
4287 case OFPFC_DELETE_STRICT:
4288 delete_flow_strict(p, &fm);
4292 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4297 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4299 const struct nxt_tun_id_cookie *msg
4300 = (const struct nxt_tun_id_cookie *) oh;
4302 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4307 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4309 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4310 struct nx_role_request *reply;
4314 if (ofconn->type != OFCONN_PRIMARY) {
4315 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4317 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4320 role = ntohl(nrr->role);
4321 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4322 && role != NX_ROLE_SLAVE) {
4323 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4325 /* There's no good error code for this. */
4326 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4329 if (role == NX_ROLE_MASTER) {
4330 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4331 struct ofconn *other;
4333 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
4334 if (other->role == NX_ROLE_MASTER) {
4335 other->role = NX_ROLE_SLAVE;
4339 ofconn->role = role;
4341 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4342 reply->role = htonl(role);
4343 queue_tx(buf, ofconn, ofconn->reply_counter);
4349 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4351 const struct nxt_set_flow_format *msg
4352 = (const struct nxt_set_flow_format *) oh;
4355 format = ntohl(msg->format);
4356 if (format == NXFF_OPENFLOW10
4357 || format == NXFF_TUN_ID_FROM_COOKIE
4358 || format == NXFF_NXM) {
4359 ofconn->flow_format = format;
4362 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4367 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4369 struct ofp_header *ob;
4372 /* Currently, everything executes synchronously, so we can just
4373 * immediately send the barrier reply. */
4374 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4375 queue_tx(buf, ofconn, ofconn->reply_counter);
4380 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4382 const struct ofp_header *oh = msg->data;
4383 const struct ofputil_msg_type *type;
4386 error = ofputil_decode_msg_type(oh, &type);
4391 switch (ofputil_msg_type_code(type)) {
4392 /* OpenFlow requests. */
4393 case OFPUTIL_OFPT_ECHO_REQUEST:
4394 return handle_echo_request(ofconn, oh);
4396 case OFPUTIL_OFPT_FEATURES_REQUEST:
4397 return handle_features_request(ofconn, oh);
4399 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4400 return handle_get_config_request(ofconn, oh);
4402 case OFPUTIL_OFPT_SET_CONFIG:
4403 return handle_set_config(ofconn, msg->data);
4405 case OFPUTIL_OFPT_PACKET_OUT:
4406 return handle_packet_out(ofconn, oh);
4408 case OFPUTIL_OFPT_PORT_MOD:
4409 return handle_port_mod(ofconn, oh);
4411 case OFPUTIL_OFPT_FLOW_MOD:
4412 return handle_flow_mod(ofconn, oh);
4414 case OFPUTIL_OFPT_BARRIER_REQUEST:
4415 return handle_barrier_request(ofconn, oh);
4417 /* OpenFlow replies. */
4418 case OFPUTIL_OFPT_ECHO_REPLY:
4421 /* Nicira extension requests. */
4422 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4423 return handle_tun_id_from_cookie(ofconn, oh);
4425 case OFPUTIL_NXT_ROLE_REQUEST:
4426 return handle_role_request(ofconn, oh);
4428 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4429 return handle_nxt_set_flow_format(ofconn, oh);
4431 case OFPUTIL_NXT_FLOW_MOD:
4432 return handle_flow_mod(ofconn, oh);
4434 /* OpenFlow statistics requests. */
4435 case OFPUTIL_OFPST_DESC_REQUEST:
4436 return handle_desc_stats_request(ofconn, oh);
4438 case OFPUTIL_OFPST_FLOW_REQUEST:
4439 return handle_flow_stats_request(ofconn, oh);
4441 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4442 return handle_aggregate_stats_request(ofconn, oh);
4444 case OFPUTIL_OFPST_TABLE_REQUEST:
4445 return handle_table_stats_request(ofconn, oh);
4447 case OFPUTIL_OFPST_PORT_REQUEST:
4448 return handle_port_stats_request(ofconn, oh);
4450 case OFPUTIL_OFPST_QUEUE_REQUEST:
4451 return handle_queue_stats_request(ofconn, oh);
4453 /* Nicira extension statistics requests. */
4454 case OFPUTIL_NXST_FLOW_REQUEST:
4455 return handle_nxst_flow(ofconn, oh);
4457 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4458 return handle_nxst_aggregate(ofconn, oh);
4460 case OFPUTIL_INVALID:
4461 case OFPUTIL_OFPT_HELLO:
4462 case OFPUTIL_OFPT_ERROR:
4463 case OFPUTIL_OFPT_FEATURES_REPLY:
4464 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4465 case OFPUTIL_OFPT_PACKET_IN:
4466 case OFPUTIL_OFPT_FLOW_REMOVED:
4467 case OFPUTIL_OFPT_PORT_STATUS:
4468 case OFPUTIL_OFPT_BARRIER_REPLY:
4469 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4470 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4471 case OFPUTIL_OFPST_DESC_REPLY:
4472 case OFPUTIL_OFPST_FLOW_REPLY:
4473 case OFPUTIL_OFPST_QUEUE_REPLY:
4474 case OFPUTIL_OFPST_PORT_REPLY:
4475 case OFPUTIL_OFPST_TABLE_REPLY:
4476 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4477 case OFPUTIL_NXT_ROLE_REPLY:
4478 case OFPUTIL_NXT_FLOW_REMOVED:
4479 case OFPUTIL_NXST_FLOW_REPLY:
4480 case OFPUTIL_NXST_AGGREGATE_REPLY:
4482 if (VLOG_IS_WARN_ENABLED()) {
4483 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4484 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4487 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4488 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4490 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4496 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4498 int error = handle_openflow__(ofconn, ofp_msg);
4500 send_error_oh(ofconn, ofp_msg->data, error);
4502 COVERAGE_INC(ofproto_recv_openflow);
4506 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4508 struct facet *facet;
4511 /* Obtain in_port and tun_id, at least. */
4512 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4514 /* Set header pointers in 'flow'. */
4515 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4517 if (cfm_should_process_flow(&flow)) {
4518 ofproto_process_cfm(p, &flow, upcall->packet);
4519 ofpbuf_delete(upcall->packet);
4521 } else if (p->ofhooks->special_cb
4522 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4523 ofpbuf_delete(upcall->packet);
4527 /* Check with in-band control to see if this packet should be sent
4528 * to the local port regardless of the flow table. */
4529 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4530 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
4533 facet = facet_lookup_valid(p, &flow);
4535 struct rule *rule = rule_lookup(p, &flow);
4537 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4538 struct ofport *port = get_port(p, flow.in_port);
4540 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4541 COVERAGE_INC(ofproto_no_packet_in);
4542 /* XXX install 'drop' flow entry */
4543 ofpbuf_delete(upcall->packet);
4547 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4551 COVERAGE_INC(ofproto_packet_in);
4552 send_packet_in(p, upcall, &flow, false);
4556 facet = facet_create(p, rule, &flow, upcall->packet);
4557 } else if (!facet->may_install) {
4558 /* The facet is not installable, that is, we need to process every
4559 * packet, so process the current packet's actions into 'facet'. */
4560 facet_make_actions(p, facet, upcall->packet);
4563 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4565 * Extra-special case for fail-open mode.
4567 * We are in fail-open mode and the packet matched the fail-open rule,
4568 * but we are connected to a controller too. We should send the packet
4569 * up to the controller in the hope that it will try to set up a flow
4570 * and thereby allow us to exit fail-open.
4572 * See the top-level comment in fail-open.c for more information.
4574 send_packet_in(p, upcall, &flow, true);
4577 facet_execute(p, facet, upcall->packet);
4578 facet_install(p, facet, false);
4582 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4586 switch (upcall->type) {
4587 case DPIF_UC_ACTION:
4588 COVERAGE_INC(ofproto_ctlr_action);
4589 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4590 send_packet_in(p, upcall, &flow, false);
4593 case DPIF_UC_SAMPLE:
4595 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4596 ofproto_sflow_received(p->sflow, upcall, &flow);
4598 ofpbuf_delete(upcall->packet);
4602 handle_miss_upcall(p, upcall);
4605 case DPIF_N_UC_TYPES:
4607 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4612 /* Flow expiration. */
4614 static int ofproto_dp_max_idle(const struct ofproto *);
4615 static void ofproto_update_stats(struct ofproto *);
4616 static void rule_expire(struct ofproto *, struct rule *);
4617 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4619 /* This function is called periodically by ofproto_run(). Its job is to
4620 * collect updates for the flows that have been installed into the datapath,
4621 * most importantly when they last were used, and then use that information to
4622 * expire flows that have not been used recently.
4624 * Returns the number of milliseconds after which it should be called again. */
4626 ofproto_expire(struct ofproto *ofproto)
4628 struct rule *rule, *next_rule;
4629 struct cls_cursor cursor;
4632 /* Update stats for each flow in the datapath. */
4633 ofproto_update_stats(ofproto);
4635 /* Expire facets that have been idle too long. */
4636 dp_max_idle = ofproto_dp_max_idle(ofproto);
4637 ofproto_expire_facets(ofproto, dp_max_idle);
4639 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4640 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4641 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4642 rule_expire(ofproto, rule);
4645 /* Let the hook know that we're at a stable point: all outstanding data
4646 * in existing flows has been accounted to the account_cb. Thus, the
4647 * hook can now reasonably do operations that depend on having accurate
4648 * flow volume accounting (currently, that's just bond rebalancing). */
4649 if (ofproto->ofhooks->account_checkpoint_cb) {
4650 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4653 return MIN(dp_max_idle, 1000);
4656 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4658 * This function also pushes statistics updates to rules which each facet
4659 * resubmits into. Generally these statistics will be accurate. However, if a
4660 * facet changes the rule it resubmits into at some time in between
4661 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4662 * old rule will be incorrectly attributed to the new rule. This could be
4663 * avoided by calling ofproto_update_stats() whenever rules are created or
4664 * deleted. However, the performance impact of making so many calls to the
4665 * datapath do not justify the benefit of having perfectly accurate statistics.
4668 ofproto_update_stats(struct ofproto *p)
4670 const struct dpif_flow_stats *stats;
4671 struct dpif_flow_dump dump;
4672 const struct nlattr *key;
4675 dpif_flow_dump_start(&dump, p->dpif);
4676 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4677 struct facet *facet;
4680 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4684 odp_flow_key_format(key, key_len, &s);
4685 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4691 facet = facet_find(p, &flow);
4693 if (facet && facet->installed) {
4695 if (stats->n_packets >= facet->dp_packet_count) {
4696 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4698 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4701 if (stats->n_bytes >= facet->dp_byte_count) {
4702 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4704 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4707 facet->dp_packet_count = stats->n_packets;
4708 facet->dp_byte_count = stats->n_bytes;
4710 facet_update_time(p, facet, stats->used);
4711 facet_account(p, facet, stats->n_bytes);
4712 facet_push_stats(p, facet);
4714 /* There's a flow in the datapath that we know nothing about.
4716 COVERAGE_INC(ofproto_unexpected_rule);
4717 dpif_flow_del(p->dpif, key, key_len, NULL);
4720 dpif_flow_dump_done(&dump);
4723 /* Calculates and returns the number of milliseconds of idle time after which
4724 * facets should expire from the datapath and we should fold their statistics
4725 * into their parent rules in userspace. */
4727 ofproto_dp_max_idle(const struct ofproto *ofproto)
4730 * Idle time histogram.
4732 * Most of the time a switch has a relatively small number of facets. When
4733 * this is the case we might as well keep statistics for all of them in
4734 * userspace and to cache them in the kernel datapath for performance as
4737 * As the number of facets increases, the memory required to maintain
4738 * statistics about them in userspace and in the kernel becomes
4739 * significant. However, with a large number of facets it is likely that
4740 * only a few of them are "heavy hitters" that consume a large amount of
4741 * bandwidth. At this point, only heavy hitters are worth caching in the
4742 * kernel and maintaining in userspaces; other facets we can discard.
4744 * The technique used to compute the idle time is to build a histogram with
4745 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4746 * that is installed in the kernel gets dropped in the appropriate bucket.
4747 * After the histogram has been built, we compute the cutoff so that only
4748 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4749 * cached. At least the most-recently-used bucket of facets is kept, so
4750 * actually an arbitrary number of facets can be kept in any given
4751 * expiration run (though the next run will delete most of those unless
4752 * they receive additional data).
4754 * This requires a second pass through the facets, in addition to the pass
4755 * made by ofproto_update_stats(), because the former function never looks
4756 * at uninstallable facets.
4758 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4759 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4760 int buckets[N_BUCKETS] = { 0 };
4761 struct facet *facet;
4766 total = hmap_count(&ofproto->facets);
4767 if (total <= 1000) {
4768 return N_BUCKETS * BUCKET_WIDTH;
4771 /* Build histogram. */
4773 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4774 long long int idle = now - facet->used;
4775 int bucket = (idle <= 0 ? 0
4776 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4777 : (unsigned int) idle / BUCKET_WIDTH);
4781 /* Find the first bucket whose flows should be expired. */
4782 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4783 if (buckets[bucket]) {
4786 subtotal += buckets[bucket++];
4787 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4792 if (VLOG_IS_DBG_ENABLED()) {
4796 ds_put_cstr(&s, "keep");
4797 for (i = 0; i < N_BUCKETS; i++) {
4799 ds_put_cstr(&s, ", drop");
4802 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4805 VLOG_INFO("%s: %s (msec:count)",
4806 dpif_name(ofproto->dpif), ds_cstr(&s));
4810 return bucket * BUCKET_WIDTH;
4814 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4816 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4817 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4818 struct ofexpired expired;
4820 if (facet->installed) {
4821 struct dpif_flow_stats stats;
4823 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4825 facet_update_stats(ofproto, facet, &stats);
4828 expired.flow = facet->flow;
4829 expired.packet_count = facet->packet_count;
4830 expired.byte_count = facet->byte_count;
4831 expired.used = facet->used;
4832 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4837 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4839 long long int cutoff = time_msec() - dp_max_idle;
4840 struct facet *facet, *next_facet;
4842 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4843 facet_active_timeout(ofproto, facet);
4844 if (facet->used < cutoff) {
4845 facet_remove(ofproto, facet);
4850 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4851 * then delete it entirely. */
4853 rule_expire(struct ofproto *ofproto, struct rule *rule)
4855 struct facet *facet, *next_facet;
4859 /* Has 'rule' expired? */
4861 if (rule->hard_timeout
4862 && now > rule->created + rule->hard_timeout * 1000) {
4863 reason = OFPRR_HARD_TIMEOUT;
4864 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4865 && now >rule->used + rule->idle_timeout * 1000) {
4866 reason = OFPRR_IDLE_TIMEOUT;
4871 COVERAGE_INC(ofproto_expired);
4873 /* Update stats. (This is a no-op if the rule expired due to an idle
4874 * timeout, because that only happens when the rule has no facets left.) */
4875 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4876 facet_remove(ofproto, facet);
4879 /* Get rid of the rule. */
4880 if (!rule_is_hidden(rule)) {
4881 rule_send_removed(ofproto, rule, reason);
4883 rule_remove(ofproto, rule);
4887 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4889 struct ofputil_flow_removed fr;
4890 struct ofconn *ofconn;
4892 if (!rule->send_flow_removed) {
4897 fr.cookie = rule->flow_cookie;
4899 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4900 fr.idle_timeout = rule->idle_timeout;
4901 fr.packet_count = rule->packet_count;
4902 fr.byte_count = rule->byte_count;
4904 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4905 if (!rconn_is_connected(ofconn->rconn)
4906 || !ofconn_receives_async_msgs(ofconn)) {
4910 /* Account flow expirations under ofconn->reply_counter, the counter
4911 * for replies to OpenFlow requests. That works because preventing
4912 * OpenFlow requests from being processed also prevents new flows from
4913 * being added (and expiring). (It also prevents processing OpenFlow
4914 * requests that would not add new flows, so it is imperfect.) */
4915 queue_tx(ofputil_encode_flow_removed(&fr, ofconn->flow_format),
4916 ofconn, ofconn->reply_counter);
4920 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4921 * The returned statistics include statistics for all of 'rule''s facets. */
4923 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4926 struct facet *facet;
4928 /* Start from historical data for 'rule' itself that are no longer tracked
4929 * in facets. This counts, for example, facets that have expired. */
4930 p = rule->packet_count;
4931 b = rule->byte_count;
4933 /* Add any statistics that are tracked by facets. This includes
4934 * statistical data recently updated by ofproto_update_stats() as well as
4935 * stats for packets that were executed "by hand" via dpif_execute(). */
4936 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4937 p += facet->packet_count;
4938 b += facet->byte_count;
4945 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4947 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4949 struct ofconn *ofconn = ofconn_;
4951 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4952 ofconn->packet_in_counter, 100);
4955 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4956 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4957 * scheduler for sending.
4959 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4960 * Otherwise, ownership is transferred to this function. */
4962 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4963 const struct flow *flow, bool clone)
4965 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4966 struct ofputil_packet_in pin;
4969 /* Figure out the easy parts. */
4970 pin.packet = upcall->packet;
4971 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4972 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4974 /* Get OpenFlow buffer_id. */
4975 if (upcall->type == DPIF_UC_ACTION) {
4976 pin.buffer_id = UINT32_MAX;
4977 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4978 pin.buffer_id = pktbuf_get_null();
4979 } else if (!ofconn->pktbuf) {
4980 pin.buffer_id = UINT32_MAX;
4982 pin.buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet,
4986 /* Figure out how much of the packet to send. */
4987 pin.send_len = upcall->packet->size;
4988 if (pin.buffer_id != UINT32_MAX) {
4989 pin.send_len = MIN(pin.send_len, ofconn->miss_send_len);
4991 if (upcall->type == DPIF_UC_ACTION) {
4992 pin.send_len = MIN(pin.send_len, upcall->userdata);
4995 /* Make OFPT_PACKET_IN and hand over to packet scheduler. It might
4996 * immediately call into do_send_packet_in() or it might buffer it for a
4997 * while (until a later call to pinsched_run()). */
4998 msg = ofputil_encode_packet_in(&pin, clone ? NULL : upcall->packet);
4999 pinsched_send(ofconn->schedulers[upcall->type == DPIF_UC_MISS ? 0 : 1],
5000 flow->in_port, msg, do_send_packet_in, ofconn);
5003 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
5004 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
5005 * their individual configurations.
5007 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
5008 * Otherwise, ownership is transferred to this function. */
5010 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
5011 const struct flow *flow, bool clone)
5013 struct ofconn *ofconn, *prev;
5016 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5017 if (ofconn_receives_async_msgs(ofconn)) {
5019 schedule_packet_in(prev, upcall, flow, true);
5025 schedule_packet_in(prev, upcall, flow, clone);
5026 } else if (!clone) {
5027 ofpbuf_delete(upcall->packet);
5032 pick_datapath_id(const struct ofproto *ofproto)
5034 const struct ofport *port;
5036 port = get_port(ofproto, ODPP_LOCAL);
5038 uint8_t ea[ETH_ADDR_LEN];
5041 error = netdev_get_etheraddr(port->netdev, ea);
5043 return eth_addr_to_uint64(ea);
5045 VLOG_WARN("could not get MAC address for %s (%s)",
5046 netdev_get_name(port->netdev), strerror(error));
5048 return ofproto->fallback_dpid;
5052 pick_fallback_dpid(void)
5054 uint8_t ea[ETH_ADDR_LEN];
5055 eth_addr_nicira_random(ea);
5056 return eth_addr_to_uint64(ea);
5060 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
5061 void *aux OVS_UNUSED)
5063 const struct shash_node *node;
5067 SHASH_FOR_EACH (node, &all_ofprotos) {
5068 ds_put_format(&results, "%s\n", node->name);
5070 unixctl_command_reply(conn, 200, ds_cstr(&results));
5071 ds_destroy(&results);
5074 struct ofproto_trace {
5075 struct action_xlate_ctx ctx;
5081 trace_format_rule(struct ds *result, int level, const struct rule *rule)
5083 ds_put_char_multiple(result, '\t', level);
5085 ds_put_cstr(result, "No match\n");
5089 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
5090 ntohll(rule->flow_cookie));
5091 cls_rule_format(&rule->cr, result);
5092 ds_put_char(result, '\n');
5094 ds_put_char_multiple(result, '\t', level);
5095 ds_put_cstr(result, "OpenFlow ");
5096 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
5097 rule->n_actions * sizeof *rule->actions);
5098 ds_put_char(result, '\n');
5102 trace_format_flow(struct ds *result, int level, const char *title,
5103 struct ofproto_trace *trace)
5105 ds_put_char_multiple(result, '\t', level);
5106 ds_put_format(result, "%s: ", title);
5107 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5108 ds_put_cstr(result, "unchanged");
5110 flow_format(result, &trace->ctx.flow);
5111 trace->flow = trace->ctx.flow;
5113 ds_put_char(result, '\n');
5117 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
5119 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5120 struct ds *result = trace->result;
5122 ds_put_char(result, '\n');
5123 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5124 trace_format_rule(result, ctx->recurse + 1, rule);
5128 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5129 void *aux OVS_UNUSED)
5131 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5132 char *args = xstrdup(args_);
5133 char *save_ptr = NULL;
5134 struct ofproto *ofproto;
5135 struct ofpbuf packet;
5143 ofpbuf_init(&packet, strlen(args) / 2);
5146 dpname = strtok_r(args, " ", &save_ptr);
5147 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5148 in_port_s = strtok_r(NULL, " ", &save_ptr);
5149 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5150 if (!dpname || !in_port_s || !packet_s) {
5151 unixctl_command_reply(conn, 501, "Bad command syntax");
5155 ofproto = shash_find_data(&all_ofprotos, dpname);
5157 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5162 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5163 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5165 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5166 packet_s += strspn(packet_s, " ");
5167 if (*packet_s != '\0') {
5168 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5171 if (packet.size < ETH_HEADER_LEN) {
5172 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5176 ds_put_cstr(&result, "Packet: ");
5177 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5178 ds_put_cstr(&result, s);
5181 flow_extract(&packet, tun_id, in_port, &flow);
5182 ds_put_cstr(&result, "Flow: ");
5183 flow_format(&result, &flow);
5184 ds_put_char(&result, '\n');
5186 rule = rule_lookup(ofproto, &flow);
5187 trace_format_rule(&result, 0, rule);
5189 struct ofproto_trace trace;
5190 struct ofpbuf *odp_actions;
5192 trace.result = &result;
5194 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5195 trace.ctx.resubmit_hook = trace_resubmit;
5196 odp_actions = xlate_actions(&trace.ctx,
5197 rule->actions, rule->n_actions);
5199 ds_put_char(&result, '\n');
5200 trace_format_flow(&result, 0, "Final flow", &trace);
5201 ds_put_cstr(&result, "Datapath actions: ");
5202 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5203 ofpbuf_delete(odp_actions);
5206 unixctl_command_reply(conn, 200, ds_cstr(&result));
5209 ds_destroy(&result);
5210 ofpbuf_uninit(&packet);
5215 ofproto_unixctl_init(void)
5217 static bool registered;
5223 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5224 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5228 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5229 struct ofpbuf *odp_actions, tag_type *tags,
5230 uint16_t *nf_output_iface, void *ofproto_)
5232 struct ofproto *ofproto = ofproto_;
5233 struct mac_entry *dst_mac;
5235 /* Drop frames for reserved multicast addresses. */
5236 if (eth_addr_is_reserved(flow->dl_dst)) {
5240 /* Learn source MAC (but don't try to learn from revalidation). */
5242 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
5243 struct mac_entry *src_mac;
5245 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
5246 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
5247 /* The log messages here could actually be useful in debugging,
5248 * so keep the rate limit relatively high. */
5249 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5250 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5251 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5253 ofproto_revalidate(ofproto,
5254 mac_learning_changed(ofproto->ml, src_mac));
5255 src_mac->port.i = flow->in_port;
5259 /* Determine output port. */
5260 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
5262 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5263 nf_output_iface, odp_actions);
5265 int out_port = dst_mac->port.i;
5266 if (out_port != flow->in_port) {
5267 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5268 *nf_output_iface = out_port;
5277 static const struct ofhooks default_ofhooks = {
5278 default_normal_ofhook_cb,