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"
66 VLOG_DEFINE_THIS_MODULE(ofproto);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(odp_overflow);
71 COVERAGE_DEFINE(ofproto_agg_request);
72 COVERAGE_DEFINE(ofproto_costly_flags);
73 COVERAGE_DEFINE(ofproto_ctlr_action);
74 COVERAGE_DEFINE(ofproto_del_rule);
75 COVERAGE_DEFINE(ofproto_error);
76 COVERAGE_DEFINE(ofproto_expiration);
77 COVERAGE_DEFINE(ofproto_expired);
78 COVERAGE_DEFINE(ofproto_flows_req);
79 COVERAGE_DEFINE(ofproto_flush);
80 COVERAGE_DEFINE(ofproto_invalidated);
81 COVERAGE_DEFINE(ofproto_no_packet_in);
82 COVERAGE_DEFINE(ofproto_ofconn_stuck);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_unexpected_rule);
90 COVERAGE_DEFINE(ofproto_uninstallable);
91 COVERAGE_DEFINE(ofproto_update_port);
93 #include "sflow_api.h"
98 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
99 struct netdev *netdev;
100 struct ofp_phy_port opp; /* In host byte order. */
104 static void ofport_free(struct ofport *);
105 static void hton_ofp_phy_port(struct ofp_phy_port *);
107 struct action_xlate_ctx {
108 /* action_xlate_ctx_init() initializes these members. */
111 struct ofproto *ofproto;
113 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
114 * this flow when actions change header fields. */
117 /* The packet corresponding to 'flow', or a null pointer if we are
118 * revalidating without a packet to refer to. */
119 const struct ofpbuf *packet;
121 /* If nonnull, called just before executing a resubmit action.
123 * This is normally null so the client has to set it manually after
124 * calling action_xlate_ctx_init(). */
125 void (*resubmit_hook)(struct action_xlate_ctx *, const struct rule *);
127 /* xlate_actions() initializes and uses these members. The client might want
128 * to look at them after it returns. */
130 struct ofpbuf *odp_actions; /* Datapath actions. */
131 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
132 bool may_set_up_flow; /* True ordinarily; false if the actions must
133 * be reassessed for every packet. */
134 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
136 /* xlate_actions() initializes and uses these members, but the client has no
137 * reason to look at them. */
139 int recurse; /* Recursion level, via xlate_table_action. */
140 int last_pop_priority; /* Offset in 'odp_actions' just past most
141 * recently added ODPAT_SET_PRIORITY. */
144 static void action_xlate_ctx_init(struct action_xlate_ctx *,
145 struct ofproto *, const struct flow *,
146 const struct ofpbuf *);
147 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
148 const union ofp_action *in, size_t n_in);
150 /* An OpenFlow flow. */
152 long long int used; /* Time last used; time created if not used. */
153 long long int created; /* Creation time. */
157 * - Do include packets and bytes from facets that have been deleted or
158 * whose own statistics have been folded into the rule.
160 * - Do include packets and bytes sent "by hand" that were accounted to
161 * the rule without any facet being involved (this is a rare corner
162 * case in rule_execute()).
164 * - Do not include packet or bytes that can be obtained from any facet's
165 * packet_count or byte_count member or that can be obtained from the
166 * datapath by, e.g., dpif_flow_get() for any facet.
168 uint64_t packet_count; /* Number of packets received. */
169 uint64_t byte_count; /* Number of bytes received. */
171 ovs_be64 flow_cookie; /* Controller-issued identifier. */
173 struct cls_rule cr; /* In owning ofproto's classifier. */
174 uint16_t idle_timeout; /* In seconds from time of last use. */
175 uint16_t hard_timeout; /* In seconds from time of creation. */
176 bool send_flow_removed; /* Send a flow removed message? */
177 int n_actions; /* Number of elements in actions[]. */
178 union ofp_action *actions; /* OpenFlow actions. */
179 struct list facets; /* List of "struct facet"s. */
182 static struct rule *rule_from_cls_rule(const struct cls_rule *);
183 static bool rule_is_hidden(const struct rule *);
185 static struct rule *rule_create(const struct cls_rule *,
186 const union ofp_action *, size_t n_actions,
187 uint16_t idle_timeout, uint16_t hard_timeout,
188 ovs_be64 flow_cookie, bool send_flow_removed);
189 static void rule_destroy(struct ofproto *, struct rule *);
190 static void rule_free(struct rule *);
192 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
193 static void rule_insert(struct ofproto *, struct rule *);
194 static void rule_remove(struct ofproto *, struct rule *);
196 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
198 /* An exact-match instantiation of an OpenFlow flow. */
200 long long int used; /* Time last used; time created if not used. */
204 * - Do include packets and bytes sent "by hand", e.g. with
207 * - Do include packets and bytes that were obtained from the datapath
208 * when a flow was deleted (e.g. dpif_flow_del()) or when its
209 * statistics were reset (e.g. dpif_flow_put() with
210 * DPIF_FP_ZERO_STATS).
212 * - Do not include any packets or bytes that can currently be obtained
213 * from the datapath by, e.g., dpif_flow_get().
215 uint64_t packet_count; /* Number of packets received. */
216 uint64_t byte_count; /* Number of bytes received. */
218 /* Number of bytes passed to account_cb. This may include bytes that can
219 * currently obtained from the datapath (thus, it can be greater than
221 uint64_t accounted_bytes;
223 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
224 struct list list_node; /* In owning rule's 'facets' list. */
225 struct rule *rule; /* Owning rule. */
226 struct flow flow; /* Exact-match flow. */
227 bool installed; /* Installed in datapath? */
228 bool may_install; /* True ordinarily; false if actions must
229 * be reassessed for every packet. */
230 size_t actions_len; /* Number of bytes in actions[]. */
231 struct nlattr *actions; /* Datapath actions. */
232 tag_type tags; /* Tags (set only by hooks). */
233 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
236 static struct facet *facet_create(struct ofproto *, struct rule *,
238 const struct ofpbuf *packet);
239 static void facet_remove(struct ofproto *, struct facet *);
240 static void facet_free(struct facet *);
242 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
243 static bool facet_revalidate(struct ofproto *, struct facet *);
245 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
246 static void facet_uninstall(struct ofproto *, struct facet *);
247 static void facet_flush_stats(struct ofproto *, struct facet *);
249 static void facet_make_actions(struct ofproto *, struct facet *,
250 const struct ofpbuf *packet);
251 static void facet_update_stats(struct ofproto *, struct facet *,
252 const struct dpif_flow_stats *);
254 /* ofproto supports two kinds of OpenFlow connections:
256 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
257 * maintains persistent connections to these controllers and by default
258 * sends them asynchronous messages such as packet-ins.
260 * - "Service" connections, e.g. from ovs-ofctl. When these connections
261 * drop, it is the other side's responsibility to reconnect them if
262 * necessary. ofproto does not send them asynchronous messages by default.
264 * Currently, active (tcp, ssl, unix) connections are always "primary"
265 * connections and passive (ptcp, pssl, punix) connections are always "service"
266 * connections. There is no inherent reason for this, but it reflects the
270 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
271 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
274 /* A listener for incoming OpenFlow "service" connections. */
276 struct hmap_node node; /* In struct ofproto's "services" hmap. */
277 struct pvconn *pvconn; /* OpenFlow connection listener. */
279 /* These are not used by ofservice directly. They are settings for
280 * accepted "struct ofconn"s from the pvconn. */
281 int probe_interval; /* Max idle time before probing, in seconds. */
282 int rate_limit; /* Max packet-in rate in packets per second. */
283 int burst_limit; /* Limit on accumulating packet credits. */
286 static struct ofservice *ofservice_lookup(struct ofproto *,
288 static int ofservice_create(struct ofproto *,
289 const struct ofproto_controller *);
290 static void ofservice_reconfigure(struct ofservice *,
291 const struct ofproto_controller *);
292 static void ofservice_destroy(struct ofproto *, struct ofservice *);
294 /* An OpenFlow connection. */
296 struct ofproto *ofproto; /* The ofproto that owns this connection. */
297 struct list node; /* In struct ofproto's "all_conns" list. */
298 struct rconn *rconn; /* OpenFlow connection. */
299 enum ofconn_type type; /* Type. */
300 enum nx_flow_format flow_format; /* Currently selected flow format. */
302 /* OFPT_PACKET_IN related data. */
303 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
304 #define N_SCHEDULERS 2
305 struct pinsched *schedulers[N_SCHEDULERS];
306 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
307 int miss_send_len; /* Bytes to send of buffered packets. */
309 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
310 * requests, and the maximum number before we stop reading OpenFlow
312 #define OFCONN_REPLY_MAX 100
313 struct rconn_packet_counter *reply_counter;
315 /* type == OFCONN_PRIMARY only. */
316 enum nx_role role; /* Role. */
317 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
318 struct discovery *discovery; /* Controller discovery object, if enabled. */
319 struct status_category *ss; /* Switch status category. */
320 enum ofproto_band band; /* In-band or out-of-band? */
324 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
326 static void ofconn_destroy(struct ofconn *);
327 static void ofconn_run(struct ofconn *);
328 static void ofconn_wait(struct ofconn *);
329 static bool ofconn_receives_async_msgs(const struct ofconn *);
330 static char *ofconn_make_name(const struct ofproto *, const char *target);
331 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
333 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
334 struct rconn_packet_counter *counter);
336 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
337 const struct flow *, bool clone);
338 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
342 uint64_t datapath_id; /* Datapath ID. */
343 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
344 char *mfr_desc; /* Manufacturer. */
345 char *hw_desc; /* Hardware. */
346 char *sw_desc; /* Software version. */
347 char *serial_desc; /* Serial number. */
348 char *dp_desc; /* Datapath description. */
352 struct netdev_monitor *netdev_monitor;
353 struct hmap ports; /* Contains "struct ofport"s. */
354 struct shash port_by_name;
358 struct switch_status *switch_status;
359 struct fail_open *fail_open;
360 struct netflow *netflow;
361 struct ofproto_sflow *sflow;
363 /* In-band control. */
364 struct in_band *in_band;
365 long long int next_in_band_update;
366 struct sockaddr_in *extra_in_band_remotes;
367 size_t n_extra_remotes;
371 struct classifier cls;
372 long long int next_expiration;
376 bool need_revalidate;
377 struct tag_set revalidate_set;
379 /* OpenFlow connections. */
380 struct hmap controllers; /* Controller "struct ofconn"s. */
381 struct list all_conns; /* Contains "struct ofconn"s. */
382 enum ofproto_fail_mode fail_mode;
384 /* OpenFlow listeners. */
385 struct hmap services; /* Contains "struct ofservice"s. */
386 struct pvconn **snoops;
389 /* Hooks for ovs-vswitchd. */
390 const struct ofhooks *ofhooks;
393 /* Used by default ofhooks. */
394 struct mac_learning *ml;
397 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
398 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
400 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
402 static const struct ofhooks default_ofhooks;
404 static uint64_t pick_datapath_id(const struct ofproto *);
405 static uint64_t pick_fallback_dpid(void);
407 static int ofproto_expire(struct ofproto *);
409 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
411 static void handle_openflow(struct ofconn *, struct ofpbuf *);
413 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
414 static void update_port(struct ofproto *, const char *devname);
415 static int init_ports(struct ofproto *);
416 static void reinit_ports(struct ofproto *);
418 static void ofproto_unixctl_init(void);
421 ofproto_create(const char *datapath, const char *datapath_type,
422 const struct ofhooks *ofhooks, void *aux,
423 struct ofproto **ofprotop)
431 ofproto_unixctl_init();
433 /* Connect to datapath and start listening for messages. */
434 error = dpif_open(datapath, datapath_type, &dpif);
436 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
439 error = dpif_recv_set_mask(dpif,
440 ((1u << DPIF_UC_MISS) |
441 (1u << DPIF_UC_ACTION) |
442 (1u << DPIF_UC_SAMPLE)));
444 VLOG_ERR("failed to listen on datapath %s: %s",
445 datapath, strerror(error));
449 dpif_flow_flush(dpif);
450 dpif_recv_purge(dpif);
452 /* Initialize settings. */
453 p = xzalloc(sizeof *p);
454 p->fallback_dpid = pick_fallback_dpid();
455 p->datapath_id = p->fallback_dpid;
456 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
457 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
458 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
459 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
460 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
462 /* Initialize datapath. */
464 p->netdev_monitor = netdev_monitor_create();
465 hmap_init(&p->ports);
466 shash_init(&p->port_by_name);
467 p->max_ports = dpif_get_max_ports(dpif);
469 /* Initialize submodules. */
470 p->switch_status = switch_status_create(p);
475 /* Initialize in-band control. */
477 p->in_band_queue = -1;
479 /* Initialize flow table. */
480 classifier_init(&p->cls);
481 p->next_expiration = time_msec() + 1000;
483 /* Initialize facet table. */
484 hmap_init(&p->facets);
485 p->need_revalidate = false;
486 tag_set_init(&p->revalidate_set);
488 /* Initialize OpenFlow connections. */
489 list_init(&p->all_conns);
490 hmap_init(&p->controllers);
491 hmap_init(&p->services);
495 /* Initialize hooks. */
497 p->ofhooks = ofhooks;
501 p->ofhooks = &default_ofhooks;
503 p->ml = mac_learning_create();
506 /* Pick final datapath ID. */
507 p->datapath_id = pick_datapath_id(p);
508 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
510 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
517 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
519 uint64_t old_dpid = p->datapath_id;
520 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
521 if (p->datapath_id != old_dpid) {
522 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
524 /* Force all active connections to reconnect, since there is no way to
525 * notify a controller that the datapath ID has changed. */
526 ofproto_reconnect_controllers(p);
531 is_discovery_controller(const struct ofproto_controller *c)
533 return !strcmp(c->target, "discover");
537 is_in_band_controller(const struct ofproto_controller *c)
539 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
542 /* Creates a new controller in 'ofproto'. Some of the settings are initially
543 * drawn from 'c', but update_controller() needs to be called later to finish
544 * the new ofconn's configuration. */
546 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
548 struct discovery *discovery;
549 struct ofconn *ofconn;
551 if (is_discovery_controller(c)) {
552 int error = discovery_create(c->accept_re, c->update_resolv_conf,
553 ofproto->dpif, ofproto->switch_status,
562 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
563 ofconn->pktbuf = pktbuf_create();
564 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
566 ofconn->discovery = discovery;
568 char *name = ofconn_make_name(ofproto, c->target);
569 rconn_connect(ofconn->rconn, c->target, name);
572 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
573 hash_string(c->target, 0));
576 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
577 * target or turn discovery on or off (these are done by creating new ofconns
578 * and deleting old ones), but it can update the rest of an ofconn's
581 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
585 ofconn->band = (is_in_band_controller(c)
586 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
588 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
590 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
591 rconn_set_probe_interval(ofconn->rconn, probe_interval);
593 if (ofconn->discovery) {
594 discovery_set_update_resolv_conf(ofconn->discovery,
595 c->update_resolv_conf);
596 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
599 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
603 ofconn_get_target(const struct ofconn *ofconn)
605 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
608 static struct ofconn *
609 find_controller_by_target(struct ofproto *ofproto, const char *target)
611 struct ofconn *ofconn;
613 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
614 hash_string(target, 0), &ofproto->controllers) {
615 if (!strcmp(ofconn_get_target(ofconn), target)) {
623 update_in_band_remotes(struct ofproto *ofproto)
625 const struct ofconn *ofconn;
626 struct sockaddr_in *addrs;
627 size_t max_addrs, n_addrs;
631 /* Allocate enough memory for as many remotes as we could possibly have. */
632 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
633 addrs = xmalloc(max_addrs * sizeof *addrs);
636 /* Add all the remotes. */
638 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
639 struct sockaddr_in *sin = &addrs[n_addrs];
641 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
645 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
646 if (sin->sin_addr.s_addr) {
647 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
650 if (ofconn->discovery) {
654 for (i = 0; i < ofproto->n_extra_remotes; i++) {
655 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
658 /* Create or update or destroy in-band.
660 * Ordinarily we only enable in-band if there's at least one remote
661 * address, but discovery needs the in-band rules for DHCP to be installed
662 * even before we know any remote addresses. */
663 if (n_addrs || discovery) {
664 if (!ofproto->in_band) {
665 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
668 if (ofproto->in_band) {
669 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
671 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
672 ofproto->next_in_band_update = time_msec() + 1000;
674 in_band_destroy(ofproto->in_band);
675 ofproto->in_band = NULL;
683 update_fail_open(struct ofproto *p)
685 struct ofconn *ofconn;
687 if (!hmap_is_empty(&p->controllers)
688 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
689 struct rconn **rconns;
693 p->fail_open = fail_open_create(p, p->switch_status);
697 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
698 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
699 rconns[n++] = ofconn->rconn;
702 fail_open_set_controllers(p->fail_open, rconns, n);
703 /* p->fail_open takes ownership of 'rconns'. */
705 fail_open_destroy(p->fail_open);
711 ofproto_set_controllers(struct ofproto *p,
712 const struct ofproto_controller *controllers,
713 size_t n_controllers)
715 struct shash new_controllers;
716 struct ofconn *ofconn, *next_ofconn;
717 struct ofservice *ofservice, *next_ofservice;
721 /* Create newly configured controllers and services.
722 * Create a name to ofproto_controller mapping in 'new_controllers'. */
723 shash_init(&new_controllers);
724 for (i = 0; i < n_controllers; i++) {
725 const struct ofproto_controller *c = &controllers[i];
727 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
728 if (!find_controller_by_target(p, c->target)) {
729 add_controller(p, c);
731 } else if (!pvconn_verify_name(c->target)) {
732 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
736 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
737 dpif_name(p->dpif), c->target);
741 shash_add_once(&new_controllers, c->target, &controllers[i]);
744 /* Delete controllers that are no longer configured.
745 * Update configuration of all now-existing controllers. */
747 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
748 struct ofproto_controller *c;
750 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
752 ofconn_destroy(ofconn);
754 update_controller(ofconn, c);
761 /* Delete services that are no longer configured.
762 * Update configuration of all now-existing services. */
763 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
764 struct ofproto_controller *c;
766 c = shash_find_data(&new_controllers,
767 pvconn_get_name(ofservice->pvconn));
769 ofservice_destroy(p, ofservice);
771 ofservice_reconfigure(ofservice, c);
775 shash_destroy(&new_controllers);
777 update_in_band_remotes(p);
780 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
781 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
782 struct ofconn, hmap_node);
783 ofconn->ss = switch_status_register(p->switch_status, "remote",
784 rconn_status_cb, ofconn->rconn);
789 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
791 p->fail_mode = fail_mode;
795 /* Drops the connections between 'ofproto' and all of its controllers, forcing
796 * them to reconnect. */
798 ofproto_reconnect_controllers(struct ofproto *ofproto)
800 struct ofconn *ofconn;
802 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
803 rconn_reconnect(ofconn->rconn);
808 any_extras_changed(const struct ofproto *ofproto,
809 const struct sockaddr_in *extras, size_t n)
813 if (n != ofproto->n_extra_remotes) {
817 for (i = 0; i < n; i++) {
818 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
819 const struct sockaddr_in *new = &extras[i];
821 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
822 old->sin_port != new->sin_port) {
830 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
831 * in-band control should guarantee access, in the same way that in-band
832 * control guarantees access to OpenFlow controllers. */
834 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
835 const struct sockaddr_in *extras, size_t n)
837 if (!any_extras_changed(ofproto, extras, n)) {
841 free(ofproto->extra_in_band_remotes);
842 ofproto->n_extra_remotes = n;
843 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
845 update_in_band_remotes(ofproto);
848 /* Sets the OpenFlow queue used by flows set up by in-band control on
849 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
850 * flows will use the default queue. */
852 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
854 if (queue_id != ofproto->in_band_queue) {
855 ofproto->in_band_queue = queue_id;
856 update_in_band_remotes(ofproto);
861 ofproto_set_desc(struct ofproto *p,
862 const char *mfr_desc, const char *hw_desc,
863 const char *sw_desc, const char *serial_desc,
866 struct ofp_desc_stats *ods;
869 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
870 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
871 sizeof ods->mfr_desc);
874 p->mfr_desc = xstrdup(mfr_desc);
877 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
878 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
879 sizeof ods->hw_desc);
882 p->hw_desc = xstrdup(hw_desc);
885 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
886 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
887 sizeof ods->sw_desc);
890 p->sw_desc = xstrdup(sw_desc);
893 if (strlen(serial_desc) >= sizeof ods->serial_num) {
894 VLOG_WARN("truncating serial_desc, must be less than %zu "
896 sizeof ods->serial_num);
898 free(p->serial_desc);
899 p->serial_desc = xstrdup(serial_desc);
902 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
903 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
904 sizeof ods->dp_desc);
907 p->dp_desc = xstrdup(dp_desc);
912 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
913 const struct svec *svec)
915 struct pvconn **pvconns = *pvconnsp;
916 size_t n_pvconns = *n_pvconnsp;
920 for (i = 0; i < n_pvconns; i++) {
921 pvconn_close(pvconns[i]);
925 pvconns = xmalloc(svec->n * sizeof *pvconns);
927 for (i = 0; i < svec->n; i++) {
928 const char *name = svec->names[i];
929 struct pvconn *pvconn;
932 error = pvconn_open(name, &pvconn);
934 pvconns[n_pvconns++] = pvconn;
936 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
944 *n_pvconnsp = n_pvconns;
950 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
952 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
956 ofproto_set_netflow(struct ofproto *ofproto,
957 const struct netflow_options *nf_options)
959 if (nf_options && nf_options->collectors.n) {
960 if (!ofproto->netflow) {
961 ofproto->netflow = netflow_create();
963 return netflow_set_options(ofproto->netflow, nf_options);
965 netflow_destroy(ofproto->netflow);
966 ofproto->netflow = NULL;
972 ofproto_set_sflow(struct ofproto *ofproto,
973 const struct ofproto_sflow_options *oso)
975 struct ofproto_sflow *os = ofproto->sflow;
978 struct ofport *ofport;
980 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
981 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
982 ofproto_sflow_add_port(os, ofport->odp_port,
983 netdev_get_name(ofport->netdev));
986 ofproto_sflow_set_options(os, oso);
988 ofproto_sflow_destroy(os);
989 ofproto->sflow = NULL;
994 ofproto_get_datapath_id(const struct ofproto *ofproto)
996 return ofproto->datapath_id;
1000 ofproto_has_primary_controller(const struct ofproto *ofproto)
1002 return !hmap_is_empty(&ofproto->controllers);
1005 enum ofproto_fail_mode
1006 ofproto_get_fail_mode(const struct ofproto *p)
1008 return p->fail_mode;
1012 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1016 for (i = 0; i < ofproto->n_snoops; i++) {
1017 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1022 ofproto_destroy(struct ofproto *p)
1024 struct ofservice *ofservice, *next_ofservice;
1025 struct ofconn *ofconn, *next_ofconn;
1026 struct ofport *ofport, *next_ofport;
1033 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1035 /* Destroy fail-open and in-band early, since they touch the classifier. */
1036 fail_open_destroy(p->fail_open);
1037 p->fail_open = NULL;
1039 in_band_destroy(p->in_band);
1041 free(p->extra_in_band_remotes);
1043 ofproto_flush_flows(p);
1044 classifier_destroy(&p->cls);
1045 hmap_destroy(&p->facets);
1047 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1048 ofconn_destroy(ofconn);
1050 hmap_destroy(&p->controllers);
1052 dpif_close(p->dpif);
1053 netdev_monitor_destroy(p->netdev_monitor);
1054 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1055 hmap_remove(&p->ports, &ofport->hmap_node);
1056 ofport_free(ofport);
1058 shash_destroy(&p->port_by_name);
1060 switch_status_destroy(p->switch_status);
1061 netflow_destroy(p->netflow);
1062 ofproto_sflow_destroy(p->sflow);
1064 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1065 ofservice_destroy(p, ofservice);
1067 hmap_destroy(&p->services);
1069 for (i = 0; i < p->n_snoops; i++) {
1070 pvconn_close(p->snoops[i]);
1074 mac_learning_destroy(p->ml);
1079 free(p->serial_desc);
1082 hmap_destroy(&p->ports);
1088 ofproto_run(struct ofproto *p)
1090 int error = ofproto_run1(p);
1092 error = ofproto_run2(p, false);
1098 process_port_change(struct ofproto *ofproto, int error, char *devname)
1100 if (error == ENOBUFS) {
1101 reinit_ports(ofproto);
1102 } else if (!error) {
1103 update_port(ofproto, devname);
1108 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1109 * means that 'ofconn' is more interesting for monitoring than a lower return
1112 snoop_preference(const struct ofconn *ofconn)
1114 switch (ofconn->role) {
1115 case NX_ROLE_MASTER:
1122 /* Shouldn't happen. */
1127 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1128 * Connects this vconn to a controller. */
1130 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1132 struct ofconn *ofconn, *best;
1134 /* Pick a controller for monitoring. */
1136 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1137 if (ofconn->type == OFCONN_PRIMARY
1138 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1144 rconn_add_monitor(best->rconn, vconn);
1146 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1152 ofproto_run1(struct ofproto *p)
1154 struct ofconn *ofconn, *next_ofconn;
1155 struct ofservice *ofservice;
1160 if (shash_is_empty(&p->port_by_name)) {
1164 for (i = 0; i < 50; i++) {
1165 struct dpif_upcall packet;
1167 error = dpif_recv(p->dpif, &packet);
1169 if (error == ENODEV) {
1170 /* Someone destroyed the datapath behind our back. The caller
1171 * better destroy us and give up, because we're just going to
1172 * spin from here on out. */
1173 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1174 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1175 dpif_name(p->dpif));
1181 handle_upcall(p, &packet);
1184 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1185 process_port_change(p, error, devname);
1187 while ((error = netdev_monitor_poll(p->netdev_monitor,
1188 &devname)) != EAGAIN) {
1189 process_port_change(p, error, devname);
1193 if (time_msec() >= p->next_in_band_update) {
1194 update_in_band_remotes(p);
1196 in_band_run(p->in_band);
1199 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1203 /* Fail-open maintenance. Do this after processing the ofconns since
1204 * fail-open checks the status of the controller rconn. */
1206 fail_open_run(p->fail_open);
1209 HMAP_FOR_EACH (ofservice, node, &p->services) {
1210 struct vconn *vconn;
1213 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1215 struct rconn *rconn;
1218 rconn = rconn_create(ofservice->probe_interval, 0);
1219 name = ofconn_make_name(p, vconn_get_name(vconn));
1220 rconn_connect_unreliably(rconn, vconn, name);
1223 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1224 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1225 ofservice->burst_limit);
1226 } else if (retval != EAGAIN) {
1227 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1231 for (i = 0; i < p->n_snoops; i++) {
1232 struct vconn *vconn;
1235 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1237 add_snooper(p, vconn);
1238 } else if (retval != EAGAIN) {
1239 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1243 if (time_msec() >= p->next_expiration) {
1244 int delay = ofproto_expire(p);
1245 p->next_expiration = time_msec() + delay;
1246 COVERAGE_INC(ofproto_expiration);
1250 netflow_run(p->netflow);
1253 ofproto_sflow_run(p->sflow);
1260 ofproto_run2(struct ofproto *p, bool revalidate_all)
1262 /* Figure out what we need to revalidate now, if anything. */
1263 struct tag_set revalidate_set = p->revalidate_set;
1264 if (p->need_revalidate) {
1265 revalidate_all = true;
1268 /* Clear the revalidation flags. */
1269 tag_set_init(&p->revalidate_set);
1270 p->need_revalidate = false;
1272 /* Now revalidate if there's anything to do. */
1273 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1274 struct facet *facet, *next;
1276 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1278 || tag_set_intersects(&revalidate_set, facet->tags)) {
1279 facet_revalidate(p, facet);
1288 ofproto_wait(struct ofproto *p)
1290 struct ofservice *ofservice;
1291 struct ofconn *ofconn;
1294 dpif_recv_wait(p->dpif);
1295 dpif_port_poll_wait(p->dpif);
1296 netdev_monitor_poll_wait(p->netdev_monitor);
1297 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1298 ofconn_wait(ofconn);
1301 poll_timer_wait_until(p->next_in_band_update);
1302 in_band_wait(p->in_band);
1305 fail_open_wait(p->fail_open);
1308 ofproto_sflow_wait(p->sflow);
1310 if (!tag_set_is_empty(&p->revalidate_set)) {
1311 poll_immediate_wake();
1313 if (p->need_revalidate) {
1314 /* Shouldn't happen, but if it does just go around again. */
1315 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1316 poll_immediate_wake();
1317 } else if (p->next_expiration != LLONG_MAX) {
1318 poll_timer_wait_until(p->next_expiration);
1320 HMAP_FOR_EACH (ofservice, node, &p->services) {
1321 pvconn_wait(ofservice->pvconn);
1323 for (i = 0; i < p->n_snoops; i++) {
1324 pvconn_wait(p->snoops[i]);
1329 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1331 tag_set_add(&ofproto->revalidate_set, tag);
1335 ofproto_get_revalidate_set(struct ofproto *ofproto)
1337 return &ofproto->revalidate_set;
1341 ofproto_is_alive(const struct ofproto *p)
1343 return !hmap_is_empty(&p->controllers);
1347 ofproto_get_ofproto_controller_info(const struct ofproto * ofproto,
1350 const struct ofconn *ofconn;
1354 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1355 const struct rconn *rconn = ofconn->rconn;
1356 const int last_error = rconn_get_last_error(rconn);
1357 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1359 shash_add(info, rconn_get_target(rconn), cinfo);
1361 cinfo->is_connected = rconn_is_connected(rconn);
1362 cinfo->role = ofconn->role;
1366 if (last_error == EOF) {
1367 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1368 cinfo->pairs.values[cinfo->pairs.n++] = xstrdup("End of file");
1369 } else if (last_error > 0) {
1370 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1371 cinfo->pairs.values[cinfo->pairs.n++] =
1372 xstrdup(strerror(last_error));
1375 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1376 cinfo->pairs.values[cinfo->pairs.n++] =
1377 xstrdup(rconn_get_state(rconn));
1379 cinfo->pairs.keys[cinfo->pairs.n] = "time_in_state";
1380 cinfo->pairs.values[cinfo->pairs.n++] =
1381 xasprintf("%u", rconn_get_state_elapsed(rconn));
1386 ofproto_free_ofproto_controller_info(struct shash *info)
1388 struct shash_node *node;
1390 SHASH_FOR_EACH (node, info) {
1391 struct ofproto_controller_info *cinfo = node->data;
1392 while (cinfo->pairs.n) {
1393 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1397 shash_destroy(info);
1400 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1402 * This is almost the same as calling dpif_port_del() directly on the
1403 * datapath, but it also makes 'ofproto' close its open netdev for the port
1404 * (if any). This makes it possible to create a new netdev of a different
1405 * type under the same name, which otherwise the netdev library would refuse
1406 * to do because of the conflict. (The netdev would eventually get closed on
1407 * the next trip through ofproto_run(), but this interface is more direct.)
1409 * Returns 0 if successful, otherwise a positive errno. */
1411 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1413 struct ofport *ofport = get_port(ofproto, odp_port);
1414 const char *name = ofport ? ofport->opp.name : "<unknown>";
1417 error = dpif_port_del(ofproto->dpif, odp_port);
1419 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1420 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1421 } else if (ofport) {
1422 /* 'name' is ofport->opp.name and update_port() is going to destroy
1423 * 'ofport'. Just in case update_port() refers to 'name' after it
1424 * destroys 'ofport', make a copy of it around the update_port()
1426 char *devname = xstrdup(name);
1427 update_port(ofproto, devname);
1433 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1434 * true if 'odp_port' exists and should be included, false otherwise. */
1436 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1438 struct ofport *ofport = get_port(ofproto, odp_port);
1439 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1443 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1444 const union ofp_action *actions, size_t n_actions,
1445 const struct ofpbuf *packet)
1447 struct action_xlate_ctx ctx;
1448 struct ofpbuf *odp_actions;
1450 action_xlate_ctx_init(&ctx, p, flow, packet);
1451 odp_actions = xlate_actions(&ctx, actions, n_actions);
1453 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1455 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet);
1457 ofpbuf_delete(odp_actions);
1462 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1463 * performs the 'n_actions' actions in 'actions'. The new flow will not
1466 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1467 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1468 * controllers; otherwise, it will be hidden.
1470 * The caller retains ownership of 'cls_rule' and 'actions'. */
1472 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1473 const union ofp_action *actions, size_t n_actions)
1476 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1477 rule_insert(p, rule);
1481 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1485 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1488 rule_remove(ofproto, rule);
1493 ofproto_flush_flows(struct ofproto *ofproto)
1495 struct facet *facet, *next_facet;
1496 struct rule *rule, *next_rule;
1497 struct cls_cursor cursor;
1499 COVERAGE_INC(ofproto_flush);
1501 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1502 /* Mark the facet as not installed so that facet_remove() doesn't
1503 * bother trying to uninstall it. There is no point in uninstalling it
1504 * individually since we are about to blow away all the facets with
1505 * dpif_flow_flush(). */
1506 facet->installed = false;
1507 facet_remove(ofproto, facet);
1510 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1511 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1512 rule_remove(ofproto, rule);
1515 dpif_flow_flush(ofproto->dpif);
1516 if (ofproto->in_band) {
1517 in_band_flushed(ofproto->in_band);
1519 if (ofproto->fail_open) {
1520 fail_open_flushed(ofproto->fail_open);
1525 reinit_ports(struct ofproto *p)
1527 struct dpif_port_dump dump;
1528 struct shash_node *node;
1529 struct shash devnames;
1530 struct ofport *ofport;
1531 struct dpif_port dpif_port;
1533 COVERAGE_INC(ofproto_reinit_ports);
1535 shash_init(&devnames);
1536 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1537 shash_add_once (&devnames, ofport->opp.name, NULL);
1539 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1540 shash_add_once (&devnames, dpif_port.name, NULL);
1543 SHASH_FOR_EACH (node, &devnames) {
1544 update_port(p, node->name);
1546 shash_destroy(&devnames);
1549 static struct ofport *
1550 make_ofport(const struct dpif_port *dpif_port)
1552 struct netdev_options netdev_options;
1553 enum netdev_flags flags;
1554 struct ofport *ofport;
1555 struct netdev *netdev;
1558 memset(&netdev_options, 0, sizeof netdev_options);
1559 netdev_options.name = dpif_port->name;
1560 netdev_options.type = dpif_port->type;
1561 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1563 error = netdev_open(&netdev_options, &netdev);
1565 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1566 "cannot be opened (%s)",
1567 dpif_port->name, dpif_port->port_no,
1568 dpif_port->name, strerror(error));
1572 ofport = xmalloc(sizeof *ofport);
1573 ofport->netdev = netdev;
1574 ofport->odp_port = dpif_port->port_no;
1575 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1576 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1577 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1579 netdev_get_flags(netdev, &flags);
1580 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1582 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1584 netdev_get_features(netdev,
1585 &ofport->opp.curr, &ofport->opp.advertised,
1586 &ofport->opp.supported, &ofport->opp.peer);
1591 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1593 if (get_port(p, dpif_port->port_no)) {
1594 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1595 dpif_port->port_no);
1597 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1598 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1607 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1609 const struct ofp_phy_port *a = &a_->opp;
1610 const struct ofp_phy_port *b = &b_->opp;
1612 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1613 return (a->port_no == b->port_no
1614 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1615 && !strcmp(a->name, b->name)
1616 && a->state == b->state
1617 && a->config == b->config
1618 && a->curr == b->curr
1619 && a->advertised == b->advertised
1620 && a->supported == b->supported
1621 && a->peer == b->peer);
1625 send_port_status(struct ofproto *p, const struct ofport *ofport,
1628 /* XXX Should limit the number of queued port status change messages. */
1629 struct ofconn *ofconn;
1630 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1631 struct ofp_port_status *ops;
1634 /* Primary controllers, even slaves, should always get port status
1635 updates. Otherwise obey ofconn_receives_async_msgs(). */
1636 if (ofconn->type != OFCONN_PRIMARY
1637 && !ofconn_receives_async_msgs(ofconn)) {
1641 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1642 ops->reason = reason;
1643 ops->desc = ofport->opp;
1644 hton_ofp_phy_port(&ops->desc);
1645 queue_tx(b, ofconn, NULL);
1650 ofport_install(struct ofproto *p, struct ofport *ofport)
1652 const char *netdev_name = ofport->opp.name;
1654 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1655 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1656 shash_add(&p->port_by_name, netdev_name, ofport);
1658 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1663 ofport_remove(struct ofproto *p, struct ofport *ofport)
1665 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1666 hmap_remove(&p->ports, &ofport->hmap_node);
1667 shash_delete(&p->port_by_name,
1668 shash_find(&p->port_by_name, ofport->opp.name));
1670 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1675 ofport_free(struct ofport *ofport)
1678 netdev_close(ofport->netdev);
1683 static struct ofport *
1684 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1686 struct ofport *port;
1688 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1689 hash_int(odp_port, 0), &ofproto->ports) {
1690 if (port->odp_port == odp_port) {
1698 update_port(struct ofproto *p, const char *devname)
1700 struct dpif_port dpif_port;
1701 struct ofport *old_ofport;
1702 struct ofport *new_ofport;
1705 COVERAGE_INC(ofproto_update_port);
1707 /* Query the datapath for port information. */
1708 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1710 /* Find the old ofport. */
1711 old_ofport = shash_find_data(&p->port_by_name, devname);
1714 /* There's no port named 'devname' but there might be a port with
1715 * the same port number. This could happen if a port is deleted
1716 * and then a new one added in its place very quickly, or if a port
1717 * is renamed. In the former case we want to send an OFPPR_DELETE
1718 * and an OFPPR_ADD, and in the latter case we want to send a
1719 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1720 * the old port's ifindex against the new port, or perhaps less
1721 * reliably but more portably by comparing the old port's MAC
1722 * against the new port's MAC. However, this code isn't that smart
1723 * and always sends an OFPPR_MODIFY (XXX). */
1724 old_ofport = get_port(p, dpif_port.port_no);
1726 } else if (error != ENOENT && error != ENODEV) {
1727 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1728 "%s", strerror(error));
1732 /* Create a new ofport. */
1733 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1735 /* Eliminate a few pathological cases. */
1736 if (!old_ofport && !new_ofport) {
1738 } else if (old_ofport && new_ofport) {
1739 /* Most of the 'config' bits are OpenFlow soft state, but
1740 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1741 * OpenFlow bits from old_ofport. (make_ofport() only sets
1742 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1743 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1745 if (ofport_equal(old_ofport, new_ofport)) {
1746 /* False alarm--no change. */
1747 ofport_free(new_ofport);
1752 /* Now deal with the normal cases. */
1754 ofport_remove(p, old_ofport);
1757 ofport_install(p, new_ofport);
1759 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1760 (!old_ofport ? OFPPR_ADD
1761 : !new_ofport ? OFPPR_DELETE
1763 ofport_free(old_ofport);
1766 dpif_port_destroy(&dpif_port);
1770 init_ports(struct ofproto *p)
1772 struct dpif_port_dump dump;
1773 struct dpif_port dpif_port;
1775 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1776 if (!ofport_conflicts(p, &dpif_port)) {
1777 struct ofport *ofport = make_ofport(&dpif_port);
1779 ofport_install(p, ofport);
1787 static struct ofconn *
1788 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1790 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1791 ofconn->ofproto = p;
1792 list_push_back(&p->all_conns, &ofconn->node);
1793 ofconn->rconn = rconn;
1794 ofconn->type = type;
1795 ofconn->flow_format = NXFF_OPENFLOW10;
1796 ofconn->role = NX_ROLE_OTHER;
1797 ofconn->packet_in_counter = rconn_packet_counter_create ();
1798 ofconn->pktbuf = NULL;
1799 ofconn->miss_send_len = 0;
1800 ofconn->reply_counter = rconn_packet_counter_create ();
1805 ofconn_destroy(struct ofconn *ofconn)
1807 if (ofconn->type == OFCONN_PRIMARY) {
1808 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1810 discovery_destroy(ofconn->discovery);
1812 list_remove(&ofconn->node);
1813 switch_status_unregister(ofconn->ss);
1814 rconn_destroy(ofconn->rconn);
1815 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1816 rconn_packet_counter_destroy(ofconn->reply_counter);
1817 pktbuf_destroy(ofconn->pktbuf);
1822 ofconn_run(struct ofconn *ofconn)
1824 struct ofproto *p = ofconn->ofproto;
1828 if (ofconn->discovery) {
1829 char *controller_name;
1830 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1831 discovery_question_connectivity(ofconn->discovery);
1833 if (discovery_run(ofconn->discovery, &controller_name)) {
1834 if (controller_name) {
1835 char *ofconn_name = ofconn_make_name(p, controller_name);
1836 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1839 rconn_disconnect(ofconn->rconn);
1844 for (i = 0; i < N_SCHEDULERS; i++) {
1845 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1848 rconn_run(ofconn->rconn);
1850 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1851 /* Limit the number of iterations to prevent other tasks from
1853 for (iteration = 0; iteration < 50; iteration++) {
1854 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1859 fail_open_maybe_recover(p->fail_open);
1861 handle_openflow(ofconn, of_msg);
1862 ofpbuf_delete(of_msg);
1866 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1867 ofconn_destroy(ofconn);
1872 ofconn_wait(struct ofconn *ofconn)
1876 if (ofconn->discovery) {
1877 discovery_wait(ofconn->discovery);
1879 for (i = 0; i < N_SCHEDULERS; i++) {
1880 pinsched_wait(ofconn->schedulers[i]);
1882 rconn_run_wait(ofconn->rconn);
1883 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1884 rconn_recv_wait(ofconn->rconn);
1886 COVERAGE_INC(ofproto_ofconn_stuck);
1890 /* Returns true if 'ofconn' should receive asynchronous messages. */
1892 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1894 if (ofconn->type == OFCONN_PRIMARY) {
1895 /* Primary controllers always get asynchronous messages unless they
1896 * have configured themselves as "slaves". */
1897 return ofconn->role != NX_ROLE_SLAVE;
1899 /* Service connections don't get asynchronous messages unless they have
1900 * explicitly asked for them by setting a nonzero miss send length. */
1901 return ofconn->miss_send_len > 0;
1905 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1906 * and 'target', suitable for use in log messages for identifying the
1909 * The name is dynamically allocated. The caller should free it (with free())
1910 * when it is no longer needed. */
1912 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1914 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1918 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1922 for (i = 0; i < N_SCHEDULERS; i++) {
1923 struct pinsched **s = &ofconn->schedulers[i];
1927 *s = pinsched_create(rate, burst,
1928 ofconn->ofproto->switch_status);
1930 pinsched_set_limits(*s, rate, burst);
1933 pinsched_destroy(*s);
1940 ofservice_reconfigure(struct ofservice *ofservice,
1941 const struct ofproto_controller *c)
1943 ofservice->probe_interval = c->probe_interval;
1944 ofservice->rate_limit = c->rate_limit;
1945 ofservice->burst_limit = c->burst_limit;
1948 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1949 * positive errno value. */
1951 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1953 struct ofservice *ofservice;
1954 struct pvconn *pvconn;
1957 error = pvconn_open(c->target, &pvconn);
1962 ofservice = xzalloc(sizeof *ofservice);
1963 hmap_insert(&ofproto->services, &ofservice->node,
1964 hash_string(c->target, 0));
1965 ofservice->pvconn = pvconn;
1967 ofservice_reconfigure(ofservice, c);
1973 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1975 hmap_remove(&ofproto->services, &ofservice->node);
1976 pvconn_close(ofservice->pvconn);
1980 /* Finds and returns the ofservice within 'ofproto' that has the given
1981 * 'target', or a null pointer if none exists. */
1982 static struct ofservice *
1983 ofservice_lookup(struct ofproto *ofproto, const char *target)
1985 struct ofservice *ofservice;
1987 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1988 &ofproto->services) {
1989 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1996 /* Returns true if 'rule' should be hidden from the controller.
1998 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1999 * (e.g. by in-band control) and are intentionally hidden from the
2002 rule_is_hidden(const struct rule *rule)
2004 return rule->cr.priority > UINT16_MAX;
2007 /* Creates and returns a new rule initialized as specified.
2009 * The caller is responsible for inserting the rule into the classifier (with
2010 * rule_insert()). */
2011 static struct rule *
2012 rule_create(const struct cls_rule *cls_rule,
2013 const union ofp_action *actions, size_t n_actions,
2014 uint16_t idle_timeout, uint16_t hard_timeout,
2015 ovs_be64 flow_cookie, bool send_flow_removed)
2017 struct rule *rule = xzalloc(sizeof *rule);
2018 rule->cr = *cls_rule;
2019 rule->idle_timeout = idle_timeout;
2020 rule->hard_timeout = hard_timeout;
2021 rule->flow_cookie = flow_cookie;
2022 rule->used = rule->created = time_msec();
2023 rule->send_flow_removed = send_flow_removed;
2024 list_init(&rule->facets);
2025 if (n_actions > 0) {
2026 rule->n_actions = n_actions;
2027 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2033 static struct rule *
2034 rule_from_cls_rule(const struct cls_rule *cls_rule)
2036 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2040 rule_free(struct rule *rule)
2042 free(rule->actions);
2046 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2047 * destroying any that no longer has a rule (which is probably all of them).
2049 * The caller must have already removed 'rule' from the classifier. */
2051 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2053 struct facet *facet, *next_facet;
2054 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2055 facet_revalidate(ofproto, facet);
2060 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2061 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2064 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2066 const union ofp_action *oa;
2067 struct actions_iterator i;
2069 if (out_port == htons(OFPP_NONE)) {
2072 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2073 oa = actions_next(&i)) {
2074 if (action_outputs_to_port(oa, out_port)) {
2081 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2082 * 'packet', which arrived on 'in_port'.
2084 * Takes ownership of 'packet'. */
2086 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2087 const struct nlattr *odp_actions, size_t actions_len,
2088 struct ofpbuf *packet)
2090 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2091 && odp_actions->nla_type == ODPAT_CONTROLLER) {
2092 /* As an optimization, avoid a round-trip from userspace to kernel to
2093 * userspace. This also avoids possibly filling up kernel packet
2094 * buffers along the way. */
2095 struct dpif_upcall upcall;
2097 upcall.type = DPIF_UC_ACTION;
2098 upcall.packet = packet;
2101 upcall.userdata = nl_attr_get_u64(odp_actions);
2102 upcall.sample_pool = 0;
2103 upcall.actions = NULL;
2104 upcall.actions_len = 0;
2106 send_packet_in(ofproto, &upcall, flow, false);
2112 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2113 ofpbuf_delete(packet);
2118 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2119 * statistics appropriately. 'packet' must have at least sizeof(struct
2120 * ofp_packet_in) bytes of headroom.
2122 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2123 * applying flow_extract() to 'packet' would yield the same flow as
2126 * 'facet' must have accurately composed ODP actions; that is, it must not be
2127 * in need of revalidation.
2129 * Takes ownership of 'packet'. */
2131 facet_execute(struct ofproto *ofproto, struct facet *facet,
2132 struct ofpbuf *packet)
2134 struct dpif_flow_stats stats;
2136 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2138 flow_extract_stats(&facet->flow, packet, &stats);
2139 if (execute_odp_actions(ofproto, &facet->flow,
2140 facet->actions, facet->actions_len, packet)) {
2141 facet_update_stats(ofproto, facet, &stats);
2142 facet->used = time_msec();
2143 netflow_flow_update_time(ofproto->netflow,
2144 &facet->nf_flow, facet->used);
2148 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2149 * statistics (or the statistics for one of its facets) appropriately.
2150 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2152 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2153 * with statistics for 'packet' either way.
2155 * Takes ownership of 'packet'. */
2157 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2158 struct ofpbuf *packet)
2160 struct action_xlate_ctx ctx;
2161 struct ofpbuf *odp_actions;
2162 struct facet *facet;
2166 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2168 flow_extract(packet, 0, in_port, &flow);
2170 /* First look for a related facet. If we find one, account it to that. */
2171 facet = facet_lookup_valid(ofproto, &flow);
2172 if (facet && facet->rule == rule) {
2173 facet_execute(ofproto, facet, packet);
2177 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2178 * create a new facet for it and use that. */
2179 if (rule_lookup(ofproto, &flow) == rule) {
2180 facet = facet_create(ofproto, rule, &flow, packet);
2181 facet_execute(ofproto, facet, packet);
2182 facet_install(ofproto, facet, true);
2186 /* We can't account anything to a facet. If we were to try, then that
2187 * facet would have a non-matching rule, busting our invariants. */
2188 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2189 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2190 size = packet->size;
2191 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2192 odp_actions->size, packet)) {
2193 rule->used = time_msec();
2194 rule->packet_count++;
2195 rule->byte_count += size;
2197 ofpbuf_delete(odp_actions);
2200 /* Inserts 'rule' into 'p''s flow table. */
2202 rule_insert(struct ofproto *p, struct rule *rule)
2204 struct rule *displaced_rule;
2206 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2207 if (displaced_rule) {
2208 rule_destroy(p, displaced_rule);
2210 p->need_revalidate = true;
2213 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2214 * 'flow' and an example 'packet' within that flow.
2216 * The caller must already have determined that no facet with an identical
2217 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2218 * 'ofproto''s classifier table. */
2219 static struct facet *
2220 facet_create(struct ofproto *ofproto, struct rule *rule,
2221 const struct flow *flow, const struct ofpbuf *packet)
2223 struct facet *facet;
2225 facet = xzalloc(sizeof *facet);
2226 facet->used = time_msec();
2227 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2228 list_push_back(&rule->facets, &facet->list_node);
2230 facet->flow = *flow;
2231 netflow_flow_init(&facet->nf_flow);
2232 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2234 facet_make_actions(ofproto, facet, packet);
2240 facet_free(struct facet *facet)
2242 free(facet->actions);
2246 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2248 * - Removes 'rule' from the classifier.
2250 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2251 * destroys them), via rule_destroy().
2254 rule_remove(struct ofproto *ofproto, struct rule *rule)
2256 COVERAGE_INC(ofproto_del_rule);
2257 ofproto->need_revalidate = true;
2258 classifier_remove(&ofproto->cls, &rule->cr);
2259 rule_destroy(ofproto, rule);
2262 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2264 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2265 * rule's statistics, via facet_uninstall().
2267 * - Removes 'facet' from its rule and from ofproto->facets.
2270 facet_remove(struct ofproto *ofproto, struct facet *facet)
2272 facet_uninstall(ofproto, facet);
2273 facet_flush_stats(ofproto, facet);
2274 hmap_remove(&ofproto->facets, &facet->hmap_node);
2275 list_remove(&facet->list_node);
2279 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2281 facet_make_actions(struct ofproto *p, struct facet *facet,
2282 const struct ofpbuf *packet)
2284 const struct rule *rule = facet->rule;
2285 struct ofpbuf *odp_actions;
2286 struct action_xlate_ctx ctx;
2288 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2289 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2290 facet->tags = ctx.tags;
2291 facet->may_install = ctx.may_set_up_flow;
2292 facet->nf_flow.output_iface = ctx.nf_output_iface;
2294 if (facet->actions_len != odp_actions->size
2295 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2296 free(facet->actions);
2297 facet->actions_len = odp_actions->size;
2298 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2301 ofpbuf_delete(odp_actions);
2305 facet_put__(struct ofproto *ofproto, struct facet *facet,
2306 const struct nlattr *actions, size_t actions_len,
2307 struct dpif_flow_stats *stats)
2309 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2310 enum dpif_flow_put_flags flags;
2313 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2315 flags |= DPIF_FP_ZERO_STATS;
2318 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2319 odp_flow_key_from_flow(&key, &facet->flow);
2320 assert(key.base == keybuf);
2322 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2323 actions, actions_len, stats);
2326 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2327 * 'zero_stats' is true, clears any existing statistics from the datapath for
2330 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2332 struct dpif_flow_stats stats;
2334 if (facet->may_install
2335 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2336 zero_stats ? &stats : NULL)) {
2337 facet->installed = true;
2341 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2342 * to the accounting hook function in the ofhooks structure. */
2344 facet_account(struct ofproto *ofproto,
2345 struct facet *facet, uint64_t extra_bytes)
2347 uint64_t total_bytes = facet->byte_count + extra_bytes;
2349 if (ofproto->ofhooks->account_flow_cb
2350 && total_bytes > facet->accounted_bytes)
2352 ofproto->ofhooks->account_flow_cb(
2353 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2354 total_bytes - facet->accounted_bytes, ofproto->aux);
2355 facet->accounted_bytes = total_bytes;
2359 /* If 'rule' is installed in the datapath, uninstalls it. */
2361 facet_uninstall(struct ofproto *p, struct facet *facet)
2363 if (facet->installed) {
2364 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2365 struct dpif_flow_stats stats;
2368 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2369 odp_flow_key_from_flow(&key, &facet->flow);
2370 assert(key.base == keybuf);
2372 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2373 facet_update_stats(p, facet, &stats);
2375 facet->installed = false;
2379 /* Returns true if the only action for 'facet' is to send to the controller.
2380 * (We don't report NetFlow expiration messages for such facets because they
2381 * are just part of the control logic for the network, not real traffic). */
2383 facet_is_controller_flow(struct facet *facet)
2386 && facet->rule->n_actions == 1
2387 && action_outputs_to_port(&facet->rule->actions[0],
2388 htons(OFPP_CONTROLLER)));
2391 /* Folds all of 'facet''s statistics into its rule. Also updates the
2392 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2394 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2396 facet_account(ofproto, facet, 0);
2398 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2399 struct ofexpired expired;
2400 expired.flow = facet->flow;
2401 expired.packet_count = facet->packet_count;
2402 expired.byte_count = facet->byte_count;
2403 expired.used = facet->used;
2404 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2407 facet->rule->packet_count += facet->packet_count;
2408 facet->rule->byte_count += facet->byte_count;
2410 /* Reset counters to prevent double counting if 'facet' ever gets
2412 facet->packet_count = 0;
2413 facet->byte_count = 0;
2414 facet->accounted_bytes = 0;
2416 netflow_flow_clear(&facet->nf_flow);
2419 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2420 * Returns it if found, otherwise a null pointer.
2422 * The returned facet might need revalidation; use facet_lookup_valid()
2423 * instead if that is important. */
2424 static struct facet *
2425 facet_find(struct ofproto *ofproto, const struct flow *flow)
2427 struct facet *facet;
2429 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2431 if (flow_equal(flow, &facet->flow)) {
2439 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2440 * Returns it if found, otherwise a null pointer.
2442 * The returned facet is guaranteed to be valid. */
2443 static struct facet *
2444 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2446 struct facet *facet = facet_find(ofproto, flow);
2448 /* The facet we found might not be valid, since we could be in need of
2449 * revalidation. If it is not valid, don't return it. */
2451 && ofproto->need_revalidate
2452 && !facet_revalidate(ofproto, facet)) {
2453 COVERAGE_INC(ofproto_invalidated);
2460 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2462 * - If the rule found is different from 'facet''s current rule, moves
2463 * 'facet' to the new rule and recompiles its actions.
2465 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2466 * where it is and recompiles its actions anyway.
2468 * - If there is none, destroys 'facet'.
2470 * Returns true if 'facet' still exists, false if it has been destroyed. */
2472 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2474 struct action_xlate_ctx ctx;
2475 struct ofpbuf *odp_actions;
2476 struct rule *new_rule;
2477 bool actions_changed;
2479 COVERAGE_INC(facet_revalidate);
2481 /* Determine the new rule. */
2482 new_rule = rule_lookup(ofproto, &facet->flow);
2484 /* No new rule, so delete the facet. */
2485 facet_remove(ofproto, facet);
2489 /* Calculate new ODP actions.
2491 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2492 * emit a NetFlow expiration and, if so, we need to have the old state
2493 * around to properly compose it. */
2494 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2495 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2496 actions_changed = (facet->actions_len != odp_actions->size
2497 || memcmp(facet->actions, odp_actions->data,
2498 facet->actions_len));
2500 /* If the ODP actions changed or the installability changed, then we need
2501 * to talk to the datapath. */
2502 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2503 if (ctx.may_set_up_flow) {
2504 struct dpif_flow_stats stats;
2506 facet_put__(ofproto, facet,
2507 odp_actions->data, odp_actions->size, &stats);
2508 facet_update_stats(ofproto, facet, &stats);
2510 facet_uninstall(ofproto, facet);
2513 /* The datapath flow is gone or has zeroed stats, so push stats out of
2514 * 'facet' into 'rule'. */
2515 facet_flush_stats(ofproto, facet);
2518 /* Update 'facet' now that we've taken care of all the old state. */
2519 facet->tags = ctx.tags;
2520 facet->nf_flow.output_iface = ctx.nf_output_iface;
2521 facet->may_install = ctx.may_set_up_flow;
2522 if (actions_changed) {
2523 free(facet->actions);
2524 facet->actions_len = odp_actions->size;
2525 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2527 if (facet->rule != new_rule) {
2528 COVERAGE_INC(facet_changed_rule);
2529 list_remove(&facet->list_node);
2530 list_push_back(&new_rule->facets, &facet->list_node);
2531 facet->rule = new_rule;
2532 facet->used = new_rule->created;
2535 ofpbuf_delete(odp_actions);
2541 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2542 struct rconn_packet_counter *counter)
2544 update_openflow_length(msg);
2545 if (rconn_send(ofconn->rconn, msg, counter)) {
2551 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2554 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2556 COVERAGE_INC(ofproto_error);
2557 queue_tx(buf, ofconn, ofconn->reply_counter);
2562 hton_ofp_phy_port(struct ofp_phy_port *opp)
2564 opp->port_no = htons(opp->port_no);
2565 opp->config = htonl(opp->config);
2566 opp->state = htonl(opp->state);
2567 opp->curr = htonl(opp->curr);
2568 opp->advertised = htonl(opp->advertised);
2569 opp->supported = htonl(opp->supported);
2570 opp->peer = htonl(opp->peer);
2574 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2576 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2581 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2583 struct ofp_switch_features *osf;
2585 struct ofport *port;
2587 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2588 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2589 osf->n_buffers = htonl(pktbuf_capacity());
2591 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2592 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2593 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2594 (1u << OFPAT_SET_VLAN_VID) |
2595 (1u << OFPAT_SET_VLAN_PCP) |
2596 (1u << OFPAT_STRIP_VLAN) |
2597 (1u << OFPAT_SET_DL_SRC) |
2598 (1u << OFPAT_SET_DL_DST) |
2599 (1u << OFPAT_SET_NW_SRC) |
2600 (1u << OFPAT_SET_NW_DST) |
2601 (1u << OFPAT_SET_NW_TOS) |
2602 (1u << OFPAT_SET_TP_SRC) |
2603 (1u << OFPAT_SET_TP_DST) |
2604 (1u << OFPAT_ENQUEUE));
2606 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2607 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2610 queue_tx(buf, ofconn, ofconn->reply_counter);
2615 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2618 struct ofp_switch_config *osc;
2622 /* Figure out flags. */
2623 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2624 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2627 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2628 osc->flags = htons(flags);
2629 osc->miss_send_len = htons(ofconn->miss_send_len);
2630 queue_tx(buf, ofconn, ofconn->reply_counter);
2636 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2638 uint16_t flags = ntohs(osc->flags);
2640 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2641 switch (flags & OFPC_FRAG_MASK) {
2642 case OFPC_FRAG_NORMAL:
2643 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2645 case OFPC_FRAG_DROP:
2646 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2649 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2655 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2660 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2661 * flow translation. */
2662 #define MAX_RESUBMIT_RECURSION 16
2664 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2665 struct action_xlate_ctx *ctx);
2668 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2670 const struct ofport *ofport = get_port(ctx->ofproto, port);
2673 if (ofport->opp.config & OFPPC_NO_FWD) {
2674 /* Forwarding disabled on port. */
2679 * We don't have an ofport record for this port, but it doesn't hurt to
2680 * allow forwarding to it anyhow. Maybe such a port will appear later
2681 * and we're pre-populating the flow table.
2685 nl_msg_put_u32(ctx->odp_actions, ODPAT_OUTPUT, port);
2686 ctx->nf_output_iface = port;
2689 static struct rule *
2690 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2692 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2696 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2698 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2699 uint16_t old_in_port;
2702 /* Look up a flow with 'in_port' as the input port. Then restore the
2703 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2704 * have surprising behavior). */
2705 old_in_port = ctx->flow.in_port;
2706 ctx->flow.in_port = in_port;
2707 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2708 ctx->flow.in_port = old_in_port;
2710 if (ctx->resubmit_hook) {
2711 ctx->resubmit_hook(ctx, rule);
2716 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2720 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2722 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2723 MAX_RESUBMIT_RECURSION);
2728 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2729 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2731 struct ofport *ofport;
2733 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2734 uint16_t odp_port = ofport->odp_port;
2735 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2736 nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, odp_port);
2739 *nf_output_iface = NF_OUT_FLOOD;
2743 xlate_output_action__(struct action_xlate_ctx *ctx,
2744 uint16_t port, uint16_t max_len)
2747 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2749 ctx->nf_output_iface = NF_OUT_DROP;
2753 add_output_action(ctx, ctx->flow.in_port);
2756 xlate_table_action(ctx, ctx->flow.in_port);
2759 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2760 ctx->odp_actions, &ctx->tags,
2761 &ctx->nf_output_iface,
2762 ctx->ofproto->aux)) {
2763 COVERAGE_INC(ofproto_uninstallable);
2764 ctx->may_set_up_flow = false;
2768 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2769 &ctx->nf_output_iface, ctx->odp_actions);
2772 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2773 &ctx->nf_output_iface, ctx->odp_actions);
2775 case OFPP_CONTROLLER:
2776 nl_msg_put_u64(ctx->odp_actions, ODPAT_CONTROLLER, max_len);
2779 add_output_action(ctx, ODPP_LOCAL);
2782 odp_port = ofp_port_to_odp_port(port);
2783 if (odp_port != ctx->flow.in_port) {
2784 add_output_action(ctx, odp_port);
2789 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2790 ctx->nf_output_iface = NF_OUT_FLOOD;
2791 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2792 ctx->nf_output_iface = prev_nf_output_iface;
2793 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2794 ctx->nf_output_iface != NF_OUT_FLOOD) {
2795 ctx->nf_output_iface = NF_OUT_MULTI;
2800 xlate_output_action(struct action_xlate_ctx *ctx,
2801 const struct ofp_action_output *oao)
2803 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2806 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2807 * optimization, because we're going to add another action that sets the
2808 * priority immediately after, or because there are no actions following the
2811 remove_pop_action(struct action_xlate_ctx *ctx)
2813 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2814 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2815 ctx->last_pop_priority = -1;
2820 add_pop_action(struct action_xlate_ctx *ctx)
2822 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2823 nl_msg_put_flag(ctx->odp_actions, ODPAT_POP_PRIORITY);
2824 ctx->last_pop_priority = ctx->odp_actions->size;
2829 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2830 const struct ofp_action_enqueue *oae)
2832 uint16_t ofp_port, odp_port;
2836 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2839 /* Fall back to ordinary output action. */
2840 xlate_output_action__(ctx, ntohs(oae->port), 0);
2844 /* Figure out ODP output port. */
2845 ofp_port = ntohs(oae->port);
2846 if (ofp_port != OFPP_IN_PORT) {
2847 odp_port = ofp_port_to_odp_port(ofp_port);
2849 odp_port = ctx->flow.in_port;
2852 /* Add ODP actions. */
2853 remove_pop_action(ctx);
2854 nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority);
2855 add_output_action(ctx, odp_port);
2856 add_pop_action(ctx);
2858 /* Update NetFlow output port. */
2859 if (ctx->nf_output_iface == NF_OUT_DROP) {
2860 ctx->nf_output_iface = odp_port;
2861 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2862 ctx->nf_output_iface = NF_OUT_MULTI;
2867 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2868 const struct nx_action_set_queue *nasq)
2873 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2876 /* Couldn't translate queue to a priority, so ignore. A warning
2877 * has already been logged. */
2881 remove_pop_action(ctx);
2882 nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority);
2886 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2888 ovs_be16 tci = ctx->flow.vlan_tci;
2889 if (!(tci & htons(VLAN_CFI))) {
2890 nl_msg_put_flag(ctx->odp_actions, ODPAT_STRIP_VLAN);
2892 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_DL_TCI,
2893 tci & ~htons(VLAN_CFI));
2897 struct xlate_reg_state {
2903 save_reg_state(const struct action_xlate_ctx *ctx,
2904 struct xlate_reg_state *state)
2906 state->vlan_tci = ctx->flow.vlan_tci;
2907 state->tun_id = ctx->flow.tun_id;
2911 update_reg_state(struct action_xlate_ctx *ctx,
2912 const struct xlate_reg_state *state)
2914 if (ctx->flow.vlan_tci != state->vlan_tci) {
2915 xlate_set_dl_tci(ctx);
2917 if (ctx->flow.tun_id != state->tun_id) {
2918 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, ctx->flow.tun_id);
2923 xlate_nicira_action(struct action_xlate_ctx *ctx,
2924 const struct nx_action_header *nah)
2926 const struct nx_action_resubmit *nar;
2927 const struct nx_action_set_tunnel *nast;
2928 const struct nx_action_set_queue *nasq;
2929 const struct nx_action_multipath *nam;
2930 enum nx_action_subtype subtype = ntohs(nah->subtype);
2931 struct xlate_reg_state state;
2934 assert(nah->vendor == htonl(NX_VENDOR_ID));
2936 case NXAST_RESUBMIT:
2937 nar = (const struct nx_action_resubmit *) nah;
2938 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2941 case NXAST_SET_TUNNEL:
2942 nast = (const struct nx_action_set_tunnel *) nah;
2943 tun_id = htonll(ntohl(nast->tun_id));
2944 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id);
2945 ctx->flow.tun_id = tun_id;
2948 case NXAST_DROP_SPOOFED_ARP:
2949 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2950 nl_msg_put_flag(ctx->odp_actions, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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 ots->lookup_count = htonll(0); /* XXX */
3388 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 ops->rx_packets = htonll(stats.rx_packets);
3410 ops->tx_packets = htonll(stats.tx_packets);
3411 ops->rx_bytes = htonll(stats.rx_bytes);
3412 ops->tx_bytes = htonll(stats.tx_bytes);
3413 ops->rx_dropped = htonll(stats.rx_dropped);
3414 ops->tx_dropped = htonll(stats.tx_dropped);
3415 ops->rx_errors = htonll(stats.rx_errors);
3416 ops->tx_errors = htonll(stats.tx_errors);
3417 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3418 ops->rx_over_err = htonll(stats.rx_over_errors);
3419 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3420 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;
3502 size_t act_len, len;
3504 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3508 act_len = sizeof *rule->actions * rule->n_actions;
3509 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3511 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3513 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3514 ofs->length = htons(len);
3517 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3518 rule->flow_cookie, &ofs->cookie);
3519 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3520 ofs->priority = htons(rule->cr.priority);
3521 ofs->idle_timeout = htons(rule->idle_timeout);
3522 ofs->hard_timeout = htons(rule->hard_timeout);
3523 memset(ofs->pad2, 0, sizeof ofs->pad2);
3524 ofs->packet_count = htonll(packet_count);
3525 ofs->byte_count = htonll(byte_count);
3526 if (rule->n_actions > 0) {
3527 memcpy(ofs->actions, rule->actions, act_len);
3532 is_valid_table(uint8_t table_id)
3534 return table_id == 0 || table_id == 0xff;
3538 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3540 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3541 struct ofpbuf *reply;
3543 COVERAGE_INC(ofproto_flows_req);
3544 reply = start_ofp_stats_reply(oh, 1024);
3545 if (is_valid_table(fsr->table_id)) {
3546 struct cls_cursor cursor;
3547 struct cls_rule target;
3550 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3552 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3553 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3554 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3557 queue_tx(reply, ofconn, ofconn->reply_counter);
3563 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3564 ovs_be16 out_port, struct ofpbuf **replyp)
3566 struct nx_flow_stats *nfs;
3567 uint64_t packet_count, byte_count;
3568 size_t act_len, start_len;
3569 struct ofpbuf *reply;
3571 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3575 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3577 act_len = sizeof *rule->actions * rule->n_actions;
3579 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3580 start_len = (*replyp)->size;
3583 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3586 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3587 nfs->cookie = rule->flow_cookie;
3588 nfs->priority = htons(rule->cr.priority);
3589 nfs->idle_timeout = htons(rule->idle_timeout);
3590 nfs->hard_timeout = htons(rule->hard_timeout);
3591 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3592 memset(nfs->pad2, 0, sizeof nfs->pad2);
3593 nfs->packet_count = htonll(packet_count);
3594 nfs->byte_count = htonll(byte_count);
3595 if (rule->n_actions > 0) {
3596 ofpbuf_put(reply, rule->actions, act_len);
3598 nfs->length = htons(reply->size - start_len);
3602 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3604 struct nx_flow_stats_request *nfsr;
3605 struct cls_rule target;
3606 struct ofpbuf *reply;
3610 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3612 /* Dissect the message. */
3613 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3614 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3619 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3622 COVERAGE_INC(ofproto_flows_req);
3623 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3624 if (is_valid_table(nfsr->table_id)) {
3625 struct cls_cursor cursor;
3628 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3629 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3630 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3633 queue_tx(reply, ofconn, ofconn->reply_counter);
3639 flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
3641 uint64_t packet_count, byte_count;
3642 size_t act_len = sizeof *rule->actions * rule->n_actions;
3644 query_stats(ofproto, rule, &packet_count, &byte_count);
3646 ds_put_format(results, "duration=%llds, ",
3647 (time_msec() - rule->created) / 1000);
3648 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3649 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3650 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3651 cls_rule_format(&rule->cr, results);
3653 ofp_print_actions(results, &rule->actions->header, act_len);
3655 ds_put_cstr(results, "drop");
3657 ds_put_cstr(results, "\n");
3660 /* Adds a pretty-printed description of all flows to 'results', including
3661 * those marked hidden by secchan (e.g., by in-band control). */
3663 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3665 struct cls_cursor cursor;
3668 cls_cursor_init(&cursor, &p->cls, NULL);
3669 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3670 flow_stats_ds(p, rule, results);
3675 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3676 ovs_be16 out_port, uint8_t table_id,
3677 struct ofp_aggregate_stats_reply *oasr)
3679 uint64_t total_packets = 0;
3680 uint64_t total_bytes = 0;
3683 COVERAGE_INC(ofproto_agg_request);
3685 if (is_valid_table(table_id)) {
3686 struct cls_cursor cursor;
3689 cls_cursor_init(&cursor, &ofproto->cls, target);
3690 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3691 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3692 uint64_t packet_count;
3693 uint64_t byte_count;
3695 query_stats(ofproto, rule, &packet_count, &byte_count);
3697 total_packets += packet_count;
3698 total_bytes += byte_count;
3704 oasr->flow_count = htonl(n_flows);
3705 oasr->packet_count = htonll(total_packets);
3706 oasr->byte_count = htonll(total_bytes);
3707 memset(oasr->pad, 0, sizeof oasr->pad);
3711 handle_aggregate_stats_request(struct ofconn *ofconn,
3712 const struct ofp_header *oh)
3714 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3715 struct ofp_aggregate_stats_reply *reply;
3716 struct cls_rule target;
3719 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3722 msg = start_ofp_stats_reply(oh, sizeof *reply);
3723 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3724 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3725 request->table_id, reply);
3726 queue_tx(msg, ofconn, ofconn->reply_counter);
3731 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3733 struct nx_aggregate_stats_request *request;
3734 struct ofp_aggregate_stats_reply *reply;
3735 struct cls_rule target;
3740 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3742 /* Dissect the message. */
3743 request = ofpbuf_pull(&b, sizeof *request);
3744 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3749 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3753 COVERAGE_INC(ofproto_flows_req);
3754 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3755 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3756 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3757 request->table_id, reply);
3758 queue_tx(buf, ofconn, ofconn->reply_counter);
3763 struct queue_stats_cbdata {
3764 struct ofconn *ofconn;
3765 struct ofport *ofport;
3770 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3771 const struct netdev_queue_stats *stats)
3773 struct ofp_queue_stats *reply;
3775 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3776 reply->port_no = htons(cbdata->ofport->opp.port_no);
3777 memset(reply->pad, 0, sizeof reply->pad);
3778 reply->queue_id = htonl(queue_id);
3779 reply->tx_bytes = htonll(stats->tx_bytes);
3780 reply->tx_packets = htonll(stats->tx_packets);
3781 reply->tx_errors = htonll(stats->tx_errors);
3785 handle_queue_stats_dump_cb(uint32_t queue_id,
3786 struct netdev_queue_stats *stats,
3789 struct queue_stats_cbdata *cbdata = cbdata_;
3791 put_queue_stats(cbdata, queue_id, stats);
3795 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3796 struct queue_stats_cbdata *cbdata)
3798 cbdata->ofport = port;
3799 if (queue_id == OFPQ_ALL) {
3800 netdev_dump_queue_stats(port->netdev,
3801 handle_queue_stats_dump_cb, cbdata);
3803 struct netdev_queue_stats stats;
3805 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3806 put_queue_stats(cbdata, queue_id, &stats);
3812 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3814 struct ofproto *ofproto = ofconn->ofproto;
3815 const struct ofp_queue_stats_request *qsr;
3816 struct queue_stats_cbdata cbdata;
3817 struct ofport *port;
3818 unsigned int port_no;
3821 qsr = ofputil_stats_body(oh);
3823 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3826 COVERAGE_INC(ofproto_queue_req);
3828 cbdata.ofconn = ofconn;
3829 cbdata.msg = start_ofp_stats_reply(oh, 128);
3831 port_no = ntohs(qsr->port_no);
3832 queue_id = ntohl(qsr->queue_id);
3833 if (port_no == OFPP_ALL) {
3834 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3835 handle_queue_stats_for_port(port, queue_id, &cbdata);
3837 } else if (port_no < ofproto->max_ports) {
3838 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3840 handle_queue_stats_for_port(port, queue_id, &cbdata);
3843 ofpbuf_delete(cbdata.msg);
3844 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3846 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3852 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3853 const struct dpif_flow_stats *stats)
3855 long long int used = stats->used;
3856 if (used > facet->used) {
3858 if (used > facet->rule->used) {
3859 facet->rule->used = used;
3861 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3865 /* Folds the statistics from 'stats' into the counters in 'facet'.
3867 * Because of the meaning of a facet's counters, it only makes sense to do this
3868 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3869 * packet that was sent by hand or if it represents statistics that have been
3870 * cleared out of the datapath. */
3872 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3873 const struct dpif_flow_stats *stats)
3875 if (stats->n_packets) {
3876 facet_update_time(ofproto, facet, stats);
3877 facet->packet_count += stats->n_packets;
3878 facet->byte_count += stats->n_bytes;
3879 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3883 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3884 * in which no matching flow already exists in the flow table.
3886 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3887 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3888 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3890 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3893 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3895 struct ofproto *p = ofconn->ofproto;
3896 struct ofpbuf *packet;
3901 if (fm->flags & OFPFF_CHECK_OVERLAP
3902 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3903 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3907 if (fm->buffer_id != UINT32_MAX) {
3908 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3912 in_port = UINT16_MAX;
3915 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3916 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3917 fm->flags & OFPFF_SEND_FLOW_REM);
3918 rule_insert(p, rule);
3920 rule_execute(p, rule, in_port, packet);
3925 static struct rule *
3926 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3928 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3932 send_buffered_packet(struct ofconn *ofconn,
3933 struct rule *rule, uint32_t buffer_id)
3935 struct ofpbuf *packet;
3939 if (buffer_id == UINT32_MAX) {
3943 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3948 rule_execute(ofconn->ofproto, rule, in_port, packet);
3953 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3955 struct modify_flows_cbdata {
3956 struct ofproto *ofproto;
3957 const struct flow_mod *fm;
3961 static int modify_flow(struct ofproto *, const struct flow_mod *,
3964 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3965 * encoded by ofp_mkerr() on failure.
3967 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3970 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3972 struct ofproto *p = ofconn->ofproto;
3973 struct rule *match = NULL;
3974 struct cls_cursor cursor;
3977 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3978 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3979 if (!rule_is_hidden(rule)) {
3981 modify_flow(p, fm, rule);
3986 /* This credits the packet to whichever flow happened to match last.
3987 * That's weird. Maybe we should do a lookup for the flow that
3988 * actually matches the packet? Who knows. */
3989 send_buffered_packet(ofconn, match, fm->buffer_id);
3992 return add_flow(ofconn, fm);
3996 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3997 * code as encoded by ofp_mkerr() on failure.
3999 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4002 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4004 struct ofproto *p = ofconn->ofproto;
4005 struct rule *rule = find_flow_strict(p, fm);
4006 if (rule && !rule_is_hidden(rule)) {
4007 modify_flow(p, fm, rule);
4008 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4010 return add_flow(ofconn, fm);
4014 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4015 * been identified as a flow in 'p''s flow table to be modified, by changing
4016 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4017 * ofp_action[] structures). */
4019 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4021 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4023 rule->flow_cookie = fm->cookie;
4025 /* If the actions are the same, do nothing. */
4026 if (fm->n_actions == rule->n_actions
4028 || !memcmp(fm->actions, rule->actions, actions_len))) {
4032 /* Replace actions. */
4033 free(rule->actions);
4034 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4035 rule->n_actions = fm->n_actions;
4037 p->need_revalidate = true;
4042 /* OFPFC_DELETE implementation. */
4044 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4046 /* Implements OFPFC_DELETE. */
4048 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4050 struct rule *rule, *next_rule;
4051 struct cls_cursor cursor;
4053 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4054 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4055 delete_flow(p, rule, htons(fm->out_port));
4059 /* Implements OFPFC_DELETE_STRICT. */
4061 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4063 struct rule *rule = find_flow_strict(p, fm);
4065 delete_flow(p, rule, htons(fm->out_port));
4069 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4070 * been identified as a flow to delete from 'p''s flow table, by deleting the
4071 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4074 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4075 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4076 * specified 'out_port'. */
4078 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4080 if (rule_is_hidden(rule)) {
4084 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4088 rule_send_removed(p, rule, OFPRR_DELETE);
4089 rule_remove(p, rule);
4093 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4095 struct ofproto *p = ofconn->ofproto;
4099 error = reject_slave_controller(ofconn, "flow_mod");
4104 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4109 /* We do not support the emergency flow cache. It will hopefully get
4110 * dropped from OpenFlow in the near future. */
4111 if (fm.flags & OFPFF_EMERG) {
4112 /* There isn't a good fit for an error code, so just state that the
4113 * flow table is full. */
4114 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4117 error = validate_actions(fm.actions, fm.n_actions,
4118 &fm.cr.flow, p->max_ports);
4123 switch (fm.command) {
4125 return add_flow(ofconn, &fm);
4128 return modify_flows_loose(ofconn, &fm);
4130 case OFPFC_MODIFY_STRICT:
4131 return modify_flow_strict(ofconn, &fm);
4134 delete_flows_loose(p, &fm);
4137 case OFPFC_DELETE_STRICT:
4138 delete_flow_strict(p, &fm);
4142 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4147 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4149 const struct nxt_tun_id_cookie *msg
4150 = (const struct nxt_tun_id_cookie *) oh;
4152 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4157 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4159 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4160 struct nx_role_request *reply;
4164 if (ofconn->type != OFCONN_PRIMARY) {
4165 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4167 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4170 role = ntohl(nrr->role);
4171 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4172 && role != NX_ROLE_SLAVE) {
4173 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4175 /* There's no good error code for this. */
4176 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4179 if (role == NX_ROLE_MASTER) {
4180 struct ofconn *other;
4182 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4183 if (other->role == NX_ROLE_MASTER) {
4184 other->role = NX_ROLE_SLAVE;
4188 ofconn->role = role;
4190 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4191 reply->role = htonl(role);
4192 queue_tx(buf, ofconn, ofconn->reply_counter);
4198 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4200 const struct nxt_set_flow_format *msg
4201 = (const struct nxt_set_flow_format *) oh;
4204 format = ntohl(msg->format);
4205 if (format == NXFF_OPENFLOW10
4206 || format == NXFF_TUN_ID_FROM_COOKIE
4207 || format == NXFF_NXM) {
4208 ofconn->flow_format = format;
4211 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4216 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4218 struct ofp_header *ob;
4221 /* Currently, everything executes synchronously, so we can just
4222 * immediately send the barrier reply. */
4223 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4224 queue_tx(buf, ofconn, ofconn->reply_counter);
4229 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4231 const struct ofp_header *oh = msg->data;
4232 const struct ofputil_msg_type *type;
4235 error = ofputil_decode_msg_type(oh, &type);
4240 switch (ofputil_msg_type_code(type)) {
4241 /* OpenFlow requests. */
4242 case OFPUTIL_OFPT_ECHO_REQUEST:
4243 return handle_echo_request(ofconn, oh);
4245 case OFPUTIL_OFPT_FEATURES_REQUEST:
4246 return handle_features_request(ofconn, oh);
4248 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4249 return handle_get_config_request(ofconn, oh);
4251 case OFPUTIL_OFPT_SET_CONFIG:
4252 return handle_set_config(ofconn, msg->data);
4254 case OFPUTIL_OFPT_PACKET_OUT:
4255 return handle_packet_out(ofconn, oh);
4257 case OFPUTIL_OFPT_PORT_MOD:
4258 return handle_port_mod(ofconn, oh);
4260 case OFPUTIL_OFPT_FLOW_MOD:
4261 return handle_flow_mod(ofconn, oh);
4263 case OFPUTIL_OFPT_BARRIER_REQUEST:
4264 return handle_barrier_request(ofconn, oh);
4266 /* OpenFlow replies. */
4267 case OFPUTIL_OFPT_ECHO_REPLY:
4270 /* Nicira extension requests. */
4271 case OFPUTIL_NXT_STATUS_REQUEST:
4272 return switch_status_handle_request(
4273 ofconn->ofproto->switch_status, ofconn->rconn, oh);
4275 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4276 return handle_tun_id_from_cookie(ofconn, oh);
4278 case OFPUTIL_NXT_ROLE_REQUEST:
4279 return handle_role_request(ofconn, oh);
4281 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4282 return handle_nxt_set_flow_format(ofconn, oh);
4284 case OFPUTIL_NXT_FLOW_MOD:
4285 return handle_flow_mod(ofconn, oh);
4287 /* OpenFlow statistics requests. */
4288 case OFPUTIL_OFPST_DESC_REQUEST:
4289 return handle_desc_stats_request(ofconn, oh);
4291 case OFPUTIL_OFPST_FLOW_REQUEST:
4292 return handle_flow_stats_request(ofconn, oh);
4294 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4295 return handle_aggregate_stats_request(ofconn, oh);
4297 case OFPUTIL_OFPST_TABLE_REQUEST:
4298 return handle_table_stats_request(ofconn, oh);
4300 case OFPUTIL_OFPST_PORT_REQUEST:
4301 return handle_port_stats_request(ofconn, oh);
4303 case OFPUTIL_OFPST_QUEUE_REQUEST:
4304 return handle_queue_stats_request(ofconn, oh);
4306 /* Nicira extension statistics requests. */
4307 case OFPUTIL_NXST_FLOW_REQUEST:
4308 return handle_nxst_flow(ofconn, oh);
4310 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4311 return handle_nxst_aggregate(ofconn, oh);
4313 case OFPUTIL_INVALID:
4314 case OFPUTIL_OFPT_HELLO:
4315 case OFPUTIL_OFPT_ERROR:
4316 case OFPUTIL_OFPT_FEATURES_REPLY:
4317 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4318 case OFPUTIL_OFPT_PACKET_IN:
4319 case OFPUTIL_OFPT_FLOW_REMOVED:
4320 case OFPUTIL_OFPT_PORT_STATUS:
4321 case OFPUTIL_OFPT_BARRIER_REPLY:
4322 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4323 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4324 case OFPUTIL_OFPST_DESC_REPLY:
4325 case OFPUTIL_OFPST_FLOW_REPLY:
4326 case OFPUTIL_OFPST_QUEUE_REPLY:
4327 case OFPUTIL_OFPST_PORT_REPLY:
4328 case OFPUTIL_OFPST_TABLE_REPLY:
4329 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4330 case OFPUTIL_NXT_STATUS_REPLY:
4331 case OFPUTIL_NXT_ROLE_REPLY:
4332 case OFPUTIL_NXT_FLOW_REMOVED:
4333 case OFPUTIL_NXST_FLOW_REPLY:
4334 case OFPUTIL_NXST_AGGREGATE_REPLY:
4336 if (VLOG_IS_WARN_ENABLED()) {
4337 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4338 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4341 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4342 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4344 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4350 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4352 int error = handle_openflow__(ofconn, ofp_msg);
4354 send_error_oh(ofconn, ofp_msg->data, error);
4356 COVERAGE_INC(ofproto_recv_openflow);
4360 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4362 struct facet *facet;
4365 /* Obtain in_port and tun_id, at least. */
4366 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4368 /* Set header pointers in 'flow'. */
4369 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4371 /* Check with in-band control to see if this packet should be sent
4372 * to the local port regardless of the flow table. */
4373 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4374 struct ofpbuf odp_actions;
4376 ofpbuf_init(&odp_actions, 32);
4377 nl_msg_put_u32(&odp_actions, ODPAT_OUTPUT, ODPP_LOCAL);
4378 dpif_execute(p->dpif, odp_actions.data, odp_actions.size,
4380 ofpbuf_uninit(&odp_actions);
4383 facet = facet_lookup_valid(p, &flow);
4385 struct rule *rule = rule_lookup(p, &flow);
4387 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4388 struct ofport *port = get_port(p, flow.in_port);
4390 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4391 COVERAGE_INC(ofproto_no_packet_in);
4392 /* XXX install 'drop' flow entry */
4393 ofpbuf_delete(upcall->packet);
4397 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4401 COVERAGE_INC(ofproto_packet_in);
4402 send_packet_in(p, upcall, &flow, false);
4406 facet = facet_create(p, rule, &flow, upcall->packet);
4407 } else if (!facet->may_install) {
4408 /* The facet is not installable, that is, we need to process every
4409 * packet, so process the current packet's actions into 'facet'. */
4410 facet_make_actions(p, facet, upcall->packet);
4413 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4415 * Extra-special case for fail-open mode.
4417 * We are in fail-open mode and the packet matched the fail-open rule,
4418 * but we are connected to a controller too. We should send the packet
4419 * up to the controller in the hope that it will try to set up a flow
4420 * and thereby allow us to exit fail-open.
4422 * See the top-level comment in fail-open.c for more information.
4424 send_packet_in(p, upcall, &flow, true);
4427 facet_execute(p, facet, upcall->packet);
4428 facet_install(p, facet, false);
4432 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4436 switch (upcall->type) {
4437 case DPIF_UC_ACTION:
4438 COVERAGE_INC(ofproto_ctlr_action);
4439 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4440 send_packet_in(p, upcall, &flow, false);
4443 case DPIF_UC_SAMPLE:
4445 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4446 ofproto_sflow_received(p->sflow, upcall, &flow);
4448 ofpbuf_delete(upcall->packet);
4452 handle_miss_upcall(p, upcall);
4456 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4461 /* Flow expiration. */
4463 static int ofproto_dp_max_idle(const struct ofproto *);
4464 static void ofproto_update_used(struct ofproto *);
4465 static void rule_expire(struct ofproto *, struct rule *);
4466 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4468 /* This function is called periodically by ofproto_run(). Its job is to
4469 * collect updates for the flows that have been installed into the datapath,
4470 * most importantly when they last were used, and then use that information to
4471 * expire flows that have not been used recently.
4473 * Returns the number of milliseconds after which it should be called again. */
4475 ofproto_expire(struct ofproto *ofproto)
4477 struct rule *rule, *next_rule;
4478 struct cls_cursor cursor;
4481 /* Update 'used' for each flow in the datapath. */
4482 ofproto_update_used(ofproto);
4484 /* Expire facets that have been idle too long. */
4485 dp_max_idle = ofproto_dp_max_idle(ofproto);
4486 ofproto_expire_facets(ofproto, dp_max_idle);
4488 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4489 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4490 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4491 rule_expire(ofproto, rule);
4494 /* Let the hook know that we're at a stable point: all outstanding data
4495 * in existing flows has been accounted to the account_cb. Thus, the
4496 * hook can now reasonably do operations that depend on having accurate
4497 * flow volume accounting (currently, that's just bond rebalancing). */
4498 if (ofproto->ofhooks->account_checkpoint_cb) {
4499 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4502 return MIN(dp_max_idle, 1000);
4505 /* Update 'used' member of installed facets. */
4507 ofproto_update_used(struct ofproto *p)
4509 const struct dpif_flow_stats *stats;
4510 struct dpif_flow_dump dump;
4511 const struct nlattr *key;
4514 dpif_flow_dump_start(&dump, p->dpif);
4515 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4516 struct facet *facet;
4519 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4523 odp_flow_key_format(key, key_len, &s);
4524 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4530 facet = facet_find(p, &flow);
4532 if (facet && facet->installed) {
4533 facet_update_time(p, facet, stats);
4534 facet_account(p, facet, stats->n_bytes);
4536 /* There's a flow in the datapath that we know nothing about.
4538 COVERAGE_INC(ofproto_unexpected_rule);
4539 dpif_flow_del(p->dpif, key, key_len, NULL);
4542 dpif_flow_dump_done(&dump);
4545 /* Calculates and returns the number of milliseconds of idle time after which
4546 * facets should expire from the datapath and we should fold their statistics
4547 * into their parent rules in userspace. */
4549 ofproto_dp_max_idle(const struct ofproto *ofproto)
4552 * Idle time histogram.
4554 * Most of the time a switch has a relatively small number of facets. When
4555 * this is the case we might as well keep statistics for all of them in
4556 * userspace and to cache them in the kernel datapath for performance as
4559 * As the number of facets increases, the memory required to maintain
4560 * statistics about them in userspace and in the kernel becomes
4561 * significant. However, with a large number of facets it is likely that
4562 * only a few of them are "heavy hitters" that consume a large amount of
4563 * bandwidth. At this point, only heavy hitters are worth caching in the
4564 * kernel and maintaining in userspaces; other facets we can discard.
4566 * The technique used to compute the idle time is to build a histogram with
4567 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4568 * that is installed in the kernel gets dropped in the appropriate bucket.
4569 * After the histogram has been built, we compute the cutoff so that only
4570 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4571 * cached. At least the most-recently-used bucket of facets is kept, so
4572 * actually an arbitrary number of facets can be kept in any given
4573 * expiration run (though the next run will delete most of those unless
4574 * they receive additional data).
4576 * This requires a second pass through the facets, in addition to the pass
4577 * made by ofproto_update_used(), because the former function never looks
4578 * at uninstallable facets.
4580 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4581 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4582 int buckets[N_BUCKETS] = { 0 };
4583 struct facet *facet;
4588 total = hmap_count(&ofproto->facets);
4589 if (total <= 1000) {
4590 return N_BUCKETS * BUCKET_WIDTH;
4593 /* Build histogram. */
4595 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4596 long long int idle = now - facet->used;
4597 int bucket = (idle <= 0 ? 0
4598 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4599 : (unsigned int) idle / BUCKET_WIDTH);
4603 /* Find the first bucket whose flows should be expired. */
4604 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4605 if (buckets[bucket]) {
4608 subtotal += buckets[bucket++];
4609 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4614 if (VLOG_IS_DBG_ENABLED()) {
4618 ds_put_cstr(&s, "keep");
4619 for (i = 0; i < N_BUCKETS; i++) {
4621 ds_put_cstr(&s, ", drop");
4624 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4627 VLOG_INFO("%s: %s (msec:count)",
4628 dpif_name(ofproto->dpif), ds_cstr(&s));
4632 return bucket * BUCKET_WIDTH;
4636 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4638 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4639 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4640 struct ofexpired expired;
4642 if (facet->installed) {
4643 struct dpif_flow_stats stats;
4645 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4647 facet_update_stats(ofproto, facet, &stats);
4650 expired.flow = facet->flow;
4651 expired.packet_count = facet->packet_count;
4652 expired.byte_count = facet->byte_count;
4653 expired.used = facet->used;
4654 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4659 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4661 long long int cutoff = time_msec() - dp_max_idle;
4662 struct facet *facet, *next_facet;
4664 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4665 facet_active_timeout(ofproto, facet);
4666 if (facet->used < cutoff) {
4667 facet_remove(ofproto, facet);
4672 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4673 * then delete it entirely. */
4675 rule_expire(struct ofproto *ofproto, struct rule *rule)
4677 struct facet *facet, *next_facet;
4681 /* Has 'rule' expired? */
4683 if (rule->hard_timeout
4684 && now > rule->created + rule->hard_timeout * 1000) {
4685 reason = OFPRR_HARD_TIMEOUT;
4686 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4687 && now >rule->used + rule->idle_timeout * 1000) {
4688 reason = OFPRR_IDLE_TIMEOUT;
4693 COVERAGE_INC(ofproto_expired);
4695 /* Update stats. (This is a no-op if the rule expired due to an idle
4696 * timeout, because that only happens when the rule has no facets left.) */
4697 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4698 facet_remove(ofproto, facet);
4701 /* Get rid of the rule. */
4702 if (!rule_is_hidden(rule)) {
4703 rule_send_removed(ofproto, rule, reason);
4705 rule_remove(ofproto, rule);
4708 static struct ofpbuf *
4709 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4712 struct ofp_flow_removed *ofr;
4715 ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
4716 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4717 rule->flow_cookie, &ofr->cookie);
4718 ofr->priority = htons(rule->cr.priority);
4719 ofr->reason = reason;
4720 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4721 ofr->idle_timeout = htons(rule->idle_timeout);
4722 ofr->packet_count = htonll(rule->packet_count);
4723 ofr->byte_count = htonll(rule->byte_count);
4728 static struct ofpbuf *
4729 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4731 struct nx_flow_removed *nfr;
4735 make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
4736 match_len = nx_put_match(buf, &rule->cr);
4739 nfr->cookie = rule->flow_cookie;
4740 nfr->priority = htons(rule->cr.priority);
4741 nfr->reason = reason;
4742 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4743 nfr->idle_timeout = htons(rule->idle_timeout);
4744 nfr->match_len = htons(match_len);
4745 nfr->packet_count = htonll(rule->packet_count);
4746 nfr->byte_count = htonll(rule->byte_count);
4752 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4754 struct ofconn *ofconn;
4756 if (!rule->send_flow_removed) {
4760 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4763 if (!rconn_is_connected(ofconn->rconn)
4764 || !ofconn_receives_async_msgs(ofconn)) {
4768 msg = (ofconn->flow_format == NXFF_NXM
4769 ? compose_nx_flow_removed(rule, reason)
4770 : compose_ofp_flow_removed(ofconn, rule, reason));
4772 /* Account flow expirations under ofconn->reply_counter, the counter
4773 * for replies to OpenFlow requests. That works because preventing
4774 * OpenFlow requests from being processed also prevents new flows from
4775 * being added (and expiring). (It also prevents processing OpenFlow
4776 * requests that would not add new flows, so it is imperfect.) */
4777 queue_tx(msg, ofconn, ofconn->reply_counter);
4781 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4783 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4785 struct ofconn *ofconn = ofconn_;
4787 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4788 ofconn->packet_in_counter, 100);
4791 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4792 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4793 * scheduler for sending.
4795 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4796 * Otherwise, ownership is transferred to this function. */
4798 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4799 const struct flow *flow, bool clone)
4801 enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) };
4802 struct ofproto *ofproto = ofconn->ofproto;
4803 struct ofp_packet_in *opi;
4804 int total_len, send_len;
4805 struct ofpbuf *packet;
4809 /* Get OpenFlow buffer_id. */
4810 if (upcall->type == DPIF_UC_ACTION) {
4811 buffer_id = UINT32_MAX;
4812 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4813 buffer_id = pktbuf_get_null();
4814 } else if (!ofconn->pktbuf) {
4815 buffer_id = UINT32_MAX;
4817 buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port);
4820 /* Figure out how much of the packet to send. */
4821 total_len = send_len = upcall->packet->size;
4822 if (buffer_id != UINT32_MAX) {
4823 send_len = MIN(send_len, ofconn->miss_send_len);
4825 if (upcall->type == DPIF_UC_ACTION) {
4826 send_len = MIN(send_len, upcall->userdata);
4829 /* Copy or steal buffer for OFPT_PACKET_IN. */
4831 packet = ofpbuf_clone_data_with_headroom(upcall->packet->data,
4832 send_len, OPI_SIZE);
4834 packet = upcall->packet;
4835 packet->size = send_len;
4838 /* Add OFPT_PACKET_IN. */
4839 opi = ofpbuf_push_zeros(packet, OPI_SIZE);
4840 opi->header.version = OFP_VERSION;
4841 opi->header.type = OFPT_PACKET_IN;
4842 opi->total_len = htons(total_len);
4843 opi->in_port = htons(odp_port_to_ofp_port(flow->in_port));
4844 opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4845 opi->buffer_id = htonl(buffer_id);
4846 update_openflow_length(packet);
4848 /* Hand over to packet scheduler. It might immediately call into
4849 * do_send_packet_in() or it might buffer it for a while (until a later
4850 * call to pinsched_run()). */
4851 idx = upcall->type == DPIF_UC_MISS ? 0 : 1;
4852 pinsched_send(ofconn->schedulers[idx], flow->in_port,
4853 packet, do_send_packet_in, ofconn);
4856 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4857 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4858 * their individual configurations.
4860 * Takes ownership of 'packet'. */
4862 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4863 const struct flow *flow, bool clone)
4865 struct ofconn *ofconn, *prev;
4868 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4869 if (ofconn_receives_async_msgs(ofconn)) {
4871 schedule_packet_in(prev, upcall, flow, true);
4877 schedule_packet_in(prev, upcall, flow, clone);
4878 } else if (!clone) {
4879 ofpbuf_delete(upcall->packet);
4884 pick_datapath_id(const struct ofproto *ofproto)
4886 const struct ofport *port;
4888 port = get_port(ofproto, ODPP_LOCAL);
4890 uint8_t ea[ETH_ADDR_LEN];
4893 error = netdev_get_etheraddr(port->netdev, ea);
4895 return eth_addr_to_uint64(ea);
4897 VLOG_WARN("could not get MAC address for %s (%s)",
4898 netdev_get_name(port->netdev), strerror(error));
4900 return ofproto->fallback_dpid;
4904 pick_fallback_dpid(void)
4906 uint8_t ea[ETH_ADDR_LEN];
4907 eth_addr_nicira_random(ea);
4908 return eth_addr_to_uint64(ea);
4912 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4913 void *aux OVS_UNUSED)
4915 const struct shash_node *node;
4919 SHASH_FOR_EACH (node, &all_ofprotos) {
4920 ds_put_format(&results, "%s\n", node->name);
4922 unixctl_command_reply(conn, 200, ds_cstr(&results));
4923 ds_destroy(&results);
4926 struct ofproto_trace {
4927 struct action_xlate_ctx ctx;
4933 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4935 ds_put_char_multiple(result, '\t', level);
4937 ds_put_cstr(result, "No match\n");
4941 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4942 ntohll(rule->flow_cookie));
4943 cls_rule_format(&rule->cr, result);
4944 ds_put_char(result, '\n');
4946 ds_put_char_multiple(result, '\t', level);
4947 ds_put_cstr(result, "OpenFlow ");
4948 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4949 rule->n_actions * sizeof *rule->actions);
4950 ds_put_char(result, '\n');
4954 trace_format_flow(struct ds *result, int level, const char *title,
4955 struct ofproto_trace *trace)
4957 ds_put_char_multiple(result, '\t', level);
4958 ds_put_format(result, "%s: ", title);
4959 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4960 ds_put_cstr(result, "unchanged");
4962 flow_format(result, &trace->ctx.flow);
4963 trace->flow = trace->ctx.flow;
4965 ds_put_char(result, '\n');
4969 trace_resubmit(struct action_xlate_ctx *ctx, const struct rule *rule)
4971 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4972 struct ds *result = trace->result;
4974 ds_put_char(result, '\n');
4975 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4976 trace_format_rule(result, ctx->recurse + 1, rule);
4980 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4981 void *aux OVS_UNUSED)
4983 char *dpname, *in_port_s, *tun_id_s, *packet_s;
4984 char *args = xstrdup(args_);
4985 char *save_ptr = NULL;
4986 struct ofproto *ofproto;
4987 struct ofpbuf packet;
4995 ofpbuf_init(&packet, strlen(args) / 2);
4998 dpname = strtok_r(args, " ", &save_ptr);
4999 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5000 in_port_s = strtok_r(NULL, " ", &save_ptr);
5001 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5002 if (!dpname || !in_port_s || !packet_s) {
5003 unixctl_command_reply(conn, 501, "Bad command syntax");
5007 ofproto = shash_find_data(&all_ofprotos, dpname);
5009 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5014 tun_id = htonll(strtoull(tun_id_s, NULL, 10));
5015 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5017 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5018 packet_s += strspn(packet_s, " ");
5019 if (*packet_s != '\0') {
5020 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5023 if (packet.size < ETH_HEADER_LEN) {
5024 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5028 ds_put_cstr(&result, "Packet: ");
5029 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5030 ds_put_cstr(&result, s);
5033 flow_extract(&packet, tun_id, in_port, &flow);
5034 ds_put_cstr(&result, "Flow: ");
5035 flow_format(&result, &flow);
5036 ds_put_char(&result, '\n');
5038 rule = rule_lookup(ofproto, &flow);
5039 trace_format_rule(&result, 0, rule);
5041 struct ofproto_trace trace;
5042 struct ofpbuf *odp_actions;
5044 trace.result = &result;
5046 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5047 trace.ctx.resubmit_hook = trace_resubmit;
5048 odp_actions = xlate_actions(&trace.ctx,
5049 rule->actions, rule->n_actions);
5051 ds_put_char(&result, '\n');
5052 trace_format_flow(&result, 0, "Final flow", &trace);
5053 ds_put_cstr(&result, "Datapath actions: ");
5054 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5055 ofpbuf_delete(odp_actions);
5058 unixctl_command_reply(conn, 200, ds_cstr(&result));
5061 ds_destroy(&result);
5062 ofpbuf_uninit(&packet);
5067 ofproto_unixctl_init(void)
5069 static bool registered;
5075 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5076 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5080 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5081 struct ofpbuf *odp_actions, tag_type *tags,
5082 uint16_t *nf_output_iface, void *ofproto_)
5084 struct ofproto *ofproto = ofproto_;
5087 /* Drop frames for reserved multicast addresses. */
5088 if (eth_addr_is_reserved(flow->dl_dst)) {
5092 /* Learn source MAC (but don't try to learn from revalidation). */
5093 if (packet != NULL) {
5094 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5096 GRAT_ARP_LOCK_NONE);
5098 /* The log messages here could actually be useful in debugging,
5099 * so keep the rate limit relatively high. */
5100 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5101 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5102 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5103 ofproto_revalidate(ofproto, rev_tag);
5107 /* Determine output port. */
5108 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5111 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5112 nf_output_iface, odp_actions);
5113 } else if (out_port != flow->in_port) {
5114 nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, out_port);
5115 *nf_output_iface = out_port;
5123 static const struct ofhooks default_ofhooks = {
5124 default_normal_ofhook_cb,