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 *);
352 static enum nx_flow_format ofconn_get_flow_format(struct ofconn *);
353 static void ofconn_set_flow_format(struct ofconn *, enum nx_flow_format);
354 static int ofconn_get_miss_send_len(const struct ofconn *);
355 static void ofconn_set_miss_send_len(struct ofconn *, int miss_send_len);
357 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
358 struct rconn_packet_counter *counter);
360 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
361 const struct flow *, bool clone);
362 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
366 uint64_t datapath_id; /* Datapath ID. */
367 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
368 char *mfr_desc; /* Manufacturer. */
369 char *hw_desc; /* Hardware. */
370 char *sw_desc; /* Software version. */
371 char *serial_desc; /* Serial number. */
372 char *dp_desc; /* Datapath description. */
376 struct netdev_monitor *netdev_monitor;
377 struct hmap ports; /* Contains "struct ofport"s. */
378 struct shash port_by_name;
382 struct fail_open *fail_open;
383 struct netflow *netflow;
384 struct ofproto_sflow *sflow;
386 /* In-band control. */
387 struct in_band *in_band;
388 long long int next_in_band_update;
389 struct sockaddr_in *extra_in_band_remotes;
390 size_t n_extra_remotes;
394 struct classifier cls;
395 long long int next_expiration;
399 bool need_revalidate;
400 struct tag_set revalidate_set;
402 /* OpenFlow connections. */
403 struct hmap controllers; /* Controller "struct ofconn"s. */
404 struct list all_conns; /* Contains "struct ofconn"s. */
405 enum ofproto_fail_mode fail_mode;
407 /* OpenFlow listeners. */
408 struct hmap services; /* Contains "struct ofservice"s. */
409 struct pvconn **snoops;
412 /* Hooks for ovs-vswitchd. */
413 const struct ofhooks *ofhooks;
416 /* Used by default ofhooks. */
417 struct mac_learning *ml;
420 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
421 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
423 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
425 static const struct ofhooks default_ofhooks;
427 static uint64_t pick_datapath_id(const struct ofproto *);
428 static uint64_t pick_fallback_dpid(void);
430 static int ofproto_expire(struct ofproto *);
431 static void flow_push_stats(struct ofproto *, const struct rule *,
432 struct flow *, uint64_t packets, uint64_t bytes,
435 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
437 static void handle_openflow(struct ofconn *, struct ofpbuf *);
439 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
440 static void update_port(struct ofproto *, const char *devname);
441 static int init_ports(struct ofproto *);
442 static void reinit_ports(struct ofproto *);
444 static void ofproto_unixctl_init(void);
447 ofproto_create(const char *datapath, const char *datapath_type,
448 const struct ofhooks *ofhooks, void *aux,
449 struct ofproto **ofprotop)
457 ofproto_unixctl_init();
459 /* Connect to datapath and start listening for messages. */
460 error = dpif_open(datapath, datapath_type, &dpif);
462 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
465 error = dpif_recv_set_mask(dpif,
466 ((1u << DPIF_UC_MISS) |
467 (1u << DPIF_UC_ACTION) |
468 (1u << DPIF_UC_SAMPLE)));
470 VLOG_ERR("failed to listen on datapath %s: %s",
471 datapath, strerror(error));
475 dpif_flow_flush(dpif);
476 dpif_recv_purge(dpif);
478 /* Initialize settings. */
479 p = xzalloc(sizeof *p);
480 p->fallback_dpid = pick_fallback_dpid();
481 p->datapath_id = p->fallback_dpid;
482 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
483 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
484 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
485 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
486 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
488 /* Initialize datapath. */
490 p->netdev_monitor = netdev_monitor_create();
491 hmap_init(&p->ports);
492 shash_init(&p->port_by_name);
493 p->max_ports = dpif_get_max_ports(dpif);
495 /* Initialize submodules. */
500 /* Initialize in-band control. */
502 p->in_band_queue = -1;
504 /* Initialize flow table. */
505 classifier_init(&p->cls);
506 p->next_expiration = time_msec() + 1000;
508 /* Initialize facet table. */
509 hmap_init(&p->facets);
510 p->need_revalidate = false;
511 tag_set_init(&p->revalidate_set);
513 /* Initialize OpenFlow connections. */
514 list_init(&p->all_conns);
515 hmap_init(&p->controllers);
516 hmap_init(&p->services);
520 /* Initialize hooks. */
522 p->ofhooks = ofhooks;
526 p->ofhooks = &default_ofhooks;
528 p->ml = mac_learning_create();
531 /* Pick final datapath ID. */
532 p->datapath_id = pick_datapath_id(p);
533 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
535 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
542 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
544 uint64_t old_dpid = p->datapath_id;
545 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
546 if (p->datapath_id != old_dpid) {
547 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
549 /* Force all active connections to reconnect, since there is no way to
550 * notify a controller that the datapath ID has changed. */
551 ofproto_reconnect_controllers(p);
555 /* Creates a new controller in 'ofproto'. Some of the settings are initially
556 * drawn from 'c', but update_controller() needs to be called later to finish
557 * the new ofconn's configuration. */
559 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
561 char *name = ofconn_make_name(ofproto, c->target);
562 struct ofconn *ofconn;
564 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
565 ofconn->pktbuf = pktbuf_create();
566 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
567 rconn_connect(ofconn->rconn, c->target, name);
568 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
569 hash_string(c->target, 0));
574 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
575 * target (this is done by creating new ofconns and deleting old ones), but it
576 * can update the rest of an ofconn's settings. */
578 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
582 ofconn->band = c->band;
584 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
586 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
587 rconn_set_probe_interval(ofconn->rconn, probe_interval);
589 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
593 ofconn_get_target(const struct ofconn *ofconn)
595 return rconn_get_target(ofconn->rconn);
598 static struct ofconn *
599 find_controller_by_target(struct ofproto *ofproto, const char *target)
601 struct ofconn *ofconn;
603 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
604 hash_string(target, 0), &ofproto->controllers) {
605 if (!strcmp(ofconn_get_target(ofconn), target)) {
613 update_in_band_remotes(struct ofproto *ofproto)
615 const struct ofconn *ofconn;
616 struct sockaddr_in *addrs;
617 size_t max_addrs, n_addrs;
620 /* Allocate enough memory for as many remotes as we could possibly have. */
621 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
622 addrs = xmalloc(max_addrs * sizeof *addrs);
625 /* Add all the remotes. */
626 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
627 struct sockaddr_in *sin = &addrs[n_addrs];
629 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
633 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
634 if (sin->sin_addr.s_addr) {
635 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
639 for (i = 0; i < ofproto->n_extra_remotes; i++) {
640 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
643 /* Create or update or destroy in-band. */
645 if (!ofproto->in_band) {
646 in_band_create(ofproto, ofproto->dpif, &ofproto->in_band);
648 if (ofproto->in_band) {
649 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
651 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
652 ofproto->next_in_band_update = time_msec() + 1000;
654 in_band_destroy(ofproto->in_band);
655 ofproto->in_band = NULL;
663 update_fail_open(struct ofproto *p)
665 struct ofconn *ofconn;
667 if (!hmap_is_empty(&p->controllers)
668 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
669 struct rconn **rconns;
673 p->fail_open = fail_open_create(p);
677 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
678 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
679 rconns[n++] = ofconn->rconn;
682 fail_open_set_controllers(p->fail_open, rconns, n);
683 /* p->fail_open takes ownership of 'rconns'. */
685 fail_open_destroy(p->fail_open);
691 ofproto_set_controllers(struct ofproto *p,
692 const struct ofproto_controller *controllers,
693 size_t n_controllers)
695 struct shash new_controllers;
696 struct ofconn *ofconn, *next_ofconn;
697 struct ofservice *ofservice, *next_ofservice;
700 /* Create newly configured controllers and services.
701 * Create a name to ofproto_controller mapping in 'new_controllers'. */
702 shash_init(&new_controllers);
703 for (i = 0; i < n_controllers; i++) {
704 const struct ofproto_controller *c = &controllers[i];
706 if (!vconn_verify_name(c->target)) {
707 if (!find_controller_by_target(p, c->target)) {
708 add_controller(p, c);
710 } else if (!pvconn_verify_name(c->target)) {
711 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
715 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
716 dpif_name(p->dpif), c->target);
720 shash_add_once(&new_controllers, c->target, &controllers[i]);
723 /* Delete controllers that are no longer configured.
724 * Update configuration of all now-existing controllers. */
725 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
726 struct ofproto_controller *c;
728 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
730 ofconn_destroy(ofconn);
732 update_controller(ofconn, c);
736 /* Delete services that are no longer configured.
737 * Update configuration of all now-existing services. */
738 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
739 struct ofproto_controller *c;
741 c = shash_find_data(&new_controllers,
742 pvconn_get_name(ofservice->pvconn));
744 ofservice_destroy(p, ofservice);
746 ofservice_reconfigure(ofservice, c);
750 shash_destroy(&new_controllers);
752 update_in_band_remotes(p);
757 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
759 p->fail_mode = fail_mode;
763 /* Drops the connections between 'ofproto' and all of its controllers, forcing
764 * them to reconnect. */
766 ofproto_reconnect_controllers(struct ofproto *ofproto)
768 struct ofconn *ofconn;
770 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
771 rconn_reconnect(ofconn->rconn);
776 any_extras_changed(const struct ofproto *ofproto,
777 const struct sockaddr_in *extras, size_t n)
781 if (n != ofproto->n_extra_remotes) {
785 for (i = 0; i < n; i++) {
786 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
787 const struct sockaddr_in *new = &extras[i];
789 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
790 old->sin_port != new->sin_port) {
798 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
799 * in-band control should guarantee access, in the same way that in-band
800 * control guarantees access to OpenFlow controllers. */
802 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
803 const struct sockaddr_in *extras, size_t n)
805 if (!any_extras_changed(ofproto, extras, n)) {
809 free(ofproto->extra_in_band_remotes);
810 ofproto->n_extra_remotes = n;
811 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
813 update_in_band_remotes(ofproto);
816 /* Sets the OpenFlow queue used by flows set up by in-band control on
817 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
818 * flows will use the default queue. */
820 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
822 if (queue_id != ofproto->in_band_queue) {
823 ofproto->in_band_queue = queue_id;
824 update_in_band_remotes(ofproto);
829 ofproto_set_desc(struct ofproto *p,
830 const char *mfr_desc, const char *hw_desc,
831 const char *sw_desc, const char *serial_desc,
834 struct ofp_desc_stats *ods;
837 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
838 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
839 sizeof ods->mfr_desc);
842 p->mfr_desc = xstrdup(mfr_desc);
845 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
846 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
847 sizeof ods->hw_desc);
850 p->hw_desc = xstrdup(hw_desc);
853 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
854 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
855 sizeof ods->sw_desc);
858 p->sw_desc = xstrdup(sw_desc);
861 if (strlen(serial_desc) >= sizeof ods->serial_num) {
862 VLOG_WARN("truncating serial_desc, must be less than %zu "
864 sizeof ods->serial_num);
866 free(p->serial_desc);
867 p->serial_desc = xstrdup(serial_desc);
870 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
871 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
872 sizeof ods->dp_desc);
875 p->dp_desc = xstrdup(dp_desc);
880 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
881 const struct svec *svec)
883 struct pvconn **pvconns = *pvconnsp;
884 size_t n_pvconns = *n_pvconnsp;
888 for (i = 0; i < n_pvconns; i++) {
889 pvconn_close(pvconns[i]);
893 pvconns = xmalloc(svec->n * sizeof *pvconns);
895 for (i = 0; i < svec->n; i++) {
896 const char *name = svec->names[i];
897 struct pvconn *pvconn;
900 error = pvconn_open(name, &pvconn);
902 pvconns[n_pvconns++] = pvconn;
904 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
912 *n_pvconnsp = n_pvconns;
918 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
920 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
924 ofproto_set_netflow(struct ofproto *ofproto,
925 const struct netflow_options *nf_options)
927 if (nf_options && nf_options->collectors.n) {
928 if (!ofproto->netflow) {
929 ofproto->netflow = netflow_create();
931 return netflow_set_options(ofproto->netflow, nf_options);
933 netflow_destroy(ofproto->netflow);
934 ofproto->netflow = NULL;
940 ofproto_set_sflow(struct ofproto *ofproto,
941 const struct ofproto_sflow_options *oso)
943 struct ofproto_sflow *os = ofproto->sflow;
946 struct ofport *ofport;
948 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
949 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
950 ofproto_sflow_add_port(os, ofport->odp_port,
951 netdev_get_name(ofport->netdev));
954 ofproto_sflow_set_options(os, oso);
956 ofproto_sflow_destroy(os);
957 ofproto->sflow = NULL;
961 /* Connectivity Fault Management configuration. */
963 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
965 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
967 struct ofport *ofport = get_port(ofproto, port_no);
968 if (ofport && ofport->cfm){
969 cfm_destroy(ofport->cfm);
974 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
975 * basic configuration from the configuration members in 'cfm', and the set of
976 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
977 * Ignores the statistics members of 'cfm'.
979 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
981 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
982 const struct cfm *cfm,
983 const uint16_t *remote_mps, size_t n_remote_mps)
985 struct ofport *ofport;
987 ofport = get_port(ofproto, port_no);
989 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
990 dpif_name(ofproto->dpif), port_no);
995 ofport->cfm = cfm_create();
998 ofport->cfm->mpid = cfm->mpid;
999 ofport->cfm->interval = cfm->interval;
1000 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
1002 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
1004 if (!cfm_configure(ofport->cfm)) {
1005 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
1006 dpif_name(ofproto->dpif), port_no,
1007 netdev_get_name(ofport->netdev));
1008 cfm_destroy(ofport->cfm);
1013 /* Returns the connectivity fault management object associated with 'port_no'
1014 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
1015 * 'port_no' or if that port does not have CFM configured. The caller must not
1016 * modify or destroy the returned object. */
1018 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
1020 struct ofport *ofport = get_port(ofproto, port_no);
1021 return ofport ? ofport->cfm : NULL;
1025 ofproto_get_datapath_id(const struct ofproto *ofproto)
1027 return ofproto->datapath_id;
1031 ofproto_has_primary_controller(const struct ofproto *ofproto)
1033 return !hmap_is_empty(&ofproto->controllers);
1036 enum ofproto_fail_mode
1037 ofproto_get_fail_mode(const struct ofproto *p)
1039 return p->fail_mode;
1043 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1047 for (i = 0; i < ofproto->n_snoops; i++) {
1048 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1053 ofproto_destroy(struct ofproto *p)
1055 struct ofservice *ofservice, *next_ofservice;
1056 struct ofconn *ofconn, *next_ofconn;
1057 struct ofport *ofport, *next_ofport;
1064 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1066 /* Destroy fail-open and in-band early, since they touch the classifier. */
1067 fail_open_destroy(p->fail_open);
1068 p->fail_open = NULL;
1070 in_band_destroy(p->in_band);
1072 free(p->extra_in_band_remotes);
1074 ofproto_flush_flows(p);
1075 classifier_destroy(&p->cls);
1076 hmap_destroy(&p->facets);
1078 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1079 ofconn_destroy(ofconn);
1081 hmap_destroy(&p->controllers);
1083 dpif_close(p->dpif);
1084 netdev_monitor_destroy(p->netdev_monitor);
1085 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1086 hmap_remove(&p->ports, &ofport->hmap_node);
1087 ofport_free(ofport);
1089 shash_destroy(&p->port_by_name);
1091 netflow_destroy(p->netflow);
1092 ofproto_sflow_destroy(p->sflow);
1094 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1095 ofservice_destroy(p, ofservice);
1097 hmap_destroy(&p->services);
1099 for (i = 0; i < p->n_snoops; i++) {
1100 pvconn_close(p->snoops[i]);
1104 mac_learning_destroy(p->ml);
1109 free(p->serial_desc);
1112 hmap_destroy(&p->ports);
1118 ofproto_run(struct ofproto *p)
1120 int error = ofproto_run1(p);
1122 error = ofproto_run2(p, false);
1128 process_port_change(struct ofproto *ofproto, int error, char *devname)
1130 if (error == ENOBUFS) {
1131 reinit_ports(ofproto);
1132 } else if (!error) {
1133 update_port(ofproto, devname);
1138 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1139 * means that 'ofconn' is more interesting for monitoring than a lower return
1142 snoop_preference(const struct ofconn *ofconn)
1144 switch (ofconn->role) {
1145 case NX_ROLE_MASTER:
1152 /* Shouldn't happen. */
1157 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1158 * Connects this vconn to a controller. */
1160 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1162 struct ofconn *ofconn, *best;
1164 /* Pick a controller for monitoring. */
1166 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1167 if (ofconn->type == OFCONN_PRIMARY
1168 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1174 rconn_add_monitor(best->rconn, vconn);
1176 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1182 ofproto_run1(struct ofproto *p)
1184 struct ofconn *ofconn, *next_ofconn;
1185 struct ofservice *ofservice;
1186 struct ofport *ofport;
1191 if (shash_is_empty(&p->port_by_name)) {
1195 for (i = 0; i < 50; i++) {
1196 struct dpif_upcall packet;
1198 error = dpif_recv(p->dpif, &packet);
1200 if (error == ENODEV) {
1201 /* Someone destroyed the datapath behind our back. The caller
1202 * better destroy us and give up, because we're just going to
1203 * spin from here on out. */
1204 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1205 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1206 dpif_name(p->dpif));
1212 handle_upcall(p, &packet);
1215 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1216 process_port_change(p, error, devname);
1218 while ((error = netdev_monitor_poll(p->netdev_monitor,
1219 &devname)) != EAGAIN) {
1220 process_port_change(p, error, devname);
1223 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1224 ofport_run(p, ofport);
1228 if (time_msec() >= p->next_in_band_update) {
1229 update_in_band_remotes(p);
1231 in_band_run(p->in_band);
1234 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1238 /* Fail-open maintenance. Do this after processing the ofconns since
1239 * fail-open checks the status of the controller rconn. */
1241 fail_open_run(p->fail_open);
1244 HMAP_FOR_EACH (ofservice, node, &p->services) {
1245 struct vconn *vconn;
1248 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1250 struct rconn *rconn;
1253 rconn = rconn_create(ofservice->probe_interval, 0);
1254 name = ofconn_make_name(p, vconn_get_name(vconn));
1255 rconn_connect_unreliably(rconn, vconn, name);
1258 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1259 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1260 ofservice->burst_limit);
1261 } else if (retval != EAGAIN) {
1262 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1266 for (i = 0; i < p->n_snoops; i++) {
1267 struct vconn *vconn;
1270 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1272 add_snooper(p, vconn);
1273 } else if (retval != EAGAIN) {
1274 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1278 if (time_msec() >= p->next_expiration) {
1279 int delay = ofproto_expire(p);
1280 p->next_expiration = time_msec() + delay;
1281 COVERAGE_INC(ofproto_expiration);
1285 netflow_run(p->netflow);
1288 ofproto_sflow_run(p->sflow);
1295 ofproto_run2(struct ofproto *p, bool revalidate_all)
1297 /* Figure out what we need to revalidate now, if anything. */
1298 struct tag_set revalidate_set = p->revalidate_set;
1299 if (p->need_revalidate) {
1300 revalidate_all = true;
1303 /* Clear the revalidation flags. */
1304 tag_set_init(&p->revalidate_set);
1305 p->need_revalidate = false;
1307 /* Now revalidate if there's anything to do. */
1308 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1309 struct facet *facet, *next;
1311 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1313 || tag_set_intersects(&revalidate_set, facet->tags)) {
1314 facet_revalidate(p, facet);
1323 ofproto_wait(struct ofproto *p)
1325 struct ofservice *ofservice;
1326 struct ofconn *ofconn;
1327 struct ofport *ofport;
1330 dpif_recv_wait(p->dpif);
1331 dpif_port_poll_wait(p->dpif);
1332 netdev_monitor_poll_wait(p->netdev_monitor);
1333 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1334 ofport_wait(ofport);
1336 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1337 ofconn_wait(ofconn);
1340 poll_timer_wait_until(p->next_in_band_update);
1341 in_band_wait(p->in_band);
1344 fail_open_wait(p->fail_open);
1347 ofproto_sflow_wait(p->sflow);
1349 if (!tag_set_is_empty(&p->revalidate_set)) {
1350 poll_immediate_wake();
1352 if (p->need_revalidate) {
1353 /* Shouldn't happen, but if it does just go around again. */
1354 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1355 poll_immediate_wake();
1356 } else if (p->next_expiration != LLONG_MAX) {
1357 poll_timer_wait_until(p->next_expiration);
1359 HMAP_FOR_EACH (ofservice, node, &p->services) {
1360 pvconn_wait(ofservice->pvconn);
1362 for (i = 0; i < p->n_snoops; i++) {
1363 pvconn_wait(p->snoops[i]);
1368 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1370 tag_set_add(&ofproto->revalidate_set, tag);
1374 ofproto_get_revalidate_set(struct ofproto *ofproto)
1376 return &ofproto->revalidate_set;
1380 ofproto_is_alive(const struct ofproto *p)
1382 return !hmap_is_empty(&p->controllers);
1386 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1389 const struct ofconn *ofconn;
1393 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1394 const struct rconn *rconn = ofconn->rconn;
1395 time_t now = time_now();
1396 time_t last_connection = rconn_get_last_connection(rconn);
1397 time_t last_disconnect = rconn_get_last_disconnect(rconn);
1398 const int last_error = rconn_get_last_error(rconn);
1399 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1401 shash_add(info, rconn_get_target(rconn), cinfo);
1403 cinfo->is_connected = rconn_is_connected(rconn);
1404 cinfo->role = ofconn->role;
1409 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1410 cinfo->pairs.values[cinfo->pairs.n++] =
1411 xstrdup(ovs_retval_to_string(last_error));
1414 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1415 cinfo->pairs.values[cinfo->pairs.n++] =
1416 xstrdup(rconn_get_state(rconn));
1418 if (last_connection != TIME_MIN) {
1419 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
1420 cinfo->pairs.values[cinfo->pairs.n++]
1421 = xasprintf("%ld", (long int) (now - last_connection));
1424 if (last_disconnect != TIME_MIN) {
1425 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
1426 cinfo->pairs.values[cinfo->pairs.n++]
1427 = xasprintf("%ld", (long int) (now - last_disconnect));
1433 ofproto_free_ofproto_controller_info(struct shash *info)
1435 struct shash_node *node;
1437 SHASH_FOR_EACH (node, info) {
1438 struct ofproto_controller_info *cinfo = node->data;
1439 while (cinfo->pairs.n) {
1440 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1444 shash_destroy(info);
1447 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1449 * This is almost the same as calling dpif_port_del() directly on the
1450 * datapath, but it also makes 'ofproto' close its open netdev for the port
1451 * (if any). This makes it possible to create a new netdev of a different
1452 * type under the same name, which otherwise the netdev library would refuse
1453 * to do because of the conflict. (The netdev would eventually get closed on
1454 * the next trip through ofproto_run(), but this interface is more direct.)
1456 * Returns 0 if successful, otherwise a positive errno. */
1458 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1460 struct ofport *ofport = get_port(ofproto, odp_port);
1461 const char *name = ofport ? ofport->opp.name : "<unknown>";
1464 error = dpif_port_del(ofproto->dpif, odp_port);
1466 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1467 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1468 } else if (ofport) {
1469 /* 'name' is ofport->opp.name and update_port() is going to destroy
1470 * 'ofport'. Just in case update_port() refers to 'name' after it
1471 * destroys 'ofport', make a copy of it around the update_port()
1473 char *devname = xstrdup(name);
1474 update_port(ofproto, devname);
1480 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1481 * true if 'odp_port' exists and should be included, false otherwise. */
1483 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1485 struct ofport *ofport = get_port(ofproto, odp_port);
1486 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1489 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1490 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1491 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1493 * Returns 0 if successful, otherwise a positive errno value. */
1495 ofproto_send_packet(struct ofproto *ofproto,
1496 uint32_t port_no, uint16_t vlan_tci,
1497 const struct ofpbuf *packet)
1499 struct ofpbuf odp_actions;
1502 ofpbuf_init(&odp_actions, 32);
1503 if (vlan_tci != 0) {
1504 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1505 ntohs(vlan_tci & ~VLAN_CFI));
1507 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1508 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1510 ofpbuf_uninit(&odp_actions);
1513 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1514 dpif_name(ofproto->dpif), port_no, strerror(error));
1519 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1520 * performs the 'n_actions' actions in 'actions'. The new flow will not
1523 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1524 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1525 * controllers; otherwise, it will be hidden.
1527 * The caller retains ownership of 'cls_rule' and 'actions'. */
1529 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1530 const union ofp_action *actions, size_t n_actions)
1533 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1534 rule_insert(p, rule);
1538 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1542 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1545 rule_remove(ofproto, rule);
1550 ofproto_flush_flows(struct ofproto *ofproto)
1552 struct facet *facet, *next_facet;
1553 struct rule *rule, *next_rule;
1554 struct cls_cursor cursor;
1556 COVERAGE_INC(ofproto_flush);
1558 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1559 /* Mark the facet as not installed so that facet_remove() doesn't
1560 * bother trying to uninstall it. There is no point in uninstalling it
1561 * individually since we are about to blow away all the facets with
1562 * dpif_flow_flush(). */
1563 facet->installed = false;
1564 facet->dp_packet_count = 0;
1565 facet->dp_byte_count = 0;
1566 facet_remove(ofproto, facet);
1569 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1570 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1571 rule_remove(ofproto, rule);
1574 dpif_flow_flush(ofproto->dpif);
1575 if (ofproto->in_band) {
1576 in_band_flushed(ofproto->in_band);
1578 if (ofproto->fail_open) {
1579 fail_open_flushed(ofproto->fail_open);
1584 reinit_ports(struct ofproto *p)
1586 struct dpif_port_dump dump;
1587 struct shash_node *node;
1588 struct shash devnames;
1589 struct ofport *ofport;
1590 struct dpif_port dpif_port;
1592 COVERAGE_INC(ofproto_reinit_ports);
1594 shash_init(&devnames);
1595 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1596 shash_add_once (&devnames, ofport->opp.name, NULL);
1598 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1599 shash_add_once (&devnames, dpif_port.name, NULL);
1602 SHASH_FOR_EACH (node, &devnames) {
1603 update_port(p, node->name);
1605 shash_destroy(&devnames);
1608 static struct ofport *
1609 make_ofport(const struct dpif_port *dpif_port)
1611 struct netdev_options netdev_options;
1612 enum netdev_flags flags;
1613 struct ofport *ofport;
1614 struct netdev *netdev;
1617 memset(&netdev_options, 0, sizeof netdev_options);
1618 netdev_options.name = dpif_port->name;
1619 netdev_options.type = dpif_port->type;
1620 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1622 error = netdev_open(&netdev_options, &netdev);
1624 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1625 "cannot be opened (%s)",
1626 dpif_port->name, dpif_port->port_no,
1627 dpif_port->name, strerror(error));
1631 ofport = xzalloc(sizeof *ofport);
1632 ofport->netdev = netdev;
1633 ofport->odp_port = dpif_port->port_no;
1634 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1635 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1636 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1638 netdev_get_flags(netdev, &flags);
1639 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1641 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1643 netdev_get_features(netdev,
1644 &ofport->opp.curr, &ofport->opp.advertised,
1645 &ofport->opp.supported, &ofport->opp.peer);
1650 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1652 if (get_port(p, dpif_port->port_no)) {
1653 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1654 dpif_port->port_no);
1656 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1657 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1666 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1668 const struct ofp_phy_port *a = &a_->opp;
1669 const struct ofp_phy_port *b = &b_->opp;
1671 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1672 return (a->port_no == b->port_no
1673 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1674 && !strcmp(a->name, b->name)
1675 && a->state == b->state
1676 && a->config == b->config
1677 && a->curr == b->curr
1678 && a->advertised == b->advertised
1679 && a->supported == b->supported
1680 && a->peer == b->peer);
1684 send_port_status(struct ofproto *p, const struct ofport *ofport,
1687 /* XXX Should limit the number of queued port status change messages. */
1688 struct ofconn *ofconn;
1689 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1690 struct ofp_port_status *ops;
1693 /* Primary controllers, even slaves, should always get port status
1694 updates. Otherwise obey ofconn_receives_async_msgs(). */
1695 if (ofconn->type != OFCONN_PRIMARY
1696 && !ofconn_receives_async_msgs(ofconn)) {
1700 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1701 ops->reason = reason;
1702 ops->desc = ofport->opp;
1703 hton_ofp_phy_port(&ops->desc);
1704 queue_tx(b, ofconn, NULL);
1709 ofport_install(struct ofproto *p, struct ofport *ofport)
1711 const char *netdev_name = ofport->opp.name;
1713 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1714 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1715 shash_add(&p->port_by_name, netdev_name, ofport);
1717 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1722 ofport_remove(struct ofproto *p, struct ofport *ofport)
1724 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1725 hmap_remove(&p->ports, &ofport->hmap_node);
1726 shash_delete(&p->port_by_name,
1727 shash_find(&p->port_by_name, ofport->opp.name));
1729 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1734 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1737 cfm_run(ofport->cfm);
1739 if (cfm_should_send_ccm(ofport->cfm)) {
1740 struct ofpbuf packet;
1743 ofpbuf_init(&packet, 0);
1744 ccm = compose_packet(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1745 ETH_TYPE_CFM, sizeof *ccm);
1746 cfm_compose_ccm(ofport->cfm, ccm);
1747 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1748 ofpbuf_uninit(&packet);
1754 ofport_wait(struct ofport *ofport)
1757 cfm_wait(ofport->cfm);
1762 ofport_free(struct ofport *ofport)
1765 cfm_destroy(ofport->cfm);
1766 netdev_close(ofport->netdev);
1771 static struct ofport *
1772 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1774 struct ofport *port;
1776 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1777 hash_int(odp_port, 0), &ofproto->ports) {
1778 if (port->odp_port == odp_port) {
1786 update_port(struct ofproto *p, const char *devname)
1788 struct dpif_port dpif_port;
1789 struct ofport *old_ofport;
1790 struct ofport *new_ofport;
1793 COVERAGE_INC(ofproto_update_port);
1795 /* Query the datapath for port information. */
1796 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1798 /* Find the old ofport. */
1799 old_ofport = shash_find_data(&p->port_by_name, devname);
1802 /* There's no port named 'devname' but there might be a port with
1803 * the same port number. This could happen if a port is deleted
1804 * and then a new one added in its place very quickly, or if a port
1805 * is renamed. In the former case we want to send an OFPPR_DELETE
1806 * and an OFPPR_ADD, and in the latter case we want to send a
1807 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1808 * the old port's ifindex against the new port, or perhaps less
1809 * reliably but more portably by comparing the old port's MAC
1810 * against the new port's MAC. However, this code isn't that smart
1811 * and always sends an OFPPR_MODIFY (XXX). */
1812 old_ofport = get_port(p, dpif_port.port_no);
1814 } else if (error != ENOENT && error != ENODEV) {
1815 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1816 "%s", strerror(error));
1820 /* Create a new ofport. */
1821 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1823 /* Eliminate a few pathological cases. */
1824 if (!old_ofport && !new_ofport) {
1826 } else if (old_ofport && new_ofport) {
1827 /* Most of the 'config' bits are OpenFlow soft state, but
1828 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1829 * OpenFlow bits from old_ofport. (make_ofport() only sets
1830 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1831 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1833 if (ofport_equal(old_ofport, new_ofport)) {
1834 /* False alarm--no change. */
1835 ofport_free(new_ofport);
1840 /* Now deal with the normal cases. */
1842 ofport_remove(p, old_ofport);
1845 ofport_install(p, new_ofport);
1847 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1848 (!old_ofport ? OFPPR_ADD
1849 : !new_ofport ? OFPPR_DELETE
1851 ofport_free(old_ofport);
1854 dpif_port_destroy(&dpif_port);
1858 init_ports(struct ofproto *p)
1860 struct dpif_port_dump dump;
1861 struct dpif_port dpif_port;
1863 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1864 if (!ofport_conflicts(p, &dpif_port)) {
1865 struct ofport *ofport = make_ofport(&dpif_port);
1867 ofport_install(p, ofport);
1875 static struct ofconn *
1876 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1878 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1879 ofconn->ofproto = p;
1880 list_push_back(&p->all_conns, &ofconn->node);
1881 ofconn->rconn = rconn;
1882 ofconn->type = type;
1883 ofconn->flow_format = NXFF_OPENFLOW10;
1884 ofconn->role = NX_ROLE_OTHER;
1885 ofconn->packet_in_counter = rconn_packet_counter_create ();
1886 ofconn->pktbuf = NULL;
1887 ofconn->miss_send_len = 0;
1888 ofconn->reply_counter = rconn_packet_counter_create ();
1893 ofconn_destroy(struct ofconn *ofconn)
1895 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1897 if (ofconn->type == OFCONN_PRIMARY) {
1898 hmap_remove(&ofproto->controllers, &ofconn->hmap_node);
1901 list_remove(&ofconn->node);
1902 rconn_destroy(ofconn->rconn);
1903 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1904 rconn_packet_counter_destroy(ofconn->reply_counter);
1905 pktbuf_destroy(ofconn->pktbuf);
1910 ofconn_run(struct ofconn *ofconn)
1912 struct ofproto *p = ofconn_get_ofproto(ofconn);
1916 for (i = 0; i < N_SCHEDULERS; i++) {
1917 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1920 rconn_run(ofconn->rconn);
1922 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1923 /* Limit the number of iterations to prevent other tasks from
1925 for (iteration = 0; iteration < 50; iteration++) {
1926 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1931 fail_open_maybe_recover(p->fail_open);
1933 handle_openflow(ofconn, of_msg);
1934 ofpbuf_delete(of_msg);
1938 if (!rconn_is_alive(ofconn->rconn)) {
1939 ofconn_destroy(ofconn);
1944 ofconn_wait(struct ofconn *ofconn)
1948 for (i = 0; i < N_SCHEDULERS; i++) {
1949 pinsched_wait(ofconn->schedulers[i]);
1951 rconn_run_wait(ofconn->rconn);
1952 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1953 rconn_recv_wait(ofconn->rconn);
1955 COVERAGE_INC(ofproto_ofconn_stuck);
1959 /* Returns true if 'ofconn' should receive asynchronous messages. */
1961 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1963 if (ofconn->type == OFCONN_PRIMARY) {
1964 /* Primary controllers always get asynchronous messages unless they
1965 * have configured themselves as "slaves". */
1966 return ofconn->role != NX_ROLE_SLAVE;
1968 /* Service connections don't get asynchronous messages unless they have
1969 * explicitly asked for them by setting a nonzero miss send length. */
1970 return ofconn->miss_send_len > 0;
1974 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1975 * and 'target', suitable for use in log messages for identifying the
1978 * The name is dynamically allocated. The caller should free it (with free())
1979 * when it is no longer needed. */
1981 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1983 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1987 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1991 for (i = 0; i < N_SCHEDULERS; i++) {
1992 struct pinsched **s = &ofconn->schedulers[i];
1996 *s = pinsched_create(rate, burst);
1998 pinsched_set_limits(*s, rate, burst);
2001 pinsched_destroy(*s);
2007 static struct ofproto *
2008 ofconn_get_ofproto(struct ofconn *ofconn)
2010 return ofconn->ofproto;
2013 static enum nx_flow_format
2014 ofconn_get_flow_format(struct ofconn *ofconn)
2016 return ofconn->flow_format;
2020 ofconn_set_flow_format(struct ofconn *ofconn, enum nx_flow_format flow_format)
2022 ofconn->flow_format = flow_format;
2026 ofconn_get_miss_send_len(const struct ofconn *ofconn)
2028 return ofconn->miss_send_len;
2032 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
2034 ofconn->miss_send_len = miss_send_len;
2038 ofservice_reconfigure(struct ofservice *ofservice,
2039 const struct ofproto_controller *c)
2041 ofservice->probe_interval = c->probe_interval;
2042 ofservice->rate_limit = c->rate_limit;
2043 ofservice->burst_limit = c->burst_limit;
2046 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
2047 * positive errno value. */
2049 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
2051 struct ofservice *ofservice;
2052 struct pvconn *pvconn;
2055 error = pvconn_open(c->target, &pvconn);
2060 ofservice = xzalloc(sizeof *ofservice);
2061 hmap_insert(&ofproto->services, &ofservice->node,
2062 hash_string(c->target, 0));
2063 ofservice->pvconn = pvconn;
2065 ofservice_reconfigure(ofservice, c);
2071 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
2073 hmap_remove(&ofproto->services, &ofservice->node);
2074 pvconn_close(ofservice->pvconn);
2078 /* Finds and returns the ofservice within 'ofproto' that has the given
2079 * 'target', or a null pointer if none exists. */
2080 static struct ofservice *
2081 ofservice_lookup(struct ofproto *ofproto, const char *target)
2083 struct ofservice *ofservice;
2085 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
2086 &ofproto->services) {
2087 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2094 /* Returns true if 'rule' should be hidden from the controller.
2096 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2097 * (e.g. by in-band control) and are intentionally hidden from the
2100 rule_is_hidden(const struct rule *rule)
2102 return rule->cr.priority > UINT16_MAX;
2105 /* Creates and returns a new rule initialized as specified.
2107 * The caller is responsible for inserting the rule into the classifier (with
2108 * rule_insert()). */
2109 static struct rule *
2110 rule_create(const struct cls_rule *cls_rule,
2111 const union ofp_action *actions, size_t n_actions,
2112 uint16_t idle_timeout, uint16_t hard_timeout,
2113 ovs_be64 flow_cookie, bool send_flow_removed)
2115 struct rule *rule = xzalloc(sizeof *rule);
2116 rule->cr = *cls_rule;
2117 rule->idle_timeout = idle_timeout;
2118 rule->hard_timeout = hard_timeout;
2119 rule->flow_cookie = flow_cookie;
2120 rule->used = rule->created = time_msec();
2121 rule->send_flow_removed = send_flow_removed;
2122 list_init(&rule->facets);
2123 if (n_actions > 0) {
2124 rule->n_actions = n_actions;
2125 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2131 static struct rule *
2132 rule_from_cls_rule(const struct cls_rule *cls_rule)
2134 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2138 rule_free(struct rule *rule)
2140 free(rule->actions);
2144 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2145 * destroying any that no longer has a rule (which is probably all of them).
2147 * The caller must have already removed 'rule' from the classifier. */
2149 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2151 struct facet *facet, *next_facet;
2152 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2153 facet_revalidate(ofproto, facet);
2158 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2159 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2162 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2164 const union ofp_action *oa;
2165 struct actions_iterator i;
2167 if (out_port == htons(OFPP_NONE)) {
2170 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2171 oa = actions_next(&i)) {
2172 if (action_outputs_to_port(oa, out_port)) {
2179 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2180 * 'packet', which arrived on 'in_port'.
2182 * Takes ownership of 'packet'. */
2184 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2185 const struct nlattr *odp_actions, size_t actions_len,
2186 struct ofpbuf *packet)
2188 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2189 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2190 /* As an optimization, avoid a round-trip from userspace to kernel to
2191 * userspace. This also avoids possibly filling up kernel packet
2192 * buffers along the way. */
2193 struct dpif_upcall upcall;
2195 upcall.type = DPIF_UC_ACTION;
2196 upcall.packet = packet;
2199 upcall.userdata = nl_attr_get_u64(odp_actions);
2200 upcall.sample_pool = 0;
2201 upcall.actions = NULL;
2202 upcall.actions_len = 0;
2204 send_packet_in(ofproto, &upcall, flow, false);
2210 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2211 ofpbuf_delete(packet);
2216 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2217 * statistics appropriately. 'packet' must have at least sizeof(struct
2218 * ofp_packet_in) bytes of headroom.
2220 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2221 * applying flow_extract() to 'packet' would yield the same flow as
2224 * 'facet' must have accurately composed ODP actions; that is, it must not be
2225 * in need of revalidation.
2227 * Takes ownership of 'packet'. */
2229 facet_execute(struct ofproto *ofproto, struct facet *facet,
2230 struct ofpbuf *packet)
2232 struct dpif_flow_stats stats;
2234 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2236 flow_extract_stats(&facet->flow, packet, &stats);
2237 stats.used = time_msec();
2238 if (execute_odp_actions(ofproto, &facet->flow,
2239 facet->actions, facet->actions_len, packet)) {
2240 facet_update_stats(ofproto, facet, &stats);
2244 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2245 * statistics (or the statistics for one of its facets) appropriately.
2246 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2248 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2249 * with statistics for 'packet' either way.
2251 * Takes ownership of 'packet'. */
2253 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2254 struct ofpbuf *packet)
2256 struct action_xlate_ctx ctx;
2257 struct ofpbuf *odp_actions;
2258 struct facet *facet;
2262 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2264 flow_extract(packet, 0, in_port, &flow);
2266 /* First look for a related facet. If we find one, account it to that. */
2267 facet = facet_lookup_valid(ofproto, &flow);
2268 if (facet && facet->rule == rule) {
2269 facet_execute(ofproto, facet, packet);
2273 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2274 * create a new facet for it and use that. */
2275 if (rule_lookup(ofproto, &flow) == rule) {
2276 facet = facet_create(ofproto, rule, &flow, packet);
2277 facet_execute(ofproto, facet, packet);
2278 facet_install(ofproto, facet, true);
2282 /* We can't account anything to a facet. If we were to try, then that
2283 * facet would have a non-matching rule, busting our invariants. */
2284 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2285 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2286 size = packet->size;
2287 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2288 odp_actions->size, packet)) {
2289 rule->used = time_msec();
2290 rule->packet_count++;
2291 rule->byte_count += size;
2292 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
2294 ofpbuf_delete(odp_actions);
2297 /* Inserts 'rule' into 'p''s flow table. */
2299 rule_insert(struct ofproto *p, struct rule *rule)
2301 struct rule *displaced_rule;
2303 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2304 if (displaced_rule) {
2305 rule_destroy(p, displaced_rule);
2307 p->need_revalidate = true;
2310 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2311 * 'flow' and an example 'packet' within that flow.
2313 * The caller must already have determined that no facet with an identical
2314 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2315 * 'ofproto''s classifier table. */
2316 static struct facet *
2317 facet_create(struct ofproto *ofproto, struct rule *rule,
2318 const struct flow *flow, const struct ofpbuf *packet)
2320 struct facet *facet;
2322 facet = xzalloc(sizeof *facet);
2323 facet->used = time_msec();
2324 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2325 list_push_back(&rule->facets, &facet->list_node);
2327 facet->flow = *flow;
2328 netflow_flow_init(&facet->nf_flow);
2329 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2331 facet_make_actions(ofproto, facet, packet);
2337 facet_free(struct facet *facet)
2339 free(facet->actions);
2343 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2345 * - Removes 'rule' from the classifier.
2347 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2348 * destroys them), via rule_destroy().
2351 rule_remove(struct ofproto *ofproto, struct rule *rule)
2353 COVERAGE_INC(ofproto_del_rule);
2354 ofproto->need_revalidate = true;
2355 classifier_remove(&ofproto->cls, &rule->cr);
2356 rule_destroy(ofproto, rule);
2359 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2361 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2362 * rule's statistics, via facet_uninstall().
2364 * - Removes 'facet' from its rule and from ofproto->facets.
2367 facet_remove(struct ofproto *ofproto, struct facet *facet)
2369 facet_uninstall(ofproto, facet);
2370 facet_flush_stats(ofproto, facet);
2371 hmap_remove(&ofproto->facets, &facet->hmap_node);
2372 list_remove(&facet->list_node);
2376 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2378 facet_make_actions(struct ofproto *p, struct facet *facet,
2379 const struct ofpbuf *packet)
2381 const struct rule *rule = facet->rule;
2382 struct ofpbuf *odp_actions;
2383 struct action_xlate_ctx ctx;
2385 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2386 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2387 facet->tags = ctx.tags;
2388 facet->may_install = ctx.may_set_up_flow;
2389 facet->nf_flow.output_iface = ctx.nf_output_iface;
2391 if (facet->actions_len != odp_actions->size
2392 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2393 free(facet->actions);
2394 facet->actions_len = odp_actions->size;
2395 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2398 ofpbuf_delete(odp_actions);
2402 facet_put__(struct ofproto *ofproto, struct facet *facet,
2403 const struct nlattr *actions, size_t actions_len,
2404 struct dpif_flow_stats *stats)
2406 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2407 enum dpif_flow_put_flags flags;
2410 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2412 flags |= DPIF_FP_ZERO_STATS;
2413 facet->dp_packet_count = 0;
2414 facet->dp_byte_count = 0;
2417 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2418 odp_flow_key_from_flow(&key, &facet->flow);
2419 assert(key.base == keybuf);
2421 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2422 actions, actions_len, stats);
2425 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2426 * 'zero_stats' is true, clears any existing statistics from the datapath for
2429 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2431 struct dpif_flow_stats stats;
2433 if (facet->may_install
2434 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2435 zero_stats ? &stats : NULL)) {
2436 facet->installed = true;
2440 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2441 * to the accounting hook function in the ofhooks structure. */
2443 facet_account(struct ofproto *ofproto,
2444 struct facet *facet, uint64_t extra_bytes)
2446 uint64_t total_bytes = facet->byte_count + extra_bytes;
2448 if (ofproto->ofhooks->account_flow_cb
2449 && total_bytes > facet->accounted_bytes)
2451 ofproto->ofhooks->account_flow_cb(
2452 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2453 total_bytes - facet->accounted_bytes, ofproto->aux);
2454 facet->accounted_bytes = total_bytes;
2458 /* If 'rule' is installed in the datapath, uninstalls it. */
2460 facet_uninstall(struct ofproto *p, struct facet *facet)
2462 if (facet->installed) {
2463 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2464 struct dpif_flow_stats stats;
2467 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2468 odp_flow_key_from_flow(&key, &facet->flow);
2469 assert(key.base == keybuf);
2471 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2472 facet_update_stats(p, facet, &stats);
2474 facet->installed = false;
2475 facet->dp_packet_count = 0;
2476 facet->dp_byte_count = 0;
2478 assert(facet->dp_packet_count == 0);
2479 assert(facet->dp_byte_count == 0);
2483 /* Returns true if the only action for 'facet' is to send to the controller.
2484 * (We don't report NetFlow expiration messages for such facets because they
2485 * are just part of the control logic for the network, not real traffic). */
2487 facet_is_controller_flow(struct facet *facet)
2490 && facet->rule->n_actions == 1
2491 && action_outputs_to_port(&facet->rule->actions[0],
2492 htons(OFPP_CONTROLLER)));
2495 /* Folds all of 'facet''s statistics into its rule. Also updates the
2496 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2497 * 'facet''s statistics in the datapath should have been zeroed and folded into
2498 * its packet and byte counts before this function is called. */
2500 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2502 assert(!facet->dp_byte_count);
2503 assert(!facet->dp_packet_count);
2505 facet_push_stats(ofproto, facet);
2506 facet_account(ofproto, facet, 0);
2508 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2509 struct ofexpired expired;
2510 expired.flow = facet->flow;
2511 expired.packet_count = facet->packet_count;
2512 expired.byte_count = facet->byte_count;
2513 expired.used = facet->used;
2514 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2517 facet->rule->packet_count += facet->packet_count;
2518 facet->rule->byte_count += facet->byte_count;
2520 /* Reset counters to prevent double counting if 'facet' ever gets
2522 facet->packet_count = 0;
2523 facet->byte_count = 0;
2524 facet->rs_packet_count = 0;
2525 facet->rs_byte_count = 0;
2526 facet->accounted_bytes = 0;
2528 netflow_flow_clear(&facet->nf_flow);
2531 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2532 * Returns it if found, otherwise a null pointer.
2534 * The returned facet might need revalidation; use facet_lookup_valid()
2535 * instead if that is important. */
2536 static struct facet *
2537 facet_find(struct ofproto *ofproto, const struct flow *flow)
2539 struct facet *facet;
2541 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2543 if (flow_equal(flow, &facet->flow)) {
2551 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2552 * Returns it if found, otherwise a null pointer.
2554 * The returned facet is guaranteed to be valid. */
2555 static struct facet *
2556 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2558 struct facet *facet = facet_find(ofproto, flow);
2560 /* The facet we found might not be valid, since we could be in need of
2561 * revalidation. If it is not valid, don't return it. */
2563 && ofproto->need_revalidate
2564 && !facet_revalidate(ofproto, facet)) {
2565 COVERAGE_INC(ofproto_invalidated);
2572 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2574 * - If the rule found is different from 'facet''s current rule, moves
2575 * 'facet' to the new rule and recompiles its actions.
2577 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2578 * where it is and recompiles its actions anyway.
2580 * - If there is none, destroys 'facet'.
2582 * Returns true if 'facet' still exists, false if it has been destroyed. */
2584 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2586 struct action_xlate_ctx ctx;
2587 struct ofpbuf *odp_actions;
2588 struct rule *new_rule;
2589 bool actions_changed;
2591 COVERAGE_INC(facet_revalidate);
2593 /* Determine the new rule. */
2594 new_rule = rule_lookup(ofproto, &facet->flow);
2596 /* No new rule, so delete the facet. */
2597 facet_remove(ofproto, facet);
2601 /* Calculate new ODP actions.
2603 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2604 * emit a NetFlow expiration and, if so, we need to have the old state
2605 * around to properly compose it. */
2606 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2607 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2608 actions_changed = (facet->actions_len != odp_actions->size
2609 || memcmp(facet->actions, odp_actions->data,
2610 facet->actions_len));
2612 /* If the ODP actions changed or the installability changed, then we need
2613 * to talk to the datapath. */
2614 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2615 if (ctx.may_set_up_flow) {
2616 struct dpif_flow_stats stats;
2618 facet_put__(ofproto, facet,
2619 odp_actions->data, odp_actions->size, &stats);
2620 facet_update_stats(ofproto, facet, &stats);
2622 facet_uninstall(ofproto, facet);
2625 /* The datapath flow is gone or has zeroed stats, so push stats out of
2626 * 'facet' into 'rule'. */
2627 facet_flush_stats(ofproto, facet);
2630 /* Update 'facet' now that we've taken care of all the old state. */
2631 facet->tags = ctx.tags;
2632 facet->nf_flow.output_iface = ctx.nf_output_iface;
2633 facet->may_install = ctx.may_set_up_flow;
2634 if (actions_changed) {
2635 free(facet->actions);
2636 facet->actions_len = odp_actions->size;
2637 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2639 if (facet->rule != new_rule) {
2640 COVERAGE_INC(facet_changed_rule);
2641 list_remove(&facet->list_node);
2642 list_push_back(&new_rule->facets, &facet->list_node);
2643 facet->rule = new_rule;
2644 facet->used = new_rule->created;
2645 facet->rs_used = facet->used;
2648 ofpbuf_delete(odp_actions);
2654 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2655 struct rconn_packet_counter *counter)
2657 update_openflow_length(msg);
2658 if (rconn_send(ofconn->rconn, msg, counter)) {
2664 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
2666 queue_tx(msg, ofconn, ofconn->reply_counter);
2670 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2673 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2675 COVERAGE_INC(ofproto_error);
2676 ofconn_send_reply(ofconn, buf);
2681 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2683 ofconn_send_reply(ofconn, make_echo_reply(oh));
2688 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2690 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2691 struct ofp_switch_features *osf;
2693 struct ofport *port;
2695 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2696 osf->datapath_id = htonll(ofproto->datapath_id);
2697 osf->n_buffers = htonl(pktbuf_capacity());
2699 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2700 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2701 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2702 (1u << OFPAT_SET_VLAN_VID) |
2703 (1u << OFPAT_SET_VLAN_PCP) |
2704 (1u << OFPAT_STRIP_VLAN) |
2705 (1u << OFPAT_SET_DL_SRC) |
2706 (1u << OFPAT_SET_DL_DST) |
2707 (1u << OFPAT_SET_NW_SRC) |
2708 (1u << OFPAT_SET_NW_DST) |
2709 (1u << OFPAT_SET_NW_TOS) |
2710 (1u << OFPAT_SET_TP_SRC) |
2711 (1u << OFPAT_SET_TP_DST) |
2712 (1u << OFPAT_ENQUEUE));
2714 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2715 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2718 ofconn_send_reply(ofconn, buf);
2723 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2725 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2727 struct ofp_switch_config *osc;
2731 /* Figure out flags. */
2732 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2733 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2736 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2737 osc->flags = htons(flags);
2738 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2739 ofconn_send_reply(ofconn, buf);
2745 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2747 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2748 uint16_t flags = ntohs(osc->flags);
2750 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2751 switch (flags & OFPC_FRAG_MASK) {
2752 case OFPC_FRAG_NORMAL:
2753 dpif_set_drop_frags(ofproto->dpif, false);
2755 case OFPC_FRAG_DROP:
2756 dpif_set_drop_frags(ofproto->dpif, true);
2759 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2765 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2770 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2771 struct action_xlate_ctx *ctx);
2774 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2776 const struct ofport *ofport = get_port(ctx->ofproto, port);
2779 if (ofport->opp.config & OFPPC_NO_FWD) {
2780 /* Forwarding disabled on port. */
2785 * We don't have an ofport record for this port, but it doesn't hurt to
2786 * allow forwarding to it anyhow. Maybe such a port will appear later
2787 * and we're pre-populating the flow table.
2791 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2792 ctx->nf_output_iface = port;
2795 static struct rule *
2796 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2798 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2802 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2804 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2805 uint16_t old_in_port;
2808 /* Look up a flow with 'in_port' as the input port. Then restore the
2809 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2810 * have surprising behavior). */
2811 old_in_port = ctx->flow.in_port;
2812 ctx->flow.in_port = in_port;
2813 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2814 ctx->flow.in_port = old_in_port;
2816 if (ctx->resubmit_hook) {
2817 ctx->resubmit_hook(ctx, rule);
2822 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2826 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2828 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2829 MAX_RESUBMIT_RECURSION);
2834 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2835 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2837 struct ofport *ofport;
2839 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2840 uint16_t odp_port = ofport->odp_port;
2841 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2842 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2845 *nf_output_iface = NF_OUT_FLOOD;
2849 xlate_output_action__(struct action_xlate_ctx *ctx,
2850 uint16_t port, uint16_t max_len)
2853 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2855 ctx->nf_output_iface = NF_OUT_DROP;
2859 add_output_action(ctx, ctx->flow.in_port);
2862 xlate_table_action(ctx, ctx->flow.in_port);
2865 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2866 ctx->odp_actions, &ctx->tags,
2867 &ctx->nf_output_iface,
2868 ctx->ofproto->aux)) {
2869 COVERAGE_INC(ofproto_uninstallable);
2870 ctx->may_set_up_flow = false;
2874 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2875 &ctx->nf_output_iface, ctx->odp_actions);
2878 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2879 &ctx->nf_output_iface, ctx->odp_actions);
2881 case OFPP_CONTROLLER:
2882 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2885 add_output_action(ctx, ODPP_LOCAL);
2888 odp_port = ofp_port_to_odp_port(port);
2889 if (odp_port != ctx->flow.in_port) {
2890 add_output_action(ctx, odp_port);
2895 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2896 ctx->nf_output_iface = NF_OUT_FLOOD;
2897 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2898 ctx->nf_output_iface = prev_nf_output_iface;
2899 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2900 ctx->nf_output_iface != NF_OUT_FLOOD) {
2901 ctx->nf_output_iface = NF_OUT_MULTI;
2906 xlate_output_action(struct action_xlate_ctx *ctx,
2907 const struct ofp_action_output *oao)
2909 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2912 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2913 * optimization, because we're going to add another action that sets the
2914 * priority immediately after, or because there are no actions following the
2917 remove_pop_action(struct action_xlate_ctx *ctx)
2919 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2920 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2921 ctx->last_pop_priority = -1;
2926 add_pop_action(struct action_xlate_ctx *ctx)
2928 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2929 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2930 ctx->last_pop_priority = ctx->odp_actions->size;
2935 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2936 const struct ofp_action_enqueue *oae)
2938 uint16_t ofp_port, odp_port;
2942 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2945 /* Fall back to ordinary output action. */
2946 xlate_output_action__(ctx, ntohs(oae->port), 0);
2950 /* Figure out ODP output port. */
2951 ofp_port = ntohs(oae->port);
2952 if (ofp_port != OFPP_IN_PORT) {
2953 odp_port = ofp_port_to_odp_port(ofp_port);
2955 odp_port = ctx->flow.in_port;
2958 /* Add ODP actions. */
2959 remove_pop_action(ctx);
2960 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2961 add_output_action(ctx, odp_port);
2962 add_pop_action(ctx);
2964 /* Update NetFlow output port. */
2965 if (ctx->nf_output_iface == NF_OUT_DROP) {
2966 ctx->nf_output_iface = odp_port;
2967 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2968 ctx->nf_output_iface = NF_OUT_MULTI;
2973 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2974 const struct nx_action_set_queue *nasq)
2979 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2982 /* Couldn't translate queue to a priority, so ignore. A warning
2983 * has already been logged. */
2987 remove_pop_action(ctx);
2988 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2992 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2994 ovs_be16 tci = ctx->flow.vlan_tci;
2995 if (!(tci & htons(VLAN_CFI))) {
2996 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2998 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2999 tci & ~htons(VLAN_CFI));
3003 struct xlate_reg_state {
3009 save_reg_state(const struct action_xlate_ctx *ctx,
3010 struct xlate_reg_state *state)
3012 state->vlan_tci = ctx->flow.vlan_tci;
3013 state->tun_id = ctx->flow.tun_id;
3017 update_reg_state(struct action_xlate_ctx *ctx,
3018 const struct xlate_reg_state *state)
3020 if (ctx->flow.vlan_tci != state->vlan_tci) {
3021 xlate_set_dl_tci(ctx);
3023 if (ctx->flow.tun_id != state->tun_id) {
3024 nl_msg_put_be64(ctx->odp_actions,
3025 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
3030 xlate_nicira_action(struct action_xlate_ctx *ctx,
3031 const struct nx_action_header *nah)
3033 const struct nx_action_resubmit *nar;
3034 const struct nx_action_set_tunnel *nast;
3035 const struct nx_action_set_queue *nasq;
3036 const struct nx_action_multipath *nam;
3037 enum nx_action_subtype subtype = ntohs(nah->subtype);
3038 struct xlate_reg_state state;
3041 assert(nah->vendor == htonl(NX_VENDOR_ID));
3043 case NXAST_RESUBMIT:
3044 nar = (const struct nx_action_resubmit *) nah;
3045 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
3048 case NXAST_SET_TUNNEL:
3049 nast = (const struct nx_action_set_tunnel *) nah;
3050 tun_id = htonll(ntohl(nast->tun_id));
3051 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3052 ctx->flow.tun_id = tun_id;
3055 case NXAST_DROP_SPOOFED_ARP:
3056 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
3057 nl_msg_put_flag(ctx->odp_actions,
3058 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
3062 case NXAST_SET_QUEUE:
3063 nasq = (const struct nx_action_set_queue *) nah;
3064 xlate_set_queue_action(ctx, nasq);
3067 case NXAST_POP_QUEUE:
3068 add_pop_action(ctx);
3071 case NXAST_REG_MOVE:
3072 save_reg_state(ctx, &state);
3073 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
3075 update_reg_state(ctx, &state);
3078 case NXAST_REG_LOAD:
3079 save_reg_state(ctx, &state);
3080 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
3082 update_reg_state(ctx, &state);
3086 /* Nothing to do. */
3089 case NXAST_SET_TUNNEL64:
3090 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
3091 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3092 ctx->flow.tun_id = tun_id;
3095 case NXAST_MULTIPATH:
3096 nam = (const struct nx_action_multipath *) nah;
3097 multipath_execute(nam, &ctx->flow);
3100 /* If you add a new action here that modifies flow data, don't forget to
3101 * update the flow key in ctx->flow at the same time. */
3103 case NXAST_SNAT__OBSOLETE:
3105 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3111 do_xlate_actions(const union ofp_action *in, size_t n_in,
3112 struct action_xlate_ctx *ctx)
3114 struct actions_iterator iter;
3115 const union ofp_action *ia;
3116 const struct ofport *port;
3118 port = get_port(ctx->ofproto, ctx->flow.in_port);
3119 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3120 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3121 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3122 /* Drop this flow. */
3126 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3127 enum ofp_action_type type = ntohs(ia->type);
3128 const struct ofp_action_dl_addr *oada;
3132 xlate_output_action(ctx, &ia->output);
3135 case OFPAT_SET_VLAN_VID:
3136 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3137 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3138 xlate_set_dl_tci(ctx);
3141 case OFPAT_SET_VLAN_PCP:
3142 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3143 ctx->flow.vlan_tci |= htons(
3144 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3145 xlate_set_dl_tci(ctx);
3148 case OFPAT_STRIP_VLAN:
3149 ctx->flow.vlan_tci = htons(0);
3150 xlate_set_dl_tci(ctx);
3153 case OFPAT_SET_DL_SRC:
3154 oada = ((struct ofp_action_dl_addr *) ia);
3155 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3156 oada->dl_addr, ETH_ADDR_LEN);
3157 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3160 case OFPAT_SET_DL_DST:
3161 oada = ((struct ofp_action_dl_addr *) ia);
3162 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3163 oada->dl_addr, ETH_ADDR_LEN);
3164 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3167 case OFPAT_SET_NW_SRC:
3168 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3169 ia->nw_addr.nw_addr);
3170 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3173 case OFPAT_SET_NW_DST:
3174 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3175 ia->nw_addr.nw_addr);
3176 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3179 case OFPAT_SET_NW_TOS:
3180 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3182 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3185 case OFPAT_SET_TP_SRC:
3186 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3187 ia->tp_port.tp_port);
3188 ctx->flow.tp_src = ia->tp_port.tp_port;
3191 case OFPAT_SET_TP_DST:
3192 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3193 ia->tp_port.tp_port);
3194 ctx->flow.tp_dst = ia->tp_port.tp_port;
3198 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3202 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3206 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3213 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3214 struct ofproto *ofproto, const struct flow *flow,
3215 const struct ofpbuf *packet)
3217 ctx->ofproto = ofproto;
3219 ctx->packet = packet;
3220 ctx->resubmit_hook = NULL;
3221 ctx->check_special = true;
3225 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
3226 const struct ofpbuf *packet)
3228 struct ofport *ofport;
3230 ofport = get_port(ofproto, flow->in_port);
3231 if (ofport && ofport->cfm) {
3232 cfm_process_heartbeat(ofport->cfm, packet);
3236 static struct ofpbuf *
3237 xlate_actions(struct action_xlate_ctx *ctx,
3238 const union ofp_action *in, size_t n_in)
3240 COVERAGE_INC(ofproto_ofp2odp);
3242 ctx->odp_actions = ofpbuf_new(512);
3244 ctx->may_set_up_flow = true;
3245 ctx->nf_output_iface = NF_OUT_DROP;
3247 ctx->last_pop_priority = -1;
3249 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
3251 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
3253 ctx->may_set_up_flow = false;
3254 } else if (ctx->check_special
3255 && ctx->ofproto->ofhooks->special_cb
3256 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3257 ctx->ofproto->aux)) {
3258 ctx->may_set_up_flow = false;
3260 do_xlate_actions(in, n_in, ctx);
3263 remove_pop_action(ctx);
3265 /* Check with in-band control to see if we're allowed to set up this
3267 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3268 ctx->odp_actions->data, ctx->odp_actions->size)) {
3269 ctx->may_set_up_flow = false;
3272 return ctx->odp_actions;
3275 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3276 * error message code (composed with ofp_mkerr()) for the caller to propagate
3277 * upward. Otherwise, returns 0.
3279 * The log message mentions 'msg_type'. */
3281 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3283 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3284 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3285 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3288 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3295 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3297 struct ofproto *p = ofconn_get_ofproto(ofconn);
3298 struct ofp_packet_out *opo;
3299 struct ofpbuf payload, *buffer;
3300 union ofp_action *ofp_actions;
3301 struct action_xlate_ctx ctx;
3302 struct ofpbuf *odp_actions;
3303 struct ofpbuf request;
3305 size_t n_ofp_actions;
3309 COVERAGE_INC(ofproto_packet_out);
3311 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3316 /* Get ofp_packet_out. */
3317 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3318 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3321 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3322 &ofp_actions, &n_ofp_actions);
3328 if (opo->buffer_id != htonl(UINT32_MAX)) {
3329 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3331 if (error || !buffer) {
3340 /* Extract flow, check actions. */
3341 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3343 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3349 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3350 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3351 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3352 ofpbuf_delete(odp_actions);
3355 ofpbuf_delete(buffer);
3360 update_port_config(struct ofproto *p, struct ofport *port,
3361 uint32_t config, uint32_t mask)
3363 mask &= config ^ port->opp.config;
3364 if (mask & OFPPC_PORT_DOWN) {
3365 if (config & OFPPC_PORT_DOWN) {
3366 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3368 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3371 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3372 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3373 if (mask & REVALIDATE_BITS) {
3374 COVERAGE_INC(ofproto_costly_flags);
3375 port->opp.config ^= mask & REVALIDATE_BITS;
3376 p->need_revalidate = true;
3378 #undef REVALIDATE_BITS
3379 if (mask & OFPPC_NO_PACKET_IN) {
3380 port->opp.config ^= OFPPC_NO_PACKET_IN;
3385 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3387 struct ofproto *p = ofconn_get_ofproto(ofconn);
3388 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3389 struct ofport *port;
3392 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3397 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3399 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3400 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3401 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3403 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3404 if (opm->advertise) {
3405 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3411 static struct ofpbuf *
3412 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3414 struct ofp_stats_reply *osr;
3417 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3418 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3420 osr->flags = htons(0);
3424 static struct ofpbuf *
3425 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3427 const struct ofp_stats_request *osr
3428 = (const struct ofp_stats_request *) request;
3429 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3433 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3434 struct ofpbuf **msgp)
3436 struct ofpbuf *msg = *msgp;
3437 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3438 if (nbytes + msg->size > UINT16_MAX) {
3439 struct ofp_stats_reply *reply = msg->data;
3440 reply->flags = htons(OFPSF_REPLY_MORE);
3441 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3442 ofconn_send_reply(ofconn, msg);
3444 return ofpbuf_put_uninit(*msgp, nbytes);
3447 static struct ofpbuf *
3448 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3450 struct nicira_stats_msg *nsm;
3453 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3454 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3455 nsm->type = htons(OFPST_VENDOR);
3456 nsm->flags = htons(0);
3457 nsm->vendor = htonl(NX_VENDOR_ID);
3458 nsm->subtype = subtype;
3462 static struct ofpbuf *
3463 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3465 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3469 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3470 struct ofpbuf **msgp)
3472 struct ofpbuf *msg = *msgp;
3473 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3474 if (nbytes + msg->size > UINT16_MAX) {
3475 struct nicira_stats_msg *reply = msg->data;
3476 reply->flags = htons(OFPSF_REPLY_MORE);
3477 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3478 ofconn_send_reply(ofconn, msg);
3480 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3484 handle_desc_stats_request(struct ofconn *ofconn,
3485 const struct ofp_header *request)
3487 struct ofproto *p = ofconn_get_ofproto(ofconn);
3488 struct ofp_desc_stats *ods;
3491 msg = start_ofp_stats_reply(request, sizeof *ods);
3492 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3493 memset(ods, 0, sizeof *ods);
3494 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3495 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3496 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3497 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3498 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3499 ofconn_send_reply(ofconn, msg);
3505 handle_table_stats_request(struct ofconn *ofconn,
3506 const struct ofp_header *request)
3508 struct ofproto *p = ofconn_get_ofproto(ofconn);
3509 struct ofp_table_stats *ots;
3512 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3514 /* Classifier table. */
3515 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3516 memset(ots, 0, sizeof *ots);
3517 strcpy(ots->name, "classifier");
3518 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
3519 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3520 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3521 ots->active_count = htonl(classifier_count(&p->cls));
3522 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3523 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3525 ofconn_send_reply(ofconn, msg);
3530 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3531 struct ofpbuf **msgp)
3533 struct netdev_stats stats;
3534 struct ofp_port_stats *ops;
3536 /* Intentionally ignore return value, since errors will set
3537 * 'stats' to all-1s, which is correct for OpenFlow, and
3538 * netdev_get_stats() will log errors. */
3539 netdev_get_stats(port->netdev, &stats);
3541 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3542 ops->port_no = htons(port->opp.port_no);
3543 memset(ops->pad, 0, sizeof ops->pad);
3544 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3545 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3546 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3547 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3548 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3549 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3550 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3551 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3552 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3553 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3554 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3555 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3559 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3561 struct ofproto *p = ofconn_get_ofproto(ofconn);
3562 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3563 struct ofp_port_stats *ops;
3565 struct ofport *port;
3567 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3568 if (psr->port_no != htons(OFPP_NONE)) {
3569 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3571 append_port_stat(port, ofconn, &msg);
3574 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3575 append_port_stat(port, ofconn, &msg);
3579 ofconn_send_reply(ofconn, msg);
3584 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
3586 long long int msecs = time_msec() - start;
3587 *sec = msecs / 1000;
3588 *nsec = (msecs % 1000) * (1000 * 1000);
3592 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
3596 calc_flow_duration__(start, &sec, &nsec);
3597 *sec_be = htonl(sec);
3598 *nsec_be = htonl(nsec);
3602 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3603 ovs_be16 out_port, struct ofpbuf **replyp)
3605 struct ofp_flow_stats *ofs;
3606 uint64_t packet_count, byte_count;
3608 size_t act_len, len;
3610 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3614 act_len = sizeof *rule->actions * rule->n_actions;
3615 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3617 rule_get_stats(rule, &packet_count, &byte_count);
3619 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3620 ofs->length = htons(len);
3623 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
3624 &ofs->match, rule->flow_cookie, &cookie);
3625 put_32aligned_be64(&ofs->cookie, cookie);
3626 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3627 ofs->priority = htons(rule->cr.priority);
3628 ofs->idle_timeout = htons(rule->idle_timeout);
3629 ofs->hard_timeout = htons(rule->hard_timeout);
3630 memset(ofs->pad2, 0, sizeof ofs->pad2);
3631 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3632 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3633 if (rule->n_actions > 0) {
3634 memcpy(ofs->actions, rule->actions, act_len);
3639 is_valid_table(uint8_t table_id)
3641 if (table_id == 0 || table_id == 0xff) {
3644 /* It would probably be better to reply with an error but there doesn't
3645 * seem to be any appropriate value, so that might just be
3647 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3654 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3656 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3657 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3658 struct ofpbuf *reply;
3660 COVERAGE_INC(ofproto_flows_req);
3661 reply = start_ofp_stats_reply(oh, 1024);
3662 if (is_valid_table(fsr->table_id)) {
3663 struct cls_cursor cursor;
3664 struct cls_rule target;
3667 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3669 cls_cursor_init(&cursor, &ofproto->cls, &target);
3670 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3671 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3674 ofconn_send_reply(ofconn, reply);
3680 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3681 ovs_be16 out_port, struct ofpbuf **replyp)
3683 struct nx_flow_stats *nfs;
3684 uint64_t packet_count, byte_count;
3685 size_t act_len, start_len;
3686 struct ofpbuf *reply;
3688 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3692 rule_get_stats(rule, &packet_count, &byte_count);
3694 act_len = sizeof *rule->actions * rule->n_actions;
3696 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3697 start_len = (*replyp)->size;
3700 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3703 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3704 nfs->cookie = rule->flow_cookie;
3705 nfs->priority = htons(rule->cr.priority);
3706 nfs->idle_timeout = htons(rule->idle_timeout);
3707 nfs->hard_timeout = htons(rule->hard_timeout);
3708 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3709 memset(nfs->pad2, 0, sizeof nfs->pad2);
3710 nfs->packet_count = htonll(packet_count);
3711 nfs->byte_count = htonll(byte_count);
3712 if (rule->n_actions > 0) {
3713 ofpbuf_put(reply, rule->actions, act_len);
3715 nfs->length = htons(reply->size - start_len);
3719 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3721 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3722 struct nx_flow_stats_request *nfsr;
3723 struct cls_rule target;
3724 struct ofpbuf *reply;
3728 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3730 /* Dissect the message. */
3731 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3732 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3737 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3740 COVERAGE_INC(ofproto_flows_req);
3741 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3742 if (is_valid_table(nfsr->table_id)) {
3743 struct cls_cursor cursor;
3746 cls_cursor_init(&cursor, &ofproto->cls, &target);
3747 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3748 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3751 ofconn_send_reply(ofconn, reply);
3757 flow_stats_ds(struct rule *rule, struct ds *results)
3759 uint64_t packet_count, byte_count;
3760 size_t act_len = sizeof *rule->actions * rule->n_actions;
3762 rule_get_stats(rule, &packet_count, &byte_count);
3764 ds_put_format(results, "duration=%llds, ",
3765 (time_msec() - rule->created) / 1000);
3766 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3767 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3768 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3769 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3770 cls_rule_format(&rule->cr, results);
3771 ds_put_char(results, ',');
3773 ofp_print_actions(results, &rule->actions->header, act_len);
3775 ds_put_cstr(results, "drop");
3777 ds_put_cstr(results, "\n");
3780 /* Adds a pretty-printed description of all flows to 'results', including
3781 * hidden flows (e.g., set up by in-band control). */
3783 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3785 struct cls_cursor cursor;
3788 cls_cursor_init(&cursor, &p->cls, NULL);
3789 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3790 flow_stats_ds(rule, results);
3795 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3796 ovs_be16 out_port, uint8_t table_id,
3797 struct ofp_aggregate_stats_reply *oasr)
3799 uint64_t total_packets = 0;
3800 uint64_t total_bytes = 0;
3803 COVERAGE_INC(ofproto_agg_request);
3805 if (is_valid_table(table_id)) {
3806 struct cls_cursor cursor;
3809 cls_cursor_init(&cursor, &ofproto->cls, target);
3810 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3811 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3812 uint64_t packet_count;
3813 uint64_t byte_count;
3815 rule_get_stats(rule, &packet_count, &byte_count);
3817 total_packets += packet_count;
3818 total_bytes += byte_count;
3824 oasr->flow_count = htonl(n_flows);
3825 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3826 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3827 memset(oasr->pad, 0, sizeof oasr->pad);
3831 handle_aggregate_stats_request(struct ofconn *ofconn,
3832 const struct ofp_header *oh)
3834 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3835 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3836 struct ofp_aggregate_stats_reply *reply;
3837 struct cls_rule target;
3840 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3843 msg = start_ofp_stats_reply(oh, sizeof *reply);
3844 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3845 query_aggregate_stats(ofproto, &target, request->out_port,
3846 request->table_id, reply);
3847 ofconn_send_reply(ofconn, msg);
3852 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3854 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3855 struct nx_aggregate_stats_request *request;
3856 struct ofp_aggregate_stats_reply *reply;
3857 struct cls_rule target;
3862 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3864 /* Dissect the message. */
3865 request = ofpbuf_pull(&b, sizeof *request);
3866 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3871 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3875 COVERAGE_INC(ofproto_flows_req);
3876 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3877 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3878 query_aggregate_stats(ofproto, &target, request->out_port,
3879 request->table_id, reply);
3880 ofconn_send_reply(ofconn, buf);
3885 struct queue_stats_cbdata {
3886 struct ofconn *ofconn;
3887 struct ofport *ofport;
3892 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3893 const struct netdev_queue_stats *stats)
3895 struct ofp_queue_stats *reply;
3897 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3898 reply->port_no = htons(cbdata->ofport->opp.port_no);
3899 memset(reply->pad, 0, sizeof reply->pad);
3900 reply->queue_id = htonl(queue_id);
3901 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3902 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3903 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3907 handle_queue_stats_dump_cb(uint32_t queue_id,
3908 struct netdev_queue_stats *stats,
3911 struct queue_stats_cbdata *cbdata = cbdata_;
3913 put_queue_stats(cbdata, queue_id, stats);
3917 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3918 struct queue_stats_cbdata *cbdata)
3920 cbdata->ofport = port;
3921 if (queue_id == OFPQ_ALL) {
3922 netdev_dump_queue_stats(port->netdev,
3923 handle_queue_stats_dump_cb, cbdata);
3925 struct netdev_queue_stats stats;
3927 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3928 put_queue_stats(cbdata, queue_id, &stats);
3934 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3936 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3937 const struct ofp_queue_stats_request *qsr;
3938 struct queue_stats_cbdata cbdata;
3939 struct ofport *port;
3940 unsigned int port_no;
3943 qsr = ofputil_stats_body(oh);
3945 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3948 COVERAGE_INC(ofproto_queue_req);
3950 cbdata.ofconn = ofconn;
3951 cbdata.msg = start_ofp_stats_reply(oh, 128);
3953 port_no = ntohs(qsr->port_no);
3954 queue_id = ntohl(qsr->queue_id);
3955 if (port_no == OFPP_ALL) {
3956 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3957 handle_queue_stats_for_port(port, queue_id, &cbdata);
3959 } else if (port_no < ofproto->max_ports) {
3960 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3962 handle_queue_stats_for_port(port, queue_id, &cbdata);
3965 ofpbuf_delete(cbdata.msg);
3966 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3968 ofconn_send_reply(ofconn, cbdata.msg);
3973 /* Updates 'facet''s used time. Caller is responsible for calling
3974 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3976 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3979 if (used > facet->used) {
3981 if (used > facet->rule->used) {
3982 facet->rule->used = used;
3984 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3988 /* Folds the statistics from 'stats' into the counters in 'facet'.
3990 * Because of the meaning of a facet's counters, it only makes sense to do this
3991 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3992 * packet that was sent by hand or if it represents statistics that have been
3993 * cleared out of the datapath. */
3995 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3996 const struct dpif_flow_stats *stats)
3998 if (stats->n_packets || stats->used > facet->used) {
3999 facet_update_time(ofproto, facet, stats->used);
4000 facet->packet_count += stats->n_packets;
4001 facet->byte_count += stats->n_bytes;
4002 facet_push_stats(ofproto, facet);
4003 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4008 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
4010 uint64_t rs_packets, rs_bytes;
4012 assert(facet->packet_count >= facet->rs_packet_count);
4013 assert(facet->byte_count >= facet->rs_byte_count);
4014 assert(facet->used >= facet->rs_used);
4016 rs_packets = facet->packet_count - facet->rs_packet_count;
4017 rs_bytes = facet->byte_count - facet->rs_byte_count;
4019 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
4020 facet->rs_packet_count = facet->packet_count;
4021 facet->rs_byte_count = facet->byte_count;
4022 facet->rs_used = facet->used;
4024 flow_push_stats(ofproto, facet->rule, &facet->flow,
4025 rs_packets, rs_bytes, facet->used);
4029 struct ofproto_push {
4030 struct action_xlate_ctx ctx;
4037 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4039 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
4042 rule->packet_count += push->packets;
4043 rule->byte_count += push->bytes;
4044 rule->used = MAX(push->used, rule->used);
4048 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4049 * 'rule''s actions. */
4051 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
4052 struct flow *flow, uint64_t packets, uint64_t bytes,
4055 struct ofproto_push push;
4057 push.packets = packets;
4061 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
4062 push.ctx.resubmit_hook = push_resubmit;
4063 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
4066 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4067 * in which no matching flow already exists in the flow table.
4069 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4070 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
4071 * OpenFlow error code as encoded by ofp_mkerr() on failure.
4073 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4076 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
4078 struct ofproto *p = ofconn_get_ofproto(ofconn);
4079 struct ofpbuf *packet;
4084 if (fm->flags & OFPFF_CHECK_OVERLAP
4085 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
4086 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
4090 if (fm->buffer_id != UINT32_MAX) {
4091 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
4095 in_port = UINT16_MAX;
4098 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
4099 fm->idle_timeout, fm->hard_timeout, fm->cookie,
4100 fm->flags & OFPFF_SEND_FLOW_REM);
4101 rule_insert(p, rule);
4103 rule_execute(p, rule, in_port, packet);
4108 static struct rule *
4109 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
4111 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
4115 send_buffered_packet(struct ofconn *ofconn,
4116 struct rule *rule, uint32_t buffer_id)
4118 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4119 struct ofpbuf *packet;
4123 if (buffer_id == UINT32_MAX) {
4127 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
4132 rule_execute(ofproto, rule, in_port, packet);
4137 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4139 struct modify_flows_cbdata {
4140 struct ofproto *ofproto;
4141 const struct flow_mod *fm;
4145 static int modify_flow(struct ofproto *, const struct flow_mod *,
4148 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4149 * encoded by ofp_mkerr() on failure.
4151 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4154 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
4156 struct ofproto *p = ofconn_get_ofproto(ofconn);
4157 struct rule *match = NULL;
4158 struct cls_cursor cursor;
4161 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4162 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4163 if (!rule_is_hidden(rule)) {
4165 modify_flow(p, fm, rule);
4170 /* This credits the packet to whichever flow happened to match last.
4171 * That's weird. Maybe we should do a lookup for the flow that
4172 * actually matches the packet? Who knows. */
4173 send_buffered_packet(ofconn, match, fm->buffer_id);
4176 return add_flow(ofconn, fm);
4180 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4181 * code as encoded by ofp_mkerr() on failure.
4183 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4186 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4188 struct ofproto *p = ofconn_get_ofproto(ofconn);
4189 struct rule *rule = find_flow_strict(p, fm);
4190 if (rule && !rule_is_hidden(rule)) {
4191 modify_flow(p, fm, rule);
4192 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4194 return add_flow(ofconn, fm);
4198 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4199 * been identified as a flow in 'p''s flow table to be modified, by changing
4200 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4201 * ofp_action[] structures). */
4203 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4205 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4207 rule->flow_cookie = fm->cookie;
4209 /* If the actions are the same, do nothing. */
4210 if (fm->n_actions == rule->n_actions
4212 || !memcmp(fm->actions, rule->actions, actions_len))) {
4216 /* Replace actions. */
4217 free(rule->actions);
4218 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4219 rule->n_actions = fm->n_actions;
4221 p->need_revalidate = true;
4226 /* OFPFC_DELETE implementation. */
4228 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4230 /* Implements OFPFC_DELETE. */
4232 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4234 struct rule *rule, *next_rule;
4235 struct cls_cursor cursor;
4237 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4238 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4239 delete_flow(p, rule, htons(fm->out_port));
4243 /* Implements OFPFC_DELETE_STRICT. */
4245 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4247 struct rule *rule = find_flow_strict(p, fm);
4249 delete_flow(p, rule, htons(fm->out_port));
4253 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4254 * been identified as a flow to delete from 'p''s flow table, by deleting the
4255 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4258 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4259 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4260 * specified 'out_port'. */
4262 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4264 if (rule_is_hidden(rule)) {
4268 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4272 rule_send_removed(p, rule, OFPRR_DELETE);
4273 rule_remove(p, rule);
4277 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4279 struct ofproto *p = ofconn_get_ofproto(ofconn);
4283 error = reject_slave_controller(ofconn, "flow_mod");
4288 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
4293 /* We do not support the emergency flow cache. It will hopefully get
4294 * dropped from OpenFlow in the near future. */
4295 if (fm.flags & OFPFF_EMERG) {
4296 /* There isn't a good fit for an error code, so just state that the
4297 * flow table is full. */
4298 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4301 error = validate_actions(fm.actions, fm.n_actions,
4302 &fm.cr.flow, p->max_ports);
4307 switch (fm.command) {
4309 return add_flow(ofconn, &fm);
4312 return modify_flows_loose(ofconn, &fm);
4314 case OFPFC_MODIFY_STRICT:
4315 return modify_flow_strict(ofconn, &fm);
4318 delete_flows_loose(p, &fm);
4321 case OFPFC_DELETE_STRICT:
4322 delete_flow_strict(p, &fm);
4326 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4331 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4333 const struct nxt_tun_id_cookie *msg
4334 = (const struct nxt_tun_id_cookie *) oh;
4335 enum nx_flow_format flow_format;
4337 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4338 ofconn_set_flow_format(ofconn, flow_format);
4344 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4346 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4347 struct nx_role_request *reply;
4351 if (ofconn->type != OFCONN_PRIMARY) {
4352 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4354 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4357 role = ntohl(nrr->role);
4358 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4359 && role != NX_ROLE_SLAVE) {
4360 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4362 /* There's no good error code for this. */
4363 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4366 if (role == NX_ROLE_MASTER) {
4367 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4368 struct ofconn *other;
4370 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
4371 if (other->role == NX_ROLE_MASTER) {
4372 other->role = NX_ROLE_SLAVE;
4376 ofconn->role = role;
4378 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4379 reply->role = htonl(role);
4380 ofconn_send_reply(ofconn, buf);
4386 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4388 const struct nxt_set_flow_format *msg
4389 = (const struct nxt_set_flow_format *) oh;
4392 format = ntohl(msg->format);
4393 if (format == NXFF_OPENFLOW10
4394 || format == NXFF_TUN_ID_FROM_COOKIE
4395 || format == NXFF_NXM) {
4396 ofconn_set_flow_format(ofconn, format);
4399 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4404 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4406 struct ofp_header *ob;
4409 /* Currently, everything executes synchronously, so we can just
4410 * immediately send the barrier reply. */
4411 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4412 ofconn_send_reply(ofconn, buf);
4417 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4419 const struct ofp_header *oh = msg->data;
4420 const struct ofputil_msg_type *type;
4423 error = ofputil_decode_msg_type(oh, &type);
4428 switch (ofputil_msg_type_code(type)) {
4429 /* OpenFlow requests. */
4430 case OFPUTIL_OFPT_ECHO_REQUEST:
4431 return handle_echo_request(ofconn, oh);
4433 case OFPUTIL_OFPT_FEATURES_REQUEST:
4434 return handle_features_request(ofconn, oh);
4436 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4437 return handle_get_config_request(ofconn, oh);
4439 case OFPUTIL_OFPT_SET_CONFIG:
4440 return handle_set_config(ofconn, msg->data);
4442 case OFPUTIL_OFPT_PACKET_OUT:
4443 return handle_packet_out(ofconn, oh);
4445 case OFPUTIL_OFPT_PORT_MOD:
4446 return handle_port_mod(ofconn, oh);
4448 case OFPUTIL_OFPT_FLOW_MOD:
4449 return handle_flow_mod(ofconn, oh);
4451 case OFPUTIL_OFPT_BARRIER_REQUEST:
4452 return handle_barrier_request(ofconn, oh);
4454 /* OpenFlow replies. */
4455 case OFPUTIL_OFPT_ECHO_REPLY:
4458 /* Nicira extension requests. */
4459 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4460 return handle_tun_id_from_cookie(ofconn, oh);
4462 case OFPUTIL_NXT_ROLE_REQUEST:
4463 return handle_role_request(ofconn, oh);
4465 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4466 return handle_nxt_set_flow_format(ofconn, oh);
4468 case OFPUTIL_NXT_FLOW_MOD:
4469 return handle_flow_mod(ofconn, oh);
4471 /* OpenFlow statistics requests. */
4472 case OFPUTIL_OFPST_DESC_REQUEST:
4473 return handle_desc_stats_request(ofconn, oh);
4475 case OFPUTIL_OFPST_FLOW_REQUEST:
4476 return handle_flow_stats_request(ofconn, oh);
4478 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4479 return handle_aggregate_stats_request(ofconn, oh);
4481 case OFPUTIL_OFPST_TABLE_REQUEST:
4482 return handle_table_stats_request(ofconn, oh);
4484 case OFPUTIL_OFPST_PORT_REQUEST:
4485 return handle_port_stats_request(ofconn, oh);
4487 case OFPUTIL_OFPST_QUEUE_REQUEST:
4488 return handle_queue_stats_request(ofconn, oh);
4490 /* Nicira extension statistics requests. */
4491 case OFPUTIL_NXST_FLOW_REQUEST:
4492 return handle_nxst_flow(ofconn, oh);
4494 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4495 return handle_nxst_aggregate(ofconn, oh);
4497 case OFPUTIL_INVALID:
4498 case OFPUTIL_OFPT_HELLO:
4499 case OFPUTIL_OFPT_ERROR:
4500 case OFPUTIL_OFPT_FEATURES_REPLY:
4501 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4502 case OFPUTIL_OFPT_PACKET_IN:
4503 case OFPUTIL_OFPT_FLOW_REMOVED:
4504 case OFPUTIL_OFPT_PORT_STATUS:
4505 case OFPUTIL_OFPT_BARRIER_REPLY:
4506 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4507 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4508 case OFPUTIL_OFPST_DESC_REPLY:
4509 case OFPUTIL_OFPST_FLOW_REPLY:
4510 case OFPUTIL_OFPST_QUEUE_REPLY:
4511 case OFPUTIL_OFPST_PORT_REPLY:
4512 case OFPUTIL_OFPST_TABLE_REPLY:
4513 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4514 case OFPUTIL_NXT_ROLE_REPLY:
4515 case OFPUTIL_NXT_FLOW_REMOVED:
4516 case OFPUTIL_NXST_FLOW_REPLY:
4517 case OFPUTIL_NXST_AGGREGATE_REPLY:
4519 if (VLOG_IS_WARN_ENABLED()) {
4520 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4521 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4524 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4525 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4527 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4533 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4535 int error = handle_openflow__(ofconn, ofp_msg);
4537 send_error_oh(ofconn, ofp_msg->data, error);
4539 COVERAGE_INC(ofproto_recv_openflow);
4543 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4545 struct facet *facet;
4548 /* Obtain in_port and tun_id, at least. */
4549 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4551 /* Set header pointers in 'flow'. */
4552 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4554 if (cfm_should_process_flow(&flow)) {
4555 ofproto_process_cfm(p, &flow, upcall->packet);
4556 ofpbuf_delete(upcall->packet);
4558 } else if (p->ofhooks->special_cb
4559 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4560 ofpbuf_delete(upcall->packet);
4564 /* Check with in-band control to see if this packet should be sent
4565 * to the local port regardless of the flow table. */
4566 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4567 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
4570 facet = facet_lookup_valid(p, &flow);
4572 struct rule *rule = rule_lookup(p, &flow);
4574 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4575 struct ofport *port = get_port(p, flow.in_port);
4577 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4578 COVERAGE_INC(ofproto_no_packet_in);
4579 /* XXX install 'drop' flow entry */
4580 ofpbuf_delete(upcall->packet);
4584 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4588 COVERAGE_INC(ofproto_packet_in);
4589 send_packet_in(p, upcall, &flow, false);
4593 facet = facet_create(p, rule, &flow, upcall->packet);
4594 } else if (!facet->may_install) {
4595 /* The facet is not installable, that is, we need to process every
4596 * packet, so process the current packet's actions into 'facet'. */
4597 facet_make_actions(p, facet, upcall->packet);
4600 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4602 * Extra-special case for fail-open mode.
4604 * We are in fail-open mode and the packet matched the fail-open rule,
4605 * but we are connected to a controller too. We should send the packet
4606 * up to the controller in the hope that it will try to set up a flow
4607 * and thereby allow us to exit fail-open.
4609 * See the top-level comment in fail-open.c for more information.
4611 send_packet_in(p, upcall, &flow, true);
4614 facet_execute(p, facet, upcall->packet);
4615 facet_install(p, facet, false);
4619 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4623 switch (upcall->type) {
4624 case DPIF_UC_ACTION:
4625 COVERAGE_INC(ofproto_ctlr_action);
4626 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4627 send_packet_in(p, upcall, &flow, false);
4630 case DPIF_UC_SAMPLE:
4632 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4633 ofproto_sflow_received(p->sflow, upcall, &flow);
4635 ofpbuf_delete(upcall->packet);
4639 handle_miss_upcall(p, upcall);
4642 case DPIF_N_UC_TYPES:
4644 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4649 /* Flow expiration. */
4651 static int ofproto_dp_max_idle(const struct ofproto *);
4652 static void ofproto_update_stats(struct ofproto *);
4653 static void rule_expire(struct ofproto *, struct rule *);
4654 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4656 /* This function is called periodically by ofproto_run(). Its job is to
4657 * collect updates for the flows that have been installed into the datapath,
4658 * most importantly when they last were used, and then use that information to
4659 * expire flows that have not been used recently.
4661 * Returns the number of milliseconds after which it should be called again. */
4663 ofproto_expire(struct ofproto *ofproto)
4665 struct rule *rule, *next_rule;
4666 struct cls_cursor cursor;
4669 /* Update stats for each flow in the datapath. */
4670 ofproto_update_stats(ofproto);
4672 /* Expire facets that have been idle too long. */
4673 dp_max_idle = ofproto_dp_max_idle(ofproto);
4674 ofproto_expire_facets(ofproto, dp_max_idle);
4676 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4677 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4678 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4679 rule_expire(ofproto, rule);
4682 /* Let the hook know that we're at a stable point: all outstanding data
4683 * in existing flows has been accounted to the account_cb. Thus, the
4684 * hook can now reasonably do operations that depend on having accurate
4685 * flow volume accounting (currently, that's just bond rebalancing). */
4686 if (ofproto->ofhooks->account_checkpoint_cb) {
4687 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4690 return MIN(dp_max_idle, 1000);
4693 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4695 * This function also pushes statistics updates to rules which each facet
4696 * resubmits into. Generally these statistics will be accurate. However, if a
4697 * facet changes the rule it resubmits into at some time in between
4698 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4699 * old rule will be incorrectly attributed to the new rule. This could be
4700 * avoided by calling ofproto_update_stats() whenever rules are created or
4701 * deleted. However, the performance impact of making so many calls to the
4702 * datapath do not justify the benefit of having perfectly accurate statistics.
4705 ofproto_update_stats(struct ofproto *p)
4707 const struct dpif_flow_stats *stats;
4708 struct dpif_flow_dump dump;
4709 const struct nlattr *key;
4712 dpif_flow_dump_start(&dump, p->dpif);
4713 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4714 struct facet *facet;
4717 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4721 odp_flow_key_format(key, key_len, &s);
4722 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4728 facet = facet_find(p, &flow);
4730 if (facet && facet->installed) {
4732 if (stats->n_packets >= facet->dp_packet_count) {
4733 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4735 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4738 if (stats->n_bytes >= facet->dp_byte_count) {
4739 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4741 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4744 facet->dp_packet_count = stats->n_packets;
4745 facet->dp_byte_count = stats->n_bytes;
4747 facet_update_time(p, facet, stats->used);
4748 facet_account(p, facet, stats->n_bytes);
4749 facet_push_stats(p, facet);
4751 /* There's a flow in the datapath that we know nothing about.
4753 COVERAGE_INC(ofproto_unexpected_rule);
4754 dpif_flow_del(p->dpif, key, key_len, NULL);
4757 dpif_flow_dump_done(&dump);
4760 /* Calculates and returns the number of milliseconds of idle time after which
4761 * facets should expire from the datapath and we should fold their statistics
4762 * into their parent rules in userspace. */
4764 ofproto_dp_max_idle(const struct ofproto *ofproto)
4767 * Idle time histogram.
4769 * Most of the time a switch has a relatively small number of facets. When
4770 * this is the case we might as well keep statistics for all of them in
4771 * userspace and to cache them in the kernel datapath for performance as
4774 * As the number of facets increases, the memory required to maintain
4775 * statistics about them in userspace and in the kernel becomes
4776 * significant. However, with a large number of facets it is likely that
4777 * only a few of them are "heavy hitters" that consume a large amount of
4778 * bandwidth. At this point, only heavy hitters are worth caching in the
4779 * kernel and maintaining in userspaces; other facets we can discard.
4781 * The technique used to compute the idle time is to build a histogram with
4782 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4783 * that is installed in the kernel gets dropped in the appropriate bucket.
4784 * After the histogram has been built, we compute the cutoff so that only
4785 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4786 * cached. At least the most-recently-used bucket of facets is kept, so
4787 * actually an arbitrary number of facets can be kept in any given
4788 * expiration run (though the next run will delete most of those unless
4789 * they receive additional data).
4791 * This requires a second pass through the facets, in addition to the pass
4792 * made by ofproto_update_stats(), because the former function never looks
4793 * at uninstallable facets.
4795 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4796 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4797 int buckets[N_BUCKETS] = { 0 };
4798 struct facet *facet;
4803 total = hmap_count(&ofproto->facets);
4804 if (total <= 1000) {
4805 return N_BUCKETS * BUCKET_WIDTH;
4808 /* Build histogram. */
4810 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4811 long long int idle = now - facet->used;
4812 int bucket = (idle <= 0 ? 0
4813 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4814 : (unsigned int) idle / BUCKET_WIDTH);
4818 /* Find the first bucket whose flows should be expired. */
4819 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4820 if (buckets[bucket]) {
4823 subtotal += buckets[bucket++];
4824 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4829 if (VLOG_IS_DBG_ENABLED()) {
4833 ds_put_cstr(&s, "keep");
4834 for (i = 0; i < N_BUCKETS; i++) {
4836 ds_put_cstr(&s, ", drop");
4839 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4842 VLOG_INFO("%s: %s (msec:count)",
4843 dpif_name(ofproto->dpif), ds_cstr(&s));
4847 return bucket * BUCKET_WIDTH;
4851 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4853 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4854 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4855 struct ofexpired expired;
4857 if (facet->installed) {
4858 struct dpif_flow_stats stats;
4860 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4862 facet_update_stats(ofproto, facet, &stats);
4865 expired.flow = facet->flow;
4866 expired.packet_count = facet->packet_count;
4867 expired.byte_count = facet->byte_count;
4868 expired.used = facet->used;
4869 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4874 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4876 long long int cutoff = time_msec() - dp_max_idle;
4877 struct facet *facet, *next_facet;
4879 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4880 facet_active_timeout(ofproto, facet);
4881 if (facet->used < cutoff) {
4882 facet_remove(ofproto, facet);
4887 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4888 * then delete it entirely. */
4890 rule_expire(struct ofproto *ofproto, struct rule *rule)
4892 struct facet *facet, *next_facet;
4896 /* Has 'rule' expired? */
4898 if (rule->hard_timeout
4899 && now > rule->created + rule->hard_timeout * 1000) {
4900 reason = OFPRR_HARD_TIMEOUT;
4901 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4902 && now >rule->used + rule->idle_timeout * 1000) {
4903 reason = OFPRR_IDLE_TIMEOUT;
4908 COVERAGE_INC(ofproto_expired);
4910 /* Update stats. (This is a no-op if the rule expired due to an idle
4911 * timeout, because that only happens when the rule has no facets left.) */
4912 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4913 facet_remove(ofproto, facet);
4916 /* Get rid of the rule. */
4917 if (!rule_is_hidden(rule)) {
4918 rule_send_removed(ofproto, rule, reason);
4920 rule_remove(ofproto, rule);
4924 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4926 struct ofputil_flow_removed fr;
4927 struct ofconn *ofconn;
4929 if (!rule->send_flow_removed) {
4934 fr.cookie = rule->flow_cookie;
4936 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4937 fr.idle_timeout = rule->idle_timeout;
4938 fr.packet_count = rule->packet_count;
4939 fr.byte_count = rule->byte_count;
4941 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4944 if (!rconn_is_connected(ofconn->rconn)
4945 || !ofconn_receives_async_msgs(ofconn)) {
4949 /* This accounts flow expirations as if they were replies to OpenFlow
4950 * requests. That works because preventing OpenFlow requests from
4951 * being processed also prevents new flows from being added (and
4952 * expiring). (It also prevents processing OpenFlow requests that
4953 * would not add new flows, so it is imperfect.) */
4954 msg = ofputil_encode_flow_removed(&fr, ofconn_get_flow_format(ofconn));
4955 ofconn_send_reply(ofconn, msg);
4959 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4960 * The returned statistics include statistics for all of 'rule''s facets. */
4962 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4965 struct facet *facet;
4967 /* Start from historical data for 'rule' itself that are no longer tracked
4968 * in facets. This counts, for example, facets that have expired. */
4969 p = rule->packet_count;
4970 b = rule->byte_count;
4972 /* Add any statistics that are tracked by facets. This includes
4973 * statistical data recently updated by ofproto_update_stats() as well as
4974 * stats for packets that were executed "by hand" via dpif_execute(). */
4975 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4976 p += facet->packet_count;
4977 b += facet->byte_count;
4984 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4986 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4988 struct ofconn *ofconn = ofconn_;
4990 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4991 ofconn->packet_in_counter, 100);
4994 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4995 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4996 * scheduler for sending.
4998 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4999 * Otherwise, ownership is transferred to this function. */
5001 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
5002 const struct flow *flow, bool clone)
5004 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5005 struct ofputil_packet_in pin;
5008 /* Figure out the easy parts. */
5009 pin.packet = upcall->packet;
5010 pin.in_port = odp_port_to_ofp_port(flow->in_port);
5011 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
5013 /* Get OpenFlow buffer_id. */
5014 if (upcall->type == DPIF_UC_ACTION) {
5015 pin.buffer_id = UINT32_MAX;
5016 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
5017 pin.buffer_id = pktbuf_get_null();
5018 } else if (!ofconn->pktbuf) {
5019 pin.buffer_id = UINT32_MAX;
5021 pin.buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet,
5025 /* Figure out how much of the packet to send. */
5026 pin.send_len = upcall->packet->size;
5027 if (pin.buffer_id != UINT32_MAX) {
5028 pin.send_len = MIN(pin.send_len, ofconn->miss_send_len);
5030 if (upcall->type == DPIF_UC_ACTION) {
5031 pin.send_len = MIN(pin.send_len, upcall->userdata);
5034 /* Make OFPT_PACKET_IN and hand over to packet scheduler. It might
5035 * immediately call into do_send_packet_in() or it might buffer it for a
5036 * while (until a later call to pinsched_run()). */
5037 msg = ofputil_encode_packet_in(&pin, clone ? NULL : upcall->packet);
5038 pinsched_send(ofconn->schedulers[upcall->type == DPIF_UC_MISS ? 0 : 1],
5039 flow->in_port, msg, do_send_packet_in, ofconn);
5042 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
5043 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
5044 * their individual configurations.
5046 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
5047 * Otherwise, ownership is transferred to this function. */
5049 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
5050 const struct flow *flow, bool clone)
5052 struct ofconn *ofconn, *prev;
5055 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5056 if (ofconn_receives_async_msgs(ofconn)) {
5058 schedule_packet_in(prev, upcall, flow, true);
5064 schedule_packet_in(prev, upcall, flow, clone);
5065 } else if (!clone) {
5066 ofpbuf_delete(upcall->packet);
5071 pick_datapath_id(const struct ofproto *ofproto)
5073 const struct ofport *port;
5075 port = get_port(ofproto, ODPP_LOCAL);
5077 uint8_t ea[ETH_ADDR_LEN];
5080 error = netdev_get_etheraddr(port->netdev, ea);
5082 return eth_addr_to_uint64(ea);
5084 VLOG_WARN("could not get MAC address for %s (%s)",
5085 netdev_get_name(port->netdev), strerror(error));
5087 return ofproto->fallback_dpid;
5091 pick_fallback_dpid(void)
5093 uint8_t ea[ETH_ADDR_LEN];
5094 eth_addr_nicira_random(ea);
5095 return eth_addr_to_uint64(ea);
5099 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
5100 void *aux OVS_UNUSED)
5102 const struct shash_node *node;
5106 SHASH_FOR_EACH (node, &all_ofprotos) {
5107 ds_put_format(&results, "%s\n", node->name);
5109 unixctl_command_reply(conn, 200, ds_cstr(&results));
5110 ds_destroy(&results);
5113 struct ofproto_trace {
5114 struct action_xlate_ctx ctx;
5120 trace_format_rule(struct ds *result, int level, const struct rule *rule)
5122 ds_put_char_multiple(result, '\t', level);
5124 ds_put_cstr(result, "No match\n");
5128 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
5129 ntohll(rule->flow_cookie));
5130 cls_rule_format(&rule->cr, result);
5131 ds_put_char(result, '\n');
5133 ds_put_char_multiple(result, '\t', level);
5134 ds_put_cstr(result, "OpenFlow ");
5135 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
5136 rule->n_actions * sizeof *rule->actions);
5137 ds_put_char(result, '\n');
5141 trace_format_flow(struct ds *result, int level, const char *title,
5142 struct ofproto_trace *trace)
5144 ds_put_char_multiple(result, '\t', level);
5145 ds_put_format(result, "%s: ", title);
5146 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5147 ds_put_cstr(result, "unchanged");
5149 flow_format(result, &trace->ctx.flow);
5150 trace->flow = trace->ctx.flow;
5152 ds_put_char(result, '\n');
5156 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
5158 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5159 struct ds *result = trace->result;
5161 ds_put_char(result, '\n');
5162 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5163 trace_format_rule(result, ctx->recurse + 1, rule);
5167 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5168 void *aux OVS_UNUSED)
5170 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5171 char *args = xstrdup(args_);
5172 char *save_ptr = NULL;
5173 struct ofproto *ofproto;
5174 struct ofpbuf packet;
5182 ofpbuf_init(&packet, strlen(args) / 2);
5185 dpname = strtok_r(args, " ", &save_ptr);
5186 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5187 in_port_s = strtok_r(NULL, " ", &save_ptr);
5188 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5189 if (!dpname || !in_port_s || !packet_s) {
5190 unixctl_command_reply(conn, 501, "Bad command syntax");
5194 ofproto = shash_find_data(&all_ofprotos, dpname);
5196 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5201 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5202 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5204 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5205 packet_s += strspn(packet_s, " ");
5206 if (*packet_s != '\0') {
5207 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5210 if (packet.size < ETH_HEADER_LEN) {
5211 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5215 ds_put_cstr(&result, "Packet: ");
5216 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5217 ds_put_cstr(&result, s);
5220 flow_extract(&packet, tun_id, in_port, &flow);
5221 ds_put_cstr(&result, "Flow: ");
5222 flow_format(&result, &flow);
5223 ds_put_char(&result, '\n');
5225 rule = rule_lookup(ofproto, &flow);
5226 trace_format_rule(&result, 0, rule);
5228 struct ofproto_trace trace;
5229 struct ofpbuf *odp_actions;
5231 trace.result = &result;
5233 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5234 trace.ctx.resubmit_hook = trace_resubmit;
5235 odp_actions = xlate_actions(&trace.ctx,
5236 rule->actions, rule->n_actions);
5238 ds_put_char(&result, '\n');
5239 trace_format_flow(&result, 0, "Final flow", &trace);
5240 ds_put_cstr(&result, "Datapath actions: ");
5241 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5242 ofpbuf_delete(odp_actions);
5245 unixctl_command_reply(conn, 200, ds_cstr(&result));
5248 ds_destroy(&result);
5249 ofpbuf_uninit(&packet);
5254 ofproto_unixctl_init(void)
5256 static bool registered;
5262 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5263 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5267 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5268 struct ofpbuf *odp_actions, tag_type *tags,
5269 uint16_t *nf_output_iface, void *ofproto_)
5271 struct ofproto *ofproto = ofproto_;
5272 struct mac_entry *dst_mac;
5274 /* Drop frames for reserved multicast addresses. */
5275 if (eth_addr_is_reserved(flow->dl_dst)) {
5279 /* Learn source MAC (but don't try to learn from revalidation). */
5281 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
5282 struct mac_entry *src_mac;
5284 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
5285 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
5286 /* The log messages here could actually be useful in debugging,
5287 * so keep the rate limit relatively high. */
5288 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5289 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5290 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5292 ofproto_revalidate(ofproto,
5293 mac_learning_changed(ofproto->ml, src_mac));
5294 src_mac->port.i = flow->in_port;
5298 /* Determine output port. */
5299 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
5301 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5302 nf_output_iface, odp_actions);
5304 int out_port = dst_mac->port.i;
5305 if (out_port != flow->in_port) {
5306 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5307 *nf_output_iface = out_port;
5316 static const struct ofhooks default_ofhooks = {
5317 default_normal_ofhook_cb,