2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
29 #include "classifier.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "multipath.h"
44 #include "ofp-print.h"
46 #include "ofproto-sflow.h"
48 #include "openflow/nicira-ext.h"
49 #include "openflow/openflow.h"
50 #include "openvswitch/datapath-protocol.h"
54 #include "poll-loop.h"
57 #include "stream-ssl.h"
61 #include "unaligned.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(odp_overflow);
71 COVERAGE_DEFINE(ofproto_agg_request);
72 COVERAGE_DEFINE(ofproto_costly_flags);
73 COVERAGE_DEFINE(ofproto_ctlr_action);
74 COVERAGE_DEFINE(ofproto_del_rule);
75 COVERAGE_DEFINE(ofproto_error);
76 COVERAGE_DEFINE(ofproto_expiration);
77 COVERAGE_DEFINE(ofproto_expired);
78 COVERAGE_DEFINE(ofproto_flows_req);
79 COVERAGE_DEFINE(ofproto_flush);
80 COVERAGE_DEFINE(ofproto_invalidated);
81 COVERAGE_DEFINE(ofproto_no_packet_in);
82 COVERAGE_DEFINE(ofproto_ofconn_stuck);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_unexpected_rule);
90 COVERAGE_DEFINE(ofproto_uninstallable);
91 COVERAGE_DEFINE(ofproto_update_port);
93 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
94 * flow translation. */
95 #define MAX_RESUBMIT_RECURSION 16
100 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
101 struct netdev *netdev;
102 struct ofp_phy_port opp; /* In host byte order. */
104 struct cfm *cfm; /* Connectivity Fault Management, if any. */
107 static void ofport_free(struct ofport *);
108 static void ofport_run(struct ofproto *, struct ofport *);
109 static void ofport_wait(struct ofport *);
111 struct action_xlate_ctx {
112 /* action_xlate_ctx_init() initializes these members. */
115 struct ofproto *ofproto;
117 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
118 * this flow when actions change header fields. */
121 /* The packet corresponding to 'flow', or a null pointer if we are
122 * revalidating without a packet to refer to. */
123 const struct ofpbuf *packet;
125 /* If nonnull, called just before executing a resubmit action.
127 * This is normally null so the client has to set it manually after
128 * calling action_xlate_ctx_init(). */
129 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
131 /* If true, the speciality of 'flow' should be checked before executing
132 * its actions. If special_cb returns false on 'flow' rendered
133 * uninstallable and no actions will be executed. */
136 /* xlate_actions() initializes and uses these members. The client might want
137 * to look at them after it returns. */
139 struct ofpbuf *odp_actions; /* Datapath actions. */
140 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
141 bool may_set_up_flow; /* True ordinarily; false if the actions must
142 * be reassessed for every packet. */
143 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
145 /* xlate_actions() initializes and uses these members, but the client has no
146 * reason to look at them. */
148 int recurse; /* Recursion level, via xlate_table_action. */
149 int last_pop_priority; /* Offset in 'odp_actions' just past most
150 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
153 static void action_xlate_ctx_init(struct action_xlate_ctx *,
154 struct ofproto *, const struct flow *,
155 const struct ofpbuf *);
156 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
157 const union ofp_action *in, size_t n_in);
159 /* An OpenFlow flow. */
161 long long int used; /* Time last used; time created if not used. */
162 long long int created; /* Creation time. */
166 * - Do include packets and bytes from facets that have been deleted or
167 * whose own statistics have been folded into the rule.
169 * - Do include packets and bytes sent "by hand" that were accounted to
170 * the rule without any facet being involved (this is a rare corner
171 * case in rule_execute()).
173 * - Do not include packet or bytes that can be obtained from any facet's
174 * packet_count or byte_count member or that can be obtained from the
175 * datapath by, e.g., dpif_flow_get() for any facet.
177 uint64_t packet_count; /* Number of packets received. */
178 uint64_t byte_count; /* Number of bytes received. */
180 ovs_be64 flow_cookie; /* Controller-issued identifier. */
182 struct cls_rule cr; /* In owning ofproto's classifier. */
183 uint16_t idle_timeout; /* In seconds from time of last use. */
184 uint16_t hard_timeout; /* In seconds from time of creation. */
185 bool send_flow_removed; /* Send a flow removed message? */
186 int n_actions; /* Number of elements in actions[]. */
187 union ofp_action *actions; /* OpenFlow actions. */
188 struct list facets; /* List of "struct facet"s. */
191 static struct rule *rule_from_cls_rule(const struct cls_rule *);
192 static bool rule_is_hidden(const struct rule *);
194 static struct rule *rule_create(const struct cls_rule *,
195 const union ofp_action *, size_t n_actions,
196 uint16_t idle_timeout, uint16_t hard_timeout,
197 ovs_be64 flow_cookie, bool send_flow_removed);
198 static void rule_destroy(struct ofproto *, struct rule *);
199 static void rule_free(struct rule *);
201 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
202 static void rule_insert(struct ofproto *, struct rule *);
203 static void rule_remove(struct ofproto *, struct rule *);
205 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
206 static void rule_get_stats(const struct rule *, uint64_t *packets,
209 /* An exact-match instantiation of an OpenFlow flow. */
211 long long int used; /* Time last used; time created if not used. */
215 * - Do include packets and bytes sent "by hand", e.g. with
218 * - Do include packets and bytes that were obtained from the datapath
219 * when a flow was deleted (e.g. dpif_flow_del()) or when its
220 * statistics were reset (e.g. dpif_flow_put() with
221 * DPIF_FP_ZERO_STATS).
223 * - Do not include any packets or bytes that can currently be obtained
224 * from the datapath by, e.g., dpif_flow_get().
226 uint64_t packet_count; /* Number of packets received. */
227 uint64_t byte_count; /* Number of bytes received. */
229 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
230 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
232 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
233 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
234 long long int rs_used; /* Used time pushed to resubmit children. */
236 /* Number of bytes passed to account_cb. This may include bytes that can
237 * currently obtained from the datapath (thus, it can be greater than
239 uint64_t accounted_bytes;
241 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
242 struct list list_node; /* In owning rule's 'facets' list. */
243 struct rule *rule; /* Owning rule. */
244 struct flow flow; /* Exact-match flow. */
245 bool installed; /* Installed in datapath? */
246 bool may_install; /* True ordinarily; false if actions must
247 * be reassessed for every packet. */
248 size_t actions_len; /* Number of bytes in actions[]. */
249 struct nlattr *actions; /* Datapath actions. */
250 tag_type tags; /* Tags (set only by hooks). */
251 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
254 static struct facet *facet_create(struct ofproto *, struct rule *,
256 const struct ofpbuf *packet);
257 static void facet_remove(struct ofproto *, struct facet *);
258 static void facet_free(struct facet *);
260 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
261 static bool facet_revalidate(struct ofproto *, struct facet *);
263 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
264 static void facet_uninstall(struct ofproto *, struct facet *);
265 static void facet_flush_stats(struct ofproto *, struct facet *);
267 static void facet_make_actions(struct ofproto *, struct facet *,
268 const struct ofpbuf *packet);
269 static void facet_update_stats(struct ofproto *, struct facet *,
270 const struct dpif_flow_stats *);
271 static void facet_push_stats(struct ofproto *, struct facet *);
273 /* ofproto supports two kinds of OpenFlow connections:
275 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
276 * maintains persistent connections to these controllers and by default
277 * sends them asynchronous messages such as packet-ins.
279 * - "Service" connections, e.g. from ovs-ofctl. When these connections
280 * drop, it is the other side's responsibility to reconnect them if
281 * necessary. ofproto does not send them asynchronous messages by default.
283 * Currently, active (tcp, ssl, unix) connections are always "primary"
284 * connections and passive (ptcp, pssl, punix) connections are always "service"
285 * connections. There is no inherent reason for this, but it reflects the
289 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
290 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
293 /* A listener for incoming OpenFlow "service" connections. */
295 struct hmap_node node; /* In struct ofproto's "services" hmap. */
296 struct pvconn *pvconn; /* OpenFlow connection listener. */
298 /* These are not used by ofservice directly. They are settings for
299 * accepted "struct ofconn"s from the pvconn. */
300 int probe_interval; /* Max idle time before probing, in seconds. */
301 int rate_limit; /* Max packet-in rate in packets per second. */
302 int burst_limit; /* Limit on accumulating packet credits. */
305 static struct ofservice *ofservice_lookup(struct ofproto *,
307 static int ofservice_create(struct ofproto *,
308 const struct ofproto_controller *);
309 static void ofservice_reconfigure(struct ofservice *,
310 const struct ofproto_controller *);
311 static void ofservice_destroy(struct ofproto *, struct ofservice *);
313 /* An OpenFlow connection. */
315 struct ofproto *ofproto; /* The ofproto that owns this connection. */
316 struct list node; /* In struct ofproto's "all_conns" list. */
317 struct rconn *rconn; /* OpenFlow connection. */
318 enum ofconn_type type; /* Type. */
319 enum nx_flow_format flow_format; /* Currently selected flow format. */
321 /* OFPT_PACKET_IN related data. */
322 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
323 #define N_SCHEDULERS 2
324 struct pinsched *schedulers[N_SCHEDULERS];
325 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
326 int miss_send_len; /* Bytes to send of buffered packets. */
328 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
329 * requests, and the maximum number before we stop reading OpenFlow
331 #define OFCONN_REPLY_MAX 100
332 struct rconn_packet_counter *reply_counter;
334 /* type == OFCONN_PRIMARY only. */
335 enum nx_role role; /* Role. */
336 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
337 enum ofproto_band band; /* In-band or out-of-band? */
341 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
343 static void ofconn_destroy(struct ofconn *);
344 static void ofconn_run(struct ofconn *);
345 static void ofconn_wait(struct ofconn *);
347 static bool ofconn_receives_async_msgs(const struct ofconn *);
348 static char *ofconn_make_name(const struct ofproto *, const char *target);
349 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
351 static struct ofproto *ofconn_get_ofproto(struct ofconn *);
353 static enum nx_flow_format ofconn_get_flow_format(struct ofconn *);
354 static void ofconn_set_flow_format(struct ofconn *, enum nx_flow_format);
356 static int ofconn_get_miss_send_len(const struct ofconn *);
357 static void ofconn_set_miss_send_len(struct ofconn *, int miss_send_len);
359 static enum ofconn_type ofconn_get_type(const struct ofconn *);
361 static enum nx_role ofconn_get_role(const struct ofconn *);
362 static void ofconn_set_role(struct ofconn *, enum nx_role);
364 static int ofconn_pktbuf_retrieve(struct ofconn *, uint32_t id,
365 struct ofpbuf **bufferp, uint16_t *in_port);
368 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
369 struct rconn_packet_counter *counter);
371 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
372 const struct flow *, bool clone);
373 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
377 uint64_t datapath_id; /* Datapath ID. */
378 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
379 char *mfr_desc; /* Manufacturer. */
380 char *hw_desc; /* Hardware. */
381 char *sw_desc; /* Software version. */
382 char *serial_desc; /* Serial number. */
383 char *dp_desc; /* Datapath description. */
387 struct netdev_monitor *netdev_monitor;
388 struct hmap ports; /* Contains "struct ofport"s. */
389 struct shash port_by_name;
393 struct fail_open *fail_open;
394 struct netflow *netflow;
395 struct ofproto_sflow *sflow;
397 /* In-band control. */
398 struct in_band *in_band;
399 long long int next_in_band_update;
400 struct sockaddr_in *extra_in_band_remotes;
401 size_t n_extra_remotes;
405 struct classifier cls;
406 long long int next_expiration;
410 bool need_revalidate;
411 struct tag_set revalidate_set;
413 /* OpenFlow connections. */
414 struct hmap controllers; /* Controller "struct ofconn"s. */
415 struct list all_conns; /* Contains "struct ofconn"s. */
416 enum ofproto_fail_mode fail_mode;
418 /* OpenFlow listeners. */
419 struct hmap services; /* Contains "struct ofservice"s. */
420 struct pvconn **snoops;
423 /* Hooks for ovs-vswitchd. */
424 const struct ofhooks *ofhooks;
427 /* Used by default ofhooks. */
428 struct mac_learning *ml;
431 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
432 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
434 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
436 static const struct ofhooks default_ofhooks;
438 static uint64_t pick_datapath_id(const struct ofproto *);
439 static uint64_t pick_fallback_dpid(void);
441 static int ofproto_expire(struct ofproto *);
442 static void flow_push_stats(struct ofproto *, const struct rule *,
443 struct flow *, uint64_t packets, uint64_t bytes,
446 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
448 static void handle_openflow(struct ofconn *, struct ofpbuf *);
450 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
451 static void update_port(struct ofproto *, const char *devname);
452 static int init_ports(struct ofproto *);
453 static void reinit_ports(struct ofproto *);
455 static void ofproto_unixctl_init(void);
458 ofproto_create(const char *datapath, const char *datapath_type,
459 const struct ofhooks *ofhooks, void *aux,
460 struct ofproto **ofprotop)
468 ofproto_unixctl_init();
470 /* Connect to datapath and start listening for messages. */
471 error = dpif_open(datapath, datapath_type, &dpif);
473 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
476 error = dpif_recv_set_mask(dpif,
477 ((1u << DPIF_UC_MISS) |
478 (1u << DPIF_UC_ACTION) |
479 (1u << DPIF_UC_SAMPLE)));
481 VLOG_ERR("failed to listen on datapath %s: %s",
482 datapath, strerror(error));
486 dpif_flow_flush(dpif);
487 dpif_recv_purge(dpif);
489 /* Initialize settings. */
490 p = xzalloc(sizeof *p);
491 p->fallback_dpid = pick_fallback_dpid();
492 p->datapath_id = p->fallback_dpid;
493 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
494 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
495 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
496 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
497 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
499 /* Initialize datapath. */
501 p->netdev_monitor = netdev_monitor_create();
502 hmap_init(&p->ports);
503 shash_init(&p->port_by_name);
504 p->max_ports = dpif_get_max_ports(dpif);
506 /* Initialize submodules. */
511 /* Initialize in-band control. */
513 p->in_band_queue = -1;
515 /* Initialize flow table. */
516 classifier_init(&p->cls);
517 p->next_expiration = time_msec() + 1000;
519 /* Initialize facet table. */
520 hmap_init(&p->facets);
521 p->need_revalidate = false;
522 tag_set_init(&p->revalidate_set);
524 /* Initialize OpenFlow connections. */
525 list_init(&p->all_conns);
526 hmap_init(&p->controllers);
527 hmap_init(&p->services);
531 /* Initialize hooks. */
533 p->ofhooks = ofhooks;
537 p->ofhooks = &default_ofhooks;
539 p->ml = mac_learning_create();
542 /* Pick final datapath ID. */
543 p->datapath_id = pick_datapath_id(p);
544 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
546 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
553 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
555 uint64_t old_dpid = p->datapath_id;
556 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
557 if (p->datapath_id != old_dpid) {
558 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
560 /* Force all active connections to reconnect, since there is no way to
561 * notify a controller that the datapath ID has changed. */
562 ofproto_reconnect_controllers(p);
566 /* Creates a new controller in 'ofproto'. Some of the settings are initially
567 * drawn from 'c', but update_controller() needs to be called later to finish
568 * the new ofconn's configuration. */
570 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
572 char *name = ofconn_make_name(ofproto, c->target);
573 struct ofconn *ofconn;
575 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
576 ofconn->pktbuf = pktbuf_create();
577 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
578 rconn_connect(ofconn->rconn, c->target, name);
579 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
580 hash_string(c->target, 0));
585 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
586 * target (this is done by creating new ofconns and deleting old ones), but it
587 * can update the rest of an ofconn's settings. */
589 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
593 ofconn->band = c->band;
595 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
597 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
598 rconn_set_probe_interval(ofconn->rconn, probe_interval);
600 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
604 ofconn_get_target(const struct ofconn *ofconn)
606 return rconn_get_target(ofconn->rconn);
609 static struct ofconn *
610 find_controller_by_target(struct ofproto *ofproto, const char *target)
612 struct ofconn *ofconn;
614 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
615 hash_string(target, 0), &ofproto->controllers) {
616 if (!strcmp(ofconn_get_target(ofconn), target)) {
624 update_in_band_remotes(struct ofproto *ofproto)
626 const struct ofconn *ofconn;
627 struct sockaddr_in *addrs;
628 size_t max_addrs, n_addrs;
631 /* Allocate enough memory for as many remotes as we could possibly have. */
632 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
633 addrs = xmalloc(max_addrs * sizeof *addrs);
636 /* Add all the remotes. */
637 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
638 struct sockaddr_in *sin = &addrs[n_addrs];
640 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
644 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
645 if (sin->sin_addr.s_addr) {
646 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
650 for (i = 0; i < ofproto->n_extra_remotes; i++) {
651 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
654 /* Create or update or destroy in-band. */
656 if (!ofproto->in_band) {
657 in_band_create(ofproto, ofproto->dpif, &ofproto->in_band);
659 if (ofproto->in_band) {
660 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
662 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
663 ofproto->next_in_band_update = time_msec() + 1000;
665 in_band_destroy(ofproto->in_band);
666 ofproto->in_band = NULL;
674 update_fail_open(struct ofproto *p)
676 struct ofconn *ofconn;
678 if (!hmap_is_empty(&p->controllers)
679 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
680 struct rconn **rconns;
684 p->fail_open = fail_open_create(p);
688 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
689 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
690 rconns[n++] = ofconn->rconn;
693 fail_open_set_controllers(p->fail_open, rconns, n);
694 /* p->fail_open takes ownership of 'rconns'. */
696 fail_open_destroy(p->fail_open);
702 ofproto_set_controllers(struct ofproto *p,
703 const struct ofproto_controller *controllers,
704 size_t n_controllers)
706 struct shash new_controllers;
707 struct ofconn *ofconn, *next_ofconn;
708 struct ofservice *ofservice, *next_ofservice;
711 /* Create newly configured controllers and services.
712 * Create a name to ofproto_controller mapping in 'new_controllers'. */
713 shash_init(&new_controllers);
714 for (i = 0; i < n_controllers; i++) {
715 const struct ofproto_controller *c = &controllers[i];
717 if (!vconn_verify_name(c->target)) {
718 if (!find_controller_by_target(p, c->target)) {
719 add_controller(p, c);
721 } else if (!pvconn_verify_name(c->target)) {
722 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
726 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
727 dpif_name(p->dpif), c->target);
731 shash_add_once(&new_controllers, c->target, &controllers[i]);
734 /* Delete controllers that are no longer configured.
735 * Update configuration of all now-existing controllers. */
736 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
737 struct ofproto_controller *c;
739 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
741 ofconn_destroy(ofconn);
743 update_controller(ofconn, c);
747 /* Delete services that are no longer configured.
748 * Update configuration of all now-existing services. */
749 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
750 struct ofproto_controller *c;
752 c = shash_find_data(&new_controllers,
753 pvconn_get_name(ofservice->pvconn));
755 ofservice_destroy(p, ofservice);
757 ofservice_reconfigure(ofservice, c);
761 shash_destroy(&new_controllers);
763 update_in_band_remotes(p);
768 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
770 p->fail_mode = fail_mode;
774 /* Drops the connections between 'ofproto' and all of its controllers, forcing
775 * them to reconnect. */
777 ofproto_reconnect_controllers(struct ofproto *ofproto)
779 struct ofconn *ofconn;
781 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
782 rconn_reconnect(ofconn->rconn);
787 any_extras_changed(const struct ofproto *ofproto,
788 const struct sockaddr_in *extras, size_t n)
792 if (n != ofproto->n_extra_remotes) {
796 for (i = 0; i < n; i++) {
797 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
798 const struct sockaddr_in *new = &extras[i];
800 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
801 old->sin_port != new->sin_port) {
809 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
810 * in-band control should guarantee access, in the same way that in-band
811 * control guarantees access to OpenFlow controllers. */
813 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
814 const struct sockaddr_in *extras, size_t n)
816 if (!any_extras_changed(ofproto, extras, n)) {
820 free(ofproto->extra_in_band_remotes);
821 ofproto->n_extra_remotes = n;
822 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
824 update_in_band_remotes(ofproto);
827 /* Sets the OpenFlow queue used by flows set up by in-band control on
828 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
829 * flows will use the default queue. */
831 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
833 if (queue_id != ofproto->in_band_queue) {
834 ofproto->in_band_queue = queue_id;
835 update_in_band_remotes(ofproto);
840 ofproto_set_desc(struct ofproto *p,
841 const char *mfr_desc, const char *hw_desc,
842 const char *sw_desc, const char *serial_desc,
845 struct ofp_desc_stats *ods;
848 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
849 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
850 sizeof ods->mfr_desc);
853 p->mfr_desc = xstrdup(mfr_desc);
856 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
857 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
858 sizeof ods->hw_desc);
861 p->hw_desc = xstrdup(hw_desc);
864 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
865 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
866 sizeof ods->sw_desc);
869 p->sw_desc = xstrdup(sw_desc);
872 if (strlen(serial_desc) >= sizeof ods->serial_num) {
873 VLOG_WARN("truncating serial_desc, must be less than %zu "
875 sizeof ods->serial_num);
877 free(p->serial_desc);
878 p->serial_desc = xstrdup(serial_desc);
881 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
882 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
883 sizeof ods->dp_desc);
886 p->dp_desc = xstrdup(dp_desc);
891 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
892 const struct svec *svec)
894 struct pvconn **pvconns = *pvconnsp;
895 size_t n_pvconns = *n_pvconnsp;
899 for (i = 0; i < n_pvconns; i++) {
900 pvconn_close(pvconns[i]);
904 pvconns = xmalloc(svec->n * sizeof *pvconns);
906 for (i = 0; i < svec->n; i++) {
907 const char *name = svec->names[i];
908 struct pvconn *pvconn;
911 error = pvconn_open(name, &pvconn);
913 pvconns[n_pvconns++] = pvconn;
915 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
923 *n_pvconnsp = n_pvconns;
929 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
931 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
935 ofproto_set_netflow(struct ofproto *ofproto,
936 const struct netflow_options *nf_options)
938 if (nf_options && nf_options->collectors.n) {
939 if (!ofproto->netflow) {
940 ofproto->netflow = netflow_create();
942 return netflow_set_options(ofproto->netflow, nf_options);
944 netflow_destroy(ofproto->netflow);
945 ofproto->netflow = NULL;
951 ofproto_set_sflow(struct ofproto *ofproto,
952 const struct ofproto_sflow_options *oso)
954 struct ofproto_sflow *os = ofproto->sflow;
957 struct ofport *ofport;
959 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
960 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
961 ofproto_sflow_add_port(os, ofport->odp_port,
962 netdev_get_name(ofport->netdev));
965 ofproto_sflow_set_options(os, oso);
967 ofproto_sflow_destroy(os);
968 ofproto->sflow = NULL;
972 /* Connectivity Fault Management configuration. */
974 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
976 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
978 struct ofport *ofport = get_port(ofproto, port_no);
979 if (ofport && ofport->cfm){
980 cfm_destroy(ofport->cfm);
985 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
986 * basic configuration from the configuration members in 'cfm', and the set of
987 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
988 * Ignores the statistics members of 'cfm'.
990 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
992 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
993 const struct cfm *cfm,
994 const uint16_t *remote_mps, size_t n_remote_mps)
996 struct ofport *ofport;
998 ofport = get_port(ofproto, port_no);
1000 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
1001 dpif_name(ofproto->dpif), port_no);
1006 ofport->cfm = cfm_create();
1009 ofport->cfm->mpid = cfm->mpid;
1010 ofport->cfm->interval = cfm->interval;
1011 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
1013 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
1015 if (!cfm_configure(ofport->cfm)) {
1016 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
1017 dpif_name(ofproto->dpif), port_no,
1018 netdev_get_name(ofport->netdev));
1019 cfm_destroy(ofport->cfm);
1024 /* Returns the connectivity fault management object associated with 'port_no'
1025 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
1026 * 'port_no' or if that port does not have CFM configured. The caller must not
1027 * modify or destroy the returned object. */
1029 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
1031 struct ofport *ofport = get_port(ofproto, port_no);
1032 return ofport ? ofport->cfm : NULL;
1036 ofproto_get_datapath_id(const struct ofproto *ofproto)
1038 return ofproto->datapath_id;
1042 ofproto_has_primary_controller(const struct ofproto *ofproto)
1044 return !hmap_is_empty(&ofproto->controllers);
1047 enum ofproto_fail_mode
1048 ofproto_get_fail_mode(const struct ofproto *p)
1050 return p->fail_mode;
1054 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1058 for (i = 0; i < ofproto->n_snoops; i++) {
1059 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1064 ofproto_destroy(struct ofproto *p)
1066 struct ofservice *ofservice, *next_ofservice;
1067 struct ofconn *ofconn, *next_ofconn;
1068 struct ofport *ofport, *next_ofport;
1075 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1077 /* Destroy fail-open and in-band early, since they touch the classifier. */
1078 fail_open_destroy(p->fail_open);
1079 p->fail_open = NULL;
1081 in_band_destroy(p->in_band);
1083 free(p->extra_in_band_remotes);
1085 ofproto_flush_flows(p);
1086 classifier_destroy(&p->cls);
1087 hmap_destroy(&p->facets);
1089 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1090 ofconn_destroy(ofconn);
1092 hmap_destroy(&p->controllers);
1094 dpif_close(p->dpif);
1095 netdev_monitor_destroy(p->netdev_monitor);
1096 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1097 hmap_remove(&p->ports, &ofport->hmap_node);
1098 ofport_free(ofport);
1100 shash_destroy(&p->port_by_name);
1102 netflow_destroy(p->netflow);
1103 ofproto_sflow_destroy(p->sflow);
1105 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1106 ofservice_destroy(p, ofservice);
1108 hmap_destroy(&p->services);
1110 for (i = 0; i < p->n_snoops; i++) {
1111 pvconn_close(p->snoops[i]);
1115 mac_learning_destroy(p->ml);
1120 free(p->serial_desc);
1123 hmap_destroy(&p->ports);
1129 ofproto_run(struct ofproto *p)
1131 int error = ofproto_run1(p);
1133 error = ofproto_run2(p, false);
1139 process_port_change(struct ofproto *ofproto, int error, char *devname)
1141 if (error == ENOBUFS) {
1142 reinit_ports(ofproto);
1143 } else if (!error) {
1144 update_port(ofproto, devname);
1149 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1150 * means that 'ofconn' is more interesting for monitoring than a lower return
1153 snoop_preference(const struct ofconn *ofconn)
1155 switch (ofconn_get_role(ofconn)) {
1156 case NX_ROLE_MASTER:
1163 /* Shouldn't happen. */
1168 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1169 * Connects this vconn to a controller. */
1171 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1173 struct ofconn *ofconn, *best;
1175 /* Pick a controller for monitoring. */
1177 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1178 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1179 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1185 rconn_add_monitor(best->rconn, vconn);
1187 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1193 ofproto_run1(struct ofproto *p)
1195 struct ofconn *ofconn, *next_ofconn;
1196 struct ofservice *ofservice;
1197 struct ofport *ofport;
1202 if (shash_is_empty(&p->port_by_name)) {
1206 for (i = 0; i < 50; i++) {
1207 struct dpif_upcall packet;
1209 error = dpif_recv(p->dpif, &packet);
1211 if (error == ENODEV) {
1212 /* Someone destroyed the datapath behind our back. The caller
1213 * better destroy us and give up, because we're just going to
1214 * spin from here on out. */
1215 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1216 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1217 dpif_name(p->dpif));
1223 handle_upcall(p, &packet);
1226 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1227 process_port_change(p, error, devname);
1229 while ((error = netdev_monitor_poll(p->netdev_monitor,
1230 &devname)) != EAGAIN) {
1231 process_port_change(p, error, devname);
1234 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1235 ofport_run(p, ofport);
1239 if (time_msec() >= p->next_in_band_update) {
1240 update_in_band_remotes(p);
1242 in_band_run(p->in_band);
1245 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1249 /* Fail-open maintenance. Do this after processing the ofconns since
1250 * fail-open checks the status of the controller rconn. */
1252 fail_open_run(p->fail_open);
1255 HMAP_FOR_EACH (ofservice, node, &p->services) {
1256 struct vconn *vconn;
1259 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1261 struct rconn *rconn;
1264 rconn = rconn_create(ofservice->probe_interval, 0);
1265 name = ofconn_make_name(p, vconn_get_name(vconn));
1266 rconn_connect_unreliably(rconn, vconn, name);
1269 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1270 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1271 ofservice->burst_limit);
1272 } else if (retval != EAGAIN) {
1273 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1277 for (i = 0; i < p->n_snoops; i++) {
1278 struct vconn *vconn;
1281 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1283 add_snooper(p, vconn);
1284 } else if (retval != EAGAIN) {
1285 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1289 if (time_msec() >= p->next_expiration) {
1290 int delay = ofproto_expire(p);
1291 p->next_expiration = time_msec() + delay;
1292 COVERAGE_INC(ofproto_expiration);
1296 netflow_run(p->netflow);
1299 ofproto_sflow_run(p->sflow);
1306 ofproto_run2(struct ofproto *p, bool revalidate_all)
1308 /* Figure out what we need to revalidate now, if anything. */
1309 struct tag_set revalidate_set = p->revalidate_set;
1310 if (p->need_revalidate) {
1311 revalidate_all = true;
1314 /* Clear the revalidation flags. */
1315 tag_set_init(&p->revalidate_set);
1316 p->need_revalidate = false;
1318 /* Now revalidate if there's anything to do. */
1319 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1320 struct facet *facet, *next;
1322 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1324 || tag_set_intersects(&revalidate_set, facet->tags)) {
1325 facet_revalidate(p, facet);
1334 ofproto_wait(struct ofproto *p)
1336 struct ofservice *ofservice;
1337 struct ofconn *ofconn;
1338 struct ofport *ofport;
1341 dpif_recv_wait(p->dpif);
1342 dpif_port_poll_wait(p->dpif);
1343 netdev_monitor_poll_wait(p->netdev_monitor);
1344 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1345 ofport_wait(ofport);
1347 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1348 ofconn_wait(ofconn);
1351 poll_timer_wait_until(p->next_in_band_update);
1352 in_band_wait(p->in_band);
1355 fail_open_wait(p->fail_open);
1358 ofproto_sflow_wait(p->sflow);
1360 if (!tag_set_is_empty(&p->revalidate_set)) {
1361 poll_immediate_wake();
1363 if (p->need_revalidate) {
1364 /* Shouldn't happen, but if it does just go around again. */
1365 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1366 poll_immediate_wake();
1367 } else if (p->next_expiration != LLONG_MAX) {
1368 poll_timer_wait_until(p->next_expiration);
1370 HMAP_FOR_EACH (ofservice, node, &p->services) {
1371 pvconn_wait(ofservice->pvconn);
1373 for (i = 0; i < p->n_snoops; i++) {
1374 pvconn_wait(p->snoops[i]);
1379 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1381 tag_set_add(&ofproto->revalidate_set, tag);
1385 ofproto_get_revalidate_set(struct ofproto *ofproto)
1387 return &ofproto->revalidate_set;
1391 ofproto_is_alive(const struct ofproto *p)
1393 return !hmap_is_empty(&p->controllers);
1397 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1400 const struct ofconn *ofconn;
1404 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1405 const struct rconn *rconn = ofconn->rconn;
1406 time_t now = time_now();
1407 time_t last_connection = rconn_get_last_connection(rconn);
1408 time_t last_disconnect = rconn_get_last_disconnect(rconn);
1409 const int last_error = rconn_get_last_error(rconn);
1410 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1412 shash_add(info, rconn_get_target(rconn), cinfo);
1414 cinfo->is_connected = rconn_is_connected(rconn);
1415 cinfo->role = ofconn_get_role(ofconn);
1420 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1421 cinfo->pairs.values[cinfo->pairs.n++] =
1422 xstrdup(ovs_retval_to_string(last_error));
1425 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1426 cinfo->pairs.values[cinfo->pairs.n++] =
1427 xstrdup(rconn_get_state(rconn));
1429 if (last_connection != TIME_MIN) {
1430 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
1431 cinfo->pairs.values[cinfo->pairs.n++]
1432 = xasprintf("%ld", (long int) (now - last_connection));
1435 if (last_disconnect != TIME_MIN) {
1436 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
1437 cinfo->pairs.values[cinfo->pairs.n++]
1438 = xasprintf("%ld", (long int) (now - last_disconnect));
1444 ofproto_free_ofproto_controller_info(struct shash *info)
1446 struct shash_node *node;
1448 SHASH_FOR_EACH (node, info) {
1449 struct ofproto_controller_info *cinfo = node->data;
1450 while (cinfo->pairs.n) {
1451 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1455 shash_destroy(info);
1458 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1460 * This is almost the same as calling dpif_port_del() directly on the
1461 * datapath, but it also makes 'ofproto' close its open netdev for the port
1462 * (if any). This makes it possible to create a new netdev of a different
1463 * type under the same name, which otherwise the netdev library would refuse
1464 * to do because of the conflict. (The netdev would eventually get closed on
1465 * the next trip through ofproto_run(), but this interface is more direct.)
1467 * Returns 0 if successful, otherwise a positive errno. */
1469 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1471 struct ofport *ofport = get_port(ofproto, odp_port);
1472 const char *name = ofport ? ofport->opp.name : "<unknown>";
1475 error = dpif_port_del(ofproto->dpif, odp_port);
1477 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1478 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1479 } else if (ofport) {
1480 /* 'name' is ofport->opp.name and update_port() is going to destroy
1481 * 'ofport'. Just in case update_port() refers to 'name' after it
1482 * destroys 'ofport', make a copy of it around the update_port()
1484 char *devname = xstrdup(name);
1485 update_port(ofproto, devname);
1491 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1492 * true if 'odp_port' exists and should be included, false otherwise. */
1494 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1496 struct ofport *ofport = get_port(ofproto, odp_port);
1497 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1500 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1501 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1502 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1504 * Returns 0 if successful, otherwise a positive errno value. */
1506 ofproto_send_packet(struct ofproto *ofproto,
1507 uint32_t port_no, uint16_t vlan_tci,
1508 const struct ofpbuf *packet)
1510 struct ofpbuf odp_actions;
1513 ofpbuf_init(&odp_actions, 32);
1514 if (vlan_tci != 0) {
1515 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1516 ntohs(vlan_tci & ~VLAN_CFI));
1518 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1519 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1521 ofpbuf_uninit(&odp_actions);
1524 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1525 dpif_name(ofproto->dpif), port_no, strerror(error));
1530 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1531 * performs the 'n_actions' actions in 'actions'. The new flow will not
1534 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1535 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1536 * controllers; otherwise, it will be hidden.
1538 * The caller retains ownership of 'cls_rule' and 'actions'. */
1540 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1541 const union ofp_action *actions, size_t n_actions)
1544 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1545 rule_insert(p, rule);
1549 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1553 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1556 rule_remove(ofproto, rule);
1561 ofproto_flush_flows(struct ofproto *ofproto)
1563 struct facet *facet, *next_facet;
1564 struct rule *rule, *next_rule;
1565 struct cls_cursor cursor;
1567 COVERAGE_INC(ofproto_flush);
1569 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1570 /* Mark the facet as not installed so that facet_remove() doesn't
1571 * bother trying to uninstall it. There is no point in uninstalling it
1572 * individually since we are about to blow away all the facets with
1573 * dpif_flow_flush(). */
1574 facet->installed = false;
1575 facet->dp_packet_count = 0;
1576 facet->dp_byte_count = 0;
1577 facet_remove(ofproto, facet);
1580 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1581 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1582 rule_remove(ofproto, rule);
1585 dpif_flow_flush(ofproto->dpif);
1586 if (ofproto->in_band) {
1587 in_band_flushed(ofproto->in_band);
1589 if (ofproto->fail_open) {
1590 fail_open_flushed(ofproto->fail_open);
1595 reinit_ports(struct ofproto *p)
1597 struct dpif_port_dump dump;
1598 struct shash_node *node;
1599 struct shash devnames;
1600 struct ofport *ofport;
1601 struct dpif_port dpif_port;
1603 COVERAGE_INC(ofproto_reinit_ports);
1605 shash_init(&devnames);
1606 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1607 shash_add_once (&devnames, ofport->opp.name, NULL);
1609 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1610 shash_add_once (&devnames, dpif_port.name, NULL);
1613 SHASH_FOR_EACH (node, &devnames) {
1614 update_port(p, node->name);
1616 shash_destroy(&devnames);
1619 static struct ofport *
1620 make_ofport(const struct dpif_port *dpif_port)
1622 struct netdev_options netdev_options;
1623 enum netdev_flags flags;
1624 struct ofport *ofport;
1625 struct netdev *netdev;
1628 memset(&netdev_options, 0, sizeof netdev_options);
1629 netdev_options.name = dpif_port->name;
1630 netdev_options.type = dpif_port->type;
1631 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1633 error = netdev_open(&netdev_options, &netdev);
1635 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1636 "cannot be opened (%s)",
1637 dpif_port->name, dpif_port->port_no,
1638 dpif_port->name, strerror(error));
1642 ofport = xzalloc(sizeof *ofport);
1643 ofport->netdev = netdev;
1644 ofport->odp_port = dpif_port->port_no;
1645 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1646 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1647 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1649 netdev_get_flags(netdev, &flags);
1650 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1652 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1654 netdev_get_features(netdev,
1655 &ofport->opp.curr, &ofport->opp.advertised,
1656 &ofport->opp.supported, &ofport->opp.peer);
1661 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1663 if (get_port(p, dpif_port->port_no)) {
1664 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1665 dpif_port->port_no);
1667 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1668 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1677 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1679 const struct ofp_phy_port *a = &a_->opp;
1680 const struct ofp_phy_port *b = &b_->opp;
1682 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1683 return (a->port_no == b->port_no
1684 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1685 && !strcmp(a->name, b->name)
1686 && a->state == b->state
1687 && a->config == b->config
1688 && a->curr == b->curr
1689 && a->advertised == b->advertised
1690 && a->supported == b->supported
1691 && a->peer == b->peer);
1695 send_port_status(struct ofproto *p, const struct ofport *ofport,
1698 /* XXX Should limit the number of queued port status change messages. */
1699 struct ofconn *ofconn;
1700 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1701 struct ofp_port_status *ops;
1704 /* Primary controllers, even slaves, should always get port status
1705 updates. Otherwise obey ofconn_receives_async_msgs(). */
1706 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
1707 && !ofconn_receives_async_msgs(ofconn)) {
1711 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1712 ops->reason = reason;
1713 ops->desc = ofport->opp;
1714 hton_ofp_phy_port(&ops->desc);
1715 queue_tx(b, ofconn, NULL);
1720 ofport_install(struct ofproto *p, struct ofport *ofport)
1722 const char *netdev_name = ofport->opp.name;
1724 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1725 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1726 shash_add(&p->port_by_name, netdev_name, ofport);
1728 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1733 ofport_remove(struct ofproto *p, struct ofport *ofport)
1735 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1736 hmap_remove(&p->ports, &ofport->hmap_node);
1737 shash_delete(&p->port_by_name,
1738 shash_find(&p->port_by_name, ofport->opp.name));
1740 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1745 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1748 cfm_run(ofport->cfm);
1750 if (cfm_should_send_ccm(ofport->cfm)) {
1751 struct ofpbuf packet;
1754 ofpbuf_init(&packet, 0);
1755 ccm = compose_packet(&packet, eth_addr_ccm, ofport->opp.hw_addr,
1756 ETH_TYPE_CFM, sizeof *ccm);
1757 cfm_compose_ccm(ofport->cfm, ccm);
1758 ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet);
1759 ofpbuf_uninit(&packet);
1765 ofport_wait(struct ofport *ofport)
1768 cfm_wait(ofport->cfm);
1773 ofport_free(struct ofport *ofport)
1776 cfm_destroy(ofport->cfm);
1777 netdev_close(ofport->netdev);
1782 static struct ofport *
1783 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1785 struct ofport *port;
1787 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1788 hash_int(odp_port, 0), &ofproto->ports) {
1789 if (port->odp_port == odp_port) {
1797 update_port(struct ofproto *p, const char *devname)
1799 struct dpif_port dpif_port;
1800 struct ofport *old_ofport;
1801 struct ofport *new_ofport;
1804 COVERAGE_INC(ofproto_update_port);
1806 /* Query the datapath for port information. */
1807 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1809 /* Find the old ofport. */
1810 old_ofport = shash_find_data(&p->port_by_name, devname);
1813 /* There's no port named 'devname' but there might be a port with
1814 * the same port number. This could happen if a port is deleted
1815 * and then a new one added in its place very quickly, or if a port
1816 * is renamed. In the former case we want to send an OFPPR_DELETE
1817 * and an OFPPR_ADD, and in the latter case we want to send a
1818 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1819 * the old port's ifindex against the new port, or perhaps less
1820 * reliably but more portably by comparing the old port's MAC
1821 * against the new port's MAC. However, this code isn't that smart
1822 * and always sends an OFPPR_MODIFY (XXX). */
1823 old_ofport = get_port(p, dpif_port.port_no);
1825 } else if (error != ENOENT && error != ENODEV) {
1826 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1827 "%s", strerror(error));
1831 /* Create a new ofport. */
1832 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1834 /* Eliminate a few pathological cases. */
1835 if (!old_ofport && !new_ofport) {
1837 } else if (old_ofport && new_ofport) {
1838 /* Most of the 'config' bits are OpenFlow soft state, but
1839 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1840 * OpenFlow bits from old_ofport. (make_ofport() only sets
1841 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1842 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1844 if (ofport_equal(old_ofport, new_ofport)) {
1845 /* False alarm--no change. */
1846 ofport_free(new_ofport);
1851 /* Now deal with the normal cases. */
1853 ofport_remove(p, old_ofport);
1856 ofport_install(p, new_ofport);
1858 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1859 (!old_ofport ? OFPPR_ADD
1860 : !new_ofport ? OFPPR_DELETE
1862 ofport_free(old_ofport);
1865 dpif_port_destroy(&dpif_port);
1869 init_ports(struct ofproto *p)
1871 struct dpif_port_dump dump;
1872 struct dpif_port dpif_port;
1874 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1875 if (!ofport_conflicts(p, &dpif_port)) {
1876 struct ofport *ofport = make_ofport(&dpif_port);
1878 ofport_install(p, ofport);
1886 static struct ofconn *
1887 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1889 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1890 ofconn->ofproto = p;
1891 list_push_back(&p->all_conns, &ofconn->node);
1892 ofconn->rconn = rconn;
1893 ofconn->type = type;
1894 ofconn->flow_format = NXFF_OPENFLOW10;
1895 ofconn->role = NX_ROLE_OTHER;
1896 ofconn->packet_in_counter = rconn_packet_counter_create ();
1897 ofconn->pktbuf = NULL;
1898 ofconn->miss_send_len = 0;
1899 ofconn->reply_counter = rconn_packet_counter_create ();
1904 ofconn_destroy(struct ofconn *ofconn)
1906 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1908 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY) {
1909 hmap_remove(&ofproto->controllers, &ofconn->hmap_node);
1912 list_remove(&ofconn->node);
1913 rconn_destroy(ofconn->rconn);
1914 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1915 rconn_packet_counter_destroy(ofconn->reply_counter);
1916 pktbuf_destroy(ofconn->pktbuf);
1921 ofconn_run(struct ofconn *ofconn)
1923 struct ofproto *p = ofconn_get_ofproto(ofconn);
1927 for (i = 0; i < N_SCHEDULERS; i++) {
1928 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1931 rconn_run(ofconn->rconn);
1933 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1934 /* Limit the number of iterations to prevent other tasks from
1936 for (iteration = 0; iteration < 50; iteration++) {
1937 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1942 fail_open_maybe_recover(p->fail_open);
1944 handle_openflow(ofconn, of_msg);
1945 ofpbuf_delete(of_msg);
1949 if (!rconn_is_alive(ofconn->rconn)) {
1950 ofconn_destroy(ofconn);
1955 ofconn_wait(struct ofconn *ofconn)
1959 for (i = 0; i < N_SCHEDULERS; i++) {
1960 pinsched_wait(ofconn->schedulers[i]);
1962 rconn_run_wait(ofconn->rconn);
1963 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1964 rconn_recv_wait(ofconn->rconn);
1966 COVERAGE_INC(ofproto_ofconn_stuck);
1970 /* Returns true if 'ofconn' should receive asynchronous messages. */
1972 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1974 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY) {
1975 /* Primary controllers always get asynchronous messages unless they
1976 * have configured themselves as "slaves". */
1977 return ofconn_get_role(ofconn) != NX_ROLE_SLAVE;
1979 /* Service connections don't get asynchronous messages unless they have
1980 * explicitly asked for them by setting a nonzero miss send length. */
1981 return ofconn->miss_send_len > 0;
1985 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1986 * and 'target', suitable for use in log messages for identifying the
1989 * The name is dynamically allocated. The caller should free it (with free())
1990 * when it is no longer needed. */
1992 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1994 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1998 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
2002 for (i = 0; i < N_SCHEDULERS; i++) {
2003 struct pinsched **s = &ofconn->schedulers[i];
2007 *s = pinsched_create(rate, burst);
2009 pinsched_set_limits(*s, rate, burst);
2012 pinsched_destroy(*s);
2018 static struct ofproto *
2019 ofconn_get_ofproto(struct ofconn *ofconn)
2021 return ofconn->ofproto;
2024 static enum nx_flow_format
2025 ofconn_get_flow_format(struct ofconn *ofconn)
2027 return ofconn->flow_format;
2031 ofconn_set_flow_format(struct ofconn *ofconn, enum nx_flow_format flow_format)
2033 ofconn->flow_format = flow_format;
2037 ofconn_get_miss_send_len(const struct ofconn *ofconn)
2039 return ofconn->miss_send_len;
2043 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
2045 ofconn->miss_send_len = miss_send_len;
2048 static enum ofconn_type
2049 ofconn_get_type(const struct ofconn *ofconn)
2051 return ofconn->type;
2055 ofconn_get_role(const struct ofconn *ofconn)
2057 return ofconn->role;
2061 ofconn_set_role(struct ofconn *ofconn, enum nx_role role)
2063 ofconn->role = role;
2067 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
2068 struct ofpbuf **bufferp, uint16_t *in_port)
2070 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
2074 ofservice_reconfigure(struct ofservice *ofservice,
2075 const struct ofproto_controller *c)
2077 ofservice->probe_interval = c->probe_interval;
2078 ofservice->rate_limit = c->rate_limit;
2079 ofservice->burst_limit = c->burst_limit;
2082 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
2083 * positive errno value. */
2085 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
2087 struct ofservice *ofservice;
2088 struct pvconn *pvconn;
2091 error = pvconn_open(c->target, &pvconn);
2096 ofservice = xzalloc(sizeof *ofservice);
2097 hmap_insert(&ofproto->services, &ofservice->node,
2098 hash_string(c->target, 0));
2099 ofservice->pvconn = pvconn;
2101 ofservice_reconfigure(ofservice, c);
2107 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
2109 hmap_remove(&ofproto->services, &ofservice->node);
2110 pvconn_close(ofservice->pvconn);
2114 /* Finds and returns the ofservice within 'ofproto' that has the given
2115 * 'target', or a null pointer if none exists. */
2116 static struct ofservice *
2117 ofservice_lookup(struct ofproto *ofproto, const char *target)
2119 struct ofservice *ofservice;
2121 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
2122 &ofproto->services) {
2123 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2130 /* Returns true if 'rule' should be hidden from the controller.
2132 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2133 * (e.g. by in-band control) and are intentionally hidden from the
2136 rule_is_hidden(const struct rule *rule)
2138 return rule->cr.priority > UINT16_MAX;
2141 /* Creates and returns a new rule initialized as specified.
2143 * The caller is responsible for inserting the rule into the classifier (with
2144 * rule_insert()). */
2145 static struct rule *
2146 rule_create(const struct cls_rule *cls_rule,
2147 const union ofp_action *actions, size_t n_actions,
2148 uint16_t idle_timeout, uint16_t hard_timeout,
2149 ovs_be64 flow_cookie, bool send_flow_removed)
2151 struct rule *rule = xzalloc(sizeof *rule);
2152 rule->cr = *cls_rule;
2153 rule->idle_timeout = idle_timeout;
2154 rule->hard_timeout = hard_timeout;
2155 rule->flow_cookie = flow_cookie;
2156 rule->used = rule->created = time_msec();
2157 rule->send_flow_removed = send_flow_removed;
2158 list_init(&rule->facets);
2159 if (n_actions > 0) {
2160 rule->n_actions = n_actions;
2161 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2167 static struct rule *
2168 rule_from_cls_rule(const struct cls_rule *cls_rule)
2170 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2174 rule_free(struct rule *rule)
2176 free(rule->actions);
2180 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2181 * destroying any that no longer has a rule (which is probably all of them).
2183 * The caller must have already removed 'rule' from the classifier. */
2185 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2187 struct facet *facet, *next_facet;
2188 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2189 facet_revalidate(ofproto, facet);
2194 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2195 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2198 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2200 const union ofp_action *oa;
2201 struct actions_iterator i;
2203 if (out_port == htons(OFPP_NONE)) {
2206 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2207 oa = actions_next(&i)) {
2208 if (action_outputs_to_port(oa, out_port)) {
2215 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2216 * 'packet', which arrived on 'in_port'.
2218 * Takes ownership of 'packet'. */
2220 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2221 const struct nlattr *odp_actions, size_t actions_len,
2222 struct ofpbuf *packet)
2224 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2225 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2226 /* As an optimization, avoid a round-trip from userspace to kernel to
2227 * userspace. This also avoids possibly filling up kernel packet
2228 * buffers along the way. */
2229 struct dpif_upcall upcall;
2231 upcall.type = DPIF_UC_ACTION;
2232 upcall.packet = packet;
2235 upcall.userdata = nl_attr_get_u64(odp_actions);
2236 upcall.sample_pool = 0;
2237 upcall.actions = NULL;
2238 upcall.actions_len = 0;
2240 send_packet_in(ofproto, &upcall, flow, false);
2246 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2247 ofpbuf_delete(packet);
2252 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2253 * statistics appropriately. 'packet' must have at least sizeof(struct
2254 * ofp_packet_in) bytes of headroom.
2256 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2257 * applying flow_extract() to 'packet' would yield the same flow as
2260 * 'facet' must have accurately composed ODP actions; that is, it must not be
2261 * in need of revalidation.
2263 * Takes ownership of 'packet'. */
2265 facet_execute(struct ofproto *ofproto, struct facet *facet,
2266 struct ofpbuf *packet)
2268 struct dpif_flow_stats stats;
2270 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2272 flow_extract_stats(&facet->flow, packet, &stats);
2273 stats.used = time_msec();
2274 if (execute_odp_actions(ofproto, &facet->flow,
2275 facet->actions, facet->actions_len, packet)) {
2276 facet_update_stats(ofproto, facet, &stats);
2280 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2281 * statistics (or the statistics for one of its facets) appropriately.
2282 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2284 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2285 * with statistics for 'packet' either way.
2287 * Takes ownership of 'packet'. */
2289 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2290 struct ofpbuf *packet)
2292 struct action_xlate_ctx ctx;
2293 struct ofpbuf *odp_actions;
2294 struct facet *facet;
2298 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2300 flow_extract(packet, 0, in_port, &flow);
2302 /* First look for a related facet. If we find one, account it to that. */
2303 facet = facet_lookup_valid(ofproto, &flow);
2304 if (facet && facet->rule == rule) {
2305 facet_execute(ofproto, facet, packet);
2309 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2310 * create a new facet for it and use that. */
2311 if (rule_lookup(ofproto, &flow) == rule) {
2312 facet = facet_create(ofproto, rule, &flow, packet);
2313 facet_execute(ofproto, facet, packet);
2314 facet_install(ofproto, facet, true);
2318 /* We can't account anything to a facet. If we were to try, then that
2319 * facet would have a non-matching rule, busting our invariants. */
2320 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2321 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2322 size = packet->size;
2323 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2324 odp_actions->size, packet)) {
2325 rule->used = time_msec();
2326 rule->packet_count++;
2327 rule->byte_count += size;
2328 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
2330 ofpbuf_delete(odp_actions);
2333 /* Inserts 'rule' into 'p''s flow table. */
2335 rule_insert(struct ofproto *p, struct rule *rule)
2337 struct rule *displaced_rule;
2339 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2340 if (displaced_rule) {
2341 rule_destroy(p, displaced_rule);
2343 p->need_revalidate = true;
2346 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2347 * 'flow' and an example 'packet' within that flow.
2349 * The caller must already have determined that no facet with an identical
2350 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2351 * 'ofproto''s classifier table. */
2352 static struct facet *
2353 facet_create(struct ofproto *ofproto, struct rule *rule,
2354 const struct flow *flow, const struct ofpbuf *packet)
2356 struct facet *facet;
2358 facet = xzalloc(sizeof *facet);
2359 facet->used = time_msec();
2360 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2361 list_push_back(&rule->facets, &facet->list_node);
2363 facet->flow = *flow;
2364 netflow_flow_init(&facet->nf_flow);
2365 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2367 facet_make_actions(ofproto, facet, packet);
2373 facet_free(struct facet *facet)
2375 free(facet->actions);
2379 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2381 * - Removes 'rule' from the classifier.
2383 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2384 * destroys them), via rule_destroy().
2387 rule_remove(struct ofproto *ofproto, struct rule *rule)
2389 COVERAGE_INC(ofproto_del_rule);
2390 ofproto->need_revalidate = true;
2391 classifier_remove(&ofproto->cls, &rule->cr);
2392 rule_destroy(ofproto, rule);
2395 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2397 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2398 * rule's statistics, via facet_uninstall().
2400 * - Removes 'facet' from its rule and from ofproto->facets.
2403 facet_remove(struct ofproto *ofproto, struct facet *facet)
2405 facet_uninstall(ofproto, facet);
2406 facet_flush_stats(ofproto, facet);
2407 hmap_remove(&ofproto->facets, &facet->hmap_node);
2408 list_remove(&facet->list_node);
2412 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2414 facet_make_actions(struct ofproto *p, struct facet *facet,
2415 const struct ofpbuf *packet)
2417 const struct rule *rule = facet->rule;
2418 struct ofpbuf *odp_actions;
2419 struct action_xlate_ctx ctx;
2421 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2422 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2423 facet->tags = ctx.tags;
2424 facet->may_install = ctx.may_set_up_flow;
2425 facet->nf_flow.output_iface = ctx.nf_output_iface;
2427 if (facet->actions_len != odp_actions->size
2428 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2429 free(facet->actions);
2430 facet->actions_len = odp_actions->size;
2431 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2434 ofpbuf_delete(odp_actions);
2438 facet_put__(struct ofproto *ofproto, struct facet *facet,
2439 const struct nlattr *actions, size_t actions_len,
2440 struct dpif_flow_stats *stats)
2442 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2443 enum dpif_flow_put_flags flags;
2446 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2448 flags |= DPIF_FP_ZERO_STATS;
2449 facet->dp_packet_count = 0;
2450 facet->dp_byte_count = 0;
2453 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2454 odp_flow_key_from_flow(&key, &facet->flow);
2455 assert(key.base == keybuf);
2457 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2458 actions, actions_len, stats);
2461 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2462 * 'zero_stats' is true, clears any existing statistics from the datapath for
2465 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2467 struct dpif_flow_stats stats;
2469 if (facet->may_install
2470 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2471 zero_stats ? &stats : NULL)) {
2472 facet->installed = true;
2476 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2477 * to the accounting hook function in the ofhooks structure. */
2479 facet_account(struct ofproto *ofproto,
2480 struct facet *facet, uint64_t extra_bytes)
2482 uint64_t total_bytes = facet->byte_count + extra_bytes;
2484 if (ofproto->ofhooks->account_flow_cb
2485 && total_bytes > facet->accounted_bytes)
2487 ofproto->ofhooks->account_flow_cb(
2488 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2489 total_bytes - facet->accounted_bytes, ofproto->aux);
2490 facet->accounted_bytes = total_bytes;
2494 /* If 'rule' is installed in the datapath, uninstalls it. */
2496 facet_uninstall(struct ofproto *p, struct facet *facet)
2498 if (facet->installed) {
2499 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2500 struct dpif_flow_stats stats;
2503 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2504 odp_flow_key_from_flow(&key, &facet->flow);
2505 assert(key.base == keybuf);
2507 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2508 facet_update_stats(p, facet, &stats);
2510 facet->installed = false;
2511 facet->dp_packet_count = 0;
2512 facet->dp_byte_count = 0;
2514 assert(facet->dp_packet_count == 0);
2515 assert(facet->dp_byte_count == 0);
2519 /* Returns true if the only action for 'facet' is to send to the controller.
2520 * (We don't report NetFlow expiration messages for such facets because they
2521 * are just part of the control logic for the network, not real traffic). */
2523 facet_is_controller_flow(struct facet *facet)
2526 && facet->rule->n_actions == 1
2527 && action_outputs_to_port(&facet->rule->actions[0],
2528 htons(OFPP_CONTROLLER)));
2531 /* Folds all of 'facet''s statistics into its rule. Also updates the
2532 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2533 * 'facet''s statistics in the datapath should have been zeroed and folded into
2534 * its packet and byte counts before this function is called. */
2536 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2538 assert(!facet->dp_byte_count);
2539 assert(!facet->dp_packet_count);
2541 facet_push_stats(ofproto, facet);
2542 facet_account(ofproto, facet, 0);
2544 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2545 struct ofexpired expired;
2546 expired.flow = facet->flow;
2547 expired.packet_count = facet->packet_count;
2548 expired.byte_count = facet->byte_count;
2549 expired.used = facet->used;
2550 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2553 facet->rule->packet_count += facet->packet_count;
2554 facet->rule->byte_count += facet->byte_count;
2556 /* Reset counters to prevent double counting if 'facet' ever gets
2558 facet->packet_count = 0;
2559 facet->byte_count = 0;
2560 facet->rs_packet_count = 0;
2561 facet->rs_byte_count = 0;
2562 facet->accounted_bytes = 0;
2564 netflow_flow_clear(&facet->nf_flow);
2567 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2568 * Returns it if found, otherwise a null pointer.
2570 * The returned facet might need revalidation; use facet_lookup_valid()
2571 * instead if that is important. */
2572 static struct facet *
2573 facet_find(struct ofproto *ofproto, const struct flow *flow)
2575 struct facet *facet;
2577 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2579 if (flow_equal(flow, &facet->flow)) {
2587 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2588 * Returns it if found, otherwise a null pointer.
2590 * The returned facet is guaranteed to be valid. */
2591 static struct facet *
2592 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2594 struct facet *facet = facet_find(ofproto, flow);
2596 /* The facet we found might not be valid, since we could be in need of
2597 * revalidation. If it is not valid, don't return it. */
2599 && ofproto->need_revalidate
2600 && !facet_revalidate(ofproto, facet)) {
2601 COVERAGE_INC(ofproto_invalidated);
2608 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2610 * - If the rule found is different from 'facet''s current rule, moves
2611 * 'facet' to the new rule and recompiles its actions.
2613 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2614 * where it is and recompiles its actions anyway.
2616 * - If there is none, destroys 'facet'.
2618 * Returns true if 'facet' still exists, false if it has been destroyed. */
2620 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2622 struct action_xlate_ctx ctx;
2623 struct ofpbuf *odp_actions;
2624 struct rule *new_rule;
2625 bool actions_changed;
2627 COVERAGE_INC(facet_revalidate);
2629 /* Determine the new rule. */
2630 new_rule = rule_lookup(ofproto, &facet->flow);
2632 /* No new rule, so delete the facet. */
2633 facet_remove(ofproto, facet);
2637 /* Calculate new ODP actions.
2639 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2640 * emit a NetFlow expiration and, if so, we need to have the old state
2641 * around to properly compose it. */
2642 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2643 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2644 actions_changed = (facet->actions_len != odp_actions->size
2645 || memcmp(facet->actions, odp_actions->data,
2646 facet->actions_len));
2648 /* If the ODP actions changed or the installability changed, then we need
2649 * to talk to the datapath. */
2650 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2651 if (ctx.may_set_up_flow) {
2652 struct dpif_flow_stats stats;
2654 facet_put__(ofproto, facet,
2655 odp_actions->data, odp_actions->size, &stats);
2656 facet_update_stats(ofproto, facet, &stats);
2658 facet_uninstall(ofproto, facet);
2661 /* The datapath flow is gone or has zeroed stats, so push stats out of
2662 * 'facet' into 'rule'. */
2663 facet_flush_stats(ofproto, facet);
2666 /* Update 'facet' now that we've taken care of all the old state. */
2667 facet->tags = ctx.tags;
2668 facet->nf_flow.output_iface = ctx.nf_output_iface;
2669 facet->may_install = ctx.may_set_up_flow;
2670 if (actions_changed) {
2671 free(facet->actions);
2672 facet->actions_len = odp_actions->size;
2673 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2675 if (facet->rule != new_rule) {
2676 COVERAGE_INC(facet_changed_rule);
2677 list_remove(&facet->list_node);
2678 list_push_back(&new_rule->facets, &facet->list_node);
2679 facet->rule = new_rule;
2680 facet->used = new_rule->created;
2681 facet->rs_used = facet->used;
2684 ofpbuf_delete(odp_actions);
2690 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2691 struct rconn_packet_counter *counter)
2693 update_openflow_length(msg);
2694 if (rconn_send(ofconn->rconn, msg, counter)) {
2700 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
2702 queue_tx(msg, ofconn, ofconn->reply_counter);
2706 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2709 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2711 COVERAGE_INC(ofproto_error);
2712 ofconn_send_reply(ofconn, buf);
2717 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2719 ofconn_send_reply(ofconn, make_echo_reply(oh));
2724 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2726 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2727 struct ofp_switch_features *osf;
2729 struct ofport *port;
2731 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2732 osf->datapath_id = htonll(ofproto->datapath_id);
2733 osf->n_buffers = htonl(pktbuf_capacity());
2735 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2736 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2737 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2738 (1u << OFPAT_SET_VLAN_VID) |
2739 (1u << OFPAT_SET_VLAN_PCP) |
2740 (1u << OFPAT_STRIP_VLAN) |
2741 (1u << OFPAT_SET_DL_SRC) |
2742 (1u << OFPAT_SET_DL_DST) |
2743 (1u << OFPAT_SET_NW_SRC) |
2744 (1u << OFPAT_SET_NW_DST) |
2745 (1u << OFPAT_SET_NW_TOS) |
2746 (1u << OFPAT_SET_TP_SRC) |
2747 (1u << OFPAT_SET_TP_DST) |
2748 (1u << OFPAT_ENQUEUE));
2750 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2751 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2754 ofconn_send_reply(ofconn, buf);
2759 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2761 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2763 struct ofp_switch_config *osc;
2767 /* Figure out flags. */
2768 dpif_get_drop_frags(ofproto->dpif, &drop_frags);
2769 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2772 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2773 osc->flags = htons(flags);
2774 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2775 ofconn_send_reply(ofconn, buf);
2781 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2783 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2784 uint16_t flags = ntohs(osc->flags);
2786 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2787 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2788 switch (flags & OFPC_FRAG_MASK) {
2789 case OFPC_FRAG_NORMAL:
2790 dpif_set_drop_frags(ofproto->dpif, false);
2792 case OFPC_FRAG_DROP:
2793 dpif_set_drop_frags(ofproto->dpif, true);
2796 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2802 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2807 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2808 struct action_xlate_ctx *ctx);
2811 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2813 const struct ofport *ofport = get_port(ctx->ofproto, port);
2816 if (ofport->opp.config & OFPPC_NO_FWD) {
2817 /* Forwarding disabled on port. */
2822 * We don't have an ofport record for this port, but it doesn't hurt to
2823 * allow forwarding to it anyhow. Maybe such a port will appear later
2824 * and we're pre-populating the flow table.
2828 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2829 ctx->nf_output_iface = port;
2832 static struct rule *
2833 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2835 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2839 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2841 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2842 uint16_t old_in_port;
2845 /* Look up a flow with 'in_port' as the input port. Then restore the
2846 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2847 * have surprising behavior). */
2848 old_in_port = ctx->flow.in_port;
2849 ctx->flow.in_port = in_port;
2850 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2851 ctx->flow.in_port = old_in_port;
2853 if (ctx->resubmit_hook) {
2854 ctx->resubmit_hook(ctx, rule);
2859 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2863 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2865 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2866 MAX_RESUBMIT_RECURSION);
2871 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2872 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2874 struct ofport *ofport;
2876 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2877 uint16_t odp_port = ofport->odp_port;
2878 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2879 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2882 *nf_output_iface = NF_OUT_FLOOD;
2886 xlate_output_action__(struct action_xlate_ctx *ctx,
2887 uint16_t port, uint16_t max_len)
2890 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2892 ctx->nf_output_iface = NF_OUT_DROP;
2896 add_output_action(ctx, ctx->flow.in_port);
2899 xlate_table_action(ctx, ctx->flow.in_port);
2902 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2903 ctx->odp_actions, &ctx->tags,
2904 &ctx->nf_output_iface,
2905 ctx->ofproto->aux)) {
2906 COVERAGE_INC(ofproto_uninstallable);
2907 ctx->may_set_up_flow = false;
2911 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2912 &ctx->nf_output_iface, ctx->odp_actions);
2915 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2916 &ctx->nf_output_iface, ctx->odp_actions);
2918 case OFPP_CONTROLLER:
2919 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2922 add_output_action(ctx, ODPP_LOCAL);
2925 odp_port = ofp_port_to_odp_port(port);
2926 if (odp_port != ctx->flow.in_port) {
2927 add_output_action(ctx, odp_port);
2932 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2933 ctx->nf_output_iface = NF_OUT_FLOOD;
2934 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2935 ctx->nf_output_iface = prev_nf_output_iface;
2936 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2937 ctx->nf_output_iface != NF_OUT_FLOOD) {
2938 ctx->nf_output_iface = NF_OUT_MULTI;
2943 xlate_output_action(struct action_xlate_ctx *ctx,
2944 const struct ofp_action_output *oao)
2946 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2949 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2950 * optimization, because we're going to add another action that sets the
2951 * priority immediately after, or because there are no actions following the
2954 remove_pop_action(struct action_xlate_ctx *ctx)
2956 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2957 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2958 ctx->last_pop_priority = -1;
2963 add_pop_action(struct action_xlate_ctx *ctx)
2965 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2966 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2967 ctx->last_pop_priority = ctx->odp_actions->size;
2972 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2973 const struct ofp_action_enqueue *oae)
2975 uint16_t ofp_port, odp_port;
2979 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2982 /* Fall back to ordinary output action. */
2983 xlate_output_action__(ctx, ntohs(oae->port), 0);
2987 /* Figure out ODP output port. */
2988 ofp_port = ntohs(oae->port);
2989 if (ofp_port != OFPP_IN_PORT) {
2990 odp_port = ofp_port_to_odp_port(ofp_port);
2992 odp_port = ctx->flow.in_port;
2995 /* Add ODP actions. */
2996 remove_pop_action(ctx);
2997 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2998 add_output_action(ctx, odp_port);
2999 add_pop_action(ctx);
3001 /* Update NetFlow output port. */
3002 if (ctx->nf_output_iface == NF_OUT_DROP) {
3003 ctx->nf_output_iface = odp_port;
3004 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3005 ctx->nf_output_iface = NF_OUT_MULTI;
3010 xlate_set_queue_action(struct action_xlate_ctx *ctx,
3011 const struct nx_action_set_queue *nasq)
3016 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
3019 /* Couldn't translate queue to a priority, so ignore. A warning
3020 * has already been logged. */
3024 remove_pop_action(ctx);
3025 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
3029 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
3031 ovs_be16 tci = ctx->flow.vlan_tci;
3032 if (!(tci & htons(VLAN_CFI))) {
3033 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
3035 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
3036 tci & ~htons(VLAN_CFI));
3040 struct xlate_reg_state {
3046 save_reg_state(const struct action_xlate_ctx *ctx,
3047 struct xlate_reg_state *state)
3049 state->vlan_tci = ctx->flow.vlan_tci;
3050 state->tun_id = ctx->flow.tun_id;
3054 update_reg_state(struct action_xlate_ctx *ctx,
3055 const struct xlate_reg_state *state)
3057 if (ctx->flow.vlan_tci != state->vlan_tci) {
3058 xlate_set_dl_tci(ctx);
3060 if (ctx->flow.tun_id != state->tun_id) {
3061 nl_msg_put_be64(ctx->odp_actions,
3062 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
3067 xlate_nicira_action(struct action_xlate_ctx *ctx,
3068 const struct nx_action_header *nah)
3070 const struct nx_action_resubmit *nar;
3071 const struct nx_action_set_tunnel *nast;
3072 const struct nx_action_set_queue *nasq;
3073 const struct nx_action_multipath *nam;
3074 enum nx_action_subtype subtype = ntohs(nah->subtype);
3075 struct xlate_reg_state state;
3078 assert(nah->vendor == htonl(NX_VENDOR_ID));
3080 case NXAST_RESUBMIT:
3081 nar = (const struct nx_action_resubmit *) nah;
3082 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
3085 case NXAST_SET_TUNNEL:
3086 nast = (const struct nx_action_set_tunnel *) nah;
3087 tun_id = htonll(ntohl(nast->tun_id));
3088 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3089 ctx->flow.tun_id = tun_id;
3092 case NXAST_DROP_SPOOFED_ARP:
3093 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
3094 nl_msg_put_flag(ctx->odp_actions,
3095 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
3099 case NXAST_SET_QUEUE:
3100 nasq = (const struct nx_action_set_queue *) nah;
3101 xlate_set_queue_action(ctx, nasq);
3104 case NXAST_POP_QUEUE:
3105 add_pop_action(ctx);
3108 case NXAST_REG_MOVE:
3109 save_reg_state(ctx, &state);
3110 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
3112 update_reg_state(ctx, &state);
3115 case NXAST_REG_LOAD:
3116 save_reg_state(ctx, &state);
3117 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
3119 update_reg_state(ctx, &state);
3123 /* Nothing to do. */
3126 case NXAST_SET_TUNNEL64:
3127 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
3128 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3129 ctx->flow.tun_id = tun_id;
3132 case NXAST_MULTIPATH:
3133 nam = (const struct nx_action_multipath *) nah;
3134 multipath_execute(nam, &ctx->flow);
3137 /* If you add a new action here that modifies flow data, don't forget to
3138 * update the flow key in ctx->flow at the same time. */
3140 case NXAST_SNAT__OBSOLETE:
3142 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3148 do_xlate_actions(const union ofp_action *in, size_t n_in,
3149 struct action_xlate_ctx *ctx)
3151 struct actions_iterator iter;
3152 const union ofp_action *ia;
3153 const struct ofport *port;
3155 port = get_port(ctx->ofproto, ctx->flow.in_port);
3156 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3157 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3158 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3159 /* Drop this flow. */
3163 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3164 enum ofp_action_type type = ntohs(ia->type);
3165 const struct ofp_action_dl_addr *oada;
3169 xlate_output_action(ctx, &ia->output);
3172 case OFPAT_SET_VLAN_VID:
3173 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3174 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3175 xlate_set_dl_tci(ctx);
3178 case OFPAT_SET_VLAN_PCP:
3179 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3180 ctx->flow.vlan_tci |= htons(
3181 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3182 xlate_set_dl_tci(ctx);
3185 case OFPAT_STRIP_VLAN:
3186 ctx->flow.vlan_tci = htons(0);
3187 xlate_set_dl_tci(ctx);
3190 case OFPAT_SET_DL_SRC:
3191 oada = ((struct ofp_action_dl_addr *) ia);
3192 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3193 oada->dl_addr, ETH_ADDR_LEN);
3194 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3197 case OFPAT_SET_DL_DST:
3198 oada = ((struct ofp_action_dl_addr *) ia);
3199 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3200 oada->dl_addr, ETH_ADDR_LEN);
3201 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3204 case OFPAT_SET_NW_SRC:
3205 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3206 ia->nw_addr.nw_addr);
3207 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3210 case OFPAT_SET_NW_DST:
3211 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3212 ia->nw_addr.nw_addr);
3213 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3216 case OFPAT_SET_NW_TOS:
3217 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3219 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3222 case OFPAT_SET_TP_SRC:
3223 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3224 ia->tp_port.tp_port);
3225 ctx->flow.tp_src = ia->tp_port.tp_port;
3228 case OFPAT_SET_TP_DST:
3229 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3230 ia->tp_port.tp_port);
3231 ctx->flow.tp_dst = ia->tp_port.tp_port;
3235 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3239 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3243 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3250 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3251 struct ofproto *ofproto, const struct flow *flow,
3252 const struct ofpbuf *packet)
3254 ctx->ofproto = ofproto;
3256 ctx->packet = packet;
3257 ctx->resubmit_hook = NULL;
3258 ctx->check_special = true;
3262 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
3263 const struct ofpbuf *packet)
3265 struct ofport *ofport;
3267 ofport = get_port(ofproto, flow->in_port);
3268 if (ofport && ofport->cfm) {
3269 cfm_process_heartbeat(ofport->cfm, packet);
3273 static struct ofpbuf *
3274 xlate_actions(struct action_xlate_ctx *ctx,
3275 const union ofp_action *in, size_t n_in)
3277 COVERAGE_INC(ofproto_ofp2odp);
3279 ctx->odp_actions = ofpbuf_new(512);
3281 ctx->may_set_up_flow = true;
3282 ctx->nf_output_iface = NF_OUT_DROP;
3284 ctx->last_pop_priority = -1;
3286 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
3288 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
3290 ctx->may_set_up_flow = false;
3291 } else if (ctx->check_special
3292 && ctx->ofproto->ofhooks->special_cb
3293 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3294 ctx->ofproto->aux)) {
3295 ctx->may_set_up_flow = false;
3297 do_xlate_actions(in, n_in, ctx);
3300 remove_pop_action(ctx);
3302 /* Check with in-band control to see if we're allowed to set up this
3304 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3305 ctx->odp_actions->data, ctx->odp_actions->size)) {
3306 ctx->may_set_up_flow = false;
3309 return ctx->odp_actions;
3312 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3313 * error message code (composed with ofp_mkerr()) for the caller to propagate
3314 * upward. Otherwise, returns 0.
3316 * The log message mentions 'msg_type'. */
3318 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3320 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
3321 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
3322 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3323 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3326 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3333 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3335 struct ofproto *p = ofconn_get_ofproto(ofconn);
3336 struct ofp_packet_out *opo;
3337 struct ofpbuf payload, *buffer;
3338 union ofp_action *ofp_actions;
3339 struct action_xlate_ctx ctx;
3340 struct ofpbuf *odp_actions;
3341 struct ofpbuf request;
3343 size_t n_ofp_actions;
3347 COVERAGE_INC(ofproto_packet_out);
3349 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3354 /* Get ofp_packet_out. */
3355 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3356 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3359 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3360 &ofp_actions, &n_ofp_actions);
3366 if (opo->buffer_id != htonl(UINT32_MAX)) {
3367 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
3369 if (error || !buffer) {
3378 /* Extract flow, check actions. */
3379 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3381 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3387 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3388 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3389 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3390 ofpbuf_delete(odp_actions);
3393 ofpbuf_delete(buffer);
3398 update_port_config(struct ofproto *p, struct ofport *port,
3399 uint32_t config, uint32_t mask)
3401 mask &= config ^ port->opp.config;
3402 if (mask & OFPPC_PORT_DOWN) {
3403 if (config & OFPPC_PORT_DOWN) {
3404 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3406 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3409 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3410 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3411 if (mask & REVALIDATE_BITS) {
3412 COVERAGE_INC(ofproto_costly_flags);
3413 port->opp.config ^= mask & REVALIDATE_BITS;
3414 p->need_revalidate = true;
3416 #undef REVALIDATE_BITS
3417 if (mask & OFPPC_NO_PACKET_IN) {
3418 port->opp.config ^= OFPPC_NO_PACKET_IN;
3423 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3425 struct ofproto *p = ofconn_get_ofproto(ofconn);
3426 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3427 struct ofport *port;
3430 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3435 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3437 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3438 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3439 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3441 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3442 if (opm->advertise) {
3443 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3449 static struct ofpbuf *
3450 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3452 struct ofp_stats_reply *osr;
3455 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3456 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3458 osr->flags = htons(0);
3462 static struct ofpbuf *
3463 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3465 const struct ofp_stats_request *osr
3466 = (const struct ofp_stats_request *) request;
3467 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3471 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3472 struct ofpbuf **msgp)
3474 struct ofpbuf *msg = *msgp;
3475 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3476 if (nbytes + msg->size > UINT16_MAX) {
3477 struct ofp_stats_reply *reply = msg->data;
3478 reply->flags = htons(OFPSF_REPLY_MORE);
3479 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3480 ofconn_send_reply(ofconn, msg);
3482 return ofpbuf_put_uninit(*msgp, nbytes);
3485 static struct ofpbuf *
3486 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3488 struct nicira_stats_msg *nsm;
3491 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3492 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3493 nsm->type = htons(OFPST_VENDOR);
3494 nsm->flags = htons(0);
3495 nsm->vendor = htonl(NX_VENDOR_ID);
3496 nsm->subtype = subtype;
3500 static struct ofpbuf *
3501 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3503 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3507 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3508 struct ofpbuf **msgp)
3510 struct ofpbuf *msg = *msgp;
3511 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3512 if (nbytes + msg->size > UINT16_MAX) {
3513 struct nicira_stats_msg *reply = msg->data;
3514 reply->flags = htons(OFPSF_REPLY_MORE);
3515 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3516 ofconn_send_reply(ofconn, msg);
3518 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3522 handle_desc_stats_request(struct ofconn *ofconn,
3523 const struct ofp_header *request)
3525 struct ofproto *p = ofconn_get_ofproto(ofconn);
3526 struct ofp_desc_stats *ods;
3529 msg = start_ofp_stats_reply(request, sizeof *ods);
3530 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3531 memset(ods, 0, sizeof *ods);
3532 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3533 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3534 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3535 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3536 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3537 ofconn_send_reply(ofconn, msg);
3543 handle_table_stats_request(struct ofconn *ofconn,
3544 const struct ofp_header *request)
3546 struct ofproto *p = ofconn_get_ofproto(ofconn);
3547 struct ofp_table_stats *ots;
3550 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3552 /* Classifier table. */
3553 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3554 memset(ots, 0, sizeof *ots);
3555 strcpy(ots->name, "classifier");
3556 ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10
3557 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3558 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3559 ots->active_count = htonl(classifier_count(&p->cls));
3560 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3561 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3563 ofconn_send_reply(ofconn, msg);
3568 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3569 struct ofpbuf **msgp)
3571 struct netdev_stats stats;
3572 struct ofp_port_stats *ops;
3574 /* Intentionally ignore return value, since errors will set
3575 * 'stats' to all-1s, which is correct for OpenFlow, and
3576 * netdev_get_stats() will log errors. */
3577 netdev_get_stats(port->netdev, &stats);
3579 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3580 ops->port_no = htons(port->opp.port_no);
3581 memset(ops->pad, 0, sizeof ops->pad);
3582 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3583 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3584 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3585 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3586 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3587 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3588 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3589 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3590 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3591 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3592 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3593 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3597 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3599 struct ofproto *p = ofconn_get_ofproto(ofconn);
3600 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3601 struct ofp_port_stats *ops;
3603 struct ofport *port;
3605 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3606 if (psr->port_no != htons(OFPP_NONE)) {
3607 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3609 append_port_stat(port, ofconn, &msg);
3612 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3613 append_port_stat(port, ofconn, &msg);
3617 ofconn_send_reply(ofconn, msg);
3622 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
3624 long long int msecs = time_msec() - start;
3625 *sec = msecs / 1000;
3626 *nsec = (msecs % 1000) * (1000 * 1000);
3630 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
3634 calc_flow_duration__(start, &sec, &nsec);
3635 *sec_be = htonl(sec);
3636 *nsec_be = htonl(nsec);
3640 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3641 ovs_be16 out_port, struct ofpbuf **replyp)
3643 struct ofp_flow_stats *ofs;
3644 uint64_t packet_count, byte_count;
3646 size_t act_len, len;
3648 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3652 act_len = sizeof *rule->actions * rule->n_actions;
3653 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3655 rule_get_stats(rule, &packet_count, &byte_count);
3657 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3658 ofs->length = htons(len);
3661 ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn),
3662 &ofs->match, rule->flow_cookie, &cookie);
3663 put_32aligned_be64(&ofs->cookie, cookie);
3664 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3665 ofs->priority = htons(rule->cr.priority);
3666 ofs->idle_timeout = htons(rule->idle_timeout);
3667 ofs->hard_timeout = htons(rule->hard_timeout);
3668 memset(ofs->pad2, 0, sizeof ofs->pad2);
3669 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3670 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3671 if (rule->n_actions > 0) {
3672 memcpy(ofs->actions, rule->actions, act_len);
3677 is_valid_table(uint8_t table_id)
3679 if (table_id == 0 || table_id == 0xff) {
3682 /* It would probably be better to reply with an error but there doesn't
3683 * seem to be any appropriate value, so that might just be
3685 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3692 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3694 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3695 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3696 struct ofpbuf *reply;
3698 COVERAGE_INC(ofproto_flows_req);
3699 reply = start_ofp_stats_reply(oh, 1024);
3700 if (is_valid_table(fsr->table_id)) {
3701 struct cls_cursor cursor;
3702 struct cls_rule target;
3705 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3707 cls_cursor_init(&cursor, &ofproto->cls, &target);
3708 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3709 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3712 ofconn_send_reply(ofconn, reply);
3718 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3719 ovs_be16 out_port, struct ofpbuf **replyp)
3721 struct nx_flow_stats *nfs;
3722 uint64_t packet_count, byte_count;
3723 size_t act_len, start_len;
3724 struct ofpbuf *reply;
3726 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3730 rule_get_stats(rule, &packet_count, &byte_count);
3732 act_len = sizeof *rule->actions * rule->n_actions;
3734 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3735 start_len = (*replyp)->size;
3738 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3741 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3742 nfs->cookie = rule->flow_cookie;
3743 nfs->priority = htons(rule->cr.priority);
3744 nfs->idle_timeout = htons(rule->idle_timeout);
3745 nfs->hard_timeout = htons(rule->hard_timeout);
3746 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3747 memset(nfs->pad2, 0, sizeof nfs->pad2);
3748 nfs->packet_count = htonll(packet_count);
3749 nfs->byte_count = htonll(byte_count);
3750 if (rule->n_actions > 0) {
3751 ofpbuf_put(reply, rule->actions, act_len);
3753 nfs->length = htons(reply->size - start_len);
3757 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3759 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3760 struct nx_flow_stats_request *nfsr;
3761 struct cls_rule target;
3762 struct ofpbuf *reply;
3766 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3768 /* Dissect the message. */
3769 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3770 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3775 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3778 COVERAGE_INC(ofproto_flows_req);
3779 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3780 if (is_valid_table(nfsr->table_id)) {
3781 struct cls_cursor cursor;
3784 cls_cursor_init(&cursor, &ofproto->cls, &target);
3785 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3786 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3789 ofconn_send_reply(ofconn, reply);
3795 flow_stats_ds(struct rule *rule, struct ds *results)
3797 uint64_t packet_count, byte_count;
3798 size_t act_len = sizeof *rule->actions * rule->n_actions;
3800 rule_get_stats(rule, &packet_count, &byte_count);
3802 ds_put_format(results, "duration=%llds, ",
3803 (time_msec() - rule->created) / 1000);
3804 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3805 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3806 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3807 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3808 cls_rule_format(&rule->cr, results);
3809 ds_put_char(results, ',');
3811 ofp_print_actions(results, &rule->actions->header, act_len);
3813 ds_put_cstr(results, "drop");
3815 ds_put_cstr(results, "\n");
3818 /* Adds a pretty-printed description of all flows to 'results', including
3819 * hidden flows (e.g., set up by in-band control). */
3821 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3823 struct cls_cursor cursor;
3826 cls_cursor_init(&cursor, &p->cls, NULL);
3827 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3828 flow_stats_ds(rule, results);
3833 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3834 ovs_be16 out_port, uint8_t table_id,
3835 struct ofp_aggregate_stats_reply *oasr)
3837 uint64_t total_packets = 0;
3838 uint64_t total_bytes = 0;
3841 COVERAGE_INC(ofproto_agg_request);
3843 if (is_valid_table(table_id)) {
3844 struct cls_cursor cursor;
3847 cls_cursor_init(&cursor, &ofproto->cls, target);
3848 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3849 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3850 uint64_t packet_count;
3851 uint64_t byte_count;
3853 rule_get_stats(rule, &packet_count, &byte_count);
3855 total_packets += packet_count;
3856 total_bytes += byte_count;
3862 oasr->flow_count = htonl(n_flows);
3863 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3864 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3865 memset(oasr->pad, 0, sizeof oasr->pad);
3869 handle_aggregate_stats_request(struct ofconn *ofconn,
3870 const struct ofp_header *oh)
3872 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3873 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3874 struct ofp_aggregate_stats_reply *reply;
3875 struct cls_rule target;
3878 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3881 msg = start_ofp_stats_reply(oh, sizeof *reply);
3882 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3883 query_aggregate_stats(ofproto, &target, request->out_port,
3884 request->table_id, reply);
3885 ofconn_send_reply(ofconn, msg);
3890 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3892 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3893 struct nx_aggregate_stats_request *request;
3894 struct ofp_aggregate_stats_reply *reply;
3895 struct cls_rule target;
3900 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3902 /* Dissect the message. */
3903 request = ofpbuf_pull(&b, sizeof *request);
3904 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3909 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3913 COVERAGE_INC(ofproto_flows_req);
3914 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3915 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3916 query_aggregate_stats(ofproto, &target, request->out_port,
3917 request->table_id, reply);
3918 ofconn_send_reply(ofconn, buf);
3923 struct queue_stats_cbdata {
3924 struct ofconn *ofconn;
3925 struct ofport *ofport;
3930 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3931 const struct netdev_queue_stats *stats)
3933 struct ofp_queue_stats *reply;
3935 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3936 reply->port_no = htons(cbdata->ofport->opp.port_no);
3937 memset(reply->pad, 0, sizeof reply->pad);
3938 reply->queue_id = htonl(queue_id);
3939 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3940 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3941 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3945 handle_queue_stats_dump_cb(uint32_t queue_id,
3946 struct netdev_queue_stats *stats,
3949 struct queue_stats_cbdata *cbdata = cbdata_;
3951 put_queue_stats(cbdata, queue_id, stats);
3955 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3956 struct queue_stats_cbdata *cbdata)
3958 cbdata->ofport = port;
3959 if (queue_id == OFPQ_ALL) {
3960 netdev_dump_queue_stats(port->netdev,
3961 handle_queue_stats_dump_cb, cbdata);
3963 struct netdev_queue_stats stats;
3965 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3966 put_queue_stats(cbdata, queue_id, &stats);
3972 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3974 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3975 const struct ofp_queue_stats_request *qsr;
3976 struct queue_stats_cbdata cbdata;
3977 struct ofport *port;
3978 unsigned int port_no;
3981 qsr = ofputil_stats_body(oh);
3983 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3986 COVERAGE_INC(ofproto_queue_req);
3988 cbdata.ofconn = ofconn;
3989 cbdata.msg = start_ofp_stats_reply(oh, 128);
3991 port_no = ntohs(qsr->port_no);
3992 queue_id = ntohl(qsr->queue_id);
3993 if (port_no == OFPP_ALL) {
3994 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3995 handle_queue_stats_for_port(port, queue_id, &cbdata);
3997 } else if (port_no < ofproto->max_ports) {
3998 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
4000 handle_queue_stats_for_port(port, queue_id, &cbdata);
4003 ofpbuf_delete(cbdata.msg);
4004 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
4006 ofconn_send_reply(ofconn, cbdata.msg);
4011 /* Updates 'facet''s used time. Caller is responsible for calling
4012 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4014 facet_update_time(struct ofproto *ofproto, struct facet *facet,
4017 if (used > facet->used) {
4019 if (used > facet->rule->used) {
4020 facet->rule->used = used;
4022 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4026 /* Folds the statistics from 'stats' into the counters in 'facet'.
4028 * Because of the meaning of a facet's counters, it only makes sense to do this
4029 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
4030 * packet that was sent by hand or if it represents statistics that have been
4031 * cleared out of the datapath. */
4033 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
4034 const struct dpif_flow_stats *stats)
4036 if (stats->n_packets || stats->used > facet->used) {
4037 facet_update_time(ofproto, facet, stats->used);
4038 facet->packet_count += stats->n_packets;
4039 facet->byte_count += stats->n_bytes;
4040 facet_push_stats(ofproto, facet);
4041 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
4046 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
4048 uint64_t rs_packets, rs_bytes;
4050 assert(facet->packet_count >= facet->rs_packet_count);
4051 assert(facet->byte_count >= facet->rs_byte_count);
4052 assert(facet->used >= facet->rs_used);
4054 rs_packets = facet->packet_count - facet->rs_packet_count;
4055 rs_bytes = facet->byte_count - facet->rs_byte_count;
4057 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
4058 facet->rs_packet_count = facet->packet_count;
4059 facet->rs_byte_count = facet->byte_count;
4060 facet->rs_used = facet->used;
4062 flow_push_stats(ofproto, facet->rule, &facet->flow,
4063 rs_packets, rs_bytes, facet->used);
4067 struct ofproto_push {
4068 struct action_xlate_ctx ctx;
4075 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
4077 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
4080 rule->packet_count += push->packets;
4081 rule->byte_count += push->bytes;
4082 rule->used = MAX(push->used, rule->used);
4086 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4087 * 'rule''s actions. */
4089 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
4090 struct flow *flow, uint64_t packets, uint64_t bytes,
4093 struct ofproto_push push;
4095 push.packets = packets;
4099 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
4100 push.ctx.resubmit_hook = push_resubmit;
4101 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
4104 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4105 * in which no matching flow already exists in the flow table.
4107 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4108 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
4109 * OpenFlow error code as encoded by ofp_mkerr() on failure.
4111 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4114 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
4116 struct ofproto *p = ofconn_get_ofproto(ofconn);
4117 struct ofpbuf *packet;
4122 if (fm->flags & OFPFF_CHECK_OVERLAP
4123 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
4124 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
4128 if (fm->buffer_id != UINT32_MAX) {
4129 error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id,
4133 in_port = UINT16_MAX;
4136 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
4137 fm->idle_timeout, fm->hard_timeout, fm->cookie,
4138 fm->flags & OFPFF_SEND_FLOW_REM);
4139 rule_insert(p, rule);
4141 rule_execute(p, rule, in_port, packet);
4146 static struct rule *
4147 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
4149 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
4153 send_buffered_packet(struct ofconn *ofconn,
4154 struct rule *rule, uint32_t buffer_id)
4156 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4157 struct ofpbuf *packet;
4161 if (buffer_id == UINT32_MAX) {
4165 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
4170 rule_execute(ofproto, rule, in_port, packet);
4175 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4177 struct modify_flows_cbdata {
4178 struct ofproto *ofproto;
4179 const struct flow_mod *fm;
4183 static int modify_flow(struct ofproto *, const struct flow_mod *,
4186 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4187 * encoded by ofp_mkerr() on failure.
4189 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4192 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
4194 struct ofproto *p = ofconn_get_ofproto(ofconn);
4195 struct rule *match = NULL;
4196 struct cls_cursor cursor;
4199 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4200 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4201 if (!rule_is_hidden(rule)) {
4203 modify_flow(p, fm, rule);
4208 /* This credits the packet to whichever flow happened to match last.
4209 * That's weird. Maybe we should do a lookup for the flow that
4210 * actually matches the packet? Who knows. */
4211 send_buffered_packet(ofconn, match, fm->buffer_id);
4214 return add_flow(ofconn, fm);
4218 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4219 * code as encoded by ofp_mkerr() on failure.
4221 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4224 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4226 struct ofproto *p = ofconn_get_ofproto(ofconn);
4227 struct rule *rule = find_flow_strict(p, fm);
4228 if (rule && !rule_is_hidden(rule)) {
4229 modify_flow(p, fm, rule);
4230 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4232 return add_flow(ofconn, fm);
4236 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4237 * been identified as a flow in 'p''s flow table to be modified, by changing
4238 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4239 * ofp_action[] structures). */
4241 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4243 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4245 rule->flow_cookie = fm->cookie;
4247 /* If the actions are the same, do nothing. */
4248 if (fm->n_actions == rule->n_actions
4250 || !memcmp(fm->actions, rule->actions, actions_len))) {
4254 /* Replace actions. */
4255 free(rule->actions);
4256 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4257 rule->n_actions = fm->n_actions;
4259 p->need_revalidate = true;
4264 /* OFPFC_DELETE implementation. */
4266 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4268 /* Implements OFPFC_DELETE. */
4270 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4272 struct rule *rule, *next_rule;
4273 struct cls_cursor cursor;
4275 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4276 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4277 delete_flow(p, rule, htons(fm->out_port));
4281 /* Implements OFPFC_DELETE_STRICT. */
4283 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4285 struct rule *rule = find_flow_strict(p, fm);
4287 delete_flow(p, rule, htons(fm->out_port));
4291 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4292 * been identified as a flow to delete from 'p''s flow table, by deleting the
4293 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4296 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4297 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4298 * specified 'out_port'. */
4300 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4302 if (rule_is_hidden(rule)) {
4306 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4310 rule_send_removed(p, rule, OFPRR_DELETE);
4311 rule_remove(p, rule);
4315 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4317 struct ofproto *p = ofconn_get_ofproto(ofconn);
4321 error = reject_slave_controller(ofconn, "flow_mod");
4326 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn));
4331 /* We do not support the emergency flow cache. It will hopefully get
4332 * dropped from OpenFlow in the near future. */
4333 if (fm.flags & OFPFF_EMERG) {
4334 /* There isn't a good fit for an error code, so just state that the
4335 * flow table is full. */
4336 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4339 error = validate_actions(fm.actions, fm.n_actions,
4340 &fm.cr.flow, p->max_ports);
4345 switch (fm.command) {
4347 return add_flow(ofconn, &fm);
4350 return modify_flows_loose(ofconn, &fm);
4352 case OFPFC_MODIFY_STRICT:
4353 return modify_flow_strict(ofconn, &fm);
4356 delete_flows_loose(p, &fm);
4359 case OFPFC_DELETE_STRICT:
4360 delete_flow_strict(p, &fm);
4364 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4369 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4371 const struct nxt_tun_id_cookie *msg
4372 = (const struct nxt_tun_id_cookie *) oh;
4373 enum nx_flow_format flow_format;
4375 flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4376 ofconn_set_flow_format(ofconn, flow_format);
4382 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4384 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4385 struct nx_role_request *reply;
4389 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
4390 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4392 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4395 role = ntohl(nrr->role);
4396 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4397 && role != NX_ROLE_SLAVE) {
4398 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4400 /* There's no good error code for this. */
4401 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4404 if (role == NX_ROLE_MASTER) {
4405 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4406 struct ofconn *other;
4408 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
4409 if (ofconn_get_role(other) == NX_ROLE_MASTER) {
4410 ofconn_set_role(other, NX_ROLE_SLAVE);
4414 ofconn_set_role(ofconn, role);
4416 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4417 reply->role = htonl(role);
4418 ofconn_send_reply(ofconn, buf);
4424 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4426 const struct nxt_set_flow_format *msg
4427 = (const struct nxt_set_flow_format *) oh;
4430 format = ntohl(msg->format);
4431 if (format == NXFF_OPENFLOW10
4432 || format == NXFF_TUN_ID_FROM_COOKIE
4433 || format == NXFF_NXM) {
4434 ofconn_set_flow_format(ofconn, format);
4437 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4442 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4444 struct ofp_header *ob;
4447 /* Currently, everything executes synchronously, so we can just
4448 * immediately send the barrier reply. */
4449 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4450 ofconn_send_reply(ofconn, buf);
4455 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4457 const struct ofp_header *oh = msg->data;
4458 const struct ofputil_msg_type *type;
4461 error = ofputil_decode_msg_type(oh, &type);
4466 switch (ofputil_msg_type_code(type)) {
4467 /* OpenFlow requests. */
4468 case OFPUTIL_OFPT_ECHO_REQUEST:
4469 return handle_echo_request(ofconn, oh);
4471 case OFPUTIL_OFPT_FEATURES_REQUEST:
4472 return handle_features_request(ofconn, oh);
4474 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4475 return handle_get_config_request(ofconn, oh);
4477 case OFPUTIL_OFPT_SET_CONFIG:
4478 return handle_set_config(ofconn, msg->data);
4480 case OFPUTIL_OFPT_PACKET_OUT:
4481 return handle_packet_out(ofconn, oh);
4483 case OFPUTIL_OFPT_PORT_MOD:
4484 return handle_port_mod(ofconn, oh);
4486 case OFPUTIL_OFPT_FLOW_MOD:
4487 return handle_flow_mod(ofconn, oh);
4489 case OFPUTIL_OFPT_BARRIER_REQUEST:
4490 return handle_barrier_request(ofconn, oh);
4492 /* OpenFlow replies. */
4493 case OFPUTIL_OFPT_ECHO_REPLY:
4496 /* Nicira extension requests. */
4497 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4498 return handle_tun_id_from_cookie(ofconn, oh);
4500 case OFPUTIL_NXT_ROLE_REQUEST:
4501 return handle_role_request(ofconn, oh);
4503 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4504 return handle_nxt_set_flow_format(ofconn, oh);
4506 case OFPUTIL_NXT_FLOW_MOD:
4507 return handle_flow_mod(ofconn, oh);
4509 /* OpenFlow statistics requests. */
4510 case OFPUTIL_OFPST_DESC_REQUEST:
4511 return handle_desc_stats_request(ofconn, oh);
4513 case OFPUTIL_OFPST_FLOW_REQUEST:
4514 return handle_flow_stats_request(ofconn, oh);
4516 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4517 return handle_aggregate_stats_request(ofconn, oh);
4519 case OFPUTIL_OFPST_TABLE_REQUEST:
4520 return handle_table_stats_request(ofconn, oh);
4522 case OFPUTIL_OFPST_PORT_REQUEST:
4523 return handle_port_stats_request(ofconn, oh);
4525 case OFPUTIL_OFPST_QUEUE_REQUEST:
4526 return handle_queue_stats_request(ofconn, oh);
4528 /* Nicira extension statistics requests. */
4529 case OFPUTIL_NXST_FLOW_REQUEST:
4530 return handle_nxst_flow(ofconn, oh);
4532 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4533 return handle_nxst_aggregate(ofconn, oh);
4535 case OFPUTIL_INVALID:
4536 case OFPUTIL_OFPT_HELLO:
4537 case OFPUTIL_OFPT_ERROR:
4538 case OFPUTIL_OFPT_FEATURES_REPLY:
4539 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4540 case OFPUTIL_OFPT_PACKET_IN:
4541 case OFPUTIL_OFPT_FLOW_REMOVED:
4542 case OFPUTIL_OFPT_PORT_STATUS:
4543 case OFPUTIL_OFPT_BARRIER_REPLY:
4544 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4545 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4546 case OFPUTIL_OFPST_DESC_REPLY:
4547 case OFPUTIL_OFPST_FLOW_REPLY:
4548 case OFPUTIL_OFPST_QUEUE_REPLY:
4549 case OFPUTIL_OFPST_PORT_REPLY:
4550 case OFPUTIL_OFPST_TABLE_REPLY:
4551 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4552 case OFPUTIL_NXT_ROLE_REPLY:
4553 case OFPUTIL_NXT_FLOW_REMOVED:
4554 case OFPUTIL_NXST_FLOW_REPLY:
4555 case OFPUTIL_NXST_AGGREGATE_REPLY:
4557 if (VLOG_IS_WARN_ENABLED()) {
4558 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4559 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4562 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4563 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4565 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4571 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4573 int error = handle_openflow__(ofconn, ofp_msg);
4575 send_error_oh(ofconn, ofp_msg->data, error);
4577 COVERAGE_INC(ofproto_recv_openflow);
4581 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4583 struct facet *facet;
4586 /* Obtain in_port and tun_id, at least. */
4587 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4589 /* Set header pointers in 'flow'. */
4590 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4592 if (cfm_should_process_flow(&flow)) {
4593 ofproto_process_cfm(p, &flow, upcall->packet);
4594 ofpbuf_delete(upcall->packet);
4596 } else if (p->ofhooks->special_cb
4597 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4598 ofpbuf_delete(upcall->packet);
4602 /* Check with in-band control to see if this packet should be sent
4603 * to the local port regardless of the flow table. */
4604 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4605 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
4608 facet = facet_lookup_valid(p, &flow);
4610 struct rule *rule = rule_lookup(p, &flow);
4612 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4613 struct ofport *port = get_port(p, flow.in_port);
4615 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4616 COVERAGE_INC(ofproto_no_packet_in);
4617 /* XXX install 'drop' flow entry */
4618 ofpbuf_delete(upcall->packet);
4622 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4626 COVERAGE_INC(ofproto_packet_in);
4627 send_packet_in(p, upcall, &flow, false);
4631 facet = facet_create(p, rule, &flow, upcall->packet);
4632 } else if (!facet->may_install) {
4633 /* The facet is not installable, that is, we need to process every
4634 * packet, so process the current packet's actions into 'facet'. */
4635 facet_make_actions(p, facet, upcall->packet);
4638 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4640 * Extra-special case for fail-open mode.
4642 * We are in fail-open mode and the packet matched the fail-open rule,
4643 * but we are connected to a controller too. We should send the packet
4644 * up to the controller in the hope that it will try to set up a flow
4645 * and thereby allow us to exit fail-open.
4647 * See the top-level comment in fail-open.c for more information.
4649 send_packet_in(p, upcall, &flow, true);
4652 facet_execute(p, facet, upcall->packet);
4653 facet_install(p, facet, false);
4657 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4661 switch (upcall->type) {
4662 case DPIF_UC_ACTION:
4663 COVERAGE_INC(ofproto_ctlr_action);
4664 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4665 send_packet_in(p, upcall, &flow, false);
4668 case DPIF_UC_SAMPLE:
4670 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4671 ofproto_sflow_received(p->sflow, upcall, &flow);
4673 ofpbuf_delete(upcall->packet);
4677 handle_miss_upcall(p, upcall);
4680 case DPIF_N_UC_TYPES:
4682 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4687 /* Flow expiration. */
4689 static int ofproto_dp_max_idle(const struct ofproto *);
4690 static void ofproto_update_stats(struct ofproto *);
4691 static void rule_expire(struct ofproto *, struct rule *);
4692 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4694 /* This function is called periodically by ofproto_run(). Its job is to
4695 * collect updates for the flows that have been installed into the datapath,
4696 * most importantly when they last were used, and then use that information to
4697 * expire flows that have not been used recently.
4699 * Returns the number of milliseconds after which it should be called again. */
4701 ofproto_expire(struct ofproto *ofproto)
4703 struct rule *rule, *next_rule;
4704 struct cls_cursor cursor;
4707 /* Update stats for each flow in the datapath. */
4708 ofproto_update_stats(ofproto);
4710 /* Expire facets that have been idle too long. */
4711 dp_max_idle = ofproto_dp_max_idle(ofproto);
4712 ofproto_expire_facets(ofproto, dp_max_idle);
4714 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4715 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4716 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4717 rule_expire(ofproto, rule);
4720 /* Let the hook know that we're at a stable point: all outstanding data
4721 * in existing flows has been accounted to the account_cb. Thus, the
4722 * hook can now reasonably do operations that depend on having accurate
4723 * flow volume accounting (currently, that's just bond rebalancing). */
4724 if (ofproto->ofhooks->account_checkpoint_cb) {
4725 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4728 return MIN(dp_max_idle, 1000);
4731 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4733 * This function also pushes statistics updates to rules which each facet
4734 * resubmits into. Generally these statistics will be accurate. However, if a
4735 * facet changes the rule it resubmits into at some time in between
4736 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4737 * old rule will be incorrectly attributed to the new rule. This could be
4738 * avoided by calling ofproto_update_stats() whenever rules are created or
4739 * deleted. However, the performance impact of making so many calls to the
4740 * datapath do not justify the benefit of having perfectly accurate statistics.
4743 ofproto_update_stats(struct ofproto *p)
4745 const struct dpif_flow_stats *stats;
4746 struct dpif_flow_dump dump;
4747 const struct nlattr *key;
4750 dpif_flow_dump_start(&dump, p->dpif);
4751 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4752 struct facet *facet;
4755 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4759 odp_flow_key_format(key, key_len, &s);
4760 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4766 facet = facet_find(p, &flow);
4768 if (facet && facet->installed) {
4770 if (stats->n_packets >= facet->dp_packet_count) {
4771 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4773 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4776 if (stats->n_bytes >= facet->dp_byte_count) {
4777 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4779 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4782 facet->dp_packet_count = stats->n_packets;
4783 facet->dp_byte_count = stats->n_bytes;
4785 facet_update_time(p, facet, stats->used);
4786 facet_account(p, facet, stats->n_bytes);
4787 facet_push_stats(p, facet);
4789 /* There's a flow in the datapath that we know nothing about.
4791 COVERAGE_INC(ofproto_unexpected_rule);
4792 dpif_flow_del(p->dpif, key, key_len, NULL);
4795 dpif_flow_dump_done(&dump);
4798 /* Calculates and returns the number of milliseconds of idle time after which
4799 * facets should expire from the datapath and we should fold their statistics
4800 * into their parent rules in userspace. */
4802 ofproto_dp_max_idle(const struct ofproto *ofproto)
4805 * Idle time histogram.
4807 * Most of the time a switch has a relatively small number of facets. When
4808 * this is the case we might as well keep statistics for all of them in
4809 * userspace and to cache them in the kernel datapath for performance as
4812 * As the number of facets increases, the memory required to maintain
4813 * statistics about them in userspace and in the kernel becomes
4814 * significant. However, with a large number of facets it is likely that
4815 * only a few of them are "heavy hitters" that consume a large amount of
4816 * bandwidth. At this point, only heavy hitters are worth caching in the
4817 * kernel and maintaining in userspaces; other facets we can discard.
4819 * The technique used to compute the idle time is to build a histogram with
4820 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4821 * that is installed in the kernel gets dropped in the appropriate bucket.
4822 * After the histogram has been built, we compute the cutoff so that only
4823 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4824 * cached. At least the most-recently-used bucket of facets is kept, so
4825 * actually an arbitrary number of facets can be kept in any given
4826 * expiration run (though the next run will delete most of those unless
4827 * they receive additional data).
4829 * This requires a second pass through the facets, in addition to the pass
4830 * made by ofproto_update_stats(), because the former function never looks
4831 * at uninstallable facets.
4833 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4834 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4835 int buckets[N_BUCKETS] = { 0 };
4836 struct facet *facet;
4841 total = hmap_count(&ofproto->facets);
4842 if (total <= 1000) {
4843 return N_BUCKETS * BUCKET_WIDTH;
4846 /* Build histogram. */
4848 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4849 long long int idle = now - facet->used;
4850 int bucket = (idle <= 0 ? 0
4851 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4852 : (unsigned int) idle / BUCKET_WIDTH);
4856 /* Find the first bucket whose flows should be expired. */
4857 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4858 if (buckets[bucket]) {
4861 subtotal += buckets[bucket++];
4862 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4867 if (VLOG_IS_DBG_ENABLED()) {
4871 ds_put_cstr(&s, "keep");
4872 for (i = 0; i < N_BUCKETS; i++) {
4874 ds_put_cstr(&s, ", drop");
4877 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4880 VLOG_INFO("%s: %s (msec:count)",
4881 dpif_name(ofproto->dpif), ds_cstr(&s));
4885 return bucket * BUCKET_WIDTH;
4889 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4891 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4892 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4893 struct ofexpired expired;
4895 if (facet->installed) {
4896 struct dpif_flow_stats stats;
4898 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4900 facet_update_stats(ofproto, facet, &stats);
4903 expired.flow = facet->flow;
4904 expired.packet_count = facet->packet_count;
4905 expired.byte_count = facet->byte_count;
4906 expired.used = facet->used;
4907 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4912 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4914 long long int cutoff = time_msec() - dp_max_idle;
4915 struct facet *facet, *next_facet;
4917 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4918 facet_active_timeout(ofproto, facet);
4919 if (facet->used < cutoff) {
4920 facet_remove(ofproto, facet);
4925 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4926 * then delete it entirely. */
4928 rule_expire(struct ofproto *ofproto, struct rule *rule)
4930 struct facet *facet, *next_facet;
4934 /* Has 'rule' expired? */
4936 if (rule->hard_timeout
4937 && now > rule->created + rule->hard_timeout * 1000) {
4938 reason = OFPRR_HARD_TIMEOUT;
4939 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4940 && now >rule->used + rule->idle_timeout * 1000) {
4941 reason = OFPRR_IDLE_TIMEOUT;
4946 COVERAGE_INC(ofproto_expired);
4948 /* Update stats. (This is a no-op if the rule expired due to an idle
4949 * timeout, because that only happens when the rule has no facets left.) */
4950 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4951 facet_remove(ofproto, facet);
4954 /* Get rid of the rule. */
4955 if (!rule_is_hidden(rule)) {
4956 rule_send_removed(ofproto, rule, reason);
4958 rule_remove(ofproto, rule);
4962 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4964 struct ofputil_flow_removed fr;
4965 struct ofconn *ofconn;
4967 if (!rule->send_flow_removed) {
4972 fr.cookie = rule->flow_cookie;
4974 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
4975 fr.idle_timeout = rule->idle_timeout;
4976 fr.packet_count = rule->packet_count;
4977 fr.byte_count = rule->byte_count;
4979 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4982 if (!rconn_is_connected(ofconn->rconn)
4983 || !ofconn_receives_async_msgs(ofconn)) {
4987 /* This accounts flow expirations as if they were replies to OpenFlow
4988 * requests. That works because preventing OpenFlow requests from
4989 * being processed also prevents new flows from being added (and
4990 * expiring). (It also prevents processing OpenFlow requests that
4991 * would not add new flows, so it is imperfect.) */
4992 msg = ofputil_encode_flow_removed(&fr, ofconn_get_flow_format(ofconn));
4993 ofconn_send_reply(ofconn, msg);
4997 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4998 * The returned statistics include statistics for all of 'rule''s facets. */
5000 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
5003 struct facet *facet;
5005 /* Start from historical data for 'rule' itself that are no longer tracked
5006 * in facets. This counts, for example, facets that have expired. */
5007 p = rule->packet_count;
5008 b = rule->byte_count;
5010 /* Add any statistics that are tracked by facets. This includes
5011 * statistical data recently updated by ofproto_update_stats() as well as
5012 * stats for packets that were executed "by hand" via dpif_execute(). */
5013 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5014 p += facet->packet_count;
5015 b += facet->byte_count;
5022 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
5024 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
5026 struct ofconn *ofconn = ofconn_;
5028 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
5029 ofconn->packet_in_counter, 100);
5032 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
5033 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
5034 * scheduler for sending.
5036 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
5037 * Otherwise, ownership is transferred to this function. */
5039 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
5040 const struct flow *flow, bool clone)
5042 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5043 struct ofputil_packet_in pin;
5046 /* Figure out the easy parts. */
5047 pin.packet = upcall->packet;
5048 pin.in_port = odp_port_to_ofp_port(flow->in_port);
5049 pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
5051 /* Get OpenFlow buffer_id. */
5052 if (upcall->type == DPIF_UC_ACTION) {
5053 pin.buffer_id = UINT32_MAX;
5054 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
5055 pin.buffer_id = pktbuf_get_null();
5056 } else if (!ofconn->pktbuf) {
5057 pin.buffer_id = UINT32_MAX;
5059 pin.buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet,
5063 /* Figure out how much of the packet to send. */
5064 pin.send_len = upcall->packet->size;
5065 if (pin.buffer_id != UINT32_MAX) {
5066 pin.send_len = MIN(pin.send_len, ofconn->miss_send_len);
5068 if (upcall->type == DPIF_UC_ACTION) {
5069 pin.send_len = MIN(pin.send_len, upcall->userdata);
5072 /* Make OFPT_PACKET_IN and hand over to packet scheduler. It might
5073 * immediately call into do_send_packet_in() or it might buffer it for a
5074 * while (until a later call to pinsched_run()). */
5075 msg = ofputil_encode_packet_in(&pin, clone ? NULL : upcall->packet);
5076 pinsched_send(ofconn->schedulers[upcall->type == DPIF_UC_MISS ? 0 : 1],
5077 flow->in_port, msg, do_send_packet_in, ofconn);
5080 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
5081 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
5082 * their individual configurations.
5084 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
5085 * Otherwise, ownership is transferred to this function. */
5087 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
5088 const struct flow *flow, bool clone)
5090 struct ofconn *ofconn, *prev;
5093 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5094 if (ofconn_receives_async_msgs(ofconn)) {
5096 schedule_packet_in(prev, upcall, flow, true);
5102 schedule_packet_in(prev, upcall, flow, clone);
5103 } else if (!clone) {
5104 ofpbuf_delete(upcall->packet);
5109 pick_datapath_id(const struct ofproto *ofproto)
5111 const struct ofport *port;
5113 port = get_port(ofproto, ODPP_LOCAL);
5115 uint8_t ea[ETH_ADDR_LEN];
5118 error = netdev_get_etheraddr(port->netdev, ea);
5120 return eth_addr_to_uint64(ea);
5122 VLOG_WARN("could not get MAC address for %s (%s)",
5123 netdev_get_name(port->netdev), strerror(error));
5125 return ofproto->fallback_dpid;
5129 pick_fallback_dpid(void)
5131 uint8_t ea[ETH_ADDR_LEN];
5132 eth_addr_nicira_random(ea);
5133 return eth_addr_to_uint64(ea);
5137 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
5138 void *aux OVS_UNUSED)
5140 const struct shash_node *node;
5144 SHASH_FOR_EACH (node, &all_ofprotos) {
5145 ds_put_format(&results, "%s\n", node->name);
5147 unixctl_command_reply(conn, 200, ds_cstr(&results));
5148 ds_destroy(&results);
5151 struct ofproto_trace {
5152 struct action_xlate_ctx ctx;
5158 trace_format_rule(struct ds *result, int level, const struct rule *rule)
5160 ds_put_char_multiple(result, '\t', level);
5162 ds_put_cstr(result, "No match\n");
5166 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
5167 ntohll(rule->flow_cookie));
5168 cls_rule_format(&rule->cr, result);
5169 ds_put_char(result, '\n');
5171 ds_put_char_multiple(result, '\t', level);
5172 ds_put_cstr(result, "OpenFlow ");
5173 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
5174 rule->n_actions * sizeof *rule->actions);
5175 ds_put_char(result, '\n');
5179 trace_format_flow(struct ds *result, int level, const char *title,
5180 struct ofproto_trace *trace)
5182 ds_put_char_multiple(result, '\t', level);
5183 ds_put_format(result, "%s: ", title);
5184 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5185 ds_put_cstr(result, "unchanged");
5187 flow_format(result, &trace->ctx.flow);
5188 trace->flow = trace->ctx.flow;
5190 ds_put_char(result, '\n');
5194 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
5196 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5197 struct ds *result = trace->result;
5199 ds_put_char(result, '\n');
5200 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5201 trace_format_rule(result, ctx->recurse + 1, rule);
5205 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5206 void *aux OVS_UNUSED)
5208 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5209 char *args = xstrdup(args_);
5210 char *save_ptr = NULL;
5211 struct ofproto *ofproto;
5212 struct ofpbuf packet;
5220 ofpbuf_init(&packet, strlen(args) / 2);
5223 dpname = strtok_r(args, " ", &save_ptr);
5224 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5225 in_port_s = strtok_r(NULL, " ", &save_ptr);
5226 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5227 if (!dpname || !in_port_s || !packet_s) {
5228 unixctl_command_reply(conn, 501, "Bad command syntax");
5232 ofproto = shash_find_data(&all_ofprotos, dpname);
5234 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5239 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
5240 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5242 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5243 packet_s += strspn(packet_s, " ");
5244 if (*packet_s != '\0') {
5245 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5248 if (packet.size < ETH_HEADER_LEN) {
5249 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5253 ds_put_cstr(&result, "Packet: ");
5254 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5255 ds_put_cstr(&result, s);
5258 flow_extract(&packet, tun_id, in_port, &flow);
5259 ds_put_cstr(&result, "Flow: ");
5260 flow_format(&result, &flow);
5261 ds_put_char(&result, '\n');
5263 rule = rule_lookup(ofproto, &flow);
5264 trace_format_rule(&result, 0, rule);
5266 struct ofproto_trace trace;
5267 struct ofpbuf *odp_actions;
5269 trace.result = &result;
5271 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5272 trace.ctx.resubmit_hook = trace_resubmit;
5273 odp_actions = xlate_actions(&trace.ctx,
5274 rule->actions, rule->n_actions);
5276 ds_put_char(&result, '\n');
5277 trace_format_flow(&result, 0, "Final flow", &trace);
5278 ds_put_cstr(&result, "Datapath actions: ");
5279 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5280 ofpbuf_delete(odp_actions);
5283 unixctl_command_reply(conn, 200, ds_cstr(&result));
5286 ds_destroy(&result);
5287 ofpbuf_uninit(&packet);
5292 ofproto_unixctl_init(void)
5294 static bool registered;
5300 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5301 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5305 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5306 struct ofpbuf *odp_actions, tag_type *tags,
5307 uint16_t *nf_output_iface, void *ofproto_)
5309 struct ofproto *ofproto = ofproto_;
5310 struct mac_entry *dst_mac;
5312 /* Drop frames for reserved multicast addresses. */
5313 if (eth_addr_is_reserved(flow->dl_dst)) {
5317 /* Learn source MAC (but don't try to learn from revalidation). */
5319 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
5320 struct mac_entry *src_mac;
5322 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
5323 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
5324 /* The log messages here could actually be useful in debugging,
5325 * so keep the rate limit relatively high. */
5326 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5327 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5328 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5330 ofproto_revalidate(ofproto,
5331 mac_learning_changed(ofproto->ml, src_mac));
5332 src_mac->port.i = flow->in_port;
5336 /* Determine output port. */
5337 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
5339 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5340 nf_output_iface, odp_actions);
5342 int out_port = dst_mac->port.i;
5343 if (out_port != flow->in_port) {
5344 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5345 *nf_output_iface = out_port;
5354 static const struct ofhooks default_ofhooks = {
5355 default_normal_ofhook_cb,