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"
28 #include "classifier.h"
30 #include "discovery.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"
58 #include "stream-ssl.h"
62 #include "unaligned.h"
67 VLOG_DEFINE_THIS_MODULE(ofproto);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(odp_overflow);
72 COVERAGE_DEFINE(ofproto_agg_request);
73 COVERAGE_DEFINE(ofproto_costly_flags);
74 COVERAGE_DEFINE(ofproto_ctlr_action);
75 COVERAGE_DEFINE(ofproto_del_rule);
76 COVERAGE_DEFINE(ofproto_error);
77 COVERAGE_DEFINE(ofproto_expiration);
78 COVERAGE_DEFINE(ofproto_expired);
79 COVERAGE_DEFINE(ofproto_flows_req);
80 COVERAGE_DEFINE(ofproto_flush);
81 COVERAGE_DEFINE(ofproto_invalidated);
82 COVERAGE_DEFINE(ofproto_no_packet_in);
83 COVERAGE_DEFINE(ofproto_ofconn_stuck);
84 COVERAGE_DEFINE(ofproto_ofp2odp);
85 COVERAGE_DEFINE(ofproto_packet_in);
86 COVERAGE_DEFINE(ofproto_packet_out);
87 COVERAGE_DEFINE(ofproto_queue_req);
88 COVERAGE_DEFINE(ofproto_recv_openflow);
89 COVERAGE_DEFINE(ofproto_reinit_ports);
90 COVERAGE_DEFINE(ofproto_unexpected_rule);
91 COVERAGE_DEFINE(ofproto_uninstallable);
92 COVERAGE_DEFINE(ofproto_update_port);
94 #include "sflow_api.h"
99 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
100 struct netdev *netdev;
101 struct ofp_phy_port opp; /* In host byte order. */
105 static void ofport_free(struct ofport *);
106 static void hton_ofp_phy_port(struct ofp_phy_port *);
108 struct action_xlate_ctx {
109 /* action_xlate_ctx_init() initializes these members. */
112 struct ofproto *ofproto;
114 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
115 * this flow when actions change header fields. */
118 /* The packet corresponding to 'flow', or a null pointer if we are
119 * revalidating without a packet to refer to. */
120 const struct ofpbuf *packet;
122 /* If nonnull, called just before executing a resubmit action.
124 * This is normally null so the client has to set it manually after
125 * calling action_xlate_ctx_init(). */
126 void (*resubmit_hook)(struct action_xlate_ctx *, const struct rule *);
128 /* xlate_actions() initializes and uses these members. The client might want
129 * to look at them after it returns. */
131 struct ofpbuf *odp_actions; /* Datapath actions. */
132 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
133 bool may_set_up_flow; /* True ordinarily; false if the actions must
134 * be reassessed for every packet. */
135 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
137 /* xlate_actions() initializes and uses these members, but the client has no
138 * reason to look at them. */
140 int recurse; /* Recursion level, via xlate_table_action. */
141 int last_pop_priority; /* Offset in 'odp_actions' just past most
142 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
145 static void action_xlate_ctx_init(struct action_xlate_ctx *,
146 struct ofproto *, const struct flow *,
147 const struct ofpbuf *);
148 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
149 const union ofp_action *in, size_t n_in);
151 /* An OpenFlow flow. */
153 long long int used; /* Time last used; time created if not used. */
154 long long int created; /* Creation time. */
158 * - Do include packets and bytes from facets that have been deleted or
159 * whose own statistics have been folded into the rule.
161 * - Do include packets and bytes sent "by hand" that were accounted to
162 * the rule without any facet being involved (this is a rare corner
163 * case in rule_execute()).
165 * - Do not include packet or bytes that can be obtained from any facet's
166 * packet_count or byte_count member or that can be obtained from the
167 * datapath by, e.g., dpif_flow_get() for any facet.
169 uint64_t packet_count; /* Number of packets received. */
170 uint64_t byte_count; /* Number of bytes received. */
172 ovs_be64 flow_cookie; /* Controller-issued identifier. */
174 struct cls_rule cr; /* In owning ofproto's classifier. */
175 uint16_t idle_timeout; /* In seconds from time of last use. */
176 uint16_t hard_timeout; /* In seconds from time of creation. */
177 bool send_flow_removed; /* Send a flow removed message? */
178 int n_actions; /* Number of elements in actions[]. */
179 union ofp_action *actions; /* OpenFlow actions. */
180 struct list facets; /* List of "struct facet"s. */
183 static struct rule *rule_from_cls_rule(const struct cls_rule *);
184 static bool rule_is_hidden(const struct rule *);
186 static struct rule *rule_create(const struct cls_rule *,
187 const union ofp_action *, size_t n_actions,
188 uint16_t idle_timeout, uint16_t hard_timeout,
189 ovs_be64 flow_cookie, bool send_flow_removed);
190 static void rule_destroy(struct ofproto *, struct rule *);
191 static void rule_free(struct rule *);
193 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
194 static void rule_insert(struct ofproto *, struct rule *);
195 static void rule_remove(struct ofproto *, struct rule *);
197 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
199 /* An exact-match instantiation of an OpenFlow flow. */
201 long long int used; /* Time last used; time created if not used. */
205 * - Do include packets and bytes sent "by hand", e.g. with
208 * - Do include packets and bytes that were obtained from the datapath
209 * when a flow was deleted (e.g. dpif_flow_del()) or when its
210 * statistics were reset (e.g. dpif_flow_put() with
211 * DPIF_FP_ZERO_STATS).
213 * - Do not include any packets or bytes that can currently be obtained
214 * from the datapath by, e.g., dpif_flow_get().
216 uint64_t packet_count; /* Number of packets received. */
217 uint64_t byte_count; /* Number of bytes received. */
219 /* Number of bytes passed to account_cb. This may include bytes that can
220 * currently obtained from the datapath (thus, it can be greater than
222 uint64_t accounted_bytes;
224 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
225 struct list list_node; /* In owning rule's 'facets' list. */
226 struct rule *rule; /* Owning rule. */
227 struct flow flow; /* Exact-match flow. */
228 bool installed; /* Installed in datapath? */
229 bool may_install; /* True ordinarily; false if actions must
230 * be reassessed for every packet. */
231 size_t actions_len; /* Number of bytes in actions[]. */
232 struct nlattr *actions; /* Datapath actions. */
233 tag_type tags; /* Tags (set only by hooks). */
234 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
237 static struct facet *facet_create(struct ofproto *, struct rule *,
239 const struct ofpbuf *packet);
240 static void facet_remove(struct ofproto *, struct facet *);
241 static void facet_free(struct facet *);
243 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
244 static bool facet_revalidate(struct ofproto *, struct facet *);
246 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
247 static void facet_uninstall(struct ofproto *, struct facet *);
248 static void facet_flush_stats(struct ofproto *, struct facet *);
250 static void facet_make_actions(struct ofproto *, struct facet *,
251 const struct ofpbuf *packet);
252 static void facet_update_stats(struct ofproto *, struct facet *,
253 const struct dpif_flow_stats *);
255 /* ofproto supports two kinds of OpenFlow connections:
257 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
258 * maintains persistent connections to these controllers and by default
259 * sends them asynchronous messages such as packet-ins.
261 * - "Service" connections, e.g. from ovs-ofctl. When these connections
262 * drop, it is the other side's responsibility to reconnect them if
263 * necessary. ofproto does not send them asynchronous messages by default.
265 * Currently, active (tcp, ssl, unix) connections are always "primary"
266 * connections and passive (ptcp, pssl, punix) connections are always "service"
267 * connections. There is no inherent reason for this, but it reflects the
271 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
272 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
275 /* A listener for incoming OpenFlow "service" connections. */
277 struct hmap_node node; /* In struct ofproto's "services" hmap. */
278 struct pvconn *pvconn; /* OpenFlow connection listener. */
280 /* These are not used by ofservice directly. They are settings for
281 * accepted "struct ofconn"s from the pvconn. */
282 int probe_interval; /* Max idle time before probing, in seconds. */
283 int rate_limit; /* Max packet-in rate in packets per second. */
284 int burst_limit; /* Limit on accumulating packet credits. */
287 static struct ofservice *ofservice_lookup(struct ofproto *,
289 static int ofservice_create(struct ofproto *,
290 const struct ofproto_controller *);
291 static void ofservice_reconfigure(struct ofservice *,
292 const struct ofproto_controller *);
293 static void ofservice_destroy(struct ofproto *, struct ofservice *);
295 /* An OpenFlow connection. */
297 struct ofproto *ofproto; /* The ofproto that owns this connection. */
298 struct list node; /* In struct ofproto's "all_conns" list. */
299 struct rconn *rconn; /* OpenFlow connection. */
300 enum ofconn_type type; /* Type. */
301 enum nx_flow_format flow_format; /* Currently selected flow format. */
303 /* OFPT_PACKET_IN related data. */
304 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
305 #define N_SCHEDULERS 2
306 struct pinsched *schedulers[N_SCHEDULERS];
307 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
308 int miss_send_len; /* Bytes to send of buffered packets. */
310 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
311 * requests, and the maximum number before we stop reading OpenFlow
313 #define OFCONN_REPLY_MAX 100
314 struct rconn_packet_counter *reply_counter;
316 /* type == OFCONN_PRIMARY only. */
317 enum nx_role role; /* Role. */
318 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
319 struct discovery *discovery; /* Controller discovery object, if enabled. */
320 struct status_category *ss; /* Switch status category. */
321 enum ofproto_band band; /* In-band or out-of-band? */
325 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
327 static void ofconn_destroy(struct ofconn *);
328 static void ofconn_run(struct ofconn *);
329 static void ofconn_wait(struct ofconn *);
330 static bool ofconn_receives_async_msgs(const struct ofconn *);
331 static char *ofconn_make_name(const struct ofproto *, const char *target);
332 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
334 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
335 struct rconn_packet_counter *counter);
337 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
338 const struct flow *, bool clone);
339 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
343 uint64_t datapath_id; /* Datapath ID. */
344 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
345 char *mfr_desc; /* Manufacturer. */
346 char *hw_desc; /* Hardware. */
347 char *sw_desc; /* Software version. */
348 char *serial_desc; /* Serial number. */
349 char *dp_desc; /* Datapath description. */
353 struct netdev_monitor *netdev_monitor;
354 struct hmap ports; /* Contains "struct ofport"s. */
355 struct shash port_by_name;
359 struct switch_status *switch_status;
360 struct fail_open *fail_open;
361 struct netflow *netflow;
362 struct ofproto_sflow *sflow;
364 /* In-band control. */
365 struct in_band *in_band;
366 long long int next_in_band_update;
367 struct sockaddr_in *extra_in_band_remotes;
368 size_t n_extra_remotes;
372 struct classifier cls;
373 long long int next_expiration;
377 bool need_revalidate;
378 struct tag_set revalidate_set;
380 /* OpenFlow connections. */
381 struct hmap controllers; /* Controller "struct ofconn"s. */
382 struct list all_conns; /* Contains "struct ofconn"s. */
383 enum ofproto_fail_mode fail_mode;
385 /* OpenFlow listeners. */
386 struct hmap services; /* Contains "struct ofservice"s. */
387 struct pvconn **snoops;
390 /* Hooks for ovs-vswitchd. */
391 const struct ofhooks *ofhooks;
394 /* Used by default ofhooks. */
395 struct mac_learning *ml;
398 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
399 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
401 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
403 static const struct ofhooks default_ofhooks;
405 static uint64_t pick_datapath_id(const struct ofproto *);
406 static uint64_t pick_fallback_dpid(void);
408 static int ofproto_expire(struct ofproto *);
410 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
412 static void handle_openflow(struct ofconn *, struct ofpbuf *);
414 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
415 static void update_port(struct ofproto *, const char *devname);
416 static int init_ports(struct ofproto *);
417 static void reinit_ports(struct ofproto *);
419 static void ofproto_unixctl_init(void);
422 ofproto_create(const char *datapath, const char *datapath_type,
423 const struct ofhooks *ofhooks, void *aux,
424 struct ofproto **ofprotop)
432 ofproto_unixctl_init();
434 /* Connect to datapath and start listening for messages. */
435 error = dpif_open(datapath, datapath_type, &dpif);
437 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
440 error = dpif_recv_set_mask(dpif,
441 ((1u << DPIF_UC_MISS) |
442 (1u << DPIF_UC_ACTION) |
443 (1u << DPIF_UC_SAMPLE)));
445 VLOG_ERR("failed to listen on datapath %s: %s",
446 datapath, strerror(error));
450 dpif_flow_flush(dpif);
451 dpif_recv_purge(dpif);
453 /* Initialize settings. */
454 p = xzalloc(sizeof *p);
455 p->fallback_dpid = pick_fallback_dpid();
456 p->datapath_id = p->fallback_dpid;
457 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
458 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
459 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
460 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
461 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
463 /* Initialize datapath. */
465 p->netdev_monitor = netdev_monitor_create();
466 hmap_init(&p->ports);
467 shash_init(&p->port_by_name);
468 p->max_ports = dpif_get_max_ports(dpif);
470 /* Initialize submodules. */
471 p->switch_status = switch_status_create(p);
476 /* Initialize in-band control. */
478 p->in_band_queue = -1;
480 /* Initialize flow table. */
481 classifier_init(&p->cls);
482 p->next_expiration = time_msec() + 1000;
484 /* Initialize facet table. */
485 hmap_init(&p->facets);
486 p->need_revalidate = false;
487 tag_set_init(&p->revalidate_set);
489 /* Initialize OpenFlow connections. */
490 list_init(&p->all_conns);
491 hmap_init(&p->controllers);
492 hmap_init(&p->services);
496 /* Initialize hooks. */
498 p->ofhooks = ofhooks;
502 p->ofhooks = &default_ofhooks;
504 p->ml = mac_learning_create();
507 /* Pick final datapath ID. */
508 p->datapath_id = pick_datapath_id(p);
509 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
511 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
518 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
520 uint64_t old_dpid = p->datapath_id;
521 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
522 if (p->datapath_id != old_dpid) {
523 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
525 /* Force all active connections to reconnect, since there is no way to
526 * notify a controller that the datapath ID has changed. */
527 ofproto_reconnect_controllers(p);
532 is_discovery_controller(const struct ofproto_controller *c)
534 return !strcmp(c->target, "discover");
538 is_in_band_controller(const struct ofproto_controller *c)
540 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
543 /* Creates a new controller in 'ofproto'. Some of the settings are initially
544 * drawn from 'c', but update_controller() needs to be called later to finish
545 * the new ofconn's configuration. */
547 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
549 struct discovery *discovery;
550 struct ofconn *ofconn;
552 if (is_discovery_controller(c)) {
553 int error = discovery_create(c->accept_re, c->update_resolv_conf,
554 ofproto->dpif, ofproto->switch_status,
563 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
564 ofconn->pktbuf = pktbuf_create();
565 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
567 ofconn->discovery = discovery;
569 char *name = ofconn_make_name(ofproto, c->target);
570 rconn_connect(ofconn->rconn, c->target, name);
573 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
574 hash_string(c->target, 0));
577 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
578 * target or turn discovery on or off (these are done by creating new ofconns
579 * and deleting old ones), but it can update the rest of an ofconn's
582 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
586 ofconn->band = (is_in_band_controller(c)
587 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
589 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
591 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
592 rconn_set_probe_interval(ofconn->rconn, probe_interval);
594 if (ofconn->discovery) {
595 discovery_set_update_resolv_conf(ofconn->discovery,
596 c->update_resolv_conf);
597 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
600 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
604 ofconn_get_target(const struct ofconn *ofconn)
606 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
609 static struct ofconn *
610 find_controller_by_target(struct ofproto *ofproto, const char *target)
612 struct ofconn *ofconn;
614 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
615 hash_string(target, 0), &ofproto->controllers) {
616 if (!strcmp(ofconn_get_target(ofconn), target)) {
624 update_in_band_remotes(struct ofproto *ofproto)
626 const struct ofconn *ofconn;
627 struct sockaddr_in *addrs;
628 size_t max_addrs, n_addrs;
632 /* Allocate enough memory for as many remotes as we could possibly have. */
633 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
634 addrs = xmalloc(max_addrs * sizeof *addrs);
637 /* Add all the remotes. */
639 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
640 struct sockaddr_in *sin = &addrs[n_addrs];
642 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
646 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
647 if (sin->sin_addr.s_addr) {
648 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
651 if (ofconn->discovery) {
655 for (i = 0; i < ofproto->n_extra_remotes; i++) {
656 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
659 /* Create or update or destroy in-band.
661 * Ordinarily we only enable in-band if there's at least one remote
662 * address, but discovery needs the in-band rules for DHCP to be installed
663 * even before we know any remote addresses. */
664 if (n_addrs || discovery) {
665 if (!ofproto->in_band) {
666 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
669 if (ofproto->in_band) {
670 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
672 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
673 ofproto->next_in_band_update = time_msec() + 1000;
675 in_band_destroy(ofproto->in_band);
676 ofproto->in_band = NULL;
684 update_fail_open(struct ofproto *p)
686 struct ofconn *ofconn;
688 if (!hmap_is_empty(&p->controllers)
689 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
690 struct rconn **rconns;
694 p->fail_open = fail_open_create(p, p->switch_status);
698 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
699 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
700 rconns[n++] = ofconn->rconn;
703 fail_open_set_controllers(p->fail_open, rconns, n);
704 /* p->fail_open takes ownership of 'rconns'. */
706 fail_open_destroy(p->fail_open);
712 ofproto_set_controllers(struct ofproto *p,
713 const struct ofproto_controller *controllers,
714 size_t n_controllers)
716 struct shash new_controllers;
717 struct ofconn *ofconn, *next_ofconn;
718 struct ofservice *ofservice, *next_ofservice;
722 /* Create newly configured controllers and services.
723 * Create a name to ofproto_controller mapping in 'new_controllers'. */
724 shash_init(&new_controllers);
725 for (i = 0; i < n_controllers; i++) {
726 const struct ofproto_controller *c = &controllers[i];
728 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
729 if (!find_controller_by_target(p, c->target)) {
730 add_controller(p, c);
732 } else if (!pvconn_verify_name(c->target)) {
733 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
737 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
738 dpif_name(p->dpif), c->target);
742 shash_add_once(&new_controllers, c->target, &controllers[i]);
745 /* Delete controllers that are no longer configured.
746 * Update configuration of all now-existing controllers. */
748 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
749 struct ofproto_controller *c;
751 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
753 ofconn_destroy(ofconn);
755 update_controller(ofconn, c);
762 /* Delete services that are no longer configured.
763 * Update configuration of all now-existing services. */
764 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
765 struct ofproto_controller *c;
767 c = shash_find_data(&new_controllers,
768 pvconn_get_name(ofservice->pvconn));
770 ofservice_destroy(p, ofservice);
772 ofservice_reconfigure(ofservice, c);
776 shash_destroy(&new_controllers);
778 update_in_band_remotes(p);
781 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
782 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
783 struct ofconn, hmap_node);
784 ofconn->ss = switch_status_register(p->switch_status, "remote",
785 rconn_status_cb, ofconn->rconn);
790 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
792 p->fail_mode = fail_mode;
796 /* Drops the connections between 'ofproto' and all of its controllers, forcing
797 * them to reconnect. */
799 ofproto_reconnect_controllers(struct ofproto *ofproto)
801 struct ofconn *ofconn;
803 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
804 rconn_reconnect(ofconn->rconn);
809 any_extras_changed(const struct ofproto *ofproto,
810 const struct sockaddr_in *extras, size_t n)
814 if (n != ofproto->n_extra_remotes) {
818 for (i = 0; i < n; i++) {
819 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
820 const struct sockaddr_in *new = &extras[i];
822 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
823 old->sin_port != new->sin_port) {
831 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
832 * in-band control should guarantee access, in the same way that in-band
833 * control guarantees access to OpenFlow controllers. */
835 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
836 const struct sockaddr_in *extras, size_t n)
838 if (!any_extras_changed(ofproto, extras, n)) {
842 free(ofproto->extra_in_band_remotes);
843 ofproto->n_extra_remotes = n;
844 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
846 update_in_band_remotes(ofproto);
849 /* Sets the OpenFlow queue used by flows set up by in-band control on
850 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
851 * flows will use the default queue. */
853 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
855 if (queue_id != ofproto->in_band_queue) {
856 ofproto->in_band_queue = queue_id;
857 update_in_band_remotes(ofproto);
862 ofproto_set_desc(struct ofproto *p,
863 const char *mfr_desc, const char *hw_desc,
864 const char *sw_desc, const char *serial_desc,
867 struct ofp_desc_stats *ods;
870 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
871 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
872 sizeof ods->mfr_desc);
875 p->mfr_desc = xstrdup(mfr_desc);
878 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
879 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
880 sizeof ods->hw_desc);
883 p->hw_desc = xstrdup(hw_desc);
886 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
887 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
888 sizeof ods->sw_desc);
891 p->sw_desc = xstrdup(sw_desc);
894 if (strlen(serial_desc) >= sizeof ods->serial_num) {
895 VLOG_WARN("truncating serial_desc, must be less than %zu "
897 sizeof ods->serial_num);
899 free(p->serial_desc);
900 p->serial_desc = xstrdup(serial_desc);
903 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
904 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
905 sizeof ods->dp_desc);
908 p->dp_desc = xstrdup(dp_desc);
913 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
914 const struct svec *svec)
916 struct pvconn **pvconns = *pvconnsp;
917 size_t n_pvconns = *n_pvconnsp;
921 for (i = 0; i < n_pvconns; i++) {
922 pvconn_close(pvconns[i]);
926 pvconns = xmalloc(svec->n * sizeof *pvconns);
928 for (i = 0; i < svec->n; i++) {
929 const char *name = svec->names[i];
930 struct pvconn *pvconn;
933 error = pvconn_open(name, &pvconn);
935 pvconns[n_pvconns++] = pvconn;
937 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
945 *n_pvconnsp = n_pvconns;
951 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
953 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
957 ofproto_set_netflow(struct ofproto *ofproto,
958 const struct netflow_options *nf_options)
960 if (nf_options && nf_options->collectors.n) {
961 if (!ofproto->netflow) {
962 ofproto->netflow = netflow_create();
964 return netflow_set_options(ofproto->netflow, nf_options);
966 netflow_destroy(ofproto->netflow);
967 ofproto->netflow = NULL;
973 ofproto_set_sflow(struct ofproto *ofproto,
974 const struct ofproto_sflow_options *oso)
976 struct ofproto_sflow *os = ofproto->sflow;
979 struct ofport *ofport;
981 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
982 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
983 ofproto_sflow_add_port(os, ofport->odp_port,
984 netdev_get_name(ofport->netdev));
987 ofproto_sflow_set_options(os, oso);
989 ofproto_sflow_destroy(os);
990 ofproto->sflow = NULL;
995 ofproto_get_datapath_id(const struct ofproto *ofproto)
997 return ofproto->datapath_id;
1001 ofproto_has_primary_controller(const struct ofproto *ofproto)
1003 return !hmap_is_empty(&ofproto->controllers);
1006 enum ofproto_fail_mode
1007 ofproto_get_fail_mode(const struct ofproto *p)
1009 return p->fail_mode;
1013 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1017 for (i = 0; i < ofproto->n_snoops; i++) {
1018 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1023 ofproto_destroy(struct ofproto *p)
1025 struct ofservice *ofservice, *next_ofservice;
1026 struct ofconn *ofconn, *next_ofconn;
1027 struct ofport *ofport, *next_ofport;
1034 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1036 /* Destroy fail-open and in-band early, since they touch the classifier. */
1037 fail_open_destroy(p->fail_open);
1038 p->fail_open = NULL;
1040 in_band_destroy(p->in_band);
1042 free(p->extra_in_band_remotes);
1044 ofproto_flush_flows(p);
1045 classifier_destroy(&p->cls);
1046 hmap_destroy(&p->facets);
1048 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1049 ofconn_destroy(ofconn);
1051 hmap_destroy(&p->controllers);
1053 dpif_close(p->dpif);
1054 netdev_monitor_destroy(p->netdev_monitor);
1055 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1056 hmap_remove(&p->ports, &ofport->hmap_node);
1057 ofport_free(ofport);
1059 shash_destroy(&p->port_by_name);
1061 switch_status_destroy(p->switch_status);
1062 netflow_destroy(p->netflow);
1063 ofproto_sflow_destroy(p->sflow);
1065 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1066 ofservice_destroy(p, ofservice);
1068 hmap_destroy(&p->services);
1070 for (i = 0; i < p->n_snoops; i++) {
1071 pvconn_close(p->snoops[i]);
1075 mac_learning_destroy(p->ml);
1080 free(p->serial_desc);
1083 hmap_destroy(&p->ports);
1089 ofproto_run(struct ofproto *p)
1091 int error = ofproto_run1(p);
1093 error = ofproto_run2(p, false);
1099 process_port_change(struct ofproto *ofproto, int error, char *devname)
1101 if (error == ENOBUFS) {
1102 reinit_ports(ofproto);
1103 } else if (!error) {
1104 update_port(ofproto, devname);
1109 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1110 * means that 'ofconn' is more interesting for monitoring than a lower return
1113 snoop_preference(const struct ofconn *ofconn)
1115 switch (ofconn->role) {
1116 case NX_ROLE_MASTER:
1123 /* Shouldn't happen. */
1128 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1129 * Connects this vconn to a controller. */
1131 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1133 struct ofconn *ofconn, *best;
1135 /* Pick a controller for monitoring. */
1137 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1138 if (ofconn->type == OFCONN_PRIMARY
1139 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1145 rconn_add_monitor(best->rconn, vconn);
1147 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1153 ofproto_run1(struct ofproto *p)
1155 struct ofconn *ofconn, *next_ofconn;
1156 struct ofservice *ofservice;
1161 if (shash_is_empty(&p->port_by_name)) {
1165 for (i = 0; i < 50; i++) {
1166 struct dpif_upcall packet;
1168 error = dpif_recv(p->dpif, &packet);
1170 if (error == ENODEV) {
1171 /* Someone destroyed the datapath behind our back. The caller
1172 * better destroy us and give up, because we're just going to
1173 * spin from here on out. */
1174 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1175 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1176 dpif_name(p->dpif));
1182 handle_upcall(p, &packet);
1185 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1186 process_port_change(p, error, devname);
1188 while ((error = netdev_monitor_poll(p->netdev_monitor,
1189 &devname)) != EAGAIN) {
1190 process_port_change(p, error, devname);
1194 if (time_msec() >= p->next_in_band_update) {
1195 update_in_band_remotes(p);
1197 in_band_run(p->in_band);
1200 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1204 /* Fail-open maintenance. Do this after processing the ofconns since
1205 * fail-open checks the status of the controller rconn. */
1207 fail_open_run(p->fail_open);
1210 HMAP_FOR_EACH (ofservice, node, &p->services) {
1211 struct vconn *vconn;
1214 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1216 struct rconn *rconn;
1219 rconn = rconn_create(ofservice->probe_interval, 0);
1220 name = ofconn_make_name(p, vconn_get_name(vconn));
1221 rconn_connect_unreliably(rconn, vconn, name);
1224 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1225 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1226 ofservice->burst_limit);
1227 } else if (retval != EAGAIN) {
1228 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1232 for (i = 0; i < p->n_snoops; i++) {
1233 struct vconn *vconn;
1236 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1238 add_snooper(p, vconn);
1239 } else if (retval != EAGAIN) {
1240 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1244 if (time_msec() >= p->next_expiration) {
1245 int delay = ofproto_expire(p);
1246 p->next_expiration = time_msec() + delay;
1247 COVERAGE_INC(ofproto_expiration);
1251 netflow_run(p->netflow);
1254 ofproto_sflow_run(p->sflow);
1261 ofproto_run2(struct ofproto *p, bool revalidate_all)
1263 /* Figure out what we need to revalidate now, if anything. */
1264 struct tag_set revalidate_set = p->revalidate_set;
1265 if (p->need_revalidate) {
1266 revalidate_all = true;
1269 /* Clear the revalidation flags. */
1270 tag_set_init(&p->revalidate_set);
1271 p->need_revalidate = false;
1273 /* Now revalidate if there's anything to do. */
1274 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1275 struct facet *facet, *next;
1277 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1279 || tag_set_intersects(&revalidate_set, facet->tags)) {
1280 facet_revalidate(p, facet);
1289 ofproto_wait(struct ofproto *p)
1291 struct ofservice *ofservice;
1292 struct ofconn *ofconn;
1295 dpif_recv_wait(p->dpif);
1296 dpif_port_poll_wait(p->dpif);
1297 netdev_monitor_poll_wait(p->netdev_monitor);
1298 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1299 ofconn_wait(ofconn);
1302 poll_timer_wait_until(p->next_in_band_update);
1303 in_band_wait(p->in_band);
1306 fail_open_wait(p->fail_open);
1309 ofproto_sflow_wait(p->sflow);
1311 if (!tag_set_is_empty(&p->revalidate_set)) {
1312 poll_immediate_wake();
1314 if (p->need_revalidate) {
1315 /* Shouldn't happen, but if it does just go around again. */
1316 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1317 poll_immediate_wake();
1318 } else if (p->next_expiration != LLONG_MAX) {
1319 poll_timer_wait_until(p->next_expiration);
1321 HMAP_FOR_EACH (ofservice, node, &p->services) {
1322 pvconn_wait(ofservice->pvconn);
1324 for (i = 0; i < p->n_snoops; i++) {
1325 pvconn_wait(p->snoops[i]);
1330 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1332 tag_set_add(&ofproto->revalidate_set, tag);
1336 ofproto_get_revalidate_set(struct ofproto *ofproto)
1338 return &ofproto->revalidate_set;
1342 ofproto_is_alive(const struct ofproto *p)
1344 return !hmap_is_empty(&p->controllers);
1348 ofproto_get_ofproto_controller_info(const struct ofproto * ofproto,
1351 const struct ofconn *ofconn;
1355 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1356 const struct rconn *rconn = ofconn->rconn;
1357 const int last_error = rconn_get_last_error(rconn);
1358 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1360 shash_add(info, rconn_get_target(rconn), cinfo);
1362 cinfo->is_connected = rconn_is_connected(rconn);
1363 cinfo->role = ofconn->role;
1368 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1369 cinfo->pairs.values[cinfo->pairs.n++] =
1370 xstrdup(ovs_retval_to_string(last_error));
1373 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1374 cinfo->pairs.values[cinfo->pairs.n++] =
1375 xstrdup(rconn_get_state(rconn));
1377 cinfo->pairs.keys[cinfo->pairs.n] = "time_in_state";
1378 cinfo->pairs.values[cinfo->pairs.n++] =
1379 xasprintf("%u", rconn_get_state_elapsed(rconn));
1384 ofproto_free_ofproto_controller_info(struct shash *info)
1386 struct shash_node *node;
1388 SHASH_FOR_EACH (node, info) {
1389 struct ofproto_controller_info *cinfo = node->data;
1390 while (cinfo->pairs.n) {
1391 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1395 shash_destroy(info);
1398 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1400 * This is almost the same as calling dpif_port_del() directly on the
1401 * datapath, but it also makes 'ofproto' close its open netdev for the port
1402 * (if any). This makes it possible to create a new netdev of a different
1403 * type under the same name, which otherwise the netdev library would refuse
1404 * to do because of the conflict. (The netdev would eventually get closed on
1405 * the next trip through ofproto_run(), but this interface is more direct.)
1407 * Returns 0 if successful, otherwise a positive errno. */
1409 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1411 struct ofport *ofport = get_port(ofproto, odp_port);
1412 const char *name = ofport ? ofport->opp.name : "<unknown>";
1415 error = dpif_port_del(ofproto->dpif, odp_port);
1417 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1418 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1419 } else if (ofport) {
1420 /* 'name' is ofport->opp.name and update_port() is going to destroy
1421 * 'ofport'. Just in case update_port() refers to 'name' after it
1422 * destroys 'ofport', make a copy of it around the update_port()
1424 char *devname = xstrdup(name);
1425 update_port(ofproto, devname);
1431 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1432 * true if 'odp_port' exists and should be included, false otherwise. */
1434 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1436 struct ofport *ofport = get_port(ofproto, odp_port);
1437 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1441 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1442 const union ofp_action *actions, size_t n_actions,
1443 const struct ofpbuf *packet)
1445 struct action_xlate_ctx ctx;
1446 struct ofpbuf *odp_actions;
1448 action_xlate_ctx_init(&ctx, p, flow, packet);
1449 odp_actions = xlate_actions(&ctx, actions, n_actions);
1451 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1453 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet);
1455 ofpbuf_delete(odp_actions);
1460 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1461 * performs the 'n_actions' actions in 'actions'. The new flow will not
1464 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1465 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1466 * controllers; otherwise, it will be hidden.
1468 * The caller retains ownership of 'cls_rule' and 'actions'. */
1470 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1471 const union ofp_action *actions, size_t n_actions)
1474 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1475 rule_insert(p, rule);
1479 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1483 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1486 rule_remove(ofproto, rule);
1491 ofproto_flush_flows(struct ofproto *ofproto)
1493 struct facet *facet, *next_facet;
1494 struct rule *rule, *next_rule;
1495 struct cls_cursor cursor;
1497 COVERAGE_INC(ofproto_flush);
1499 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1500 /* Mark the facet as not installed so that facet_remove() doesn't
1501 * bother trying to uninstall it. There is no point in uninstalling it
1502 * individually since we are about to blow away all the facets with
1503 * dpif_flow_flush(). */
1504 facet->installed = false;
1505 facet_remove(ofproto, facet);
1508 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1509 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1510 rule_remove(ofproto, rule);
1513 dpif_flow_flush(ofproto->dpif);
1514 if (ofproto->in_band) {
1515 in_band_flushed(ofproto->in_band);
1517 if (ofproto->fail_open) {
1518 fail_open_flushed(ofproto->fail_open);
1523 reinit_ports(struct ofproto *p)
1525 struct dpif_port_dump dump;
1526 struct shash_node *node;
1527 struct shash devnames;
1528 struct ofport *ofport;
1529 struct dpif_port dpif_port;
1531 COVERAGE_INC(ofproto_reinit_ports);
1533 shash_init(&devnames);
1534 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1535 shash_add_once (&devnames, ofport->opp.name, NULL);
1537 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1538 shash_add_once (&devnames, dpif_port.name, NULL);
1541 SHASH_FOR_EACH (node, &devnames) {
1542 update_port(p, node->name);
1544 shash_destroy(&devnames);
1547 static struct ofport *
1548 make_ofport(const struct dpif_port *dpif_port)
1550 struct netdev_options netdev_options;
1551 enum netdev_flags flags;
1552 struct ofport *ofport;
1553 struct netdev *netdev;
1556 memset(&netdev_options, 0, sizeof netdev_options);
1557 netdev_options.name = dpif_port->name;
1558 netdev_options.type = dpif_port->type;
1559 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1561 error = netdev_open(&netdev_options, &netdev);
1563 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1564 "cannot be opened (%s)",
1565 dpif_port->name, dpif_port->port_no,
1566 dpif_port->name, strerror(error));
1570 ofport = xzalloc(sizeof *ofport);
1571 ofport->netdev = netdev;
1572 ofport->odp_port = dpif_port->port_no;
1573 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1574 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1575 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1577 netdev_get_flags(netdev, &flags);
1578 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1580 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1582 netdev_get_features(netdev,
1583 &ofport->opp.curr, &ofport->opp.advertised,
1584 &ofport->opp.supported, &ofport->opp.peer);
1589 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1591 if (get_port(p, dpif_port->port_no)) {
1592 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1593 dpif_port->port_no);
1595 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1596 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1605 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1607 const struct ofp_phy_port *a = &a_->opp;
1608 const struct ofp_phy_port *b = &b_->opp;
1610 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1611 return (a->port_no == b->port_no
1612 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1613 && !strcmp(a->name, b->name)
1614 && a->state == b->state
1615 && a->config == b->config
1616 && a->curr == b->curr
1617 && a->advertised == b->advertised
1618 && a->supported == b->supported
1619 && a->peer == b->peer);
1623 send_port_status(struct ofproto *p, const struct ofport *ofport,
1626 /* XXX Should limit the number of queued port status change messages. */
1627 struct ofconn *ofconn;
1628 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1629 struct ofp_port_status *ops;
1632 /* Primary controllers, even slaves, should always get port status
1633 updates. Otherwise obey ofconn_receives_async_msgs(). */
1634 if (ofconn->type != OFCONN_PRIMARY
1635 && !ofconn_receives_async_msgs(ofconn)) {
1639 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1640 ops->reason = reason;
1641 ops->desc = ofport->opp;
1642 hton_ofp_phy_port(&ops->desc);
1643 queue_tx(b, ofconn, NULL);
1648 ofport_install(struct ofproto *p, struct ofport *ofport)
1650 const char *netdev_name = ofport->opp.name;
1652 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1653 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1654 shash_add(&p->port_by_name, netdev_name, ofport);
1656 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1661 ofport_remove(struct ofproto *p, struct ofport *ofport)
1663 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1664 hmap_remove(&p->ports, &ofport->hmap_node);
1665 shash_delete(&p->port_by_name,
1666 shash_find(&p->port_by_name, ofport->opp.name));
1668 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1673 ofport_free(struct ofport *ofport)
1676 netdev_close(ofport->netdev);
1681 static struct ofport *
1682 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1684 struct ofport *port;
1686 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1687 hash_int(odp_port, 0), &ofproto->ports) {
1688 if (port->odp_port == odp_port) {
1696 update_port(struct ofproto *p, const char *devname)
1698 struct dpif_port dpif_port;
1699 struct ofport *old_ofport;
1700 struct ofport *new_ofport;
1703 COVERAGE_INC(ofproto_update_port);
1705 /* Query the datapath for port information. */
1706 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1708 /* Find the old ofport. */
1709 old_ofport = shash_find_data(&p->port_by_name, devname);
1712 /* There's no port named 'devname' but there might be a port with
1713 * the same port number. This could happen if a port is deleted
1714 * and then a new one added in its place very quickly, or if a port
1715 * is renamed. In the former case we want to send an OFPPR_DELETE
1716 * and an OFPPR_ADD, and in the latter case we want to send a
1717 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1718 * the old port's ifindex against the new port, or perhaps less
1719 * reliably but more portably by comparing the old port's MAC
1720 * against the new port's MAC. However, this code isn't that smart
1721 * and always sends an OFPPR_MODIFY (XXX). */
1722 old_ofport = get_port(p, dpif_port.port_no);
1724 } else if (error != ENOENT && error != ENODEV) {
1725 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1726 "%s", strerror(error));
1730 /* Create a new ofport. */
1731 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1733 /* Eliminate a few pathological cases. */
1734 if (!old_ofport && !new_ofport) {
1736 } else if (old_ofport && new_ofport) {
1737 /* Most of the 'config' bits are OpenFlow soft state, but
1738 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1739 * OpenFlow bits from old_ofport. (make_ofport() only sets
1740 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1741 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1743 if (ofport_equal(old_ofport, new_ofport)) {
1744 /* False alarm--no change. */
1745 ofport_free(new_ofport);
1750 /* Now deal with the normal cases. */
1752 ofport_remove(p, old_ofport);
1755 ofport_install(p, new_ofport);
1757 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1758 (!old_ofport ? OFPPR_ADD
1759 : !new_ofport ? OFPPR_DELETE
1761 ofport_free(old_ofport);
1764 dpif_port_destroy(&dpif_port);
1768 init_ports(struct ofproto *p)
1770 struct dpif_port_dump dump;
1771 struct dpif_port dpif_port;
1773 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1774 if (!ofport_conflicts(p, &dpif_port)) {
1775 struct ofport *ofport = make_ofport(&dpif_port);
1777 ofport_install(p, ofport);
1785 static struct ofconn *
1786 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1788 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1789 ofconn->ofproto = p;
1790 list_push_back(&p->all_conns, &ofconn->node);
1791 ofconn->rconn = rconn;
1792 ofconn->type = type;
1793 ofconn->flow_format = NXFF_OPENFLOW10;
1794 ofconn->role = NX_ROLE_OTHER;
1795 ofconn->packet_in_counter = rconn_packet_counter_create ();
1796 ofconn->pktbuf = NULL;
1797 ofconn->miss_send_len = 0;
1798 ofconn->reply_counter = rconn_packet_counter_create ();
1803 ofconn_destroy(struct ofconn *ofconn)
1805 if (ofconn->type == OFCONN_PRIMARY) {
1806 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1808 discovery_destroy(ofconn->discovery);
1810 list_remove(&ofconn->node);
1811 switch_status_unregister(ofconn->ss);
1812 rconn_destroy(ofconn->rconn);
1813 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1814 rconn_packet_counter_destroy(ofconn->reply_counter);
1815 pktbuf_destroy(ofconn->pktbuf);
1820 ofconn_run(struct ofconn *ofconn)
1822 struct ofproto *p = ofconn->ofproto;
1826 if (ofconn->discovery) {
1827 char *controller_name;
1828 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1829 discovery_question_connectivity(ofconn->discovery);
1831 if (discovery_run(ofconn->discovery, &controller_name)) {
1832 if (controller_name) {
1833 char *ofconn_name = ofconn_make_name(p, controller_name);
1834 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1837 rconn_disconnect(ofconn->rconn);
1842 for (i = 0; i < N_SCHEDULERS; i++) {
1843 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1846 rconn_run(ofconn->rconn);
1848 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1849 /* Limit the number of iterations to prevent other tasks from
1851 for (iteration = 0; iteration < 50; iteration++) {
1852 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1857 fail_open_maybe_recover(p->fail_open);
1859 handle_openflow(ofconn, of_msg);
1860 ofpbuf_delete(of_msg);
1864 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1865 ofconn_destroy(ofconn);
1870 ofconn_wait(struct ofconn *ofconn)
1874 if (ofconn->discovery) {
1875 discovery_wait(ofconn->discovery);
1877 for (i = 0; i < N_SCHEDULERS; i++) {
1878 pinsched_wait(ofconn->schedulers[i]);
1880 rconn_run_wait(ofconn->rconn);
1881 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1882 rconn_recv_wait(ofconn->rconn);
1884 COVERAGE_INC(ofproto_ofconn_stuck);
1888 /* Returns true if 'ofconn' should receive asynchronous messages. */
1890 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1892 if (ofconn->type == OFCONN_PRIMARY) {
1893 /* Primary controllers always get asynchronous messages unless they
1894 * have configured themselves as "slaves". */
1895 return ofconn->role != NX_ROLE_SLAVE;
1897 /* Service connections don't get asynchronous messages unless they have
1898 * explicitly asked for them by setting a nonzero miss send length. */
1899 return ofconn->miss_send_len > 0;
1903 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1904 * and 'target', suitable for use in log messages for identifying the
1907 * The name is dynamically allocated. The caller should free it (with free())
1908 * when it is no longer needed. */
1910 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1912 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1916 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1920 for (i = 0; i < N_SCHEDULERS; i++) {
1921 struct pinsched **s = &ofconn->schedulers[i];
1925 *s = pinsched_create(rate, burst,
1926 ofconn->ofproto->switch_status);
1928 pinsched_set_limits(*s, rate, burst);
1931 pinsched_destroy(*s);
1938 ofservice_reconfigure(struct ofservice *ofservice,
1939 const struct ofproto_controller *c)
1941 ofservice->probe_interval = c->probe_interval;
1942 ofservice->rate_limit = c->rate_limit;
1943 ofservice->burst_limit = c->burst_limit;
1946 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1947 * positive errno value. */
1949 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1951 struct ofservice *ofservice;
1952 struct pvconn *pvconn;
1955 error = pvconn_open(c->target, &pvconn);
1960 ofservice = xzalloc(sizeof *ofservice);
1961 hmap_insert(&ofproto->services, &ofservice->node,
1962 hash_string(c->target, 0));
1963 ofservice->pvconn = pvconn;
1965 ofservice_reconfigure(ofservice, c);
1971 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1973 hmap_remove(&ofproto->services, &ofservice->node);
1974 pvconn_close(ofservice->pvconn);
1978 /* Finds and returns the ofservice within 'ofproto' that has the given
1979 * 'target', or a null pointer if none exists. */
1980 static struct ofservice *
1981 ofservice_lookup(struct ofproto *ofproto, const char *target)
1983 struct ofservice *ofservice;
1985 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1986 &ofproto->services) {
1987 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1994 /* Returns true if 'rule' should be hidden from the controller.
1996 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1997 * (e.g. by in-band control) and are intentionally hidden from the
2000 rule_is_hidden(const struct rule *rule)
2002 return rule->cr.priority > UINT16_MAX;
2005 /* Creates and returns a new rule initialized as specified.
2007 * The caller is responsible for inserting the rule into the classifier (with
2008 * rule_insert()). */
2009 static struct rule *
2010 rule_create(const struct cls_rule *cls_rule,
2011 const union ofp_action *actions, size_t n_actions,
2012 uint16_t idle_timeout, uint16_t hard_timeout,
2013 ovs_be64 flow_cookie, bool send_flow_removed)
2015 struct rule *rule = xzalloc(sizeof *rule);
2016 rule->cr = *cls_rule;
2017 rule->idle_timeout = idle_timeout;
2018 rule->hard_timeout = hard_timeout;
2019 rule->flow_cookie = flow_cookie;
2020 rule->used = rule->created = time_msec();
2021 rule->send_flow_removed = send_flow_removed;
2022 list_init(&rule->facets);
2023 if (n_actions > 0) {
2024 rule->n_actions = n_actions;
2025 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2031 static struct rule *
2032 rule_from_cls_rule(const struct cls_rule *cls_rule)
2034 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2038 rule_free(struct rule *rule)
2040 free(rule->actions);
2044 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2045 * destroying any that no longer has a rule (which is probably all of them).
2047 * The caller must have already removed 'rule' from the classifier. */
2049 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2051 struct facet *facet, *next_facet;
2052 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2053 facet_revalidate(ofproto, facet);
2058 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2059 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2062 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2064 const union ofp_action *oa;
2065 struct actions_iterator i;
2067 if (out_port == htons(OFPP_NONE)) {
2070 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2071 oa = actions_next(&i)) {
2072 if (action_outputs_to_port(oa, out_port)) {
2079 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2080 * 'packet', which arrived on 'in_port'.
2082 * Takes ownership of 'packet'. */
2084 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2085 const struct nlattr *odp_actions, size_t actions_len,
2086 struct ofpbuf *packet)
2088 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2089 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2090 /* As an optimization, avoid a round-trip from userspace to kernel to
2091 * userspace. This also avoids possibly filling up kernel packet
2092 * buffers along the way. */
2093 struct dpif_upcall upcall;
2095 upcall.type = DPIF_UC_ACTION;
2096 upcall.packet = packet;
2099 upcall.userdata = nl_attr_get_u64(odp_actions);
2100 upcall.sample_pool = 0;
2101 upcall.actions = NULL;
2102 upcall.actions_len = 0;
2104 send_packet_in(ofproto, &upcall, flow, false);
2110 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2111 ofpbuf_delete(packet);
2116 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2117 * statistics appropriately. 'packet' must have at least sizeof(struct
2118 * ofp_packet_in) bytes of headroom.
2120 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2121 * applying flow_extract() to 'packet' would yield the same flow as
2124 * 'facet' must have accurately composed ODP actions; that is, it must not be
2125 * in need of revalidation.
2127 * Takes ownership of 'packet'. */
2129 facet_execute(struct ofproto *ofproto, struct facet *facet,
2130 struct ofpbuf *packet)
2132 struct dpif_flow_stats stats;
2134 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2136 flow_extract_stats(&facet->flow, packet, &stats);
2137 if (execute_odp_actions(ofproto, &facet->flow,
2138 facet->actions, facet->actions_len, packet)) {
2139 facet_update_stats(ofproto, facet, &stats);
2140 facet->used = time_msec();
2141 netflow_flow_update_time(ofproto->netflow,
2142 &facet->nf_flow, facet->used);
2146 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2147 * statistics (or the statistics for one of its facets) appropriately.
2148 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2150 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2151 * with statistics for 'packet' either way.
2153 * Takes ownership of 'packet'. */
2155 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2156 struct ofpbuf *packet)
2158 struct action_xlate_ctx ctx;
2159 struct ofpbuf *odp_actions;
2160 struct facet *facet;
2164 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2166 flow_extract(packet, 0, in_port, &flow);
2168 /* First look for a related facet. If we find one, account it to that. */
2169 facet = facet_lookup_valid(ofproto, &flow);
2170 if (facet && facet->rule == rule) {
2171 facet_execute(ofproto, facet, packet);
2175 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2176 * create a new facet for it and use that. */
2177 if (rule_lookup(ofproto, &flow) == rule) {
2178 facet = facet_create(ofproto, rule, &flow, packet);
2179 facet_execute(ofproto, facet, packet);
2180 facet_install(ofproto, facet, true);
2184 /* We can't account anything to a facet. If we were to try, then that
2185 * facet would have a non-matching rule, busting our invariants. */
2186 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2187 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2188 size = packet->size;
2189 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2190 odp_actions->size, packet)) {
2191 rule->used = time_msec();
2192 rule->packet_count++;
2193 rule->byte_count += size;
2195 ofpbuf_delete(odp_actions);
2198 /* Inserts 'rule' into 'p''s flow table. */
2200 rule_insert(struct ofproto *p, struct rule *rule)
2202 struct rule *displaced_rule;
2204 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2205 if (displaced_rule) {
2206 rule_destroy(p, displaced_rule);
2208 p->need_revalidate = true;
2211 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2212 * 'flow' and an example 'packet' within that flow.
2214 * The caller must already have determined that no facet with an identical
2215 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2216 * 'ofproto''s classifier table. */
2217 static struct facet *
2218 facet_create(struct ofproto *ofproto, struct rule *rule,
2219 const struct flow *flow, const struct ofpbuf *packet)
2221 struct facet *facet;
2223 facet = xzalloc(sizeof *facet);
2224 facet->used = time_msec();
2225 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2226 list_push_back(&rule->facets, &facet->list_node);
2228 facet->flow = *flow;
2229 netflow_flow_init(&facet->nf_flow);
2230 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2232 facet_make_actions(ofproto, facet, packet);
2238 facet_free(struct facet *facet)
2240 free(facet->actions);
2244 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2246 * - Removes 'rule' from the classifier.
2248 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2249 * destroys them), via rule_destroy().
2252 rule_remove(struct ofproto *ofproto, struct rule *rule)
2254 COVERAGE_INC(ofproto_del_rule);
2255 ofproto->need_revalidate = true;
2256 classifier_remove(&ofproto->cls, &rule->cr);
2257 rule_destroy(ofproto, rule);
2260 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2262 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2263 * rule's statistics, via facet_uninstall().
2265 * - Removes 'facet' from its rule and from ofproto->facets.
2268 facet_remove(struct ofproto *ofproto, struct facet *facet)
2270 facet_uninstall(ofproto, facet);
2271 facet_flush_stats(ofproto, facet);
2272 hmap_remove(&ofproto->facets, &facet->hmap_node);
2273 list_remove(&facet->list_node);
2277 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2279 facet_make_actions(struct ofproto *p, struct facet *facet,
2280 const struct ofpbuf *packet)
2282 const struct rule *rule = facet->rule;
2283 struct ofpbuf *odp_actions;
2284 struct action_xlate_ctx ctx;
2286 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2287 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2288 facet->tags = ctx.tags;
2289 facet->may_install = ctx.may_set_up_flow;
2290 facet->nf_flow.output_iface = ctx.nf_output_iface;
2292 if (facet->actions_len != odp_actions->size
2293 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2294 free(facet->actions);
2295 facet->actions_len = odp_actions->size;
2296 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2299 ofpbuf_delete(odp_actions);
2303 facet_put__(struct ofproto *ofproto, struct facet *facet,
2304 const struct nlattr *actions, size_t actions_len,
2305 struct dpif_flow_stats *stats)
2307 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2308 enum dpif_flow_put_flags flags;
2311 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2313 flags |= DPIF_FP_ZERO_STATS;
2316 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2317 odp_flow_key_from_flow(&key, &facet->flow);
2318 assert(key.base == keybuf);
2320 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2321 actions, actions_len, stats);
2324 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2325 * 'zero_stats' is true, clears any existing statistics from the datapath for
2328 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2330 struct dpif_flow_stats stats;
2332 if (facet->may_install
2333 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2334 zero_stats ? &stats : NULL)) {
2335 facet->installed = true;
2339 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2340 * to the accounting hook function in the ofhooks structure. */
2342 facet_account(struct ofproto *ofproto,
2343 struct facet *facet, uint64_t extra_bytes)
2345 uint64_t total_bytes = facet->byte_count + extra_bytes;
2347 if (ofproto->ofhooks->account_flow_cb
2348 && total_bytes > facet->accounted_bytes)
2350 ofproto->ofhooks->account_flow_cb(
2351 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2352 total_bytes - facet->accounted_bytes, ofproto->aux);
2353 facet->accounted_bytes = total_bytes;
2357 /* If 'rule' is installed in the datapath, uninstalls it. */
2359 facet_uninstall(struct ofproto *p, struct facet *facet)
2361 if (facet->installed) {
2362 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2363 struct dpif_flow_stats stats;
2366 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2367 odp_flow_key_from_flow(&key, &facet->flow);
2368 assert(key.base == keybuf);
2370 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2371 facet_update_stats(p, facet, &stats);
2373 facet->installed = false;
2377 /* Returns true if the only action for 'facet' is to send to the controller.
2378 * (We don't report NetFlow expiration messages for such facets because they
2379 * are just part of the control logic for the network, not real traffic). */
2381 facet_is_controller_flow(struct facet *facet)
2384 && facet->rule->n_actions == 1
2385 && action_outputs_to_port(&facet->rule->actions[0],
2386 htons(OFPP_CONTROLLER)));
2389 /* Folds all of 'facet''s statistics into its rule. Also updates the
2390 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2392 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2394 facet_account(ofproto, facet, 0);
2396 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2397 struct ofexpired expired;
2398 expired.flow = facet->flow;
2399 expired.packet_count = facet->packet_count;
2400 expired.byte_count = facet->byte_count;
2401 expired.used = facet->used;
2402 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2405 facet->rule->packet_count += facet->packet_count;
2406 facet->rule->byte_count += facet->byte_count;
2408 /* Reset counters to prevent double counting if 'facet' ever gets
2410 facet->packet_count = 0;
2411 facet->byte_count = 0;
2412 facet->accounted_bytes = 0;
2414 netflow_flow_clear(&facet->nf_flow);
2417 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2418 * Returns it if found, otherwise a null pointer.
2420 * The returned facet might need revalidation; use facet_lookup_valid()
2421 * instead if that is important. */
2422 static struct facet *
2423 facet_find(struct ofproto *ofproto, const struct flow *flow)
2425 struct facet *facet;
2427 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2429 if (flow_equal(flow, &facet->flow)) {
2437 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2438 * Returns it if found, otherwise a null pointer.
2440 * The returned facet is guaranteed to be valid. */
2441 static struct facet *
2442 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2444 struct facet *facet = facet_find(ofproto, flow);
2446 /* The facet we found might not be valid, since we could be in need of
2447 * revalidation. If it is not valid, don't return it. */
2449 && ofproto->need_revalidate
2450 && !facet_revalidate(ofproto, facet)) {
2451 COVERAGE_INC(ofproto_invalidated);
2458 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2460 * - If the rule found is different from 'facet''s current rule, moves
2461 * 'facet' to the new rule and recompiles its actions.
2463 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2464 * where it is and recompiles its actions anyway.
2466 * - If there is none, destroys 'facet'.
2468 * Returns true if 'facet' still exists, false if it has been destroyed. */
2470 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2472 struct action_xlate_ctx ctx;
2473 struct ofpbuf *odp_actions;
2474 struct rule *new_rule;
2475 bool actions_changed;
2477 COVERAGE_INC(facet_revalidate);
2479 /* Determine the new rule. */
2480 new_rule = rule_lookup(ofproto, &facet->flow);
2482 /* No new rule, so delete the facet. */
2483 facet_remove(ofproto, facet);
2487 /* Calculate new ODP actions.
2489 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2490 * emit a NetFlow expiration and, if so, we need to have the old state
2491 * around to properly compose it. */
2492 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2493 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2494 actions_changed = (facet->actions_len != odp_actions->size
2495 || memcmp(facet->actions, odp_actions->data,
2496 facet->actions_len));
2498 /* If the ODP actions changed or the installability changed, then we need
2499 * to talk to the datapath. */
2500 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2501 if (ctx.may_set_up_flow) {
2502 struct dpif_flow_stats stats;
2504 facet_put__(ofproto, facet,
2505 odp_actions->data, odp_actions->size, &stats);
2506 facet_update_stats(ofproto, facet, &stats);
2508 facet_uninstall(ofproto, facet);
2511 /* The datapath flow is gone or has zeroed stats, so push stats out of
2512 * 'facet' into 'rule'. */
2513 facet_flush_stats(ofproto, facet);
2516 /* Update 'facet' now that we've taken care of all the old state. */
2517 facet->tags = ctx.tags;
2518 facet->nf_flow.output_iface = ctx.nf_output_iface;
2519 facet->may_install = ctx.may_set_up_flow;
2520 if (actions_changed) {
2521 free(facet->actions);
2522 facet->actions_len = odp_actions->size;
2523 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2525 if (facet->rule != new_rule) {
2526 COVERAGE_INC(facet_changed_rule);
2527 list_remove(&facet->list_node);
2528 list_push_back(&new_rule->facets, &facet->list_node);
2529 facet->rule = new_rule;
2530 facet->used = new_rule->created;
2533 ofpbuf_delete(odp_actions);
2539 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2540 struct rconn_packet_counter *counter)
2542 update_openflow_length(msg);
2543 if (rconn_send(ofconn->rconn, msg, counter)) {
2549 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2552 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2554 COVERAGE_INC(ofproto_error);
2555 queue_tx(buf, ofconn, ofconn->reply_counter);
2560 hton_ofp_phy_port(struct ofp_phy_port *opp)
2562 opp->port_no = htons(opp->port_no);
2563 opp->config = htonl(opp->config);
2564 opp->state = htonl(opp->state);
2565 opp->curr = htonl(opp->curr);
2566 opp->advertised = htonl(opp->advertised);
2567 opp->supported = htonl(opp->supported);
2568 opp->peer = htonl(opp->peer);
2572 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2574 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2579 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2581 struct ofp_switch_features *osf;
2583 struct ofport *port;
2585 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2586 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2587 osf->n_buffers = htonl(pktbuf_capacity());
2589 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2590 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2591 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2592 (1u << OFPAT_SET_VLAN_VID) |
2593 (1u << OFPAT_SET_VLAN_PCP) |
2594 (1u << OFPAT_STRIP_VLAN) |
2595 (1u << OFPAT_SET_DL_SRC) |
2596 (1u << OFPAT_SET_DL_DST) |
2597 (1u << OFPAT_SET_NW_SRC) |
2598 (1u << OFPAT_SET_NW_DST) |
2599 (1u << OFPAT_SET_NW_TOS) |
2600 (1u << OFPAT_SET_TP_SRC) |
2601 (1u << OFPAT_SET_TP_DST) |
2602 (1u << OFPAT_ENQUEUE));
2604 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2605 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2608 queue_tx(buf, ofconn, ofconn->reply_counter);
2613 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2616 struct ofp_switch_config *osc;
2620 /* Figure out flags. */
2621 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2622 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2625 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2626 osc->flags = htons(flags);
2627 osc->miss_send_len = htons(ofconn->miss_send_len);
2628 queue_tx(buf, ofconn, ofconn->reply_counter);
2634 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2636 uint16_t flags = ntohs(osc->flags);
2638 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2639 switch (flags & OFPC_FRAG_MASK) {
2640 case OFPC_FRAG_NORMAL:
2641 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2643 case OFPC_FRAG_DROP:
2644 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2647 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2653 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2658 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2659 * flow translation. */
2660 #define MAX_RESUBMIT_RECURSION 16
2662 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2663 struct action_xlate_ctx *ctx);
2666 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2668 const struct ofport *ofport = get_port(ctx->ofproto, port);
2671 if (ofport->opp.config & OFPPC_NO_FWD) {
2672 /* Forwarding disabled on port. */
2677 * We don't have an ofport record for this port, but it doesn't hurt to
2678 * allow forwarding to it anyhow. Maybe such a port will appear later
2679 * and we're pre-populating the flow table.
2683 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2684 ctx->nf_output_iface = port;
2687 static struct rule *
2688 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2690 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2694 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2696 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2697 uint16_t old_in_port;
2700 /* Look up a flow with 'in_port' as the input port. Then restore the
2701 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2702 * have surprising behavior). */
2703 old_in_port = ctx->flow.in_port;
2704 ctx->flow.in_port = in_port;
2705 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2706 ctx->flow.in_port = old_in_port;
2708 if (ctx->resubmit_hook) {
2709 ctx->resubmit_hook(ctx, rule);
2714 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2718 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2720 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2721 MAX_RESUBMIT_RECURSION);
2726 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2727 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2729 struct ofport *ofport;
2731 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2732 uint16_t odp_port = ofport->odp_port;
2733 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2734 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2737 *nf_output_iface = NF_OUT_FLOOD;
2741 xlate_output_action__(struct action_xlate_ctx *ctx,
2742 uint16_t port, uint16_t max_len)
2745 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2747 ctx->nf_output_iface = NF_OUT_DROP;
2751 add_output_action(ctx, ctx->flow.in_port);
2754 xlate_table_action(ctx, ctx->flow.in_port);
2757 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2758 ctx->odp_actions, &ctx->tags,
2759 &ctx->nf_output_iface,
2760 ctx->ofproto->aux)) {
2761 COVERAGE_INC(ofproto_uninstallable);
2762 ctx->may_set_up_flow = false;
2766 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2767 &ctx->nf_output_iface, ctx->odp_actions);
2770 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2771 &ctx->nf_output_iface, ctx->odp_actions);
2773 case OFPP_CONTROLLER:
2774 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2777 add_output_action(ctx, ODPP_LOCAL);
2780 odp_port = ofp_port_to_odp_port(port);
2781 if (odp_port != ctx->flow.in_port) {
2782 add_output_action(ctx, odp_port);
2787 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2788 ctx->nf_output_iface = NF_OUT_FLOOD;
2789 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2790 ctx->nf_output_iface = prev_nf_output_iface;
2791 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2792 ctx->nf_output_iface != NF_OUT_FLOOD) {
2793 ctx->nf_output_iface = NF_OUT_MULTI;
2798 xlate_output_action(struct action_xlate_ctx *ctx,
2799 const struct ofp_action_output *oao)
2801 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2804 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2805 * optimization, because we're going to add another action that sets the
2806 * priority immediately after, or because there are no actions following the
2809 remove_pop_action(struct action_xlate_ctx *ctx)
2811 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2812 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2813 ctx->last_pop_priority = -1;
2818 add_pop_action(struct action_xlate_ctx *ctx)
2820 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2821 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2822 ctx->last_pop_priority = ctx->odp_actions->size;
2827 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2828 const struct ofp_action_enqueue *oae)
2830 uint16_t ofp_port, odp_port;
2834 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2837 /* Fall back to ordinary output action. */
2838 xlate_output_action__(ctx, ntohs(oae->port), 0);
2842 /* Figure out ODP output port. */
2843 ofp_port = ntohs(oae->port);
2844 if (ofp_port != OFPP_IN_PORT) {
2845 odp_port = ofp_port_to_odp_port(ofp_port);
2847 odp_port = ctx->flow.in_port;
2850 /* Add ODP actions. */
2851 remove_pop_action(ctx);
2852 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2853 add_output_action(ctx, odp_port);
2854 add_pop_action(ctx);
2856 /* Update NetFlow output port. */
2857 if (ctx->nf_output_iface == NF_OUT_DROP) {
2858 ctx->nf_output_iface = odp_port;
2859 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2860 ctx->nf_output_iface = NF_OUT_MULTI;
2865 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2866 const struct nx_action_set_queue *nasq)
2871 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2874 /* Couldn't translate queue to a priority, so ignore. A warning
2875 * has already been logged. */
2879 remove_pop_action(ctx);
2880 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2884 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2886 ovs_be16 tci = ctx->flow.vlan_tci;
2887 if (!(tci & htons(VLAN_CFI))) {
2888 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2890 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2891 tci & ~htons(VLAN_CFI));
2895 struct xlate_reg_state {
2901 save_reg_state(const struct action_xlate_ctx *ctx,
2902 struct xlate_reg_state *state)
2904 state->vlan_tci = ctx->flow.vlan_tci;
2905 state->tun_id = ctx->flow.tun_id;
2909 update_reg_state(struct action_xlate_ctx *ctx,
2910 const struct xlate_reg_state *state)
2912 if (ctx->flow.vlan_tci != state->vlan_tci) {
2913 xlate_set_dl_tci(ctx);
2915 if (ctx->flow.tun_id != state->tun_id) {
2916 nl_msg_put_be64(ctx->odp_actions,
2917 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2922 xlate_nicira_action(struct action_xlate_ctx *ctx,
2923 const struct nx_action_header *nah)
2925 const struct nx_action_resubmit *nar;
2926 const struct nx_action_set_tunnel *nast;
2927 const struct nx_action_set_queue *nasq;
2928 const struct nx_action_multipath *nam;
2929 enum nx_action_subtype subtype = ntohs(nah->subtype);
2930 struct xlate_reg_state state;
2933 assert(nah->vendor == htonl(NX_VENDOR_ID));
2935 case NXAST_RESUBMIT:
2936 nar = (const struct nx_action_resubmit *) nah;
2937 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2940 case NXAST_SET_TUNNEL:
2941 nast = (const struct nx_action_set_tunnel *) nah;
2942 tun_id = htonll(ntohl(nast->tun_id));
2943 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2944 ctx->flow.tun_id = tun_id;
2947 case NXAST_DROP_SPOOFED_ARP:
2948 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2949 nl_msg_put_flag(ctx->odp_actions,
2950 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2954 case NXAST_SET_QUEUE:
2955 nasq = (const struct nx_action_set_queue *) nah;
2956 xlate_set_queue_action(ctx, nasq);
2959 case NXAST_POP_QUEUE:
2960 add_pop_action(ctx);
2963 case NXAST_REG_MOVE:
2964 save_reg_state(ctx, &state);
2965 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2967 update_reg_state(ctx, &state);
2970 case NXAST_REG_LOAD:
2971 save_reg_state(ctx, &state);
2972 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2974 update_reg_state(ctx, &state);
2978 /* Nothing to do. */
2981 case NXAST_SET_TUNNEL64:
2982 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2983 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2984 ctx->flow.tun_id = tun_id;
2987 case NXAST_MULTIPATH:
2988 nam = (const struct nx_action_multipath *) nah;
2989 multipath_execute(nam, &ctx->flow);
2992 /* If you add a new action here that modifies flow data, don't forget to
2993 * update the flow key in ctx->flow at the same time. */
2995 case NXAST_SNAT__OBSOLETE:
2997 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3003 do_xlate_actions(const union ofp_action *in, size_t n_in,
3004 struct action_xlate_ctx *ctx)
3006 struct actions_iterator iter;
3007 const union ofp_action *ia;
3008 const struct ofport *port;
3010 port = get_port(ctx->ofproto, ctx->flow.in_port);
3011 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3012 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3013 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3014 /* Drop this flow. */
3018 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3019 enum ofp_action_type type = ntohs(ia->type);
3020 const struct ofp_action_dl_addr *oada;
3024 xlate_output_action(ctx, &ia->output);
3027 case OFPAT_SET_VLAN_VID:
3028 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3029 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3030 xlate_set_dl_tci(ctx);
3033 case OFPAT_SET_VLAN_PCP:
3034 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3035 ctx->flow.vlan_tci |= htons(
3036 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3037 xlate_set_dl_tci(ctx);
3040 case OFPAT_STRIP_VLAN:
3041 ctx->flow.vlan_tci = htons(0);
3042 xlate_set_dl_tci(ctx);
3045 case OFPAT_SET_DL_SRC:
3046 oada = ((struct ofp_action_dl_addr *) ia);
3047 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3048 oada->dl_addr, ETH_ADDR_LEN);
3049 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3052 case OFPAT_SET_DL_DST:
3053 oada = ((struct ofp_action_dl_addr *) ia);
3054 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3055 oada->dl_addr, ETH_ADDR_LEN);
3056 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3059 case OFPAT_SET_NW_SRC:
3060 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3061 ia->nw_addr.nw_addr);
3062 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3065 case OFPAT_SET_NW_DST:
3066 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3067 ia->nw_addr.nw_addr);
3068 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3071 case OFPAT_SET_NW_TOS:
3072 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3074 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3077 case OFPAT_SET_TP_SRC:
3078 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3079 ia->tp_port.tp_port);
3080 ctx->flow.tp_src = ia->tp_port.tp_port;
3083 case OFPAT_SET_TP_DST:
3084 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3085 ia->tp_port.tp_port);
3086 ctx->flow.tp_dst = ia->tp_port.tp_port;
3090 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3094 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3098 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3105 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3106 struct ofproto *ofproto, const struct flow *flow,
3107 const struct ofpbuf *packet)
3109 ctx->ofproto = ofproto;
3111 ctx->packet = packet;
3112 ctx->resubmit_hook = NULL;
3115 static struct ofpbuf *
3116 xlate_actions(struct action_xlate_ctx *ctx,
3117 const union ofp_action *in, size_t n_in)
3119 COVERAGE_INC(ofproto_ofp2odp);
3121 ctx->odp_actions = ofpbuf_new(512);
3123 ctx->may_set_up_flow = true;
3124 ctx->nf_output_iface = NF_OUT_DROP;
3126 ctx->last_pop_priority = -1;
3127 do_xlate_actions(in, n_in, ctx);
3128 remove_pop_action(ctx);
3130 /* Check with in-band control to see if we're allowed to set up this
3132 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3133 ctx->odp_actions->data, ctx->odp_actions->size)) {
3134 ctx->may_set_up_flow = false;
3137 return ctx->odp_actions;
3140 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3141 * error message code (composed with ofp_mkerr()) for the caller to propagate
3142 * upward. Otherwise, returns 0.
3144 * The log message mentions 'msg_type'. */
3146 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3148 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3149 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3150 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3153 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3160 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3162 struct ofproto *p = ofconn->ofproto;
3163 struct ofp_packet_out *opo;
3164 struct ofpbuf payload, *buffer;
3165 union ofp_action *ofp_actions;
3166 struct action_xlate_ctx ctx;
3167 struct ofpbuf *odp_actions;
3168 struct ofpbuf request;
3170 size_t n_ofp_actions;
3174 COVERAGE_INC(ofproto_packet_out);
3176 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3181 /* Get ofp_packet_out. */
3182 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3183 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3186 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3187 &ofp_actions, &n_ofp_actions);
3193 if (opo->buffer_id != htonl(UINT32_MAX)) {
3194 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3196 if (error || !buffer) {
3205 /* Extract flow, check actions. */
3206 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3208 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3214 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3215 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3216 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3217 ofpbuf_delete(odp_actions);
3220 ofpbuf_delete(buffer);
3225 update_port_config(struct ofproto *p, struct ofport *port,
3226 uint32_t config, uint32_t mask)
3228 mask &= config ^ port->opp.config;
3229 if (mask & OFPPC_PORT_DOWN) {
3230 if (config & OFPPC_PORT_DOWN) {
3231 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3233 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3236 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3237 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3238 if (mask & REVALIDATE_BITS) {
3239 COVERAGE_INC(ofproto_costly_flags);
3240 port->opp.config ^= mask & REVALIDATE_BITS;
3241 p->need_revalidate = true;
3243 #undef REVALIDATE_BITS
3244 if (mask & OFPPC_NO_PACKET_IN) {
3245 port->opp.config ^= OFPPC_NO_PACKET_IN;
3250 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3252 struct ofproto *p = ofconn->ofproto;
3253 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3254 struct ofport *port;
3257 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3262 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3264 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3265 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3266 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3268 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3269 if (opm->advertise) {
3270 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3276 static struct ofpbuf *
3277 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3279 struct ofp_stats_reply *osr;
3282 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3283 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3285 osr->flags = htons(0);
3289 static struct ofpbuf *
3290 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3292 const struct ofp_stats_request *osr
3293 = (const struct ofp_stats_request *) request;
3294 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3298 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3299 struct ofpbuf **msgp)
3301 struct ofpbuf *msg = *msgp;
3302 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3303 if (nbytes + msg->size > UINT16_MAX) {
3304 struct ofp_stats_reply *reply = msg->data;
3305 reply->flags = htons(OFPSF_REPLY_MORE);
3306 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3307 queue_tx(msg, ofconn, ofconn->reply_counter);
3309 return ofpbuf_put_uninit(*msgp, nbytes);
3312 static struct ofpbuf *
3313 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3315 struct nicira_stats_msg *nsm;
3318 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3319 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3320 nsm->type = htons(OFPST_VENDOR);
3321 nsm->flags = htons(0);
3322 nsm->vendor = htonl(NX_VENDOR_ID);
3323 nsm->subtype = subtype;
3327 static struct ofpbuf *
3328 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3330 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3334 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3335 struct ofpbuf **msgp)
3337 struct ofpbuf *msg = *msgp;
3338 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3339 if (nbytes + msg->size > UINT16_MAX) {
3340 struct nicira_stats_msg *reply = msg->data;
3341 reply->flags = htons(OFPSF_REPLY_MORE);
3342 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3343 queue_tx(msg, ofconn, ofconn->reply_counter);
3345 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3349 handle_desc_stats_request(struct ofconn *ofconn,
3350 const struct ofp_header *request)
3352 struct ofproto *p = ofconn->ofproto;
3353 struct ofp_desc_stats *ods;
3356 msg = start_ofp_stats_reply(request, sizeof *ods);
3357 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3358 memset(ods, 0, sizeof *ods);
3359 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3360 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3361 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3362 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3363 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3364 queue_tx(msg, ofconn, ofconn->reply_counter);
3370 handle_table_stats_request(struct ofconn *ofconn,
3371 const struct ofp_header *request)
3373 struct ofproto *p = ofconn->ofproto;
3374 struct ofp_table_stats *ots;
3377 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3379 /* Classifier table. */
3380 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3381 memset(ots, 0, sizeof *ots);
3382 strcpy(ots->name, "classifier");
3383 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3384 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3385 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3386 ots->active_count = htonl(classifier_count(&p->cls));
3387 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3388 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3390 queue_tx(msg, ofconn, ofconn->reply_counter);
3395 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3396 struct ofpbuf **msgp)
3398 struct netdev_stats stats;
3399 struct ofp_port_stats *ops;
3401 /* Intentionally ignore return value, since errors will set
3402 * 'stats' to all-1s, which is correct for OpenFlow, and
3403 * netdev_get_stats() will log errors. */
3404 netdev_get_stats(port->netdev, &stats);
3406 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3407 ops->port_no = htons(port->opp.port_no);
3408 memset(ops->pad, 0, sizeof ops->pad);
3409 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3410 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3411 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3412 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3413 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3414 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3415 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3416 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3417 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3418 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3419 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3420 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3424 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3426 struct ofproto *p = ofconn->ofproto;
3427 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3428 struct ofp_port_stats *ops;
3430 struct ofport *port;
3432 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3433 if (psr->port_no != htons(OFPP_NONE)) {
3434 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3436 append_port_stat(port, ofconn, &msg);
3439 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3440 append_port_stat(port, ofconn, &msg);
3444 queue_tx(msg, ofconn, ofconn->reply_counter);
3448 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3449 * '*packet_countp' and '*byte_countp'. The returned statistics include
3450 * statistics for all of 'rule''s facets. */
3452 query_stats(struct ofproto *p, struct rule *rule,
3453 uint64_t *packet_countp, uint64_t *byte_countp)
3455 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
3456 uint64_t packet_count, byte_count;
3457 struct facet *facet;
3460 /* Start from historical data for 'rule' itself that are no longer tracked
3461 * by the datapath. This counts, for example, facets that have expired. */
3462 packet_count = rule->packet_count;
3463 byte_count = rule->byte_count;
3465 /* Ask the datapath for statistics on all of the rule's facets.
3467 * Also, add any statistics that are not tracked by the datapath for each
3468 * facet. This includes, for example, statistics for packets that were
3469 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3471 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
3472 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3473 struct dpif_flow_stats stats;
3476 odp_flow_key_from_flow(&key, &facet->flow);
3477 dpif_flow_get(p->dpif, key.data, key.size, NULL, &stats);
3479 packet_count += stats.n_packets + facet->packet_count;
3480 byte_count += stats.n_bytes + facet->byte_count;
3483 /* Return the stats to the caller. */
3484 *packet_countp = packet_count;
3485 *byte_countp = byte_count;
3489 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3491 long long int msecs = time_msec() - start;
3492 *sec = htonl(msecs / 1000);
3493 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3497 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3498 ovs_be16 out_port, struct ofpbuf **replyp)
3500 struct ofp_flow_stats *ofs;
3501 uint64_t packet_count, byte_count;
3503 size_t act_len, len;
3505 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3509 act_len = sizeof *rule->actions * rule->n_actions;
3510 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3512 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3514 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3515 ofs->length = htons(len);
3518 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3519 rule->flow_cookie, &cookie);
3520 put_32aligned_be64(&ofs->cookie, cookie);
3521 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3522 ofs->priority = htons(rule->cr.priority);
3523 ofs->idle_timeout = htons(rule->idle_timeout);
3524 ofs->hard_timeout = htons(rule->hard_timeout);
3525 memset(ofs->pad2, 0, sizeof ofs->pad2);
3526 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3527 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3528 if (rule->n_actions > 0) {
3529 memcpy(ofs->actions, rule->actions, act_len);
3534 is_valid_table(uint8_t table_id)
3536 return table_id == 0 || table_id == 0xff;
3540 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3542 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3543 struct ofpbuf *reply;
3545 COVERAGE_INC(ofproto_flows_req);
3546 reply = start_ofp_stats_reply(oh, 1024);
3547 if (is_valid_table(fsr->table_id)) {
3548 struct cls_cursor cursor;
3549 struct cls_rule target;
3552 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3554 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3555 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3556 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3559 queue_tx(reply, ofconn, ofconn->reply_counter);
3565 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3566 ovs_be16 out_port, struct ofpbuf **replyp)
3568 struct nx_flow_stats *nfs;
3569 uint64_t packet_count, byte_count;
3570 size_t act_len, start_len;
3571 struct ofpbuf *reply;
3573 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3577 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3579 act_len = sizeof *rule->actions * rule->n_actions;
3581 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3582 start_len = (*replyp)->size;
3585 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3588 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3589 nfs->cookie = rule->flow_cookie;
3590 nfs->priority = htons(rule->cr.priority);
3591 nfs->idle_timeout = htons(rule->idle_timeout);
3592 nfs->hard_timeout = htons(rule->hard_timeout);
3593 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3594 memset(nfs->pad2, 0, sizeof nfs->pad2);
3595 nfs->packet_count = htonll(packet_count);
3596 nfs->byte_count = htonll(byte_count);
3597 if (rule->n_actions > 0) {
3598 ofpbuf_put(reply, rule->actions, act_len);
3600 nfs->length = htons(reply->size - start_len);
3604 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3606 struct nx_flow_stats_request *nfsr;
3607 struct cls_rule target;
3608 struct ofpbuf *reply;
3612 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3614 /* Dissect the message. */
3615 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3616 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3621 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3624 COVERAGE_INC(ofproto_flows_req);
3625 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3626 if (is_valid_table(nfsr->table_id)) {
3627 struct cls_cursor cursor;
3630 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3631 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3632 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3635 queue_tx(reply, ofconn, ofconn->reply_counter);
3641 flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
3643 uint64_t packet_count, byte_count;
3644 size_t act_len = sizeof *rule->actions * rule->n_actions;
3646 query_stats(ofproto, rule, &packet_count, &byte_count);
3648 ds_put_format(results, "duration=%llds, ",
3649 (time_msec() - rule->created) / 1000);
3650 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3651 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3652 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3653 cls_rule_format(&rule->cr, results);
3655 ofp_print_actions(results, &rule->actions->header, act_len);
3657 ds_put_cstr(results, "drop");
3659 ds_put_cstr(results, "\n");
3662 /* Adds a pretty-printed description of all flows to 'results', including
3663 * those marked hidden by secchan (e.g., by in-band control). */
3665 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3667 struct cls_cursor cursor;
3670 cls_cursor_init(&cursor, &p->cls, NULL);
3671 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3672 flow_stats_ds(p, rule, results);
3677 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3678 ovs_be16 out_port, uint8_t table_id,
3679 struct ofp_aggregate_stats_reply *oasr)
3681 uint64_t total_packets = 0;
3682 uint64_t total_bytes = 0;
3685 COVERAGE_INC(ofproto_agg_request);
3687 if (is_valid_table(table_id)) {
3688 struct cls_cursor cursor;
3691 cls_cursor_init(&cursor, &ofproto->cls, target);
3692 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3693 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3694 uint64_t packet_count;
3695 uint64_t byte_count;
3697 query_stats(ofproto, rule, &packet_count, &byte_count);
3699 total_packets += packet_count;
3700 total_bytes += byte_count;
3706 oasr->flow_count = htonl(n_flows);
3707 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3708 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3709 memset(oasr->pad, 0, sizeof oasr->pad);
3713 handle_aggregate_stats_request(struct ofconn *ofconn,
3714 const struct ofp_header *oh)
3716 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3717 struct ofp_aggregate_stats_reply *reply;
3718 struct cls_rule target;
3721 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3724 msg = start_ofp_stats_reply(oh, sizeof *reply);
3725 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3726 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3727 request->table_id, reply);
3728 queue_tx(msg, ofconn, ofconn->reply_counter);
3733 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3735 struct nx_aggregate_stats_request *request;
3736 struct ofp_aggregate_stats_reply *reply;
3737 struct cls_rule target;
3742 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3744 /* Dissect the message. */
3745 request = ofpbuf_pull(&b, sizeof *request);
3746 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3751 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3755 COVERAGE_INC(ofproto_flows_req);
3756 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3757 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3758 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3759 request->table_id, reply);
3760 queue_tx(buf, ofconn, ofconn->reply_counter);
3765 struct queue_stats_cbdata {
3766 struct ofconn *ofconn;
3767 struct ofport *ofport;
3772 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3773 const struct netdev_queue_stats *stats)
3775 struct ofp_queue_stats *reply;
3777 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3778 reply->port_no = htons(cbdata->ofport->opp.port_no);
3779 memset(reply->pad, 0, sizeof reply->pad);
3780 reply->queue_id = htonl(queue_id);
3781 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3782 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3783 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3787 handle_queue_stats_dump_cb(uint32_t queue_id,
3788 struct netdev_queue_stats *stats,
3791 struct queue_stats_cbdata *cbdata = cbdata_;
3793 put_queue_stats(cbdata, queue_id, stats);
3797 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3798 struct queue_stats_cbdata *cbdata)
3800 cbdata->ofport = port;
3801 if (queue_id == OFPQ_ALL) {
3802 netdev_dump_queue_stats(port->netdev,
3803 handle_queue_stats_dump_cb, cbdata);
3805 struct netdev_queue_stats stats;
3807 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3808 put_queue_stats(cbdata, queue_id, &stats);
3814 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3816 struct ofproto *ofproto = ofconn->ofproto;
3817 const struct ofp_queue_stats_request *qsr;
3818 struct queue_stats_cbdata cbdata;
3819 struct ofport *port;
3820 unsigned int port_no;
3823 qsr = ofputil_stats_body(oh);
3825 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3828 COVERAGE_INC(ofproto_queue_req);
3830 cbdata.ofconn = ofconn;
3831 cbdata.msg = start_ofp_stats_reply(oh, 128);
3833 port_no = ntohs(qsr->port_no);
3834 queue_id = ntohl(qsr->queue_id);
3835 if (port_no == OFPP_ALL) {
3836 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3837 handle_queue_stats_for_port(port, queue_id, &cbdata);
3839 } else if (port_no < ofproto->max_ports) {
3840 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3842 handle_queue_stats_for_port(port, queue_id, &cbdata);
3845 ofpbuf_delete(cbdata.msg);
3846 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3848 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3854 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3855 const struct dpif_flow_stats *stats)
3857 long long int used = stats->used;
3858 if (used > facet->used) {
3860 if (used > facet->rule->used) {
3861 facet->rule->used = used;
3863 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3867 /* Folds the statistics from 'stats' into the counters in 'facet'.
3869 * Because of the meaning of a facet's counters, it only makes sense to do this
3870 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3871 * packet that was sent by hand or if it represents statistics that have been
3872 * cleared out of the datapath. */
3874 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3875 const struct dpif_flow_stats *stats)
3877 if (stats->n_packets) {
3878 facet_update_time(ofproto, facet, stats);
3879 facet->packet_count += stats->n_packets;
3880 facet->byte_count += stats->n_bytes;
3881 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3885 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3886 * in which no matching flow already exists in the flow table.
3888 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3889 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3890 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3892 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3895 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3897 struct ofproto *p = ofconn->ofproto;
3898 struct ofpbuf *packet;
3903 if (fm->flags & OFPFF_CHECK_OVERLAP
3904 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3905 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3909 if (fm->buffer_id != UINT32_MAX) {
3910 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3914 in_port = UINT16_MAX;
3917 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3918 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3919 fm->flags & OFPFF_SEND_FLOW_REM);
3920 rule_insert(p, rule);
3922 rule_execute(p, rule, in_port, packet);
3927 static struct rule *
3928 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3930 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3934 send_buffered_packet(struct ofconn *ofconn,
3935 struct rule *rule, uint32_t buffer_id)
3937 struct ofpbuf *packet;
3941 if (buffer_id == UINT32_MAX) {
3945 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3950 rule_execute(ofconn->ofproto, rule, in_port, packet);
3955 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3957 struct modify_flows_cbdata {
3958 struct ofproto *ofproto;
3959 const struct flow_mod *fm;
3963 static int modify_flow(struct ofproto *, const struct flow_mod *,
3966 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3967 * encoded by ofp_mkerr() on failure.
3969 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3972 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3974 struct ofproto *p = ofconn->ofproto;
3975 struct rule *match = NULL;
3976 struct cls_cursor cursor;
3979 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3980 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3981 if (!rule_is_hidden(rule)) {
3983 modify_flow(p, fm, rule);
3988 /* This credits the packet to whichever flow happened to match last.
3989 * That's weird. Maybe we should do a lookup for the flow that
3990 * actually matches the packet? Who knows. */
3991 send_buffered_packet(ofconn, match, fm->buffer_id);
3994 return add_flow(ofconn, fm);
3998 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3999 * code as encoded by ofp_mkerr() on failure.
4001 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4004 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4006 struct ofproto *p = ofconn->ofproto;
4007 struct rule *rule = find_flow_strict(p, fm);
4008 if (rule && !rule_is_hidden(rule)) {
4009 modify_flow(p, fm, rule);
4010 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4012 return add_flow(ofconn, fm);
4016 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4017 * been identified as a flow in 'p''s flow table to be modified, by changing
4018 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4019 * ofp_action[] structures). */
4021 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4023 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4025 rule->flow_cookie = fm->cookie;
4027 /* If the actions are the same, do nothing. */
4028 if (fm->n_actions == rule->n_actions
4030 || !memcmp(fm->actions, rule->actions, actions_len))) {
4034 /* Replace actions. */
4035 free(rule->actions);
4036 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4037 rule->n_actions = fm->n_actions;
4039 p->need_revalidate = true;
4044 /* OFPFC_DELETE implementation. */
4046 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4048 /* Implements OFPFC_DELETE. */
4050 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4052 struct rule *rule, *next_rule;
4053 struct cls_cursor cursor;
4055 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4056 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4057 delete_flow(p, rule, htons(fm->out_port));
4061 /* Implements OFPFC_DELETE_STRICT. */
4063 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4065 struct rule *rule = find_flow_strict(p, fm);
4067 delete_flow(p, rule, htons(fm->out_port));
4071 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4072 * been identified as a flow to delete from 'p''s flow table, by deleting the
4073 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4076 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4077 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4078 * specified 'out_port'. */
4080 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4082 if (rule_is_hidden(rule)) {
4086 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4090 rule_send_removed(p, rule, OFPRR_DELETE);
4091 rule_remove(p, rule);
4095 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4097 struct ofproto *p = ofconn->ofproto;
4101 error = reject_slave_controller(ofconn, "flow_mod");
4106 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4111 /* We do not support the emergency flow cache. It will hopefully get
4112 * dropped from OpenFlow in the near future. */
4113 if (fm.flags & OFPFF_EMERG) {
4114 /* There isn't a good fit for an error code, so just state that the
4115 * flow table is full. */
4116 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4119 error = validate_actions(fm.actions, fm.n_actions,
4120 &fm.cr.flow, p->max_ports);
4125 switch (fm.command) {
4127 return add_flow(ofconn, &fm);
4130 return modify_flows_loose(ofconn, &fm);
4132 case OFPFC_MODIFY_STRICT:
4133 return modify_flow_strict(ofconn, &fm);
4136 delete_flows_loose(p, &fm);
4139 case OFPFC_DELETE_STRICT:
4140 delete_flow_strict(p, &fm);
4144 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4149 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4151 const struct nxt_tun_id_cookie *msg
4152 = (const struct nxt_tun_id_cookie *) oh;
4154 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4159 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4161 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4162 struct nx_role_request *reply;
4166 if (ofconn->type != OFCONN_PRIMARY) {
4167 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4169 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4172 role = ntohl(nrr->role);
4173 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4174 && role != NX_ROLE_SLAVE) {
4175 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4177 /* There's no good error code for this. */
4178 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4181 if (role == NX_ROLE_MASTER) {
4182 struct ofconn *other;
4184 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4185 if (other->role == NX_ROLE_MASTER) {
4186 other->role = NX_ROLE_SLAVE;
4190 ofconn->role = role;
4192 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4193 reply->role = htonl(role);
4194 queue_tx(buf, ofconn, ofconn->reply_counter);
4200 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4202 const struct nxt_set_flow_format *msg
4203 = (const struct nxt_set_flow_format *) oh;
4206 format = ntohl(msg->format);
4207 if (format == NXFF_OPENFLOW10
4208 || format == NXFF_TUN_ID_FROM_COOKIE
4209 || format == NXFF_NXM) {
4210 ofconn->flow_format = format;
4213 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4218 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4220 struct ofp_header *ob;
4223 /* Currently, everything executes synchronously, so we can just
4224 * immediately send the barrier reply. */
4225 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4226 queue_tx(buf, ofconn, ofconn->reply_counter);
4231 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4233 const struct ofp_header *oh = msg->data;
4234 const struct ofputil_msg_type *type;
4237 error = ofputil_decode_msg_type(oh, &type);
4242 switch (ofputil_msg_type_code(type)) {
4243 /* OpenFlow requests. */
4244 case OFPUTIL_OFPT_ECHO_REQUEST:
4245 return handle_echo_request(ofconn, oh);
4247 case OFPUTIL_OFPT_FEATURES_REQUEST:
4248 return handle_features_request(ofconn, oh);
4250 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4251 return handle_get_config_request(ofconn, oh);
4253 case OFPUTIL_OFPT_SET_CONFIG:
4254 return handle_set_config(ofconn, msg->data);
4256 case OFPUTIL_OFPT_PACKET_OUT:
4257 return handle_packet_out(ofconn, oh);
4259 case OFPUTIL_OFPT_PORT_MOD:
4260 return handle_port_mod(ofconn, oh);
4262 case OFPUTIL_OFPT_FLOW_MOD:
4263 return handle_flow_mod(ofconn, oh);
4265 case OFPUTIL_OFPT_BARRIER_REQUEST:
4266 return handle_barrier_request(ofconn, oh);
4268 /* OpenFlow replies. */
4269 case OFPUTIL_OFPT_ECHO_REPLY:
4272 /* Nicira extension requests. */
4273 case OFPUTIL_NXT_STATUS_REQUEST:
4274 return switch_status_handle_request(
4275 ofconn->ofproto->switch_status, ofconn->rconn, oh);
4277 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4278 return handle_tun_id_from_cookie(ofconn, oh);
4280 case OFPUTIL_NXT_ROLE_REQUEST:
4281 return handle_role_request(ofconn, oh);
4283 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4284 return handle_nxt_set_flow_format(ofconn, oh);
4286 case OFPUTIL_NXT_FLOW_MOD:
4287 return handle_flow_mod(ofconn, oh);
4289 /* OpenFlow statistics requests. */
4290 case OFPUTIL_OFPST_DESC_REQUEST:
4291 return handle_desc_stats_request(ofconn, oh);
4293 case OFPUTIL_OFPST_FLOW_REQUEST:
4294 return handle_flow_stats_request(ofconn, oh);
4296 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4297 return handle_aggregate_stats_request(ofconn, oh);
4299 case OFPUTIL_OFPST_TABLE_REQUEST:
4300 return handle_table_stats_request(ofconn, oh);
4302 case OFPUTIL_OFPST_PORT_REQUEST:
4303 return handle_port_stats_request(ofconn, oh);
4305 case OFPUTIL_OFPST_QUEUE_REQUEST:
4306 return handle_queue_stats_request(ofconn, oh);
4308 /* Nicira extension statistics requests. */
4309 case OFPUTIL_NXST_FLOW_REQUEST:
4310 return handle_nxst_flow(ofconn, oh);
4312 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4313 return handle_nxst_aggregate(ofconn, oh);
4315 case OFPUTIL_INVALID:
4316 case OFPUTIL_OFPT_HELLO:
4317 case OFPUTIL_OFPT_ERROR:
4318 case OFPUTIL_OFPT_FEATURES_REPLY:
4319 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4320 case OFPUTIL_OFPT_PACKET_IN:
4321 case OFPUTIL_OFPT_FLOW_REMOVED:
4322 case OFPUTIL_OFPT_PORT_STATUS:
4323 case OFPUTIL_OFPT_BARRIER_REPLY:
4324 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4325 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4326 case OFPUTIL_OFPST_DESC_REPLY:
4327 case OFPUTIL_OFPST_FLOW_REPLY:
4328 case OFPUTIL_OFPST_QUEUE_REPLY:
4329 case OFPUTIL_OFPST_PORT_REPLY:
4330 case OFPUTIL_OFPST_TABLE_REPLY:
4331 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4332 case OFPUTIL_NXT_STATUS_REPLY:
4333 case OFPUTIL_NXT_ROLE_REPLY:
4334 case OFPUTIL_NXT_FLOW_REMOVED:
4335 case OFPUTIL_NXST_FLOW_REPLY:
4336 case OFPUTIL_NXST_AGGREGATE_REPLY:
4338 if (VLOG_IS_WARN_ENABLED()) {
4339 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4340 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4343 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4344 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4346 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4352 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4354 int error = handle_openflow__(ofconn, ofp_msg);
4356 send_error_oh(ofconn, ofp_msg->data, error);
4358 COVERAGE_INC(ofproto_recv_openflow);
4362 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4364 struct facet *facet;
4367 /* Obtain in_port and tun_id, at least. */
4368 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4370 /* Set header pointers in 'flow'. */
4371 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4373 /* Check with in-band control to see if this packet should be sent
4374 * to the local port regardless of the flow table. */
4375 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4376 struct ofpbuf odp_actions;
4378 ofpbuf_init(&odp_actions, 32);
4379 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, ODPP_LOCAL);
4380 dpif_execute(p->dpif, odp_actions.data, odp_actions.size,
4382 ofpbuf_uninit(&odp_actions);
4385 facet = facet_lookup_valid(p, &flow);
4387 struct rule *rule = rule_lookup(p, &flow);
4389 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4390 struct ofport *port = get_port(p, flow.in_port);
4392 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4393 COVERAGE_INC(ofproto_no_packet_in);
4394 /* XXX install 'drop' flow entry */
4395 ofpbuf_delete(upcall->packet);
4399 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4403 COVERAGE_INC(ofproto_packet_in);
4404 send_packet_in(p, upcall, &flow, false);
4408 facet = facet_create(p, rule, &flow, upcall->packet);
4409 } else if (!facet->may_install) {
4410 /* The facet is not installable, that is, we need to process every
4411 * packet, so process the current packet's actions into 'facet'. */
4412 facet_make_actions(p, facet, upcall->packet);
4415 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4417 * Extra-special case for fail-open mode.
4419 * We are in fail-open mode and the packet matched the fail-open rule,
4420 * but we are connected to a controller too. We should send the packet
4421 * up to the controller in the hope that it will try to set up a flow
4422 * and thereby allow us to exit fail-open.
4424 * See the top-level comment in fail-open.c for more information.
4426 send_packet_in(p, upcall, &flow, true);
4429 facet_execute(p, facet, upcall->packet);
4430 facet_install(p, facet, false);
4434 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4438 switch (upcall->type) {
4439 case DPIF_UC_ACTION:
4440 COVERAGE_INC(ofproto_ctlr_action);
4441 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4442 send_packet_in(p, upcall, &flow, false);
4445 case DPIF_UC_SAMPLE:
4447 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4448 ofproto_sflow_received(p->sflow, upcall, &flow);
4450 ofpbuf_delete(upcall->packet);
4454 handle_miss_upcall(p, upcall);
4457 case DPIF_N_UC_TYPES:
4459 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4464 /* Flow expiration. */
4466 static int ofproto_dp_max_idle(const struct ofproto *);
4467 static void ofproto_update_used(struct ofproto *);
4468 static void rule_expire(struct ofproto *, struct rule *);
4469 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4471 /* This function is called periodically by ofproto_run(). Its job is to
4472 * collect updates for the flows that have been installed into the datapath,
4473 * most importantly when they last were used, and then use that information to
4474 * expire flows that have not been used recently.
4476 * Returns the number of milliseconds after which it should be called again. */
4478 ofproto_expire(struct ofproto *ofproto)
4480 struct rule *rule, *next_rule;
4481 struct cls_cursor cursor;
4484 /* Update 'used' for each flow in the datapath. */
4485 ofproto_update_used(ofproto);
4487 /* Expire facets that have been idle too long. */
4488 dp_max_idle = ofproto_dp_max_idle(ofproto);
4489 ofproto_expire_facets(ofproto, dp_max_idle);
4491 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4492 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4493 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4494 rule_expire(ofproto, rule);
4497 /* Let the hook know that we're at a stable point: all outstanding data
4498 * in existing flows has been accounted to the account_cb. Thus, the
4499 * hook can now reasonably do operations that depend on having accurate
4500 * flow volume accounting (currently, that's just bond rebalancing). */
4501 if (ofproto->ofhooks->account_checkpoint_cb) {
4502 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4505 return MIN(dp_max_idle, 1000);
4508 /* Update 'used' member of installed facets. */
4510 ofproto_update_used(struct ofproto *p)
4512 const struct dpif_flow_stats *stats;
4513 struct dpif_flow_dump dump;
4514 const struct nlattr *key;
4517 dpif_flow_dump_start(&dump, p->dpif);
4518 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4519 struct facet *facet;
4522 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4526 odp_flow_key_format(key, key_len, &s);
4527 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4533 facet = facet_find(p, &flow);
4535 if (facet && facet->installed) {
4536 facet_update_time(p, facet, stats);
4537 facet_account(p, facet, stats->n_bytes);
4539 /* There's a flow in the datapath that we know nothing about.
4541 COVERAGE_INC(ofproto_unexpected_rule);
4542 dpif_flow_del(p->dpif, key, key_len, NULL);
4545 dpif_flow_dump_done(&dump);
4548 /* Calculates and returns the number of milliseconds of idle time after which
4549 * facets should expire from the datapath and we should fold their statistics
4550 * into their parent rules in userspace. */
4552 ofproto_dp_max_idle(const struct ofproto *ofproto)
4555 * Idle time histogram.
4557 * Most of the time a switch has a relatively small number of facets. When
4558 * this is the case we might as well keep statistics for all of them in
4559 * userspace and to cache them in the kernel datapath for performance as
4562 * As the number of facets increases, the memory required to maintain
4563 * statistics about them in userspace and in the kernel becomes
4564 * significant. However, with a large number of facets it is likely that
4565 * only a few of them are "heavy hitters" that consume a large amount of
4566 * bandwidth. At this point, only heavy hitters are worth caching in the
4567 * kernel and maintaining in userspaces; other facets we can discard.
4569 * The technique used to compute the idle time is to build a histogram with
4570 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4571 * that is installed in the kernel gets dropped in the appropriate bucket.
4572 * After the histogram has been built, we compute the cutoff so that only
4573 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4574 * cached. At least the most-recently-used bucket of facets is kept, so
4575 * actually an arbitrary number of facets can be kept in any given
4576 * expiration run (though the next run will delete most of those unless
4577 * they receive additional data).
4579 * This requires a second pass through the facets, in addition to the pass
4580 * made by ofproto_update_used(), because the former function never looks
4581 * at uninstallable facets.
4583 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4584 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4585 int buckets[N_BUCKETS] = { 0 };
4586 struct facet *facet;
4591 total = hmap_count(&ofproto->facets);
4592 if (total <= 1000) {
4593 return N_BUCKETS * BUCKET_WIDTH;
4596 /* Build histogram. */
4598 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4599 long long int idle = now - facet->used;
4600 int bucket = (idle <= 0 ? 0
4601 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4602 : (unsigned int) idle / BUCKET_WIDTH);
4606 /* Find the first bucket whose flows should be expired. */
4607 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4608 if (buckets[bucket]) {
4611 subtotal += buckets[bucket++];
4612 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4617 if (VLOG_IS_DBG_ENABLED()) {
4621 ds_put_cstr(&s, "keep");
4622 for (i = 0; i < N_BUCKETS; i++) {
4624 ds_put_cstr(&s, ", drop");
4627 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4630 VLOG_INFO("%s: %s (msec:count)",
4631 dpif_name(ofproto->dpif), ds_cstr(&s));
4635 return bucket * BUCKET_WIDTH;
4639 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4641 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4642 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4643 struct ofexpired expired;
4645 if (facet->installed) {
4646 struct dpif_flow_stats stats;
4648 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4650 facet_update_stats(ofproto, facet, &stats);
4653 expired.flow = facet->flow;
4654 expired.packet_count = facet->packet_count;
4655 expired.byte_count = facet->byte_count;
4656 expired.used = facet->used;
4657 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4662 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4664 long long int cutoff = time_msec() - dp_max_idle;
4665 struct facet *facet, *next_facet;
4667 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4668 facet_active_timeout(ofproto, facet);
4669 if (facet->used < cutoff) {
4670 facet_remove(ofproto, facet);
4675 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4676 * then delete it entirely. */
4678 rule_expire(struct ofproto *ofproto, struct rule *rule)
4680 struct facet *facet, *next_facet;
4684 /* Has 'rule' expired? */
4686 if (rule->hard_timeout
4687 && now > rule->created + rule->hard_timeout * 1000) {
4688 reason = OFPRR_HARD_TIMEOUT;
4689 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4690 && now >rule->used + rule->idle_timeout * 1000) {
4691 reason = OFPRR_IDLE_TIMEOUT;
4696 COVERAGE_INC(ofproto_expired);
4698 /* Update stats. (This is a no-op if the rule expired due to an idle
4699 * timeout, because that only happens when the rule has no facets left.) */
4700 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4701 facet_remove(ofproto, facet);
4704 /* Get rid of the rule. */
4705 if (!rule_is_hidden(rule)) {
4706 rule_send_removed(ofproto, rule, reason);
4708 rule_remove(ofproto, rule);
4711 static struct ofpbuf *
4712 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4715 struct ofp_flow_removed *ofr;
4718 ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
4719 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4720 rule->flow_cookie, &ofr->cookie);
4721 ofr->priority = htons(rule->cr.priority);
4722 ofr->reason = reason;
4723 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4724 ofr->idle_timeout = htons(rule->idle_timeout);
4725 ofr->packet_count = htonll(rule->packet_count);
4726 ofr->byte_count = htonll(rule->byte_count);
4731 static struct ofpbuf *
4732 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4734 struct nx_flow_removed *nfr;
4738 make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
4739 match_len = nx_put_match(buf, &rule->cr);
4742 nfr->cookie = rule->flow_cookie;
4743 nfr->priority = htons(rule->cr.priority);
4744 nfr->reason = reason;
4745 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4746 nfr->idle_timeout = htons(rule->idle_timeout);
4747 nfr->match_len = htons(match_len);
4748 nfr->packet_count = htonll(rule->packet_count);
4749 nfr->byte_count = htonll(rule->byte_count);
4755 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4757 struct ofconn *ofconn;
4759 if (!rule->send_flow_removed) {
4763 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4766 if (!rconn_is_connected(ofconn->rconn)
4767 || !ofconn_receives_async_msgs(ofconn)) {
4771 msg = (ofconn->flow_format == NXFF_NXM
4772 ? compose_nx_flow_removed(rule, reason)
4773 : compose_ofp_flow_removed(ofconn, rule, reason));
4775 /* Account flow expirations under ofconn->reply_counter, the counter
4776 * for replies to OpenFlow requests. That works because preventing
4777 * OpenFlow requests from being processed also prevents new flows from
4778 * being added (and expiring). (It also prevents processing OpenFlow
4779 * requests that would not add new flows, so it is imperfect.) */
4780 queue_tx(msg, ofconn, ofconn->reply_counter);
4784 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4786 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4788 struct ofconn *ofconn = ofconn_;
4790 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4791 ofconn->packet_in_counter, 100);
4794 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4795 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4796 * scheduler for sending.
4798 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4799 * Otherwise, ownership is transferred to this function. */
4801 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4802 const struct flow *flow, bool clone)
4804 enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) };
4805 struct ofproto *ofproto = ofconn->ofproto;
4806 struct ofp_packet_in *opi;
4807 int total_len, send_len;
4808 struct ofpbuf *packet;
4812 /* Get OpenFlow buffer_id. */
4813 if (upcall->type == DPIF_UC_ACTION) {
4814 buffer_id = UINT32_MAX;
4815 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4816 buffer_id = pktbuf_get_null();
4817 } else if (!ofconn->pktbuf) {
4818 buffer_id = UINT32_MAX;
4820 buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port);
4823 /* Figure out how much of the packet to send. */
4824 total_len = send_len = upcall->packet->size;
4825 if (buffer_id != UINT32_MAX) {
4826 send_len = MIN(send_len, ofconn->miss_send_len);
4828 if (upcall->type == DPIF_UC_ACTION) {
4829 send_len = MIN(send_len, upcall->userdata);
4832 /* Copy or steal buffer for OFPT_PACKET_IN. */
4834 packet = ofpbuf_clone_data_with_headroom(upcall->packet->data,
4835 send_len, OPI_SIZE);
4837 packet = upcall->packet;
4838 packet->size = send_len;
4841 /* Add OFPT_PACKET_IN. */
4842 opi = ofpbuf_push_zeros(packet, OPI_SIZE);
4843 opi->header.version = OFP_VERSION;
4844 opi->header.type = OFPT_PACKET_IN;
4845 opi->total_len = htons(total_len);
4846 opi->in_port = htons(odp_port_to_ofp_port(flow->in_port));
4847 opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4848 opi->buffer_id = htonl(buffer_id);
4849 update_openflow_length(packet);
4851 /* Hand over to packet scheduler. It might immediately call into
4852 * do_send_packet_in() or it might buffer it for a while (until a later
4853 * call to pinsched_run()). */
4854 idx = upcall->type == DPIF_UC_MISS ? 0 : 1;
4855 pinsched_send(ofconn->schedulers[idx], flow->in_port,
4856 packet, do_send_packet_in, ofconn);
4859 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4860 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4861 * their individual configurations.
4863 * Takes ownership of 'packet'. */
4865 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4866 const struct flow *flow, bool clone)
4868 struct ofconn *ofconn, *prev;
4871 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4872 if (ofconn_receives_async_msgs(ofconn)) {
4874 schedule_packet_in(prev, upcall, flow, true);
4880 schedule_packet_in(prev, upcall, flow, clone);
4881 } else if (!clone) {
4882 ofpbuf_delete(upcall->packet);
4887 pick_datapath_id(const struct ofproto *ofproto)
4889 const struct ofport *port;
4891 port = get_port(ofproto, ODPP_LOCAL);
4893 uint8_t ea[ETH_ADDR_LEN];
4896 error = netdev_get_etheraddr(port->netdev, ea);
4898 return eth_addr_to_uint64(ea);
4900 VLOG_WARN("could not get MAC address for %s (%s)",
4901 netdev_get_name(port->netdev), strerror(error));
4903 return ofproto->fallback_dpid;
4907 pick_fallback_dpid(void)
4909 uint8_t ea[ETH_ADDR_LEN];
4910 eth_addr_nicira_random(ea);
4911 return eth_addr_to_uint64(ea);
4915 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4916 void *aux OVS_UNUSED)
4918 const struct shash_node *node;
4922 SHASH_FOR_EACH (node, &all_ofprotos) {
4923 ds_put_format(&results, "%s\n", node->name);
4925 unixctl_command_reply(conn, 200, ds_cstr(&results));
4926 ds_destroy(&results);
4929 struct ofproto_trace {
4930 struct action_xlate_ctx ctx;
4936 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4938 ds_put_char_multiple(result, '\t', level);
4940 ds_put_cstr(result, "No match\n");
4944 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4945 ntohll(rule->flow_cookie));
4946 cls_rule_format(&rule->cr, result);
4947 ds_put_char(result, '\n');
4949 ds_put_char_multiple(result, '\t', level);
4950 ds_put_cstr(result, "OpenFlow ");
4951 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4952 rule->n_actions * sizeof *rule->actions);
4953 ds_put_char(result, '\n');
4957 trace_format_flow(struct ds *result, int level, const char *title,
4958 struct ofproto_trace *trace)
4960 ds_put_char_multiple(result, '\t', level);
4961 ds_put_format(result, "%s: ", title);
4962 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4963 ds_put_cstr(result, "unchanged");
4965 flow_format(result, &trace->ctx.flow);
4966 trace->flow = trace->ctx.flow;
4968 ds_put_char(result, '\n');
4972 trace_resubmit(struct action_xlate_ctx *ctx, const struct rule *rule)
4974 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4975 struct ds *result = trace->result;
4977 ds_put_char(result, '\n');
4978 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4979 trace_format_rule(result, ctx->recurse + 1, rule);
4983 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4984 void *aux OVS_UNUSED)
4986 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4987 char *args = xstrdup(args_);
4988 char *save_ptr = NULL;
4989 struct ofproto *ofproto;
4990 struct ofpbuf packet;
4998 ofpbuf_init(&packet, strlen(args) / 2);
5001 dpname = strtok_r(args, " ", &save_ptr);
5002 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5003 in_port_s = strtok_r(NULL, " ", &save_ptr);
5004 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5005 if (!dpname || !in_port_s || !packet_s) {
5006 unixctl_command_reply(conn, 501, "Bad command syntax");
5010 ofproto = shash_find_data(&all_ofprotos, dpname);
5012 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5017 tun_id = htonll(strtoull(tun_id_s, NULL, 10));
5018 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5020 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5021 packet_s += strspn(packet_s, " ");
5022 if (*packet_s != '\0') {
5023 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5026 if (packet.size < ETH_HEADER_LEN) {
5027 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5031 ds_put_cstr(&result, "Packet: ");
5032 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5033 ds_put_cstr(&result, s);
5036 flow_extract(&packet, tun_id, in_port, &flow);
5037 ds_put_cstr(&result, "Flow: ");
5038 flow_format(&result, &flow);
5039 ds_put_char(&result, '\n');
5041 rule = rule_lookup(ofproto, &flow);
5042 trace_format_rule(&result, 0, rule);
5044 struct ofproto_trace trace;
5045 struct ofpbuf *odp_actions;
5047 trace.result = &result;
5049 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5050 trace.ctx.resubmit_hook = trace_resubmit;
5051 odp_actions = xlate_actions(&trace.ctx,
5052 rule->actions, rule->n_actions);
5054 ds_put_char(&result, '\n');
5055 trace_format_flow(&result, 0, "Final flow", &trace);
5056 ds_put_cstr(&result, "Datapath actions: ");
5057 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5058 ofpbuf_delete(odp_actions);
5061 unixctl_command_reply(conn, 200, ds_cstr(&result));
5064 ds_destroy(&result);
5065 ofpbuf_uninit(&packet);
5070 ofproto_unixctl_init(void)
5072 static bool registered;
5078 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5079 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5083 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5084 struct ofpbuf *odp_actions, tag_type *tags,
5085 uint16_t *nf_output_iface, void *ofproto_)
5087 struct ofproto *ofproto = ofproto_;
5090 /* Drop frames for reserved multicast addresses. */
5091 if (eth_addr_is_reserved(flow->dl_dst)) {
5095 /* Learn source MAC (but don't try to learn from revalidation). */
5096 if (packet != NULL) {
5097 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5099 GRAT_ARP_LOCK_NONE);
5101 /* The log messages here could actually be useful in debugging,
5102 * so keep the rate limit relatively high. */
5103 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5104 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5105 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5106 ofproto_revalidate(ofproto, rev_tag);
5110 /* Determine output port. */
5111 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5114 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5115 nf_output_iface, odp_actions);
5116 } else if (out_port != flow->in_port) {
5117 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5118 *nf_output_iface = out_port;
5126 static const struct ofhooks default_ofhooks = {
5127 default_normal_ofhook_cb,