2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
29 #include "classifier.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "multipath.h"
44 #include "ofp-print.h"
46 #include "ofproto-sflow.h"
48 #include "openflow/nicira-ext.h"
49 #include "openflow/openflow.h"
50 #include "openvswitch/datapath-protocol.h"
54 #include "poll-loop.h"
57 #include "stream-ssl.h"
61 #include "unaligned.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(odp_overflow);
71 COVERAGE_DEFINE(ofproto_agg_request);
72 COVERAGE_DEFINE(ofproto_costly_flags);
73 COVERAGE_DEFINE(ofproto_ctlr_action);
74 COVERAGE_DEFINE(ofproto_del_rule);
75 COVERAGE_DEFINE(ofproto_error);
76 COVERAGE_DEFINE(ofproto_expiration);
77 COVERAGE_DEFINE(ofproto_expired);
78 COVERAGE_DEFINE(ofproto_flows_req);
79 COVERAGE_DEFINE(ofproto_flush);
80 COVERAGE_DEFINE(ofproto_invalidated);
81 COVERAGE_DEFINE(ofproto_no_packet_in);
82 COVERAGE_DEFINE(ofproto_ofconn_stuck);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_unexpected_rule);
90 COVERAGE_DEFINE(ofproto_uninstallable);
91 COVERAGE_DEFINE(ofproto_update_port);
93 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
94 * flow translation. */
95 #define MAX_RESUBMIT_RECURSION 16
100 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
101 struct netdev *netdev;
102 struct ofp_phy_port opp; /* In host byte order. */
104 struct cfm *cfm; /* Connectivity Fault Management, if any. */
107 static void ofport_free(struct ofport *);
108 static void ofport_run(struct ofproto *, struct ofport *);
109 static void ofport_wait(struct ofport *);
111 struct action_xlate_ctx {
112 /* action_xlate_ctx_init() initializes these members. */
115 struct ofproto *ofproto;
117 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
118 * this flow when actions change header fields. */
121 /* The packet corresponding to 'flow', or a null pointer if we are
122 * revalidating without a packet to refer to. */
123 const struct ofpbuf *packet;
125 /* If nonnull, called just before executing a resubmit action.
127 * This is normally null so the client has to set it manually after
128 * calling action_xlate_ctx_init(). */
129 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
131 /* If true, the speciality of 'flow' should be checked before executing
132 * its actions. If special_cb returns false on 'flow' rendered
133 * uninstallable and no actions will be executed. */
136 /* xlate_actions() initializes and uses these members. The client might want
137 * to look at them after it returns. */
139 struct ofpbuf *odp_actions; /* Datapath actions. */
140 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
141 bool may_set_up_flow; /* True ordinarily; false if the actions must
142 * be reassessed for every packet. */
143 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
145 /* xlate_actions() initializes and uses these members, but the client has no
146 * reason to look at them. */
148 int recurse; /* Recursion level, via xlate_table_action. */
149 int last_pop_priority; /* Offset in 'odp_actions' just past most
150 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
153 static void action_xlate_ctx_init(struct action_xlate_ctx *,
154 struct ofproto *, const struct flow *,
155 const struct ofpbuf *);
156 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
157 const union ofp_action *in, size_t n_in);
159 /* An OpenFlow flow. */
161 long long int used; /* Time last used; time created if not used. */
162 long long int created; /* Creation time. */
166 * - Do include packets and bytes from facets that have been deleted or
167 * whose own statistics have been folded into the rule.
169 * - Do include packets and bytes sent "by hand" that were accounted to
170 * the rule without any facet being involved (this is a rare corner
171 * case in rule_execute()).
173 * - Do not include packet or bytes that can be obtained from any facet's
174 * packet_count or byte_count member or that can be obtained from the
175 * datapath by, e.g., dpif_flow_get() for any facet.
177 uint64_t packet_count; /* Number of packets received. */
178 uint64_t byte_count; /* Number of bytes received. */
180 ovs_be64 flow_cookie; /* Controller-issued identifier. */
182 struct cls_rule cr; /* In owning ofproto's classifier. */
183 uint16_t idle_timeout; /* In seconds from time of last use. */
184 uint16_t hard_timeout; /* In seconds from time of creation. */
185 bool send_flow_removed; /* Send a flow removed message? */
186 int n_actions; /* Number of elements in actions[]. */
187 union ofp_action *actions; /* OpenFlow actions. */
188 struct list facets; /* List of "struct facet"s. */
191 static struct rule *rule_from_cls_rule(const struct cls_rule *);
192 static bool rule_is_hidden(const struct rule *);
194 static struct rule *rule_create(const struct cls_rule *,
195 const union ofp_action *, size_t n_actions,
196 uint16_t idle_timeout, uint16_t hard_timeout,
197 ovs_be64 flow_cookie, bool send_flow_removed);
198 static void rule_destroy(struct ofproto *, struct rule *);
199 static void rule_free(struct rule *);
201 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
202 static void rule_insert(struct ofproto *, struct rule *);
203 static void rule_remove(struct ofproto *, struct rule *);
205 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
206 static void rule_get_stats(const struct rule *, uint64_t *packets,
209 /* An exact-match instantiation of an OpenFlow flow. */
211 long long int used; /* Time last used; time created if not used. */
215 * - Do include packets and bytes sent "by hand", e.g. with
218 * - Do include packets and bytes that were obtained from the datapath
219 * when a flow was deleted (e.g. dpif_flow_del()) or when its
220 * statistics were reset (e.g. dpif_flow_put() with
221 * DPIF_FP_ZERO_STATS).
223 * - Do not include any packets or bytes that can currently be obtained
224 * from the datapath by, e.g., dpif_flow_get().
226 uint64_t packet_count; /* Number of packets received. */
227 uint64_t byte_count; /* Number of bytes received. */
229 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
230 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
232 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
233 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
234 long long int rs_used; /* Used time pushed to resubmit children. */
236 /* Number of bytes passed to account_cb. This may include bytes that can
237 * currently obtained from the datapath (thus, it can be greater than
239 uint64_t accounted_bytes;
241 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
242 struct list list_node; /* In owning rule's 'facets' list. */
243 struct rule *rule; /* Owning rule. */
244 struct flow flow; /* Exact-match flow. */
245 bool installed; /* Installed in datapath? */
246 bool may_install; /* True ordinarily; false if actions must
247 * be reassessed for every packet. */
248 size_t actions_len; /* Number of bytes in actions[]. */
249 struct nlattr *actions; /* Datapath actions. */
250 tag_type tags; /* Tags (set only by hooks). */
251 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
254 static struct facet *facet_create(struct ofproto *, struct rule *,
256 const struct ofpbuf *packet);
257 static void facet_remove(struct ofproto *, struct facet *);
258 static void facet_free(struct facet *);
260 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
261 static bool facet_revalidate(struct ofproto *, struct facet *);
263 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
264 static void facet_uninstall(struct ofproto *, struct facet *);
265 static void facet_flush_stats(struct ofproto *, struct facet *);
267 static void facet_make_actions(struct ofproto *, struct facet *,
268 const struct ofpbuf *packet);
269 static void facet_update_stats(struct ofproto *, struct facet *,
270 const struct dpif_flow_stats *);
271 static void facet_push_stats(struct ofproto *, struct facet *);
273 /* ofproto supports two kinds of OpenFlow connections:
275 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
276 * maintains persistent connections to these controllers and by default
277 * sends them asynchronous messages such as packet-ins.
279 * - "Service" connections, e.g. from ovs-ofctl. When these connections
280 * drop, it is the other side's responsibility to reconnect them if
281 * necessary. ofproto does not send them asynchronous messages by default.
283 * Currently, active (tcp, ssl, unix) connections are always "primary"
284 * connections and passive (ptcp, pssl, punix) connections are always "service"
285 * connections. There is no inherent reason for this, but it reflects the
289 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
290 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
293 /* A listener for incoming OpenFlow "service" connections. */
295 struct hmap_node node; /* In struct ofproto's "services" hmap. */
296 struct pvconn *pvconn; /* OpenFlow connection listener. */
298 /* These are not used by ofservice directly. They are settings for
299 * accepted "struct ofconn"s from the pvconn. */
300 int probe_interval; /* Max idle time before probing, in seconds. */
301 int rate_limit; /* Max packet-in rate in packets per second. */
302 int burst_limit; /* Limit on accumulating packet credits. */
305 static struct ofservice *ofservice_lookup(struct ofproto *,
307 static int ofservice_create(struct ofproto *,
308 const struct ofproto_controller *);
309 static void ofservice_reconfigure(struct ofservice *,
310 const struct ofproto_controller *);
311 static void ofservice_destroy(struct ofproto *, struct ofservice *);
313 /* An OpenFlow connection. */
315 struct ofproto *ofproto; /* The ofproto that owns this connection. */
316 struct list node; /* In struct ofproto's "all_conns" list. */
317 struct rconn *rconn; /* OpenFlow connection. */
318 enum ofconn_type type; /* Type. */
319 enum nx_flow_format flow_format; /* Currently selected flow format. */
321 /* OFPT_PACKET_IN related data. */
322 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
323 #define N_SCHEDULERS 2
324 struct pinsched *schedulers[N_SCHEDULERS];
325 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
326 int miss_send_len; /* Bytes to send of buffered packets. */
328 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
329 * requests, and the maximum number before we stop reading OpenFlow
331 #define OFCONN_REPLY_MAX 100
332 struct rconn_packet_counter *reply_counter;
334 /* type == OFCONN_PRIMARY only. */
335 enum nx_role role; /* Role. */
336 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
337 enum ofproto_band band; /* In-band or out-of-band? */
341 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
343 static void ofconn_destroy(struct ofconn *);
344 static void ofconn_run(struct ofconn *);
345 static void ofconn_wait(struct ofconn *);
347 static bool ofconn_receives_async_msgs(const struct ofconn *);
348 static char *ofconn_make_name(const struct ofproto *, const char *target);
349 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
351 static struct ofproto *ofconn_get_ofproto(struct ofconn *);
352 static enum nx_flow_format ofconn_get_flow_format(struct ofconn *);
353 static void ofconn_set_flow_format(struct ofconn *, enum nx_flow_format);
355 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
356 struct rconn_packet_counter *counter);
358 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
359 const struct flow *, bool clone);
360 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
364 uint64_t datapath_id; /* Datapath ID. */
365 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
366 char *mfr_desc; /* Manufacturer. */
367 char *hw_desc; /* Hardware. */
368 char *sw_desc; /* Software version. */
369 char *serial_desc; /* Serial number. */
370 char *dp_desc; /* Datapath description. */
374 struct netdev_monitor *netdev_monitor;
375 struct hmap ports; /* Contains "struct ofport"s. */
376 struct shash port_by_name;
380 struct fail_open *fail_open;
381 struct netflow *netflow;
382 struct ofproto_sflow *sflow;
384 /* In-band control. */
385 struct in_band *in_band;
386 long long int next_in_band_update;
387 struct sockaddr_in *extra_in_band_remotes;
388 size_t n_extra_remotes;
392 struct classifier cls;
393 long long int next_expiration;
397 bool need_revalidate;
398 struct tag_set revalidate_set;
400 /* OpenFlow connections. */
401 struct hmap controllers; /* Controller "struct ofconn"s. */
402 struct list all_conns; /* Contains "struct ofconn"s. */
403 enum ofproto_fail_mode fail_mode;
405 /* OpenFlow listeners. */
406 struct hmap services; /* Contains "struct ofservice"s. */
407 struct pvconn **snoops;
410 /* Hooks for ovs-vswitchd. */
411 const struct ofhooks *ofhooks;
414 /* Used by default ofhooks. */
415 struct mac_learning *ml;
418 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
419 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
421 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
423 static const struct ofhooks default_ofhooks;
425 static uint64_t pick_datapath_id(const struct ofproto *);
426 static uint64_t pick_fallback_dpid(void);
428 static int ofproto_expire(struct ofproto *);
429 static void flow_push_stats(struct ofproto *, const struct rule *,
430 struct flow *, uint64_t packets, uint64_t bytes,
433 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
435 static void handle_openflow(struct ofconn *, struct ofpbuf *);
437 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
438 static void update_port(struct ofproto *, const char *devname);
439 static int init_ports(struct ofproto *);
440 static void reinit_ports(struct ofproto *);
442 static void ofproto_unixctl_init(void);
445 ofproto_create(const char *datapath, const char *datapath_type,
446 const struct ofhooks *ofhooks, void *aux,
447 struct ofproto **ofprotop)
455 ofproto_unixctl_init();
457 /* Connect to datapath and start listening for messages. */
458 error = dpif_open(datapath, datapath_type, &dpif);
460 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
463 error = dpif_recv_set_mask(dpif,
464 ((1u << DPIF_UC_MISS) |
465 (1u << DPIF_UC_ACTION) |
466 (1u << DPIF_UC_SAMPLE)));
468 VLOG_ERR("failed to listen on datapath %s: %s",
469 datapath, strerror(error));
473 dpif_flow_flush(dpif);
474 dpif_recv_purge(dpif);
476 /* Initialize settings. */
477 p = xzalloc(sizeof *p);
478 p->fallback_dpid = pick_fallback_dpid();
479 p->datapath_id = p->fallback_dpid;
480 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
481 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
482 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
483 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
484 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
486 /* Initialize datapath. */
488 p->netdev_monitor = netdev_monitor_create();
489 hmap_init(&p->ports);
490 shash_init(&p->port_by_name);
491 p->max_ports = dpif_get_max_ports(dpif);
493 /* Initialize submodules. */
498 /* Initialize in-band control. */
500 p->in_band_queue = -1;
502 /* Initialize flow table. */
503 classifier_init(&p->cls);
504 p->next_expiration = time_msec() + 1000;
506 /* Initialize facet table. */
507 hmap_init(&p->facets);
508 p->need_revalidate = false;
509 tag_set_init(&p->revalidate_set);
511 /* Initialize OpenFlow connections. */
512 list_init(&p->all_conns);
513 hmap_init(&p->controllers);
514 hmap_init(&p->services);
518 /* Initialize hooks. */
520 p->ofhooks = ofhooks;
524 p->ofhooks = &default_ofhooks;
526 p->ml = mac_learning_create();
529 /* Pick final datapath ID. */
530 p->datapath_id = pick_datapath_id(p);
531 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
533 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
540 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
542 uint64_t old_dpid = p->datapath_id;
543 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
544 if (p->datapath_id != old_dpid) {
545 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
547 /* Force all active connections to reconnect, since there is no way to
548 * notify a controller that the datapath ID has changed. */
549 ofproto_reconnect_controllers(p);
553 /* Creates a new controller in 'ofproto'. Some of the settings are initially
554 * drawn from 'c', but update_controller() needs to be called later to finish
555 * the new ofconn's configuration. */
557 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
559 char *name = ofconn_make_name(ofproto, c->target);
560 struct ofconn *ofconn;
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;
565 rconn_connect(ofconn->rconn, c->target, name);
566 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
567 hash_string(c->target, 0));
572 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
573 * target (this is done by creating new ofconns and deleting old ones), but it
574 * can update the rest of an ofconn's settings. */
576 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
580 ofconn->band = c->band;
582 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
584 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
585 rconn_set_probe_interval(ofconn->rconn, probe_interval);
587 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
591 ofconn_get_target(const struct ofconn *ofconn)
593 return rconn_get_target(ofconn->rconn);
596 static struct ofconn *
597 find_controller_by_target(struct ofproto *ofproto, const char *target)
599 struct ofconn *ofconn;
601 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
602 hash_string(target, 0), &ofproto->controllers) {
603 if (!strcmp(ofconn_get_target(ofconn), target)) {
611 update_in_band_remotes(struct ofproto *ofproto)
613 const struct ofconn *ofconn;
614 struct sockaddr_in *addrs;
615 size_t max_addrs, n_addrs;
618 /* Allocate enough memory for as many remotes as we could possibly have. */
619 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
620 addrs = xmalloc(max_addrs * sizeof *addrs);
623 /* Add all the remotes. */
624 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
625 struct sockaddr_in *sin = &addrs[n_addrs];
627 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
631 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
632 if (sin->sin_addr.s_addr) {
633 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
637 for (i = 0; i < ofproto->n_extra_remotes; i++) {
638 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
641 /* Create or update or destroy in-band. */
643 if (!ofproto->in_band) {
644 in_band_create(ofproto, ofproto->dpif, &ofproto->in_band);
646 if (ofproto->in_band) {
647 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
649 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
650 ofproto->next_in_band_update = time_msec() + 1000;
652 in_band_destroy(ofproto->in_band);
653 ofproto->in_band = NULL;
661 update_fail_open(struct ofproto *p)
663 struct ofconn *ofconn;
665 if (!hmap_is_empty(&p->controllers)
666 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
667 struct rconn **rconns;
671 p->fail_open = fail_open_create(p);
675 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
676 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
677 rconns[n++] = ofconn->rconn;
680 fail_open_set_controllers(p->fail_open, rconns, n);
681 /* p->fail_open takes ownership of 'rconns'. */
683 fail_open_destroy(p->fail_open);
689 ofproto_set_controllers(struct ofproto *p,
690 const struct ofproto_controller *controllers,
691 size_t n_controllers)
693 struct shash new_controllers;
694 struct ofconn *ofconn, *next_ofconn;
695 struct ofservice *ofservice, *next_ofservice;
698 /* Create newly configured controllers and services.
699 * Create a name to ofproto_controller mapping in 'new_controllers'. */
700 shash_init(&new_controllers);
701 for (i = 0; i < n_controllers; i++) {
702 const struct ofproto_controller *c = &controllers[i];
704 if (!vconn_verify_name(c->target)) {
705 if (!find_controller_by_target(p, c->target)) {
706 add_controller(p, c);
708 } else if (!pvconn_verify_name(c->target)) {
709 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
713 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
714 dpif_name(p->dpif), c->target);
718 shash_add_once(&new_controllers, c->target, &controllers[i]);
721 /* Delete controllers that are no longer configured.
722 * Update configuration of all now-existing controllers. */
723 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
724 struct ofproto_controller *c;
726 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
728 ofconn_destroy(ofconn);
730 update_controller(ofconn, c);
734 /* Delete services that are no longer configured.
735 * Update configuration of all now-existing services. */
736 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
737 struct ofproto_controller *c;
739 c = shash_find_data(&new_controllers,
740 pvconn_get_name(ofservice->pvconn));
742 ofservice_destroy(p, ofservice);
744 ofservice_reconfigure(ofservice, c);
748 shash_destroy(&new_controllers);
750 update_in_band_remotes(p);
755 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
757 p->fail_mode = fail_mode;
761 /* Drops the connections between 'ofproto' and all of its controllers, forcing
762 * them to reconnect. */
764 ofproto_reconnect_controllers(struct ofproto *ofproto)
766 struct ofconn *ofconn;
768 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
769 rconn_reconnect(ofconn->rconn);
774 any_extras_changed(const struct ofproto *ofproto,
775 const struct sockaddr_in *extras, size_t n)
779 if (n != ofproto->n_extra_remotes) {
783 for (i = 0; i < n; i++) {
784 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
785 const struct sockaddr_in *new = &extras[i];
787 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
788 old->sin_port != new->sin_port) {
796 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
797 * in-band control should guarantee access, in the same way that in-band
798 * control guarantees access to OpenFlow controllers. */
800 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
801 const struct sockaddr_in *extras, size_t n)
803 if (!any_extras_changed(ofproto, extras, n)) {
807 free(ofproto->extra_in_band_remotes);
808 ofproto->n_extra_remotes = n;
809 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
811 update_in_band_remotes(ofproto);
814 /* Sets the OpenFlow queue used by flows set up by in-band control on
815 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
816 * flows will use the default queue. */
818 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
820 if (queue_id != ofproto->in_band_queue) {
821 ofproto->in_band_queue = queue_id;
822 update_in_band_remotes(ofproto);
827 ofproto_set_desc(struct ofproto *p,
828 const char *mfr_desc, const char *hw_desc,
829 const char *sw_desc, const char *serial_desc,
832 struct ofp_desc_stats *ods;
835 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
836 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
837 sizeof ods->mfr_desc);
840 p->mfr_desc = xstrdup(mfr_desc);
843 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
844 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
845 sizeof ods->hw_desc);
848 p->hw_desc = xstrdup(hw_desc);
851 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
852 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
853 sizeof ods->sw_desc);
856 p->sw_desc = xstrdup(sw_desc);
859 if (strlen(serial_desc) >= sizeof ods->serial_num) {
860 VLOG_WARN("truncating serial_desc, must be less than %zu "
862 sizeof ods->serial_num);
864 free(p->serial_desc);
865 p->serial_desc = xstrdup(serial_desc);
868 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
869 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
870 sizeof ods->dp_desc);
873 p->dp_desc = xstrdup(dp_desc);
878 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
879 const struct svec *svec)
881 struct pvconn **pvconns = *pvconnsp;
882 size_t n_pvconns = *n_pvconnsp;
886 for (i = 0; i < n_pvconns; i++) {
887 pvconn_close(pvconns[i]);
891 pvconns = xmalloc(svec->n * sizeof *pvconns);
893 for (i = 0; i < svec->n; i++) {
894 const char *name = svec->names[i];
895 struct pvconn *pvconn;
898 error = pvconn_open(name, &pvconn);
900 pvconns[n_pvconns++] = pvconn;
902 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
910 *n_pvconnsp = n_pvconns;
916 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
918 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
922 ofproto_set_netflow(struct ofproto *ofproto,
923 const struct netflow_options *nf_options)
925 if (nf_options && nf_options->collectors.n) {
926 if (!ofproto->netflow) {
927 ofproto->netflow = netflow_create();
929 return netflow_set_options(ofproto->netflow, nf_options);
931 netflow_destroy(ofproto->netflow);
932 ofproto->netflow = NULL;
938 ofproto_set_sflow(struct ofproto *ofproto,
939 const struct ofproto_sflow_options *oso)
941 struct ofproto_sflow *os = ofproto->sflow;
944 struct ofport *ofport;
946 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
947 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
948 ofproto_sflow_add_port(os, ofport->odp_port,
949 netdev_get_name(ofport->netdev));
952 ofproto_sflow_set_options(os, oso);
954 ofproto_sflow_destroy(os);
955 ofproto->sflow = NULL;
959 /* Connectivity Fault Management configuration. */
961 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
963 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
965 struct ofport *ofport = get_port(ofproto, port_no);
966 if (ofport && ofport->cfm){
967 cfm_destroy(ofport->cfm);
972 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
973 * basic configuration from the configuration members in 'cfm', and the set of
974 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
975 * Ignores the statistics members of 'cfm'.
977 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
979 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
980 const struct cfm *cfm,
981 const uint16_t *remote_mps, size_t n_remote_mps)
983 struct ofport *ofport;
985 ofport = get_port(ofproto, port_no);
987 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
988 dpif_name(ofproto->dpif), port_no);
993 ofport->cfm = cfm_create();
996 ofport->cfm->mpid = cfm->mpid;
997 ofport->cfm->interval = cfm->interval;
998 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
1000 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
1002 if (!cfm_configure(ofport->cfm)) {
1003 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
1004 dpif_name(ofproto->dpif), port_no,
1005 netdev_get_name(ofport->netdev));
1006 cfm_destroy(ofport->cfm);
1011 /* Returns the connectivity fault management object associated with 'port_no'
1012 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
1013 * 'port_no' or if that port does not have CFM configured. The caller must not
1014 * modify or destroy the returned object. */
1016 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
1018 struct ofport *ofport = get_port(ofproto, port_no);
1019 return ofport ? ofport->cfm : NULL;
1023 ofproto_get_datapath_id(const struct ofproto *ofproto)
1025 return ofproto->datapath_id;
1029 ofproto_has_primary_controller(const struct ofproto *ofproto)
1031 return !hmap_is_empty(&ofproto->controllers);
1034 enum ofproto_fail_mode
1035 ofproto_get_fail_mode(const struct ofproto *p)
1037 return p->fail_mode;
1041 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1045 for (i = 0; i < ofproto->n_snoops; i++) {
1046 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1051 ofproto_destroy(struct ofproto *p)
1053 struct ofservice *ofservice, *next_ofservice;
1054 struct ofconn *ofconn, *next_ofconn;
1055 struct ofport *ofport, *next_ofport;
1062 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1064 /* Destroy fail-open and in-band early, since they touch the classifier. */
1065 fail_open_destroy(p->fail_open);
1066 p->fail_open = NULL;
1068 in_band_destroy(p->in_band);
1070 free(p->extra_in_band_remotes);
1072 ofproto_flush_flows(p);
1073 classifier_destroy(&p->cls);
1074 hmap_destroy(&p->facets);
1076 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1077 ofconn_destroy(ofconn);
1079 hmap_destroy(&p->controllers);
1081 dpif_close(p->dpif);
1082 netdev_monitor_destroy(p->netdev_monitor);
1083 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1084 hmap_remove(&p->ports, &ofport->hmap_node);
1085 ofport_free(ofport);
1087 shash_destroy(&p->port_by_name);
1089 netflow_destroy(p->netflow);
1090 ofproto_sflow_destroy(p->sflow);
1092 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1093 ofservice_destroy(p, ofservice);
1095 hmap_destroy(&p->services);
1097 for (i = 0; i < p->n_snoops; i++) {
1098 pvconn_close(p->snoops[i]);
1102 mac_learning_destroy(p->ml);
1107 free(p->serial_desc);
1110 hmap_destroy(&p->ports);
1116 ofproto_run(struct ofproto *p)
1118 int error = ofproto_run1(p);
1120 error = ofproto_run2(p, false);
1126 process_port_change(struct ofproto *ofproto, int error, char *devname)
1128 if (error == ENOBUFS) {
1129 reinit_ports(ofproto);
1130 } else if (!error) {
1131 update_port(ofproto, devname);
1136 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1137 * means that 'ofconn' is more interesting for monitoring than a lower return
1140 snoop_preference(const struct ofconn *ofconn)
1142 switch (ofconn->role) {
1143 case NX_ROLE_MASTER:
1150 /* Shouldn't happen. */
1155 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1156 * Connects this vconn to a controller. */
1158 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1160 struct ofconn *ofconn, *best;
1162 /* Pick a controller for monitoring. */
1164 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1165 if (ofconn->type == OFCONN_PRIMARY
1166 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1172 rconn_add_monitor(best->rconn, vconn);
1174 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1180 ofproto_run1(struct ofproto *p)
1182 struct ofconn *ofconn, *next_ofconn;
1183 struct ofservice *ofservice;
1184 struct ofport *ofport;
1189 if (shash_is_empty(&p->port_by_name)) {
1193 for (i = 0; i < 50; i++) {
1194 struct dpif_upcall packet;
1196 error = dpif_recv(p->dpif, &packet);
1198 if (error == ENODEV) {
1199 /* Someone destroyed the datapath behind our back. The caller
1200 * better destroy us and give up, because we're just going to
1201 * spin from here on out. */
1202 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1203 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1204 dpif_name(p->dpif));
1210 handle_upcall(p, &packet);
1213 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1214 process_port_change(p, error, devname);
1216 while ((error = netdev_monitor_poll(p->netdev_monitor,
1217 &devname)) != EAGAIN) {
1218 process_port_change(p, error, devname);
1221 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1222 ofport_run(p, ofport);
1226 if (time_msec() >= p->next_in_band_update) {
1227 update_in_band_remotes(p);
1229 in_band_run(p->in_band);
1232 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1236 /* Fail-open maintenance. Do this after processing the ofconns since
1237 * fail-open checks the status of the controller rconn. */
1239 fail_open_run(p->fail_open);
1242 HMAP_FOR_EACH (ofservice, node, &p->services) {
1243 struct vconn *vconn;
1246 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1248 struct rconn *rconn;
1251 rconn = rconn_create(ofservice->probe_interval, 0);
1252 name = ofconn_make_name(p, vconn_get_name(vconn));
1253 rconn_connect_unreliably(rconn, vconn, name);
1256 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1257 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1258 ofservice->burst_limit);
1259 } else if (retval != EAGAIN) {
1260 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1264 for (i = 0; i < p->n_snoops; i++) {
1265 struct vconn *vconn;
1268 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1270 add_snooper(p, vconn);
1271 } else if (retval != EAGAIN) {
1272 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1276 if (time_msec() >= p->next_expiration) {
1277 int delay = ofproto_expire(p);
1278 p->next_expiration = time_msec() + delay;
1279 COVERAGE_INC(ofproto_expiration);
1283 netflow_run(p->netflow);
1286 ofproto_sflow_run(p->sflow);
1293 ofproto_run2(struct ofproto *p, bool revalidate_all)
1295 /* Figure out what we need to revalidate now, if anything. */
1296 struct tag_set revalidate_set = p->revalidate_set;
1297 if (p->need_revalidate) {
1298 revalidate_all = true;
1301 /* Clear the revalidation flags. */
1302 tag_set_init(&p->revalidate_set);
1303 p->need_revalidate = false;
1305 /* Now revalidate if there's anything to do. */
1306 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1307 struct facet *facet, *next;
1309 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1311 || tag_set_intersects(&revalidate_set, facet->tags)) {
1312 facet_revalidate(p, facet);
1321 ofproto_wait(struct ofproto *p)
1323 struct ofservice *ofservice;
1324 struct ofconn *ofconn;
1325 struct ofport *ofport;
1328 dpif_recv_wait(p->dpif);
1329 dpif_port_poll_wait(p->dpif);
1330 netdev_monitor_poll_wait(p->netdev_monitor);
1331 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1332 ofport_wait(ofport);
1334 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1335 ofconn_wait(ofconn);
1338 poll_timer_wait_until(p->next_in_band_update);
1339 in_band_wait(p->in_band);
1342 fail_open_wait(p->fail_open);
1345 ofproto_sflow_wait(p->sflow);
1347 if (!tag_set_is_empty(&p->revalidate_set)) {
1348 poll_immediate_wake();
1350 if (p->need_revalidate) {
1351 /* Shouldn't happen, but if it does just go around again. */
1352 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1353 poll_immediate_wake();
1354 } else if (p->next_expiration != LLONG_MAX) {
1355 poll_timer_wait_until(p->next_expiration);
1357 HMAP_FOR_EACH (ofservice, node, &p->services) {
1358 pvconn_wait(ofservice->pvconn);
1360 for (i = 0; i < p->n_snoops; i++) {
1361 pvconn_wait(p->snoops[i]);
1366 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1368 tag_set_add(&ofproto->revalidate_set, tag);
1372 ofproto_get_revalidate_set(struct ofproto *ofproto)
1374 return &ofproto->revalidate_set;
1378 ofproto_is_alive(const struct ofproto *p)
1380 return !hmap_is_empty(&p->controllers);
1384 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1387 const struct ofconn *ofconn;
1391 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1392 const struct rconn *rconn = ofconn->rconn;
1393 time_t now = time_now();
1394 time_t last_connection = rconn_get_last_connection(rconn);
1395 time_t last_disconnect = rconn_get_last_disconnect(rconn);
1396 const int last_error = rconn_get_last_error(rconn);
1397 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1399 shash_add(info, rconn_get_target(rconn), cinfo);
1401 cinfo->is_connected = rconn_is_connected(rconn);
1402 cinfo->role = ofconn->role;
1407 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1408 cinfo->pairs.values[cinfo->pairs.n++] =
1409 xstrdup(ovs_retval_to_string(last_error));
1412 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1413 cinfo->pairs.values[cinfo->pairs.n++] =
1414 xstrdup(rconn_get_state(rconn));
1416 if (last_connection != TIME_MIN) {
1417 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
1418 cinfo->pairs.values[cinfo->pairs.n++]
1419 = xasprintf("%ld", (long int) (now - last_connection));
1422 if (last_disconnect != TIME_MIN) {
1423 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
1424 cinfo->pairs.values[cinfo->pairs.n++]
1425 = xasprintf("%ld", (long int) (now - last_disconnect));
1431 ofproto_free_ofproto_controller_info(struct shash *info)
1433 struct shash_node *node;
1435 SHASH_FOR_EACH (node, info) {
1436 struct ofproto_controller_info *cinfo = node->data;
1437 while (cinfo->pairs.n) {
1438 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1442 shash_destroy(info);
1445 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1447 * This is almost the same as calling dpif_port_del() directly on the
1448 * datapath, but it also makes 'ofproto' close its open netdev for the port
1449 * (if any). This makes it possible to create a new netdev of a different
1450 * type under the same name, which otherwise the netdev library would refuse
1451 * to do because of the conflict. (The netdev would eventually get closed on
1452 * the next trip through ofproto_run(), but this interface is more direct.)
1454 * Returns 0 if successful, otherwise a positive errno. */
1456 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1458 struct ofport *ofport = get_port(ofproto, odp_port);
1459 const char *name = ofport ? ofport->opp.name : "<unknown>";
1462 error = dpif_port_del(ofproto->dpif, odp_port);
1464 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1465 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1466 } else if (ofport) {
1467 /* 'name' is ofport->opp.name and update_port() is going to destroy
1468 * 'ofport'. Just in case update_port() refers to 'name' after it
1469 * destroys 'ofport', make a copy of it around the update_port()
1471 char *devname = xstrdup(name);
1472 update_port(ofproto, devname);
1478 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1479 * true if 'odp_port' exists and should be included, false otherwise. */
1481 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1483 struct ofport *ofport = get_port(ofproto, odp_port);
1484 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1487 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1488 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1489 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1491 * Returns 0 if successful, otherwise a positive errno value. */
1493 ofproto_send_packet(struct ofproto *ofproto,
1494 uint32_t port_no, uint16_t vlan_tci,
1495 const struct ofpbuf *packet)
1497 struct ofpbuf odp_actions;
1500 ofpbuf_init(&odp_actions, 32);
1501 if (vlan_tci != 0) {
1502 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1503 ntohs(vlan_tci & ~VLAN_CFI));
1505 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1506 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1508 ofpbuf_uninit(&odp_actions);
1511 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1512 dpif_name(ofproto->dpif), port_no, strerror(error));
1517 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1518 * performs the 'n_actions' actions in 'actions'. The new flow will not
1521 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1522 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1523 * controllers; otherwise, it will be hidden.
1525 * The caller retains ownership of 'cls_rule' and 'actions'. */
1527 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1528 const union ofp_action *actions, size_t n_actions)
1531 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1532 rule_insert(p, rule);
1536 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1540 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1543 rule_remove(ofproto, rule);
1548 ofproto_flush_flows(struct ofproto *ofproto)
1550 struct facet *facet, *next_facet;
1551 struct rule *rule, *next_rule;
1552 struct cls_cursor cursor;
1554 COVERAGE_INC(ofproto_flush);
1556 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1557 /* Mark the facet as not installed so that facet_remove() doesn't
1558 * bother trying to uninstall it. There is no point in uninstalling it
1559 * individually since we are about to blow away all the facets with
1560 * dpif_flow_flush(). */
1561 facet->installed = false;
1562 facet->dp_packet_count = 0;
1563 facet->dp_byte_count = 0;
1564 facet_remove(ofproto, facet);
1567 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1568 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1569 rule_remove(ofproto, rule);
1572 dpif_flow_flush(ofproto->dpif);
1573 if (ofproto->in_band) {
1574 in_band_flushed(ofproto->in_band);
1576 if (ofproto->fail_open) {
1577 fail_open_flushed(ofproto->fail_open);
1582 reinit_ports(struct ofproto *p)
1584 struct dpif_port_dump dump;
1585 struct shash_node *node;
1586 struct shash devnames;
1587 struct ofport *ofport;
1588 struct dpif_port dpif_port;
1590 COVERAGE_INC(ofproto_reinit_ports);
1592 shash_init(&devnames);
1593 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1594 shash_add_once (&devnames, ofport->opp.name, NULL);
1596 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1597 shash_add_once (&devnames, dpif_port.name, NULL);
1600 SHASH_FOR_EACH (node, &devnames) {
1601 update_port(p, node->name);
1603 shash_destroy(&devnames);
1606 static struct ofport *
1607 make_ofport(const struct dpif_port *dpif_port)
1609 struct netdev_options netdev_options;
1610 enum netdev_flags flags;
1611 struct ofport *ofport;
1612 struct netdev *netdev;
1615 memset(&netdev_options, 0, sizeof netdev_options);
1616 netdev_options.name = dpif_port->name;
1617 netdev_options.type = dpif_port->type;
1618 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1620 error = netdev_open(&netdev_options, &netdev);
1622 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1623 "cannot be opened (%s)",
1624 dpif_port->name, dpif_port->port_no,
1625 dpif_port->name, strerror(error));
1629 ofport = xzalloc(sizeof *ofport);
1630 ofport->netdev = netdev;
1631 ofport->odp_port = dpif_port->port_no;
1632 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1633 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1634 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1636 netdev_get_flags(netdev, &flags);
1637 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1639 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1641 netdev_get_features(netdev,
1642 &ofport->opp.curr, &ofport->opp.advertised,
1643 &ofport->opp.supported, &ofport->opp.peer);
1648 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1650 if (get_port(p, dpif_port->port_no)) {
1651 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1652 dpif_port->port_no);
1654 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1655 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1664 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1666 const struct ofp_phy_port *a = &a_->opp;
1667 const struct ofp_phy_port *b = &b_->opp;
1669 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1670 return (a->port_no == b->port_no
1671 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1672 && !strcmp(a->name, b->name)
1673 && a->state == b->state
1674 && a->config == b->config
1675 && a->curr == b->curr
1676 && a->advertised == b->advertised
1677 && a->supported == b->supported
1678 && a->peer == b->peer);
1682 send_port_status(struct ofproto *p, const struct ofport *ofport,
1685 /* XXX Should limit the number of queued port status change messages. */
1686 struct ofconn *ofconn;
1687 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1688 struct ofp_port_status *ops;
1691 /* Primary controllers, even slaves, should always get port status
1692 updates. Otherwise obey ofconn_receives_async_msgs(). */
1693 if (ofconn->type != OFCONN_PRIMARY
1694 && !ofconn_receives_async_msgs(ofconn)) {
1698 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1699 ops->reason = reason;
1700 ops->desc = ofport->opp;
1701 hton_ofp_phy_port(&ops->desc);
1702 queue_tx(b, ofconn, NULL);
1707 ofport_install(struct ofproto *p, struct ofport *ofport)
1709 const char *netdev_name = ofport->opp.name;
1711 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1712 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1713 shash_add(&p->port_by_name, netdev_name, ofport);
1715 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1720 ofport_remove(struct ofproto *p, struct ofport *ofport)
1722 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1723 hmap_remove(&p->ports, &ofport->hmap_node);
1724 shash_delete(&p->port_by_name,
1725 shash_find(&p->port_by_name, ofport->opp.name));
1727 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1732 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1735 cfm_run(ofport->cfm);
1737 if (cfm_should_send_ccm(ofport->cfm)) {
1738 struct ofpbuf packet;
1741 ofpbuf_init(&packet, 0);
1742 ccm = compose_packet(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1743 ETH_TYPE_CFM, sizeof *ccm);
1744 cfm_compose_ccm(ofport->cfm, ccm);
1745 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1746 ofpbuf_uninit(&packet);
1752 ofport_wait(struct ofport *ofport)
1755 cfm_wait(ofport->cfm);
1760 ofport_free(struct ofport *ofport)
1763 cfm_destroy(ofport->cfm);
1764 netdev_close(ofport->netdev);
1769 static struct ofport *
1770 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1772 struct ofport *port;
1774 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1775 hash_int(odp_port, 0), &ofproto->ports) {
1776 if (port->odp_port == odp_port) {
1784 update_port(struct ofproto *p, const char *devname)
1786 struct dpif_port dpif_port;
1787 struct ofport *old_ofport;
1788 struct ofport *new_ofport;
1791 COVERAGE_INC(ofproto_update_port);
1793 /* Query the datapath for port information. */
1794 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1796 /* Find the old ofport. */
1797 old_ofport = shash_find_data(&p->port_by_name, devname);
1800 /* There's no port named 'devname' but there might be a port with
1801 * the same port number. This could happen if a port is deleted
1802 * and then a new one added in its place very quickly, or if a port
1803 * is renamed. In the former case we want to send an OFPPR_DELETE
1804 * and an OFPPR_ADD, and in the latter case we want to send a
1805 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1806 * the old port's ifindex against the new port, or perhaps less
1807 * reliably but more portably by comparing the old port's MAC
1808 * against the new port's MAC. However, this code isn't that smart
1809 * and always sends an OFPPR_MODIFY (XXX). */
1810 old_ofport = get_port(p, dpif_port.port_no);
1812 } else if (error != ENOENT && error != ENODEV) {
1813 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1814 "%s", strerror(error));
1818 /* Create a new ofport. */
1819 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1821 /* Eliminate a few pathological cases. */
1822 if (!old_ofport && !new_ofport) {
1824 } else if (old_ofport && new_ofport) {
1825 /* Most of the 'config' bits are OpenFlow soft state, but
1826 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1827 * OpenFlow bits from old_ofport. (make_ofport() only sets
1828 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1829 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1831 if (ofport_equal(old_ofport, new_ofport)) {
1832 /* False alarm--no change. */
1833 ofport_free(new_ofport);
1838 /* Now deal with the normal cases. */
1840 ofport_remove(p, old_ofport);
1843 ofport_install(p, new_ofport);
1845 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1846 (!old_ofport ? OFPPR_ADD
1847 : !new_ofport ? OFPPR_DELETE
1849 ofport_free(old_ofport);
1852 dpif_port_destroy(&dpif_port);
1856 init_ports(struct ofproto *p)
1858 struct dpif_port_dump dump;
1859 struct dpif_port dpif_port;
1861 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1862 if (!ofport_conflicts(p, &dpif_port)) {
1863 struct ofport *ofport = make_ofport(&dpif_port);
1865 ofport_install(p, ofport);
1873 static struct ofconn *
1874 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1876 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1877 ofconn->ofproto = p;
1878 list_push_back(&p->all_conns, &ofconn->node);
1879 ofconn->rconn = rconn;
1880 ofconn->type = type;
1881 ofconn->flow_format = NXFF_OPENFLOW10;
1882 ofconn->role = NX_ROLE_OTHER;
1883 ofconn->packet_in_counter = rconn_packet_counter_create ();
1884 ofconn->pktbuf = NULL;
1885 ofconn->miss_send_len = 0;
1886 ofconn->reply_counter = rconn_packet_counter_create ();
1891 ofconn_destroy(struct ofconn *ofconn)
1893 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1895 if (ofconn->type == OFCONN_PRIMARY) {
1896 hmap_remove(&ofproto->controllers, &ofconn->hmap_node);
1899 list_remove(&ofconn->node);
1900 rconn_destroy(ofconn->rconn);
1901 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1902 rconn_packet_counter_destroy(ofconn->reply_counter);
1903 pktbuf_destroy(ofconn->pktbuf);
1908 ofconn_run(struct ofconn *ofconn)
1910 struct ofproto *p = ofconn_get_ofproto(ofconn);
1914 for (i = 0; i < N_SCHEDULERS; i++) {
1915 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1918 rconn_run(ofconn->rconn);
1920 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1921 /* Limit the number of iterations to prevent other tasks from
1923 for (iteration = 0; iteration < 50; iteration++) {
1924 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1929 fail_open_maybe_recover(p->fail_open);
1931 handle_openflow(ofconn, of_msg);
1932 ofpbuf_delete(of_msg);
1936 if (!rconn_is_alive(ofconn->rconn)) {
1937 ofconn_destroy(ofconn);
1942 ofconn_wait(struct ofconn *ofconn)
1946 for (i = 0; i < N_SCHEDULERS; i++) {
1947 pinsched_wait(ofconn->schedulers[i]);
1949 rconn_run_wait(ofconn->rconn);
1950 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1951 rconn_recv_wait(ofconn->rconn);
1953 COVERAGE_INC(ofproto_ofconn_stuck);
1957 /* Returns true if 'ofconn' should receive asynchronous messages. */
1959 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1961 if (ofconn->type == OFCONN_PRIMARY) {
1962 /* Primary controllers always get asynchronous messages unless they
1963 * have configured themselves as "slaves". */
1964 return ofconn->role != NX_ROLE_SLAVE;
1966 /* Service connections don't get asynchronous messages unless they have
1967 * explicitly asked for them by setting a nonzero miss send length. */
1968 return ofconn->miss_send_len > 0;
1972 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1973 * and 'target', suitable for use in log messages for identifying the
1976 * The name is dynamically allocated. The caller should free it (with free())
1977 * when it is no longer needed. */
1979 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1981 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1985 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1989 for (i = 0; i < N_SCHEDULERS; i++) {
1990 struct pinsched **s = &ofconn->schedulers[i];
1994 *s = pinsched_create(rate, burst);
1996 pinsched_set_limits(*s, rate, burst);
1999 pinsched_destroy(*s);
2005 static struct ofproto *
2006 ofconn_get_ofproto(struct ofconn *ofconn)
2008 return ofconn->ofproto;
2011 static enum nx_flow_format
2012 ofconn_get_flow_format(struct ofconn *ofconn)
2014 return ofconn->flow_format;
2018 ofconn_set_flow_format(struct ofconn *ofconn, enum nx_flow_format flow_format)
2020 ofconn->flow_format = flow_format;
2024 ofservice_reconfigure(struct ofservice *ofservice,
2025 const struct ofproto_controller *c)
2027 ofservice->probe_interval = c->probe_interval;
2028 ofservice->rate_limit = c->rate_limit;
2029 ofservice->burst_limit = c->burst_limit;
2032 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
2033 * positive errno value. */
2035 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
2037 struct ofservice *ofservice;
2038 struct pvconn *pvconn;
2041 error = pvconn_open(c->target, &pvconn);
2046 ofservice = xzalloc(sizeof *ofservice);
2047 hmap_insert(&ofproto->services, &ofservice->node,
2048 hash_string(c->target, 0));
2049 ofservice->pvconn = pvconn;
2051 ofservice_reconfigure(ofservice, c);
2057 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
2059 hmap_remove(&ofproto->services, &ofservice->node);
2060 pvconn_close(ofservice->pvconn);
2064 /* Finds and returns the ofservice within 'ofproto' that has the given
2065 * 'target', or a null pointer if none exists. */
2066 static struct ofservice *
2067 ofservice_lookup(struct ofproto *ofproto, const char *target)
2069 struct ofservice *ofservice;
2071 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
2072 &ofproto->services) {
2073 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2080 /* Returns true if 'rule' should be hidden from the controller.
2082 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2083 * (e.g. by in-band control) and are intentionally hidden from the
2086 rule_is_hidden(const struct rule *rule)
2088 return rule->cr.priority > UINT16_MAX;
2091 /* Creates and returns a new rule initialized as specified.
2093 * The caller is responsible for inserting the rule into the classifier (with
2094 * rule_insert()). */
2095 static struct rule *
2096 rule_create(const struct cls_rule *cls_rule,
2097 const union ofp_action *actions, size_t n_actions,
2098 uint16_t idle_timeout, uint16_t hard_timeout,
2099 ovs_be64 flow_cookie, bool send_flow_removed)
2101 struct rule *rule = xzalloc(sizeof *rule);
2102 rule->cr = *cls_rule;
2103 rule->idle_timeout = idle_timeout;
2104 rule->hard_timeout = hard_timeout;
2105 rule->flow_cookie = flow_cookie;
2106 rule->used = rule->created = time_msec();
2107 rule->send_flow_removed = send_flow_removed;
2108 list_init(&rule->facets);
2109 if (n_actions > 0) {
2110 rule->n_actions = n_actions;
2111 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2117 static struct rule *
2118 rule_from_cls_rule(const struct cls_rule *cls_rule)
2120 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2124 rule_free(struct rule *rule)
2126 free(rule->actions);
2130 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2131 * destroying any that no longer has a rule (which is probably all of them).
2133 * The caller must have already removed 'rule' from the classifier. */
2135 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2137 struct facet *facet, *next_facet;
2138 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2139 facet_revalidate(ofproto, facet);
2144 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2145 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2148 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2150 const union ofp_action *oa;
2151 struct actions_iterator i;
2153 if (out_port == htons(OFPP_NONE)) {
2156 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2157 oa = actions_next(&i)) {
2158 if (action_outputs_to_port(oa, out_port)) {
2165 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2166 * 'packet', which arrived on 'in_port'.
2168 * Takes ownership of 'packet'. */
2170 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2171 const struct nlattr *odp_actions, size_t actions_len,
2172 struct ofpbuf *packet)
2174 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2175 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2176 /* As an optimization, avoid a round-trip from userspace to kernel to
2177 * userspace. This also avoids possibly filling up kernel packet
2178 * buffers along the way. */
2179 struct dpif_upcall upcall;
2181 upcall.type = DPIF_UC_ACTION;
2182 upcall.packet = packet;
2185 upcall.userdata = nl_attr_get_u64(odp_actions);
2186 upcall.sample_pool = 0;
2187 upcall.actions = NULL;
2188 upcall.actions_len = 0;
2190 send_packet_in(ofproto, &upcall, flow, false);
2196 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2197 ofpbuf_delete(packet);
2202 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2203 * statistics appropriately. 'packet' must have at least sizeof(struct
2204 * ofp_packet_in) bytes of headroom.
2206 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2207 * applying flow_extract() to 'packet' would yield the same flow as
2210 * 'facet' must have accurately composed ODP actions; that is, it must not be
2211 * in need of revalidation.
2213 * Takes ownership of 'packet'. */
2215 facet_execute(struct ofproto *ofproto, struct facet *facet,
2216 struct ofpbuf *packet)
2218 struct dpif_flow_stats stats;
2220 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2222 flow_extract_stats(&facet->flow, packet, &stats);
2223 stats.used = time_msec();
2224 if (execute_odp_actions(ofproto, &facet->flow,
2225 facet->actions, facet->actions_len, packet)) {
2226 facet_update_stats(ofproto, facet, &stats);
2230 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2231 * statistics (or the statistics for one of its facets) appropriately.
2232 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2234 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2235 * with statistics for 'packet' either way.
2237 * Takes ownership of 'packet'. */
2239 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2240 struct ofpbuf *packet)
2242 struct action_xlate_ctx ctx;
2243 struct ofpbuf *odp_actions;
2244 struct facet *facet;
2248 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2250 flow_extract(packet, 0, in_port, &flow);
2252 /* First look for a related facet. If we find one, account it to that. */
2253 facet = facet_lookup_valid(ofproto, &flow);
2254 if (facet && facet->rule == rule) {
2255 facet_execute(ofproto, facet, packet);
2259 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2260 * create a new facet for it and use that. */
2261 if (rule_lookup(ofproto, &flow) == rule) {
2262 facet = facet_create(ofproto, rule, &flow, packet);
2263 facet_execute(ofproto, facet, packet);
2264 facet_install(ofproto, facet, true);
2268 /* We can't account anything to a facet. If we were to try, then that
2269 * facet would have a non-matching rule, busting our invariants. */
2270 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2271 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2272 size = packet->size;
2273 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2274 odp_actions->size, packet)) {
2275 rule->used = time_msec();
2276 rule->packet_count++;
2277 rule->byte_count += size;
2278 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
2280 ofpbuf_delete(odp_actions);
2283 /* Inserts 'rule' into 'p''s flow table. */
2285 rule_insert(struct ofproto *p, struct rule *rule)
2287 struct rule *displaced_rule;
2289 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2290 if (displaced_rule) {
2291 rule_destroy(p, displaced_rule);
2293 p->need_revalidate = true;
2296 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2297 * 'flow' and an example 'packet' within that flow.
2299 * The caller must already have determined that no facet with an identical
2300 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2301 * 'ofproto''s classifier table. */
2302 static struct facet *
2303 facet_create(struct ofproto *ofproto, struct rule *rule,
2304 const struct flow *flow, const struct ofpbuf *packet)
2306 struct facet *facet;
2308 facet = xzalloc(sizeof *facet);
2309 facet->used = time_msec();
2310 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2311 list_push_back(&rule->facets, &facet->list_node);
2313 facet->flow = *flow;
2314 netflow_flow_init(&facet->nf_flow);
2315 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2317 facet_make_actions(ofproto, facet, packet);
2323 facet_free(struct facet *facet)
2325 free(facet->actions);
2329 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2331 * - Removes 'rule' from the classifier.
2333 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2334 * destroys them), via rule_destroy().
2337 rule_remove(struct ofproto *ofproto, struct rule *rule)
2339 COVERAGE_INC(ofproto_del_rule);
2340 ofproto->need_revalidate = true;
2341 classifier_remove(&ofproto->cls, &rule->cr);
2342 rule_destroy(ofproto, rule);
2345 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2347 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2348 * rule's statistics, via facet_uninstall().
2350 * - Removes 'facet' from its rule and from ofproto->facets.
2353 facet_remove(struct ofproto *ofproto, struct facet *facet)
2355 facet_uninstall(ofproto, facet);
2356 facet_flush_stats(ofproto, facet);
2357 hmap_remove(&ofproto->facets, &facet->hmap_node);
2358 list_remove(&facet->list_node);
2362 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2364 facet_make_actions(struct ofproto *p, struct facet *facet,
2365 const struct ofpbuf *packet)
2367 const struct rule *rule = facet->rule;
2368 struct ofpbuf *odp_actions;
2369 struct action_xlate_ctx ctx;
2371 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2372 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2373 facet->tags = ctx.tags;
2374 facet->may_install = ctx.may_set_up_flow;
2375 facet->nf_flow.output_iface = ctx.nf_output_iface;
2377 if (facet->actions_len != odp_actions->size
2378 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2379 free(facet->actions);
2380 facet->actions_len = odp_actions->size;
2381 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2384 ofpbuf_delete(odp_actions);
2388 facet_put__(struct ofproto *ofproto, struct facet *facet,
2389 const struct nlattr *actions, size_t actions_len,
2390 struct dpif_flow_stats *stats)
2392 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2393 enum dpif_flow_put_flags flags;
2396 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2398 flags |= DPIF_FP_ZERO_STATS;
2399 facet->dp_packet_count = 0;
2400 facet->dp_byte_count = 0;
2403 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2404 odp_flow_key_from_flow(&key, &facet->flow);
2405 assert(key.base == keybuf);
2407 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2408 actions, actions_len, stats);
2411 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2412 * 'zero_stats' is true, clears any existing statistics from the datapath for
2415 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2417 struct dpif_flow_stats stats;
2419 if (facet->may_install
2420 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2421 zero_stats ? &stats : NULL)) {
2422 facet->installed = true;
2426 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2427 * to the accounting hook function in the ofhooks structure. */
2429 facet_account(struct ofproto *ofproto,
2430 struct facet *facet, uint64_t extra_bytes)
2432 uint64_t total_bytes = facet->byte_count + extra_bytes;
2434 if (ofproto->ofhooks->account_flow_cb
2435 && total_bytes > facet->accounted_bytes)
2437 ofproto->ofhooks->account_flow_cb(
2438 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2439 total_bytes - facet->accounted_bytes, ofproto->aux);
2440 facet->accounted_bytes = total_bytes;
2444 /* If 'rule' is installed in the datapath, uninstalls it. */
2446 facet_uninstall(struct ofproto *p, struct facet *facet)
2448 if (facet->installed) {
2449 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2450 struct dpif_flow_stats stats;
2453 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2454 odp_flow_key_from_flow(&key, &facet->flow);
2455 assert(key.base == keybuf);
2457 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2458 facet_update_stats(p, facet, &stats);
2460 facet->installed = false;
2461 facet->dp_packet_count = 0;
2462 facet->dp_byte_count = 0;
2464 assert(facet->dp_packet_count == 0);
2465 assert(facet->dp_byte_count == 0);
2469 /* Returns true if the only action for 'facet' is to send to the controller.
2470 * (We don't report NetFlow expiration messages for such facets because they
2471 * are just part of the control logic for the network, not real traffic). */
2473 facet_is_controller_flow(struct facet *facet)
2476 && facet->rule->n_actions == 1
2477 && action_outputs_to_port(&facet->rule->actions[0],
2478 htons(OFPP_CONTROLLER)));
2481 /* Folds all of 'facet''s statistics into its rule. Also updates the
2482 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2483 * 'facet''s statistics in the datapath should have been zeroed and folded into
2484 * its packet and byte counts before this function is called. */
2486 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2488 assert(!facet->dp_byte_count);
2489 assert(!facet->dp_packet_count);
2491 facet_push_stats(ofproto, facet);
2492 facet_account(ofproto, facet, 0);
2494 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2495 struct ofexpired expired;
2496 expired.flow = facet->flow;
2497 expired.packet_count = facet->packet_count;
2498 expired.byte_count = facet->byte_count;
2499 expired.used = facet->used;
2500 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2503 facet->rule->packet_count += facet->packet_count;
2504 facet->rule->byte_count += facet->byte_count;
2506 /* Reset counters to prevent double counting if 'facet' ever gets
2508 facet->packet_count = 0;
2509 facet->byte_count = 0;
2510 facet->rs_packet_count = 0;
2511 facet->rs_byte_count = 0;
2512 facet->accounted_bytes = 0;
2514 netflow_flow_clear(&facet->nf_flow);
2517 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2518 * Returns it if found, otherwise a null pointer.
2520 * The returned facet might need revalidation; use facet_lookup_valid()
2521 * instead if that is important. */
2522 static struct facet *
2523 facet_find(struct ofproto *ofproto, const struct flow *flow)
2525 struct facet *facet;
2527 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2529 if (flow_equal(flow, &facet->flow)) {
2537 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2538 * Returns it if found, otherwise a null pointer.
2540 * The returned facet is guaranteed to be valid. */
2541 static struct facet *
2542 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2544 struct facet *facet = facet_find(ofproto, flow);
2546 /* The facet we found might not be valid, since we could be in need of
2547 * revalidation. If it is not valid, don't return it. */
2549 && ofproto->need_revalidate
2550 && !facet_revalidate(ofproto, facet)) {
2551 COVERAGE_INC(ofproto_invalidated);
2558 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2560 * - If the rule found is different from 'facet''s current rule, moves
2561 * 'facet' to the new rule and recompiles its actions.
2563 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2564 * where it is and recompiles its actions anyway.
2566 * - If there is none, destroys 'facet'.
2568 * Returns true if 'facet' still exists, false if it has been destroyed. */
2570 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2572 struct action_xlate_ctx ctx;
2573 struct ofpbuf *odp_actions;
2574 struct rule *new_rule;
2575 bool actions_changed;
2577 COVERAGE_INC(facet_revalidate);
2579 /* Determine the new rule. */
2580 new_rule = rule_lookup(ofproto, &facet->flow);
2582 /* No new rule, so delete the facet. */
2583 facet_remove(ofproto, facet);
2587 /* Calculate new ODP actions.
2589 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2590 * emit a NetFlow expiration and, if so, we need to have the old state
2591 * around to properly compose it. */
2592 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2593 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2594 actions_changed = (facet->actions_len != odp_actions->size
2595 || memcmp(facet->actions, odp_actions->data,
2596 facet->actions_len));
2598 /* If the ODP actions changed or the installability changed, then we need
2599 * to talk to the datapath. */
2600 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2601 if (ctx.may_set_up_flow) {
2602 struct dpif_flow_stats stats;
2604 facet_put__(ofproto, facet,
2605 odp_actions->data, odp_actions->size, &stats);
2606 facet_update_stats(ofproto, facet, &stats);
2608 facet_uninstall(ofproto, facet);
2611 /* The datapath flow is gone or has zeroed stats, so push stats out of
2612 * 'facet' into 'rule'. */
2613 facet_flush_stats(ofproto, facet);
2616 /* Update 'facet' now that we've taken care of all the old state. */
2617 facet->tags = ctx.tags;
2618 facet->nf_flow.output_iface = ctx.nf_output_iface;
2619 facet->may_install = ctx.may_set_up_flow;
2620 if (actions_changed) {
2621 free(facet->actions);
2622 facet->actions_len = odp_actions->size;
2623 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2625 if (facet->rule != new_rule) {
2626 COVERAGE_INC(facet_changed_rule);
2627 list_remove(&facet->list_node);
2628 list_push_back(&new_rule->facets, &facet->list_node);
2629 facet->rule = new_rule;
2630 facet->used = new_rule->created;
2631 facet->rs_used = facet->used;
2634 ofpbuf_delete(odp_actions);
2640 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2641 struct rconn_packet_counter *counter)
2643 update_openflow_length(msg);
2644 if (rconn_send(ofconn->rconn, msg, counter)) {
2650 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
2652 queue_tx(msg, ofconn, ofconn->reply_counter);
2656 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2659 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2661 COVERAGE_INC(ofproto_error);
2662 ofconn_send_reply(ofconn, buf);
2667 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2669 ofconn_send_reply(ofconn, make_echo_reply(oh));
2674 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2676 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2677 struct ofp_switch_features *osf;
2679 struct ofport *port;
2681 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2682 osf->datapath_id = htonll(ofproto->datapath_id);
2683 osf->n_buffers = htonl(pktbuf_capacity());
2685 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2686 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2687 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2688 (1u << OFPAT_SET_VLAN_VID) |
2689 (1u << OFPAT_SET_VLAN_PCP) |
2690 (1u << OFPAT_STRIP_VLAN) |
2691 (1u << OFPAT_SET_DL_SRC) |
2692 (1u << OFPAT_SET_DL_DST) |
2693 (1u << OFPAT_SET_NW_SRC) |
2694 (1u << OFPAT_SET_NW_DST) |
2695 (1u << OFPAT_SET_NW_TOS) |
2696 (1u << OFPAT_SET_TP_SRC) |
2697 (1u << OFPAT_SET_TP_DST) |
2698 (1u << OFPAT_ENQUEUE));
2700 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2701 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2704 ofconn_send_reply(ofconn, buf);
2709 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2711 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2713 struct ofp_switch_config *osc;
2717 /* Figure out flags. */
2718 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2719 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2722 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2723 osc->flags = htons(flags);
2724 osc->miss_send_len = htons(ofconn->miss_send_len);
2725 ofconn_send_reply(ofconn, buf);
2731 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2733 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2734 uint16_t flags = ntohs(osc->flags);
2736 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2737 switch (flags & OFPC_FRAG_MASK) {
2738 case OFPC_FRAG_NORMAL:
2739 dpif_set_drop_frags(ofproto->dpif, false);
2741 case OFPC_FRAG_DROP:
2742 dpif_set_drop_frags(ofproto->dpif, true);
2745 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2751 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2756 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2757 struct action_xlate_ctx *ctx);
2760 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2762 const struct ofport *ofport = get_port(ctx->ofproto, port);
2765 if (ofport->opp.config & OFPPC_NO_FWD) {
2766 /* Forwarding disabled on port. */
2771 * We don't have an ofport record for this port, but it doesn't hurt to
2772 * allow forwarding to it anyhow. Maybe such a port will appear later
2773 * and we're pre-populating the flow table.
2777 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2778 ctx->nf_output_iface = port;
2781 static struct rule *
2782 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2784 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2788 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2790 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2791 uint16_t old_in_port;
2794 /* Look up a flow with 'in_port' as the input port. Then restore the
2795 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2796 * have surprising behavior). */
2797 old_in_port = ctx->flow.in_port;
2798 ctx->flow.in_port = in_port;
2799 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2800 ctx->flow.in_port = old_in_port;
2802 if (ctx->resubmit_hook) {
2803 ctx->resubmit_hook(ctx, rule);
2808 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2812 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2814 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2815 MAX_RESUBMIT_RECURSION);
2820 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2821 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2823 struct ofport *ofport;
2825 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2826 uint16_t odp_port = ofport->odp_port;
2827 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2828 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2831 *nf_output_iface = NF_OUT_FLOOD;
2835 xlate_output_action__(struct action_xlate_ctx *ctx,
2836 uint16_t port, uint16_t max_len)
2839 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2841 ctx->nf_output_iface = NF_OUT_DROP;
2845 add_output_action(ctx, ctx->flow.in_port);
2848 xlate_table_action(ctx, ctx->flow.in_port);
2851 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2852 ctx->odp_actions, &ctx->tags,
2853 &ctx->nf_output_iface,
2854 ctx->ofproto->aux)) {
2855 COVERAGE_INC(ofproto_uninstallable);
2856 ctx->may_set_up_flow = false;
2860 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2861 &ctx->nf_output_iface, ctx->odp_actions);
2864 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2865 &ctx->nf_output_iface, ctx->odp_actions);
2867 case OFPP_CONTROLLER:
2868 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2871 add_output_action(ctx, ODPP_LOCAL);
2874 odp_port = ofp_port_to_odp_port(port);
2875 if (odp_port != ctx->flow.in_port) {
2876 add_output_action(ctx, odp_port);
2881 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2882 ctx->nf_output_iface = NF_OUT_FLOOD;
2883 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2884 ctx->nf_output_iface = prev_nf_output_iface;
2885 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2886 ctx->nf_output_iface != NF_OUT_FLOOD) {
2887 ctx->nf_output_iface = NF_OUT_MULTI;
2892 xlate_output_action(struct action_xlate_ctx *ctx,
2893 const struct ofp_action_output *oao)
2895 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2898 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2899 * optimization, because we're going to add another action that sets the
2900 * priority immediately after, or because there are no actions following the
2903 remove_pop_action(struct action_xlate_ctx *ctx)
2905 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2906 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2907 ctx->last_pop_priority = -1;
2912 add_pop_action(struct action_xlate_ctx *ctx)
2914 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2915 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2916 ctx->last_pop_priority = ctx->odp_actions->size;
2921 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2922 const struct ofp_action_enqueue *oae)
2924 uint16_t ofp_port, odp_port;
2928 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2931 /* Fall back to ordinary output action. */
2932 xlate_output_action__(ctx, ntohs(oae->port), 0);
2936 /* Figure out ODP output port. */
2937 ofp_port = ntohs(oae->port);
2938 if (ofp_port != OFPP_IN_PORT) {
2939 odp_port = ofp_port_to_odp_port(ofp_port);
2941 odp_port = ctx->flow.in_port;
2944 /* Add ODP actions. */
2945 remove_pop_action(ctx);
2946 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2947 add_output_action(ctx, odp_port);
2948 add_pop_action(ctx);
2950 /* Update NetFlow output port. */
2951 if (ctx->nf_output_iface == NF_OUT_DROP) {
2952 ctx->nf_output_iface = odp_port;
2953 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2954 ctx->nf_output_iface = NF_OUT_MULTI;
2959 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2960 const struct nx_action_set_queue *nasq)
2965 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2968 /* Couldn't translate queue to a priority, so ignore. A warning
2969 * has already been logged. */
2973 remove_pop_action(ctx);
2974 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2978 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2980 ovs_be16 tci = ctx->flow.vlan_tci;
2981 if (!(tci & htons(VLAN_CFI))) {
2982 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2984 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2985 tci & ~htons(VLAN_CFI));
2989 struct xlate_reg_state {
2995 save_reg_state(const struct action_xlate_ctx *ctx,
2996 struct xlate_reg_state *state)
2998 state->vlan_tci = ctx->flow.vlan_tci;
2999 state->tun_id = ctx->flow.tun_id;
3003 update_reg_state(struct action_xlate_ctx *ctx,
3004 const struct xlate_reg_state *state)
3006 if (ctx->flow.vlan_tci != state->vlan_tci) {
3007 xlate_set_dl_tci(ctx);
3009 if (ctx->flow.tun_id != state->tun_id) {
3010 nl_msg_put_be64(ctx->odp_actions,
3011 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
3016 xlate_nicira_action(struct action_xlate_ctx *ctx,
3017 const struct nx_action_header *nah)
3019 const struct nx_action_resubmit *nar;
3020 const struct nx_action_set_tunnel *nast;
3021 const struct nx_action_set_queue *nasq;
3022 const struct nx_action_multipath *nam;
3023 enum nx_action_subtype subtype = ntohs(nah->subtype);
3024 struct xlate_reg_state state;
3027 assert(nah->vendor == htonl(NX_VENDOR_ID));
3029 case NXAST_RESUBMIT:
3030 nar = (const struct nx_action_resubmit *) nah;
3031 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
3034 case NXAST_SET_TUNNEL:
3035 nast = (const struct nx_action_set_tunnel *) nah;
3036 tun_id = htonll(ntohl(nast->tun_id));
3037 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3038 ctx->flow.tun_id = tun_id;
3041 case NXAST_DROP_SPOOFED_ARP:
3042 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
3043 nl_msg_put_flag(ctx->odp_actions,
3044 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
3048 case NXAST_SET_QUEUE:
3049 nasq = (const struct nx_action_set_queue *) nah;
3050 xlate_set_queue_action(ctx, nasq);
3053 case NXAST_POP_QUEUE:
3054 add_pop_action(ctx);
3057 case NXAST_REG_MOVE:
3058 save_reg_state(ctx, &state);
3059 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
3061 update_reg_state(ctx, &state);
3064 case NXAST_REG_LOAD:
3065 save_reg_state(ctx, &state);
3066 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
3068 update_reg_state(ctx, &state);
3072 /* Nothing to do. */
3075 case NXAST_SET_TUNNEL64:
3076 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
3077 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3078 ctx->flow.tun_id = tun_id;
3081 case NXAST_MULTIPATH:
3082 nam = (const struct nx_action_multipath *) nah;
3083 multipath_execute(nam, &ctx->flow);
3086 /* If you add a new action here that modifies flow data, don't forget to
3087 * update the flow key in ctx->flow at the same time. */
3089 case NXAST_SNAT__OBSOLETE:
3091 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3097 do_xlate_actions(const union ofp_action *in, size_t n_in,
3098 struct action_xlate_ctx *ctx)
3100 struct actions_iterator iter;
3101 const union ofp_action *ia;
3102 const struct ofport *port;
3104 port = get_port(ctx->ofproto, ctx->flow.in_port);
3105 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3106 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3107 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3108 /* Drop this flow. */
3112 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3113 enum ofp_action_type type = ntohs(ia->type);
3114 const struct ofp_action_dl_addr *oada;
3118 xlate_output_action(ctx, &ia->output);
3121 case OFPAT_SET_VLAN_VID:
3122 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3123 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3124 xlate_set_dl_tci(ctx);
3127 case OFPAT_SET_VLAN_PCP:
3128 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3129 ctx->flow.vlan_tci |= htons(
3130 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3131 xlate_set_dl_tci(ctx);
3134 case OFPAT_STRIP_VLAN:
3135 ctx->flow.vlan_tci = htons(0);
3136 xlate_set_dl_tci(ctx);
3139 case OFPAT_SET_DL_SRC:
3140 oada = ((struct ofp_action_dl_addr *) ia);
3141 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3142 oada->dl_addr, ETH_ADDR_LEN);
3143 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3146 case OFPAT_SET_DL_DST:
3147 oada = ((struct ofp_action_dl_addr *) ia);
3148 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3149 oada->dl_addr, ETH_ADDR_LEN);
3150 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3153 case OFPAT_SET_NW_SRC:
3154 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3155 ia->nw_addr.nw_addr);
3156 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3159 case OFPAT_SET_NW_DST:
3160 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3161 ia->nw_addr.nw_addr);
3162 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3165 case OFPAT_SET_NW_TOS:
3166 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3168 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3171 case OFPAT_SET_TP_SRC:
3172 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3173 ia->tp_port.tp_port);
3174 ctx->flow.tp_src = ia->tp_port.tp_port;
3177 case OFPAT_SET_TP_DST:
3178 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3179 ia->tp_port.tp_port);
3180 ctx->flow.tp_dst = ia->tp_port.tp_port;
3184 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3188 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3192 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3199 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3200 struct ofproto *ofproto, const struct flow *flow,
3201 const struct ofpbuf *packet)
3203 ctx->ofproto = ofproto;
3205 ctx->packet = packet;
3206 ctx->resubmit_hook = NULL;
3207 ctx->check_special = true;
3211 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
3212 const struct ofpbuf *packet)
3214 struct ofport *ofport;
3216 ofport = get_port(ofproto, flow->in_port);
3217 if (ofport && ofport->cfm) {
3218 cfm_process_heartbeat(ofport->cfm, packet);
3222 static struct ofpbuf *
3223 xlate_actions(struct action_xlate_ctx *ctx,
3224 const union ofp_action *in, size_t n_in)
3226 COVERAGE_INC(ofproto_ofp2odp);
3228 ctx->odp_actions = ofpbuf_new(512);
3230 ctx->may_set_up_flow = true;
3231 ctx->nf_output_iface = NF_OUT_DROP;
3233 ctx->last_pop_priority = -1;
3235 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
3237 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
3239 ctx->may_set_up_flow = false;
3240 } else if (ctx->check_special
3241 && ctx->ofproto->ofhooks->special_cb
3242 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3243 ctx->ofproto->aux)) {
3244 ctx->may_set_up_flow = false;
3246 do_xlate_actions(in, n_in, ctx);
3249 remove_pop_action(ctx);
3251 /* Check with in-band control to see if we're allowed to set up this
3253 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3254 ctx->odp_actions->data, ctx->odp_actions->size)) {
3255 ctx->may_set_up_flow = false;
3258 return ctx->odp_actions;
3261 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3262 * error message code (composed with ofp_mkerr()) for the caller to propagate
3263 * upward. Otherwise, returns 0.
3265 * The log message mentions 'msg_type'. */
3267 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3269 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3270 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3271 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3274 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3281 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3283 struct ofproto *p = ofconn_get_ofproto(ofconn);
3284 struct ofp_packet_out *opo;
3285 struct ofpbuf payload, *buffer;
3286 union ofp_action *ofp_actions;
3287 struct action_xlate_ctx ctx;
3288 struct ofpbuf *odp_actions;
3289 struct ofpbuf request;
3291 size_t n_ofp_actions;
3295 COVERAGE_INC(ofproto_packet_out);
3297 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3302 /* Get ofp_packet_out. */
3303 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3304 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3307 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3308 &ofp_actions, &n_ofp_actions);
3314 if (opo->buffer_id != htonl(UINT32_MAX)) {
3315 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3317 if (error || !buffer) {
3326 /* Extract flow, check actions. */
3327 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3329 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3335 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3336 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3337 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3338 ofpbuf_delete(odp_actions);
3341 ofpbuf_delete(buffer);
3346 update_port_config(struct ofproto *p, struct ofport *port,
3347 uint32_t config, uint32_t mask)
3349 mask &= config ^ port->opp.config;
3350 if (mask & OFPPC_PORT_DOWN) {
3351 if (config & OFPPC_PORT_DOWN) {
3352 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3354 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3357 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3358 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3359 if (mask & REVALIDATE_BITS) {
3360 COVERAGE_INC(ofproto_costly_flags);
3361 port->opp.config ^= mask & REVALIDATE_BITS;
3362 p->need_revalidate = true;
3364 #undef REVALIDATE_BITS
3365 if (mask & OFPPC_NO_PACKET_IN) {
3366 port->opp.config ^= OFPPC_NO_PACKET_IN;
3371 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3373 struct ofproto *p = ofconn_get_ofproto(ofconn);
3374 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3375 struct ofport *port;
3378 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3383 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3385 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3386 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3387 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3389 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3390 if (opm->advertise) {
3391 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3397 static struct ofpbuf *
3398 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3400 struct ofp_stats_reply *osr;
3403 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3404 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3406 osr->flags = htons(0);
3410 static struct ofpbuf *
3411 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3413 const struct ofp_stats_request *osr
3414 = (const struct ofp_stats_request *) request;
3415 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3419 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3420 struct ofpbuf **msgp)
3422 struct ofpbuf *msg = *msgp;
3423 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3424 if (nbytes + msg->size > UINT16_MAX) {
3425 struct ofp_stats_reply *reply = msg->data;
3426 reply->flags = htons(OFPSF_REPLY_MORE);
3427 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3428 ofconn_send_reply(ofconn, msg);
3430 return ofpbuf_put_uninit(*msgp, nbytes);
3433 static struct ofpbuf *
3434 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3436 struct nicira_stats_msg *nsm;
3439 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3440 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3441 nsm->type = htons(OFPST_VENDOR);
3442 nsm->flags = htons(0);
3443 nsm->vendor = htonl(NX_VENDOR_ID);
3444 nsm->subtype = subtype;
3448 static struct ofpbuf *
3449 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3451 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3455 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3456 struct ofpbuf **msgp)
3458 struct ofpbuf *msg = *msgp;
3459 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3460 if (nbytes + msg->size > UINT16_MAX) {
3461 struct nicira_stats_msg *reply = msg->data;
3462 reply->flags = htons(OFPSF_REPLY_MORE);
3463 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3464 ofconn_send_reply(ofconn, msg);
3466 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3470 handle_desc_stats_request(struct ofconn *ofconn,
3471 const struct ofp_header *request)
3473 struct ofproto *p = ofconn_get_ofproto(ofconn);
3474 struct ofp_desc_stats *ods;
3477 msg = start_ofp_stats_reply(request, sizeof *ods);
3478 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3479 memset(ods, 0, sizeof *ods);
3480 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3481 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3482 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3483 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3484 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3485 ofconn_send_reply(ofconn, msg);
3491 handle_table_stats_request(struct ofconn *ofconn,
3492 const struct ofp_header *request)
3494 struct ofproto *p = ofconn_get_ofproto(ofconn);
3495 struct ofp_table_stats *ots;
3498 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3500 /* Classifier table. */
3501 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3502 memset(ots, 0, sizeof *ots);
3503 strcpy(ots->name, "classifier");
3504 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
3505 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3506 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3507 ots->active_count = htonl(classifier_count(&p->cls));
3508 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3509 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3511 ofconn_send_reply(ofconn, msg);
3516 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3517 struct ofpbuf **msgp)
3519 struct netdev_stats stats;
3520 struct ofp_port_stats *ops;
3522 /* Intentionally ignore return value, since errors will set
3523 * 'stats' to all-1s, which is correct for OpenFlow, and
3524 * netdev_get_stats() will log errors. */
3525 netdev_get_stats(port->netdev, &stats);
3527 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3528 ops->port_no = htons(port->opp.port_no);
3529 memset(ops->pad, 0, sizeof ops->pad);
3530 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3531 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3532 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3533 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3534 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3535 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3536 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3537 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3538 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3539 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3540 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3541 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3545 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3547 struct ofproto *p = ofconn_get_ofproto(ofconn);
3548 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3549 struct ofp_port_stats *ops;
3551 struct ofport *port;
3553 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3554 if (psr->port_no != htons(OFPP_NONE)) {
3555 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3557 append_port_stat(port, ofconn, &msg);
3560 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3561 append_port_stat(port, ofconn, &msg);
3565 ofconn_send_reply(ofconn, msg);
3570 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
3572 long long int msecs = time_msec() - start;
3573 *sec = msecs / 1000;
3574 *nsec = (msecs % 1000) * (1000 * 1000);
3578 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
3582 calc_flow_duration__(start, &sec, &nsec);
3583 *sec_be = htonl(sec);
3584 *nsec_be = htonl(nsec);
3588 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3589 ovs_be16 out_port, struct ofpbuf **replyp)
3591 struct ofp_flow_stats *ofs;
3592 uint64_t packet_count, byte_count;
3594 size_t act_len, len;
3596 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3600 act_len = sizeof *rule->actions * rule->n_actions;
3601 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3603 rule_get_stats(rule, &packet_count, &byte_count);
3605 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3606 ofs->length = htons(len);
3609 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
3610 &ofs->match, rule->flow_cookie, &cookie);
3611 put_32aligned_be64(&ofs->cookie, cookie);
3612 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3613 ofs->priority = htons(rule->cr.priority);
3614 ofs->idle_timeout = htons(rule->idle_timeout);
3615 ofs->hard_timeout = htons(rule->hard_timeout);
3616 memset(ofs->pad2, 0, sizeof ofs->pad2);
3617 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3618 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3619 if (rule->n_actions > 0) {
3620 memcpy(ofs->actions, rule->actions, act_len);
3625 is_valid_table(uint8_t table_id)
3627 if (table_id == 0 || table_id == 0xff) {
3630 /* It would probably be better to reply with an error but there doesn't
3631 * seem to be any appropriate value, so that might just be
3633 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3640 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3642 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3643 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3644 struct ofpbuf *reply;
3646 COVERAGE_INC(ofproto_flows_req);
3647 reply = start_ofp_stats_reply(oh, 1024);
3648 if (is_valid_table(fsr->table_id)) {
3649 struct cls_cursor cursor;
3650 struct cls_rule target;
3653 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3655 cls_cursor_init(&cursor, &ofproto->cls, &target);
3656 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3657 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3660 ofconn_send_reply(ofconn, reply);
3666 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3667 ovs_be16 out_port, struct ofpbuf **replyp)
3669 struct nx_flow_stats *nfs;
3670 uint64_t packet_count, byte_count;
3671 size_t act_len, start_len;
3672 struct ofpbuf *reply;
3674 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3678 rule_get_stats(rule, &packet_count, &byte_count);
3680 act_len = sizeof *rule->actions * rule->n_actions;
3682 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3683 start_len = (*replyp)->size;
3686 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3689 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3690 nfs->cookie = rule->flow_cookie;
3691 nfs->priority = htons(rule->cr.priority);
3692 nfs->idle_timeout = htons(rule->idle_timeout);
3693 nfs->hard_timeout = htons(rule->hard_timeout);
3694 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3695 memset(nfs->pad2, 0, sizeof nfs->pad2);
3696 nfs->packet_count = htonll(packet_count);
3697 nfs->byte_count = htonll(byte_count);
3698 if (rule->n_actions > 0) {
3699 ofpbuf_put(reply, rule->actions, act_len);
3701 nfs->length = htons(reply->size - start_len);
3705 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3707 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3708 struct nx_flow_stats_request *nfsr;
3709 struct cls_rule target;
3710 struct ofpbuf *reply;
3714 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3716 /* Dissect the message. */
3717 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3718 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3723 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3726 COVERAGE_INC(ofproto_flows_req);
3727 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3728 if (is_valid_table(nfsr->table_id)) {
3729 struct cls_cursor cursor;
3732 cls_cursor_init(&cursor, &ofproto->cls, &target);
3733 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3734 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3737 ofconn_send_reply(ofconn, reply);
3743 flow_stats_ds(struct rule *rule, struct ds *results)
3745 uint64_t packet_count, byte_count;
3746 size_t act_len = sizeof *rule->actions * rule->n_actions;
3748 rule_get_stats(rule, &packet_count, &byte_count);
3750 ds_put_format(results, "duration=%llds, ",
3751 (time_msec() - rule->created) / 1000);
3752 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3753 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3754 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3755 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3756 cls_rule_format(&rule->cr, results);
3757 ds_put_char(results, ',');
3759 ofp_print_actions(results, &rule->actions->header, act_len);
3761 ds_put_cstr(results, "drop");
3763 ds_put_cstr(results, "\n");
3766 /* Adds a pretty-printed description of all flows to 'results', including
3767 * hidden flows (e.g., set up by in-band control). */
3769 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3771 struct cls_cursor cursor;
3774 cls_cursor_init(&cursor, &p->cls, NULL);
3775 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3776 flow_stats_ds(rule, results);
3781 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3782 ovs_be16 out_port, uint8_t table_id,
3783 struct ofp_aggregate_stats_reply *oasr)
3785 uint64_t total_packets = 0;
3786 uint64_t total_bytes = 0;
3789 COVERAGE_INC(ofproto_agg_request);
3791 if (is_valid_table(table_id)) {
3792 struct cls_cursor cursor;
3795 cls_cursor_init(&cursor, &ofproto->cls, target);
3796 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3797 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3798 uint64_t packet_count;
3799 uint64_t byte_count;
3801 rule_get_stats(rule, &packet_count, &byte_count);
3803 total_packets += packet_count;
3804 total_bytes += byte_count;
3810 oasr->flow_count = htonl(n_flows);
3811 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3812 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3813 memset(oasr->pad, 0, sizeof oasr->pad);
3817 handle_aggregate_stats_request(struct ofconn *ofconn,
3818 const struct ofp_header *oh)
3820 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3821 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3822 struct ofp_aggregate_stats_reply *reply;
3823 struct cls_rule target;
3826 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3829 msg = start_ofp_stats_reply(oh, sizeof *reply);
3830 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3831 query_aggregate_stats(ofproto, &target, request->out_port,
3832 request->table_id, reply);
3833 ofconn_send_reply(ofconn, msg);
3838 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3840 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3841 struct nx_aggregate_stats_request *request;
3842 struct ofp_aggregate_stats_reply *reply;
3843 struct cls_rule target;
3848 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3850 /* Dissect the message. */
3851 request = ofpbuf_pull(&b, sizeof *request);
3852 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3857 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3861 COVERAGE_INC(ofproto_flows_req);
3862 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3863 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3864 query_aggregate_stats(ofproto, &target, request->out_port,
3865 request->table_id, reply);
3866 ofconn_send_reply(ofconn, buf);
3871 struct queue_stats_cbdata {
3872 struct ofconn *ofconn;
3873 struct ofport *ofport;
3878 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3879 const struct netdev_queue_stats *stats)
3881 struct ofp_queue_stats *reply;
3883 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3884 reply->port_no = htons(cbdata->ofport->opp.port_no);
3885 memset(reply->pad, 0, sizeof reply->pad);
3886 reply->queue_id = htonl(queue_id);
3887 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3888 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3889 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3893 handle_queue_stats_dump_cb(uint32_t queue_id,
3894 struct netdev_queue_stats *stats,
3897 struct queue_stats_cbdata *cbdata = cbdata_;
3899 put_queue_stats(cbdata, queue_id, stats);
3903 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3904 struct queue_stats_cbdata *cbdata)
3906 cbdata->ofport = port;
3907 if (queue_id == OFPQ_ALL) {
3908 netdev_dump_queue_stats(port->netdev,
3909 handle_queue_stats_dump_cb, cbdata);
3911 struct netdev_queue_stats stats;
3913 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3914 put_queue_stats(cbdata, queue_id, &stats);
3920 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3922 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3923 const struct ofp_queue_stats_request *qsr;
3924 struct queue_stats_cbdata cbdata;
3925 struct ofport *port;
3926 unsigned int port_no;
3929 qsr = ofputil_stats_body(oh);
3931 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3934 COVERAGE_INC(ofproto_queue_req);
3936 cbdata.ofconn = ofconn;
3937 cbdata.msg = start_ofp_stats_reply(oh, 128);
3939 port_no = ntohs(qsr->port_no);
3940 queue_id = ntohl(qsr->queue_id);
3941 if (port_no == OFPP_ALL) {
3942 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3943 handle_queue_stats_for_port(port, queue_id, &cbdata);
3945 } else if (port_no < ofproto->max_ports) {
3946 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3948 handle_queue_stats_for_port(port, queue_id, &cbdata);
3951 ofpbuf_delete(cbdata.msg);
3952 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3954 ofconn_send_reply(ofconn, cbdata.msg);
3959 /* Updates 'facet''s used time. Caller is responsible for calling
3960 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3962 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3965 if (used > facet->used) {
3967 if (used > facet->rule->used) {
3968 facet->rule->used = used;
3970 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3974 /* Folds the statistics from 'stats' into the counters in 'facet'.
3976 * Because of the meaning of a facet's counters, it only makes sense to do this
3977 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3978 * packet that was sent by hand or if it represents statistics that have been
3979 * cleared out of the datapath. */
3981 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3982 const struct dpif_flow_stats *stats)
3984 if (stats->n_packets || stats->used > facet->used) {
3985 facet_update_time(ofproto, facet, stats->used);
3986 facet->packet_count += stats->n_packets;
3987 facet->byte_count += stats->n_bytes;
3988 facet_push_stats(ofproto, facet);
3989 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3994 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3996 uint64_t rs_packets, rs_bytes;
3998 assert(facet->packet_count >= facet->rs_packet_count);
3999 assert(facet->byte_count >= facet->rs_byte_count);
4000 assert(facet->used >= facet->rs_used);
4002 rs_packets = facet->packet_count - facet->rs_packet_count;
4003 rs_bytes = facet->byte_count - facet->rs_byte_count;
4005 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
4006 facet->rs_packet_count = facet->packet_count;
4007 facet->rs_byte_count = facet->byte_count;
4008 facet->rs_used = facet->used;
4010 flow_push_stats(ofproto, facet->rule, &facet->flow,
4011 rs_packets, rs_bytes, facet->used);
4015 struct ofproto_push {
4016 struct action_xlate_ctx ctx;
4023 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4025 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
4028 rule->packet_count += push->packets;
4029 rule->byte_count += push->bytes;
4030 rule->used = MAX(push->used, rule->used);
4034 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4035 * 'rule''s actions. */
4037 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
4038 struct flow *flow, uint64_t packets, uint64_t bytes,
4041 struct ofproto_push push;
4043 push.packets = packets;
4047 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
4048 push.ctx.resubmit_hook = push_resubmit;
4049 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
4052 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4053 * in which no matching flow already exists in the flow table.
4055 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4056 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
4057 * OpenFlow error code as encoded by ofp_mkerr() on failure.
4059 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4062 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
4064 struct ofproto *p = ofconn_get_ofproto(ofconn);
4065 struct ofpbuf *packet;
4070 if (fm->flags & OFPFF_CHECK_OVERLAP
4071 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
4072 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
4076 if (fm->buffer_id != UINT32_MAX) {
4077 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
4081 in_port = UINT16_MAX;
4084 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
4085 fm->idle_timeout, fm->hard_timeout, fm->cookie,
4086 fm->flags & OFPFF_SEND_FLOW_REM);
4087 rule_insert(p, rule);
4089 rule_execute(p, rule, in_port, packet);
4094 static struct rule *
4095 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
4097 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
4101 send_buffered_packet(struct ofconn *ofconn,
4102 struct rule *rule, uint32_t buffer_id)
4104 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4105 struct ofpbuf *packet;
4109 if (buffer_id == UINT32_MAX) {
4113 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
4118 rule_execute(ofproto, rule, in_port, packet);
4123 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4125 struct modify_flows_cbdata {
4126 struct ofproto *ofproto;
4127 const struct flow_mod *fm;
4131 static int modify_flow(struct ofproto *, const struct flow_mod *,
4134 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4135 * encoded by ofp_mkerr() on failure.
4137 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4140 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
4142 struct ofproto *p = ofconn_get_ofproto(ofconn);
4143 struct rule *match = NULL;
4144 struct cls_cursor cursor;
4147 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4148 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4149 if (!rule_is_hidden(rule)) {
4151 modify_flow(p, fm, rule);
4156 /* This credits the packet to whichever flow happened to match last.
4157 * That's weird. Maybe we should do a lookup for the flow that
4158 * actually matches the packet? Who knows. */
4159 send_buffered_packet(ofconn, match, fm->buffer_id);
4162 return add_flow(ofconn, fm);
4166 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4167 * code as encoded by ofp_mkerr() on failure.
4169 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4172 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4174 struct ofproto *p = ofconn_get_ofproto(ofconn);
4175 struct rule *rule = find_flow_strict(p, fm);
4176 if (rule && !rule_is_hidden(rule)) {
4177 modify_flow(p, fm, rule);
4178 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4180 return add_flow(ofconn, fm);
4184 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4185 * been identified as a flow in 'p''s flow table to be modified, by changing
4186 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4187 * ofp_action[] structures). */
4189 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4191 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4193 rule->flow_cookie = fm->cookie;
4195 /* If the actions are the same, do nothing. */
4196 if (fm->n_actions == rule->n_actions
4198 || !memcmp(fm->actions, rule->actions, actions_len))) {
4202 /* Replace actions. */
4203 free(rule->actions);
4204 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4205 rule->n_actions = fm->n_actions;
4207 p->need_revalidate = true;
4212 /* OFPFC_DELETE implementation. */
4214 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4216 /* Implements OFPFC_DELETE. */
4218 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4220 struct rule *rule, *next_rule;
4221 struct cls_cursor cursor;
4223 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4224 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4225 delete_flow(p, rule, htons(fm->out_port));
4229 /* Implements OFPFC_DELETE_STRICT. */
4231 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4233 struct rule *rule = find_flow_strict(p, fm);
4235 delete_flow(p, rule, htons(fm->out_port));
4239 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4240 * been identified as a flow to delete from 'p''s flow table, by deleting the
4241 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4244 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4245 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4246 * specified 'out_port'. */
4248 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4250 if (rule_is_hidden(rule)) {
4254 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4258 rule_send_removed(p, rule, OFPRR_DELETE);
4259 rule_remove(p, rule);
4263 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4265 struct ofproto *p = ofconn_get_ofproto(ofconn);
4269 error = reject_slave_controller(ofconn, "flow_mod");
4274 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
4279 /* We do not support the emergency flow cache. It will hopefully get
4280 * dropped from OpenFlow in the near future. */
4281 if (fm.flags & OFPFF_EMERG) {
4282 /* There isn't a good fit for an error code, so just state that the
4283 * flow table is full. */
4284 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4287 error = validate_actions(fm.actions, fm.n_actions,
4288 &fm.cr.flow, p->max_ports);
4293 switch (fm.command) {
4295 return add_flow(ofconn, &fm);
4298 return modify_flows_loose(ofconn, &fm);
4300 case OFPFC_MODIFY_STRICT:
4301 return modify_flow_strict(ofconn, &fm);
4304 delete_flows_loose(p, &fm);
4307 case OFPFC_DELETE_STRICT:
4308 delete_flow_strict(p, &fm);
4312 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4317 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4319 const struct nxt_tun_id_cookie *msg
4320 = (const struct nxt_tun_id_cookie *) oh;
4321 enum nx_flow_format flow_format;
4323 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4324 ofconn_set_flow_format(ofconn, flow_format);
4330 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4332 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4333 struct nx_role_request *reply;
4337 if (ofconn->type != OFCONN_PRIMARY) {
4338 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4340 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4343 role = ntohl(nrr->role);
4344 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4345 && role != NX_ROLE_SLAVE) {
4346 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4348 /* There's no good error code for this. */
4349 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4352 if (role == NX_ROLE_MASTER) {
4353 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4354 struct ofconn *other;
4356 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
4357 if (other->role == NX_ROLE_MASTER) {
4358 other->role = NX_ROLE_SLAVE;
4362 ofconn->role = role;
4364 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4365 reply->role = htonl(role);
4366 ofconn_send_reply(ofconn, buf);
4372 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4374 const struct nxt_set_flow_format *msg
4375 = (const struct nxt_set_flow_format *) oh;
4378 format = ntohl(msg->format);
4379 if (format == NXFF_OPENFLOW10
4380 || format == NXFF_TUN_ID_FROM_COOKIE
4381 || format == NXFF_NXM) {
4382 ofconn_set_flow_format(ofconn, format);
4385 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4390 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4392 struct ofp_header *ob;
4395 /* Currently, everything executes synchronously, so we can just
4396 * immediately send the barrier reply. */
4397 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4398 ofconn_send_reply(ofconn, buf);
4403 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4405 const struct ofp_header *oh = msg->data;
4406 const struct ofputil_msg_type *type;
4409 error = ofputil_decode_msg_type(oh, &type);
4414 switch (ofputil_msg_type_code(type)) {
4415 /* OpenFlow requests. */
4416 case OFPUTIL_OFPT_ECHO_REQUEST:
4417 return handle_echo_request(ofconn, oh);
4419 case OFPUTIL_OFPT_FEATURES_REQUEST:
4420 return handle_features_request(ofconn, oh);
4422 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4423 return handle_get_config_request(ofconn, oh);
4425 case OFPUTIL_OFPT_SET_CONFIG:
4426 return handle_set_config(ofconn, msg->data);
4428 case OFPUTIL_OFPT_PACKET_OUT:
4429 return handle_packet_out(ofconn, oh);
4431 case OFPUTIL_OFPT_PORT_MOD:
4432 return handle_port_mod(ofconn, oh);
4434 case OFPUTIL_OFPT_FLOW_MOD:
4435 return handle_flow_mod(ofconn, oh);
4437 case OFPUTIL_OFPT_BARRIER_REQUEST:
4438 return handle_barrier_request(ofconn, oh);
4440 /* OpenFlow replies. */
4441 case OFPUTIL_OFPT_ECHO_REPLY:
4444 /* Nicira extension requests. */
4445 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4446 return handle_tun_id_from_cookie(ofconn, oh);
4448 case OFPUTIL_NXT_ROLE_REQUEST:
4449 return handle_role_request(ofconn, oh);
4451 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4452 return handle_nxt_set_flow_format(ofconn, oh);
4454 case OFPUTIL_NXT_FLOW_MOD:
4455 return handle_flow_mod(ofconn, oh);
4457 /* OpenFlow statistics requests. */
4458 case OFPUTIL_OFPST_DESC_REQUEST:
4459 return handle_desc_stats_request(ofconn, oh);
4461 case OFPUTIL_OFPST_FLOW_REQUEST:
4462 return handle_flow_stats_request(ofconn, oh);
4464 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4465 return handle_aggregate_stats_request(ofconn, oh);
4467 case OFPUTIL_OFPST_TABLE_REQUEST:
4468 return handle_table_stats_request(ofconn, oh);
4470 case OFPUTIL_OFPST_PORT_REQUEST:
4471 return handle_port_stats_request(ofconn, oh);
4473 case OFPUTIL_OFPST_QUEUE_REQUEST:
4474 return handle_queue_stats_request(ofconn, oh);
4476 /* Nicira extension statistics requests. */
4477 case OFPUTIL_NXST_FLOW_REQUEST:
4478 return handle_nxst_flow(ofconn, oh);
4480 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4481 return handle_nxst_aggregate(ofconn, oh);
4483 case OFPUTIL_INVALID:
4484 case OFPUTIL_OFPT_HELLO:
4485 case OFPUTIL_OFPT_ERROR:
4486 case OFPUTIL_OFPT_FEATURES_REPLY:
4487 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4488 case OFPUTIL_OFPT_PACKET_IN:
4489 case OFPUTIL_OFPT_FLOW_REMOVED:
4490 case OFPUTIL_OFPT_PORT_STATUS:
4491 case OFPUTIL_OFPT_BARRIER_REPLY:
4492 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4493 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4494 case OFPUTIL_OFPST_DESC_REPLY:
4495 case OFPUTIL_OFPST_FLOW_REPLY:
4496 case OFPUTIL_OFPST_QUEUE_REPLY:
4497 case OFPUTIL_OFPST_PORT_REPLY:
4498 case OFPUTIL_OFPST_TABLE_REPLY:
4499 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4500 case OFPUTIL_NXT_ROLE_REPLY:
4501 case OFPUTIL_NXT_FLOW_REMOVED:
4502 case OFPUTIL_NXST_FLOW_REPLY:
4503 case OFPUTIL_NXST_AGGREGATE_REPLY:
4505 if (VLOG_IS_WARN_ENABLED()) {
4506 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4507 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4510 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4511 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4513 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4519 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4521 int error = handle_openflow__(ofconn, ofp_msg);
4523 send_error_oh(ofconn, ofp_msg->data, error);
4525 COVERAGE_INC(ofproto_recv_openflow);
4529 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4531 struct facet *facet;
4534 /* Obtain in_port and tun_id, at least. */
4535 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4537 /* Set header pointers in 'flow'. */
4538 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4540 if (cfm_should_process_flow(&flow)) {
4541 ofproto_process_cfm(p, &flow, upcall->packet);
4542 ofpbuf_delete(upcall->packet);
4544 } else if (p->ofhooks->special_cb
4545 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4546 ofpbuf_delete(upcall->packet);
4550 /* Check with in-band control to see if this packet should be sent
4551 * to the local port regardless of the flow table. */
4552 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4553 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
4556 facet = facet_lookup_valid(p, &flow);
4558 struct rule *rule = rule_lookup(p, &flow);
4560 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4561 struct ofport *port = get_port(p, flow.in_port);
4563 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4564 COVERAGE_INC(ofproto_no_packet_in);
4565 /* XXX install 'drop' flow entry */
4566 ofpbuf_delete(upcall->packet);
4570 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4574 COVERAGE_INC(ofproto_packet_in);
4575 send_packet_in(p, upcall, &flow, false);
4579 facet = facet_create(p, rule, &flow, upcall->packet);
4580 } else if (!facet->may_install) {
4581 /* The facet is not installable, that is, we need to process every
4582 * packet, so process the current packet's actions into 'facet'. */
4583 facet_make_actions(p, facet, upcall->packet);
4586 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4588 * Extra-special case for fail-open mode.
4590 * We are in fail-open mode and the packet matched the fail-open rule,
4591 * but we are connected to a controller too. We should send the packet
4592 * up to the controller in the hope that it will try to set up a flow
4593 * and thereby allow us to exit fail-open.
4595 * See the top-level comment in fail-open.c for more information.
4597 send_packet_in(p, upcall, &flow, true);
4600 facet_execute(p, facet, upcall->packet);
4601 facet_install(p, facet, false);
4605 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4609 switch (upcall->type) {
4610 case DPIF_UC_ACTION:
4611 COVERAGE_INC(ofproto_ctlr_action);
4612 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4613 send_packet_in(p, upcall, &flow, false);
4616 case DPIF_UC_SAMPLE:
4618 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4619 ofproto_sflow_received(p->sflow, upcall, &flow);
4621 ofpbuf_delete(upcall->packet);
4625 handle_miss_upcall(p, upcall);
4628 case DPIF_N_UC_TYPES:
4630 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4635 /* Flow expiration. */
4637 static int ofproto_dp_max_idle(const struct ofproto *);
4638 static void ofproto_update_stats(struct ofproto *);
4639 static void rule_expire(struct ofproto *, struct rule *);
4640 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4642 /* This function is called periodically by ofproto_run(). Its job is to
4643 * collect updates for the flows that have been installed into the datapath,
4644 * most importantly when they last were used, and then use that information to
4645 * expire flows that have not been used recently.
4647 * Returns the number of milliseconds after which it should be called again. */
4649 ofproto_expire(struct ofproto *ofproto)
4651 struct rule *rule, *next_rule;
4652 struct cls_cursor cursor;
4655 /* Update stats for each flow in the datapath. */
4656 ofproto_update_stats(ofproto);
4658 /* Expire facets that have been idle too long. */
4659 dp_max_idle = ofproto_dp_max_idle(ofproto);
4660 ofproto_expire_facets(ofproto, dp_max_idle);
4662 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4663 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4664 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4665 rule_expire(ofproto, rule);
4668 /* Let the hook know that we're at a stable point: all outstanding data
4669 * in existing flows has been accounted to the account_cb. Thus, the
4670 * hook can now reasonably do operations that depend on having accurate
4671 * flow volume accounting (currently, that's just bond rebalancing). */
4672 if (ofproto->ofhooks->account_checkpoint_cb) {
4673 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4676 return MIN(dp_max_idle, 1000);
4679 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4681 * This function also pushes statistics updates to rules which each facet
4682 * resubmits into. Generally these statistics will be accurate. However, if a
4683 * facet changes the rule it resubmits into at some time in between
4684 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4685 * old rule will be incorrectly attributed to the new rule. This could be
4686 * avoided by calling ofproto_update_stats() whenever rules are created or
4687 * deleted. However, the performance impact of making so many calls to the
4688 * datapath do not justify the benefit of having perfectly accurate statistics.
4691 ofproto_update_stats(struct ofproto *p)
4693 const struct dpif_flow_stats *stats;
4694 struct dpif_flow_dump dump;
4695 const struct nlattr *key;
4698 dpif_flow_dump_start(&dump, p->dpif);
4699 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4700 struct facet *facet;
4703 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4707 odp_flow_key_format(key, key_len, &s);
4708 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4714 facet = facet_find(p, &flow);
4716 if (facet && facet->installed) {
4718 if (stats->n_packets >= facet->dp_packet_count) {
4719 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4721 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4724 if (stats->n_bytes >= facet->dp_byte_count) {
4725 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4727 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4730 facet->dp_packet_count = stats->n_packets;
4731 facet->dp_byte_count = stats->n_bytes;
4733 facet_update_time(p, facet, stats->used);
4734 facet_account(p, facet, stats->n_bytes);
4735 facet_push_stats(p, facet);
4737 /* There's a flow in the datapath that we know nothing about.
4739 COVERAGE_INC(ofproto_unexpected_rule);
4740 dpif_flow_del(p->dpif, key, key_len, NULL);
4743 dpif_flow_dump_done(&dump);
4746 /* Calculates and returns the number of milliseconds of idle time after which
4747 * facets should expire from the datapath and we should fold their statistics
4748 * into their parent rules in userspace. */
4750 ofproto_dp_max_idle(const struct ofproto *ofproto)
4753 * Idle time histogram.
4755 * Most of the time a switch has a relatively small number of facets. When
4756 * this is the case we might as well keep statistics for all of them in
4757 * userspace and to cache them in the kernel datapath for performance as
4760 * As the number of facets increases, the memory required to maintain
4761 * statistics about them in userspace and in the kernel becomes
4762 * significant. However, with a large number of facets it is likely that
4763 * only a few of them are "heavy hitters" that consume a large amount of
4764 * bandwidth. At this point, only heavy hitters are worth caching in the
4765 * kernel and maintaining in userspaces; other facets we can discard.
4767 * The technique used to compute the idle time is to build a histogram with
4768 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4769 * that is installed in the kernel gets dropped in the appropriate bucket.
4770 * After the histogram has been built, we compute the cutoff so that only
4771 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4772 * cached. At least the most-recently-used bucket of facets is kept, so
4773 * actually an arbitrary number of facets can be kept in any given
4774 * expiration run (though the next run will delete most of those unless
4775 * they receive additional data).
4777 * This requires a second pass through the facets, in addition to the pass
4778 * made by ofproto_update_stats(), because the former function never looks
4779 * at uninstallable facets.
4781 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4782 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4783 int buckets[N_BUCKETS] = { 0 };
4784 struct facet *facet;
4789 total = hmap_count(&ofproto->facets);
4790 if (total <= 1000) {
4791 return N_BUCKETS * BUCKET_WIDTH;
4794 /* Build histogram. */
4796 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4797 long long int idle = now - facet->used;
4798 int bucket = (idle <= 0 ? 0
4799 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4800 : (unsigned int) idle / BUCKET_WIDTH);
4804 /* Find the first bucket whose flows should be expired. */
4805 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4806 if (buckets[bucket]) {
4809 subtotal += buckets[bucket++];
4810 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4815 if (VLOG_IS_DBG_ENABLED()) {
4819 ds_put_cstr(&s, "keep");
4820 for (i = 0; i < N_BUCKETS; i++) {
4822 ds_put_cstr(&s, ", drop");
4825 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4828 VLOG_INFO("%s: %s (msec:count)",
4829 dpif_name(ofproto->dpif), ds_cstr(&s));
4833 return bucket * BUCKET_WIDTH;
4837 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4839 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4840 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4841 struct ofexpired expired;
4843 if (facet->installed) {
4844 struct dpif_flow_stats stats;
4846 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4848 facet_update_stats(ofproto, facet, &stats);
4851 expired.flow = facet->flow;
4852 expired.packet_count = facet->packet_count;
4853 expired.byte_count = facet->byte_count;
4854 expired.used = facet->used;
4855 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4860 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4862 long long int cutoff = time_msec() - dp_max_idle;
4863 struct facet *facet, *next_facet;
4865 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4866 facet_active_timeout(ofproto, facet);
4867 if (facet->used < cutoff) {
4868 facet_remove(ofproto, facet);
4873 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4874 * then delete it entirely. */
4876 rule_expire(struct ofproto *ofproto, struct rule *rule)
4878 struct facet *facet, *next_facet;
4882 /* Has 'rule' expired? */
4884 if (rule->hard_timeout
4885 && now > rule->created + rule->hard_timeout * 1000) {
4886 reason = OFPRR_HARD_TIMEOUT;
4887 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4888 && now >rule->used + rule->idle_timeout * 1000) {
4889 reason = OFPRR_IDLE_TIMEOUT;
4894 COVERAGE_INC(ofproto_expired);
4896 /* Update stats. (This is a no-op if the rule expired due to an idle
4897 * timeout, because that only happens when the rule has no facets left.) */
4898 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4899 facet_remove(ofproto, facet);
4902 /* Get rid of the rule. */
4903 if (!rule_is_hidden(rule)) {
4904 rule_send_removed(ofproto, rule, reason);
4906 rule_remove(ofproto, rule);
4910 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4912 struct ofputil_flow_removed fr;
4913 struct ofconn *ofconn;
4915 if (!rule->send_flow_removed) {
4920 fr.cookie = rule->flow_cookie;
4922 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4923 fr.idle_timeout = rule->idle_timeout;
4924 fr.packet_count = rule->packet_count;
4925 fr.byte_count = rule->byte_count;
4927 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4930 if (!rconn_is_connected(ofconn->rconn)
4931 || !ofconn_receives_async_msgs(ofconn)) {
4935 /* This accounts flow expirations as if they were replies to OpenFlow
4936 * requests. That works because preventing OpenFlow requests from
4937 * being processed also prevents new flows from being added (and
4938 * expiring). (It also prevents processing OpenFlow requests that
4939 * would not add new flows, so it is imperfect.) */
4940 msg = ofputil_encode_flow_removed(&fr, ofconn_get_flow_format(ofconn));
4941 ofconn_send_reply(ofconn, msg);
4945 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4946 * The returned statistics include statistics for all of 'rule''s facets. */
4948 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4951 struct facet *facet;
4953 /* Start from historical data for 'rule' itself that are no longer tracked
4954 * in facets. This counts, for example, facets that have expired. */
4955 p = rule->packet_count;
4956 b = rule->byte_count;
4958 /* Add any statistics that are tracked by facets. This includes
4959 * statistical data recently updated by ofproto_update_stats() as well as
4960 * stats for packets that were executed "by hand" via dpif_execute(). */
4961 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4962 p += facet->packet_count;
4963 b += facet->byte_count;
4970 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4972 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4974 struct ofconn *ofconn = ofconn_;
4976 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4977 ofconn->packet_in_counter, 100);
4980 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4981 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4982 * scheduler for sending.
4984 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4985 * Otherwise, ownership is transferred to this function. */
4987 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4988 const struct flow *flow, bool clone)
4990 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4991 struct ofputil_packet_in pin;
4994 /* Figure out the easy parts. */
4995 pin.packet = upcall->packet;
4996 pin.in_port = odp_port_to_ofp_port(flow->in_port);
4997 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4999 /* Get OpenFlow buffer_id. */
5000 if (upcall->type == DPIF_UC_ACTION) {
5001 pin.buffer_id = UINT32_MAX;
5002 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
5003 pin.buffer_id = pktbuf_get_null();
5004 } else if (!ofconn->pktbuf) {
5005 pin.buffer_id = UINT32_MAX;
5007 pin.buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet,
5011 /* Figure out how much of the packet to send. */
5012 pin.send_len = upcall->packet->size;
5013 if (pin.buffer_id != UINT32_MAX) {
5014 pin.send_len = MIN(pin.send_len, ofconn->miss_send_len);
5016 if (upcall->type == DPIF_UC_ACTION) {
5017 pin.send_len = MIN(pin.send_len, upcall->userdata);
5020 /* Make OFPT_PACKET_IN and hand over to packet scheduler. It might
5021 * immediately call into do_send_packet_in() or it might buffer it for a
5022 * while (until a later call to pinsched_run()). */
5023 msg = ofputil_encode_packet_in(&pin, clone ? NULL : upcall->packet);
5024 pinsched_send(ofconn->schedulers[upcall->type == DPIF_UC_MISS ? 0 : 1],
5025 flow->in_port, msg, do_send_packet_in, ofconn);
5028 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
5029 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
5030 * their individual configurations.
5032 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
5033 * Otherwise, ownership is transferred to this function. */
5035 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
5036 const struct flow *flow, bool clone)
5038 struct ofconn *ofconn, *prev;
5041 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5042 if (ofconn_receives_async_msgs(ofconn)) {
5044 schedule_packet_in(prev, upcall, flow, true);
5050 schedule_packet_in(prev, upcall, flow, clone);
5051 } else if (!clone) {
5052 ofpbuf_delete(upcall->packet);
5057 pick_datapath_id(const struct ofproto *ofproto)
5059 const struct ofport *port;
5061 port = get_port(ofproto, ODPP_LOCAL);
5063 uint8_t ea[ETH_ADDR_LEN];
5066 error = netdev_get_etheraddr(port->netdev, ea);
5068 return eth_addr_to_uint64(ea);
5070 VLOG_WARN("could not get MAC address for %s (%s)",
5071 netdev_get_name(port->netdev), strerror(error));
5073 return ofproto->fallback_dpid;
5077 pick_fallback_dpid(void)
5079 uint8_t ea[ETH_ADDR_LEN];
5080 eth_addr_nicira_random(ea);
5081 return eth_addr_to_uint64(ea);
5085 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
5086 void *aux OVS_UNUSED)
5088 const struct shash_node *node;
5092 SHASH_FOR_EACH (node, &all_ofprotos) {
5093 ds_put_format(&results, "%s\n", node->name);
5095 unixctl_command_reply(conn, 200, ds_cstr(&results));
5096 ds_destroy(&results);
5099 struct ofproto_trace {
5100 struct action_xlate_ctx ctx;
5106 trace_format_rule(struct ds *result, int level, const struct rule *rule)
5108 ds_put_char_multiple(result, '\t', level);
5110 ds_put_cstr(result, "No match\n");
5114 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
5115 ntohll(rule->flow_cookie));
5116 cls_rule_format(&rule->cr, result);
5117 ds_put_char(result, '\n');
5119 ds_put_char_multiple(result, '\t', level);
5120 ds_put_cstr(result, "OpenFlow ");
5121 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
5122 rule->n_actions * sizeof *rule->actions);
5123 ds_put_char(result, '\n');
5127 trace_format_flow(struct ds *result, int level, const char *title,
5128 struct ofproto_trace *trace)
5130 ds_put_char_multiple(result, '\t', level);
5131 ds_put_format(result, "%s: ", title);
5132 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5133 ds_put_cstr(result, "unchanged");
5135 flow_format(result, &trace->ctx.flow);
5136 trace->flow = trace->ctx.flow;
5138 ds_put_char(result, '\n');
5142 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
5144 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5145 struct ds *result = trace->result;
5147 ds_put_char(result, '\n');
5148 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5149 trace_format_rule(result, ctx->recurse + 1, rule);
5153 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5154 void *aux OVS_UNUSED)
5156 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5157 char *args = xstrdup(args_);
5158 char *save_ptr = NULL;
5159 struct ofproto *ofproto;
5160 struct ofpbuf packet;
5168 ofpbuf_init(&packet, strlen(args) / 2);
5171 dpname = strtok_r(args, " ", &save_ptr);
5172 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5173 in_port_s = strtok_r(NULL, " ", &save_ptr);
5174 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5175 if (!dpname || !in_port_s || !packet_s) {
5176 unixctl_command_reply(conn, 501, "Bad command syntax");
5180 ofproto = shash_find_data(&all_ofprotos, dpname);
5182 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5187 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5188 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5190 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5191 packet_s += strspn(packet_s, " ");
5192 if (*packet_s != '\0') {
5193 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5196 if (packet.size < ETH_HEADER_LEN) {
5197 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5201 ds_put_cstr(&result, "Packet: ");
5202 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5203 ds_put_cstr(&result, s);
5206 flow_extract(&packet, tun_id, in_port, &flow);
5207 ds_put_cstr(&result, "Flow: ");
5208 flow_format(&result, &flow);
5209 ds_put_char(&result, '\n');
5211 rule = rule_lookup(ofproto, &flow);
5212 trace_format_rule(&result, 0, rule);
5214 struct ofproto_trace trace;
5215 struct ofpbuf *odp_actions;
5217 trace.result = &result;
5219 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5220 trace.ctx.resubmit_hook = trace_resubmit;
5221 odp_actions = xlate_actions(&trace.ctx,
5222 rule->actions, rule->n_actions);
5224 ds_put_char(&result, '\n');
5225 trace_format_flow(&result, 0, "Final flow", &trace);
5226 ds_put_cstr(&result, "Datapath actions: ");
5227 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5228 ofpbuf_delete(odp_actions);
5231 unixctl_command_reply(conn, 200, ds_cstr(&result));
5234 ds_destroy(&result);
5235 ofpbuf_uninit(&packet);
5240 ofproto_unixctl_init(void)
5242 static bool registered;
5248 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5249 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5253 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5254 struct ofpbuf *odp_actions, tag_type *tags,
5255 uint16_t *nf_output_iface, void *ofproto_)
5257 struct ofproto *ofproto = ofproto_;
5258 struct mac_entry *dst_mac;
5260 /* Drop frames for reserved multicast addresses. */
5261 if (eth_addr_is_reserved(flow->dl_dst)) {
5265 /* Learn source MAC (but don't try to learn from revalidation). */
5267 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
5268 struct mac_entry *src_mac;
5270 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
5271 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
5272 /* The log messages here could actually be useful in debugging,
5273 * so keep the rate limit relatively high. */
5274 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5275 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5276 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5278 ofproto_revalidate(ofproto,
5279 mac_learning_changed(ofproto->ml, src_mac));
5280 src_mac->port.i = flow->in_port;
5284 /* Determine output port. */
5285 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
5287 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5288 nf_output_iface, odp_actions);
5290 int out_port = dst_mac->port.i;
5291 if (out_port != flow->in_port) {
5292 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5293 *nf_output_iface = out_port;
5302 static const struct ofhooks default_ofhooks = {
5303 default_normal_ofhook_cb,