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 *);
110 static void hton_ofp_phy_port(struct ofp_phy_port *);
112 struct action_xlate_ctx {
113 /* action_xlate_ctx_init() initializes these members. */
116 struct ofproto *ofproto;
118 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
119 * this flow when actions change header fields. */
122 /* The packet corresponding to 'flow', or a null pointer if we are
123 * revalidating without a packet to refer to. */
124 const struct ofpbuf *packet;
126 /* If nonnull, called just before executing a resubmit action.
128 * This is normally null so the client has to set it manually after
129 * calling action_xlate_ctx_init(). */
130 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *);
132 /* If true, the speciality of 'flow' should be checked before executing
133 * its actions. If special_cb returns false on 'flow' rendered
134 * uninstallable and no actions will be executed. */
137 /* xlate_actions() initializes and uses these members. The client might want
138 * to look at them after it returns. */
140 struct ofpbuf *odp_actions; /* Datapath actions. */
141 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
142 bool may_set_up_flow; /* True ordinarily; false if the actions must
143 * be reassessed for every packet. */
144 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
146 /* xlate_actions() initializes and uses these members, but the client has no
147 * reason to look at them. */
149 int recurse; /* Recursion level, via xlate_table_action. */
150 int last_pop_priority; /* Offset in 'odp_actions' just past most
151 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
154 static void action_xlate_ctx_init(struct action_xlate_ctx *,
155 struct ofproto *, const struct flow *,
156 const struct ofpbuf *);
157 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
158 const union ofp_action *in, size_t n_in);
160 /* An OpenFlow flow. */
162 long long int used; /* Time last used; time created if not used. */
163 long long int created; /* Creation time. */
167 * - Do include packets and bytes from facets that have been deleted or
168 * whose own statistics have been folded into the rule.
170 * - Do include packets and bytes sent "by hand" that were accounted to
171 * the rule without any facet being involved (this is a rare corner
172 * case in rule_execute()).
174 * - Do not include packet or bytes that can be obtained from any facet's
175 * packet_count or byte_count member or that can be obtained from the
176 * datapath by, e.g., dpif_flow_get() for any facet.
178 uint64_t packet_count; /* Number of packets received. */
179 uint64_t byte_count; /* Number of bytes received. */
181 ovs_be64 flow_cookie; /* Controller-issued identifier. */
183 struct cls_rule cr; /* In owning ofproto's classifier. */
184 uint16_t idle_timeout; /* In seconds from time of last use. */
185 uint16_t hard_timeout; /* In seconds from time of creation. */
186 bool send_flow_removed; /* Send a flow removed message? */
187 int n_actions; /* Number of elements in actions[]. */
188 union ofp_action *actions; /* OpenFlow actions. */
189 struct list facets; /* List of "struct facet"s. */
192 static struct rule *rule_from_cls_rule(const struct cls_rule *);
193 static bool rule_is_hidden(const struct rule *);
195 static struct rule *rule_create(const struct cls_rule *,
196 const union ofp_action *, size_t n_actions,
197 uint16_t idle_timeout, uint16_t hard_timeout,
198 ovs_be64 flow_cookie, bool send_flow_removed);
199 static void rule_destroy(struct ofproto *, struct rule *);
200 static void rule_free(struct rule *);
202 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
203 static void rule_insert(struct ofproto *, struct rule *);
204 static void rule_remove(struct ofproto *, struct rule *);
206 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
207 static void rule_get_stats(const struct rule *, uint64_t *packets,
210 /* An exact-match instantiation of an OpenFlow flow. */
212 long long int used; /* Time last used; time created if not used. */
216 * - Do include packets and bytes sent "by hand", e.g. with
219 * - Do include packets and bytes that were obtained from the datapath
220 * when a flow was deleted (e.g. dpif_flow_del()) or when its
221 * statistics were reset (e.g. dpif_flow_put() with
222 * DPIF_FP_ZERO_STATS).
224 * - Do not include any packets or bytes that can currently be obtained
225 * from the datapath by, e.g., dpif_flow_get().
227 uint64_t packet_count; /* Number of packets received. */
228 uint64_t byte_count; /* Number of bytes received. */
230 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
231 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
233 uint64_t rs_packet_count; /* Packets pushed to resubmit children. */
234 uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */
235 long long int rs_used; /* Used time pushed to resubmit children. */
237 /* Number of bytes passed to account_cb. This may include bytes that can
238 * currently obtained from the datapath (thus, it can be greater than
240 uint64_t accounted_bytes;
242 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
243 struct list list_node; /* In owning rule's 'facets' list. */
244 struct rule *rule; /* Owning rule. */
245 struct flow flow; /* Exact-match flow. */
246 bool installed; /* Installed in datapath? */
247 bool may_install; /* True ordinarily; false if actions must
248 * be reassessed for every packet. */
249 size_t actions_len; /* Number of bytes in actions[]. */
250 struct nlattr *actions; /* Datapath actions. */
251 tag_type tags; /* Tags (set only by hooks). */
252 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
255 static struct facet *facet_create(struct ofproto *, struct rule *,
257 const struct ofpbuf *packet);
258 static void facet_remove(struct ofproto *, struct facet *);
259 static void facet_free(struct facet *);
261 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
262 static bool facet_revalidate(struct ofproto *, struct facet *);
264 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
265 static void facet_uninstall(struct ofproto *, struct facet *);
266 static void facet_flush_stats(struct ofproto *, struct facet *);
268 static void facet_make_actions(struct ofproto *, struct facet *,
269 const struct ofpbuf *packet);
270 static void facet_update_stats(struct ofproto *, struct facet *,
271 const struct dpif_flow_stats *);
272 static void facet_push_stats(struct ofproto *, struct facet *);
274 /* ofproto supports two kinds of OpenFlow connections:
276 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
277 * maintains persistent connections to these controllers and by default
278 * sends them asynchronous messages such as packet-ins.
280 * - "Service" connections, e.g. from ovs-ofctl. When these connections
281 * drop, it is the other side's responsibility to reconnect them if
282 * necessary. ofproto does not send them asynchronous messages by default.
284 * Currently, active (tcp, ssl, unix) connections are always "primary"
285 * connections and passive (ptcp, pssl, punix) connections are always "service"
286 * connections. There is no inherent reason for this, but it reflects the
290 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
291 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
294 /* A listener for incoming OpenFlow "service" connections. */
296 struct hmap_node node; /* In struct ofproto's "services" hmap. */
297 struct pvconn *pvconn; /* OpenFlow connection listener. */
299 /* These are not used by ofservice directly. They are settings for
300 * accepted "struct ofconn"s from the pvconn. */
301 int probe_interval; /* Max idle time before probing, in seconds. */
302 int rate_limit; /* Max packet-in rate in packets per second. */
303 int burst_limit; /* Limit on accumulating packet credits. */
306 static struct ofservice *ofservice_lookup(struct ofproto *,
308 static int ofservice_create(struct ofproto *,
309 const struct ofproto_controller *);
310 static void ofservice_reconfigure(struct ofservice *,
311 const struct ofproto_controller *);
312 static void ofservice_destroy(struct ofproto *, struct ofservice *);
314 /* An OpenFlow connection. */
316 struct ofproto *ofproto; /* The ofproto that owns this connection. */
317 struct list node; /* In struct ofproto's "all_conns" list. */
318 struct rconn *rconn; /* OpenFlow connection. */
319 enum ofconn_type type; /* Type. */
320 enum nx_flow_format flow_format; /* Currently selected flow format. */
322 /* OFPT_PACKET_IN related data. */
323 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
324 #define N_SCHEDULERS 2
325 struct pinsched *schedulers[N_SCHEDULERS];
326 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
327 int miss_send_len; /* Bytes to send of buffered packets. */
329 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
330 * requests, and the maximum number before we stop reading OpenFlow
332 #define OFCONN_REPLY_MAX 100
333 struct rconn_packet_counter *reply_counter;
335 /* type == OFCONN_PRIMARY only. */
336 enum nx_role role; /* Role. */
337 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
338 enum ofproto_band band; /* In-band or out-of-band? */
342 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
344 static void ofconn_destroy(struct ofconn *);
345 static void ofconn_run(struct ofconn *);
346 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 void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
352 struct rconn_packet_counter *counter);
354 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
355 const struct flow *, bool clone);
356 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
360 uint64_t datapath_id; /* Datapath ID. */
361 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
362 char *mfr_desc; /* Manufacturer. */
363 char *hw_desc; /* Hardware. */
364 char *sw_desc; /* Software version. */
365 char *serial_desc; /* Serial number. */
366 char *dp_desc; /* Datapath description. */
370 struct netdev_monitor *netdev_monitor;
371 struct hmap ports; /* Contains "struct ofport"s. */
372 struct shash port_by_name;
376 struct fail_open *fail_open;
377 struct netflow *netflow;
378 struct ofproto_sflow *sflow;
380 /* In-band control. */
381 struct in_band *in_band;
382 long long int next_in_band_update;
383 struct sockaddr_in *extra_in_band_remotes;
384 size_t n_extra_remotes;
388 struct classifier cls;
389 long long int next_expiration;
393 bool need_revalidate;
394 struct tag_set revalidate_set;
396 /* OpenFlow connections. */
397 struct hmap controllers; /* Controller "struct ofconn"s. */
398 struct list all_conns; /* Contains "struct ofconn"s. */
399 enum ofproto_fail_mode fail_mode;
401 /* OpenFlow listeners. */
402 struct hmap services; /* Contains "struct ofservice"s. */
403 struct pvconn **snoops;
406 /* Hooks for ovs-vswitchd. */
407 const struct ofhooks *ofhooks;
410 /* Used by default ofhooks. */
411 struct mac_learning *ml;
414 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
415 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
417 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
419 static const struct ofhooks default_ofhooks;
421 static uint64_t pick_datapath_id(const struct ofproto *);
422 static uint64_t pick_fallback_dpid(void);
424 static int ofproto_expire(struct ofproto *);
425 static void flow_push_stats(struct ofproto *, const struct rule *,
426 struct flow *, uint64_t packets, uint64_t bytes,
429 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
431 static void handle_openflow(struct ofconn *, struct ofpbuf *);
433 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
434 static void update_port(struct ofproto *, const char *devname);
435 static int init_ports(struct ofproto *);
436 static void reinit_ports(struct ofproto *);
438 static void ofproto_unixctl_init(void);
441 ofproto_create(const char *datapath, const char *datapath_type,
442 const struct ofhooks *ofhooks, void *aux,
443 struct ofproto **ofprotop)
451 ofproto_unixctl_init();
453 /* Connect to datapath and start listening for messages. */
454 error = dpif_open(datapath, datapath_type, &dpif);
456 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
459 error = dpif_recv_set_mask(dpif,
460 ((1u << DPIF_UC_MISS) |
461 (1u << DPIF_UC_ACTION) |
462 (1u << DPIF_UC_SAMPLE)));
464 VLOG_ERR("failed to listen on datapath %s: %s",
465 datapath, strerror(error));
469 dpif_flow_flush(dpif);
470 dpif_recv_purge(dpif);
472 /* Initialize settings. */
473 p = xzalloc(sizeof *p);
474 p->fallback_dpid = pick_fallback_dpid();
475 p->datapath_id = p->fallback_dpid;
476 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
477 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
478 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
479 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
480 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
482 /* Initialize datapath. */
484 p->netdev_monitor = netdev_monitor_create();
485 hmap_init(&p->ports);
486 shash_init(&p->port_by_name);
487 p->max_ports = dpif_get_max_ports(dpif);
489 /* Initialize submodules. */
494 /* Initialize in-band control. */
496 p->in_band_queue = -1;
498 /* Initialize flow table. */
499 classifier_init(&p->cls);
500 p->next_expiration = time_msec() + 1000;
502 /* Initialize facet table. */
503 hmap_init(&p->facets);
504 p->need_revalidate = false;
505 tag_set_init(&p->revalidate_set);
507 /* Initialize OpenFlow connections. */
508 list_init(&p->all_conns);
509 hmap_init(&p->controllers);
510 hmap_init(&p->services);
514 /* Initialize hooks. */
516 p->ofhooks = ofhooks;
520 p->ofhooks = &default_ofhooks;
522 p->ml = mac_learning_create();
525 /* Pick final datapath ID. */
526 p->datapath_id = pick_datapath_id(p);
527 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
529 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
536 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
538 uint64_t old_dpid = p->datapath_id;
539 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
540 if (p->datapath_id != old_dpid) {
541 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
543 /* Force all active connections to reconnect, since there is no way to
544 * notify a controller that the datapath ID has changed. */
545 ofproto_reconnect_controllers(p);
549 /* Creates a new controller in 'ofproto'. Some of the settings are initially
550 * drawn from 'c', but update_controller() needs to be called later to finish
551 * the new ofconn's configuration. */
553 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
555 char *name = ofconn_make_name(ofproto, c->target);
556 struct ofconn *ofconn;
558 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
559 ofconn->pktbuf = pktbuf_create();
560 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
561 rconn_connect(ofconn->rconn, c->target, name);
562 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
563 hash_string(c->target, 0));
568 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
569 * target (this is done by creating new ofconns and deleting old ones), but it
570 * can update the rest of an ofconn's settings. */
572 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
576 ofconn->band = c->band;
578 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
580 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
581 rconn_set_probe_interval(ofconn->rconn, probe_interval);
583 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
587 ofconn_get_target(const struct ofconn *ofconn)
589 return rconn_get_target(ofconn->rconn);
592 static struct ofconn *
593 find_controller_by_target(struct ofproto *ofproto, const char *target)
595 struct ofconn *ofconn;
597 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
598 hash_string(target, 0), &ofproto->controllers) {
599 if (!strcmp(ofconn_get_target(ofconn), target)) {
607 update_in_band_remotes(struct ofproto *ofproto)
609 const struct ofconn *ofconn;
610 struct sockaddr_in *addrs;
611 size_t max_addrs, n_addrs;
614 /* Allocate enough memory for as many remotes as we could possibly have. */
615 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
616 addrs = xmalloc(max_addrs * sizeof *addrs);
619 /* Add all the remotes. */
620 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
621 struct sockaddr_in *sin = &addrs[n_addrs];
623 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
627 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
628 if (sin->sin_addr.s_addr) {
629 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
633 for (i = 0; i < ofproto->n_extra_remotes; i++) {
634 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
637 /* Create or update or destroy in-band. */
639 if (!ofproto->in_band) {
640 in_band_create(ofproto, ofproto->dpif, &ofproto->in_band);
642 if (ofproto->in_band) {
643 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
645 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
646 ofproto->next_in_band_update = time_msec() + 1000;
648 in_band_destroy(ofproto->in_band);
649 ofproto->in_band = NULL;
657 update_fail_open(struct ofproto *p)
659 struct ofconn *ofconn;
661 if (!hmap_is_empty(&p->controllers)
662 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
663 struct rconn **rconns;
667 p->fail_open = fail_open_create(p);
671 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
672 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
673 rconns[n++] = ofconn->rconn;
676 fail_open_set_controllers(p->fail_open, rconns, n);
677 /* p->fail_open takes ownership of 'rconns'. */
679 fail_open_destroy(p->fail_open);
685 ofproto_set_controllers(struct ofproto *p,
686 const struct ofproto_controller *controllers,
687 size_t n_controllers)
689 struct shash new_controllers;
690 struct ofconn *ofconn, *next_ofconn;
691 struct ofservice *ofservice, *next_ofservice;
694 /* Create newly configured controllers and services.
695 * Create a name to ofproto_controller mapping in 'new_controllers'. */
696 shash_init(&new_controllers);
697 for (i = 0; i < n_controllers; i++) {
698 const struct ofproto_controller *c = &controllers[i];
700 if (!vconn_verify_name(c->target)) {
701 if (!find_controller_by_target(p, c->target)) {
702 add_controller(p, c);
704 } else if (!pvconn_verify_name(c->target)) {
705 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
709 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
710 dpif_name(p->dpif), c->target);
714 shash_add_once(&new_controllers, c->target, &controllers[i]);
717 /* Delete controllers that are no longer configured.
718 * Update configuration of all now-existing controllers. */
719 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
720 struct ofproto_controller *c;
722 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
724 ofconn_destroy(ofconn);
726 update_controller(ofconn, c);
730 /* Delete services that are no longer configured.
731 * Update configuration of all now-existing services. */
732 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
733 struct ofproto_controller *c;
735 c = shash_find_data(&new_controllers,
736 pvconn_get_name(ofservice->pvconn));
738 ofservice_destroy(p, ofservice);
740 ofservice_reconfigure(ofservice, c);
744 shash_destroy(&new_controllers);
746 update_in_band_remotes(p);
751 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
753 p->fail_mode = fail_mode;
757 /* Drops the connections between 'ofproto' and all of its controllers, forcing
758 * them to reconnect. */
760 ofproto_reconnect_controllers(struct ofproto *ofproto)
762 struct ofconn *ofconn;
764 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
765 rconn_reconnect(ofconn->rconn);
770 any_extras_changed(const struct ofproto *ofproto,
771 const struct sockaddr_in *extras, size_t n)
775 if (n != ofproto->n_extra_remotes) {
779 for (i = 0; i < n; i++) {
780 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
781 const struct sockaddr_in *new = &extras[i];
783 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
784 old->sin_port != new->sin_port) {
792 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
793 * in-band control should guarantee access, in the same way that in-band
794 * control guarantees access to OpenFlow controllers. */
796 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
797 const struct sockaddr_in *extras, size_t n)
799 if (!any_extras_changed(ofproto, extras, n)) {
803 free(ofproto->extra_in_band_remotes);
804 ofproto->n_extra_remotes = n;
805 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
807 update_in_band_remotes(ofproto);
810 /* Sets the OpenFlow queue used by flows set up by in-band control on
811 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
812 * flows will use the default queue. */
814 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
816 if (queue_id != ofproto->in_band_queue) {
817 ofproto->in_band_queue = queue_id;
818 update_in_band_remotes(ofproto);
823 ofproto_set_desc(struct ofproto *p,
824 const char *mfr_desc, const char *hw_desc,
825 const char *sw_desc, const char *serial_desc,
828 struct ofp_desc_stats *ods;
831 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
832 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
833 sizeof ods->mfr_desc);
836 p->mfr_desc = xstrdup(mfr_desc);
839 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
840 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
841 sizeof ods->hw_desc);
844 p->hw_desc = xstrdup(hw_desc);
847 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
848 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
849 sizeof ods->sw_desc);
852 p->sw_desc = xstrdup(sw_desc);
855 if (strlen(serial_desc) >= sizeof ods->serial_num) {
856 VLOG_WARN("truncating serial_desc, must be less than %zu "
858 sizeof ods->serial_num);
860 free(p->serial_desc);
861 p->serial_desc = xstrdup(serial_desc);
864 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
865 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
866 sizeof ods->dp_desc);
869 p->dp_desc = xstrdup(dp_desc);
874 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
875 const struct svec *svec)
877 struct pvconn **pvconns = *pvconnsp;
878 size_t n_pvconns = *n_pvconnsp;
882 for (i = 0; i < n_pvconns; i++) {
883 pvconn_close(pvconns[i]);
887 pvconns = xmalloc(svec->n * sizeof *pvconns);
889 for (i = 0; i < svec->n; i++) {
890 const char *name = svec->names[i];
891 struct pvconn *pvconn;
894 error = pvconn_open(name, &pvconn);
896 pvconns[n_pvconns++] = pvconn;
898 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
906 *n_pvconnsp = n_pvconns;
912 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
914 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
918 ofproto_set_netflow(struct ofproto *ofproto,
919 const struct netflow_options *nf_options)
921 if (nf_options && nf_options->collectors.n) {
922 if (!ofproto->netflow) {
923 ofproto->netflow = netflow_create();
925 return netflow_set_options(ofproto->netflow, nf_options);
927 netflow_destroy(ofproto->netflow);
928 ofproto->netflow = NULL;
934 ofproto_set_sflow(struct ofproto *ofproto,
935 const struct ofproto_sflow_options *oso)
937 struct ofproto_sflow *os = ofproto->sflow;
940 struct ofport *ofport;
942 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
943 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
944 ofproto_sflow_add_port(os, ofport->odp_port,
945 netdev_get_name(ofport->netdev));
948 ofproto_sflow_set_options(os, oso);
950 ofproto_sflow_destroy(os);
951 ofproto->sflow = NULL;
955 /* Connectivity Fault Management configuration. */
957 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
959 ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no)
961 struct ofport *ofport = get_port(ofproto, port_no);
962 if (ofport && ofport->cfm){
963 cfm_destroy(ofport->cfm);
968 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
969 * basic configuration from the configuration members in 'cfm', and the set of
970 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
971 * Ignores the statistics members of 'cfm'.
973 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
975 ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no,
976 const struct cfm *cfm,
977 const uint16_t *remote_mps, size_t n_remote_mps)
979 struct ofport *ofport;
981 ofport = get_port(ofproto, port_no);
983 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32,
984 dpif_name(ofproto->dpif), port_no);
989 ofport->cfm = cfm_create();
992 ofport->cfm->mpid = cfm->mpid;
993 ofport->cfm->interval = cfm->interval;
994 memcpy(ofport->cfm->eth_src, cfm->eth_src, ETH_ADDR_LEN);
995 memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
997 cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
999 if (!cfm_configure(ofport->cfm)) {
1000 VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed",
1001 dpif_name(ofproto->dpif), port_no,
1002 netdev_get_name(ofport->netdev));
1003 cfm_destroy(ofport->cfm);
1008 /* Returns the connectivity fault management object associated with 'port_no'
1009 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
1010 * 'port_no' or if that port does not have CFM configured. The caller must not
1011 * modify or destroy the returned object. */
1013 ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no)
1015 struct ofport *ofport = get_port(ofproto, port_no);
1016 return ofport ? ofport->cfm : NULL;
1020 ofproto_get_datapath_id(const struct ofproto *ofproto)
1022 return ofproto->datapath_id;
1026 ofproto_has_primary_controller(const struct ofproto *ofproto)
1028 return !hmap_is_empty(&ofproto->controllers);
1031 enum ofproto_fail_mode
1032 ofproto_get_fail_mode(const struct ofproto *p)
1034 return p->fail_mode;
1038 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1042 for (i = 0; i < ofproto->n_snoops; i++) {
1043 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1048 ofproto_destroy(struct ofproto *p)
1050 struct ofservice *ofservice, *next_ofservice;
1051 struct ofconn *ofconn, *next_ofconn;
1052 struct ofport *ofport, *next_ofport;
1059 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1061 /* Destroy fail-open and in-band early, since they touch the classifier. */
1062 fail_open_destroy(p->fail_open);
1063 p->fail_open = NULL;
1065 in_band_destroy(p->in_band);
1067 free(p->extra_in_band_remotes);
1069 ofproto_flush_flows(p);
1070 classifier_destroy(&p->cls);
1071 hmap_destroy(&p->facets);
1073 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1074 ofconn_destroy(ofconn);
1076 hmap_destroy(&p->controllers);
1078 dpif_close(p->dpif);
1079 netdev_monitor_destroy(p->netdev_monitor);
1080 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1081 hmap_remove(&p->ports, &ofport->hmap_node);
1082 ofport_free(ofport);
1084 shash_destroy(&p->port_by_name);
1086 netflow_destroy(p->netflow);
1087 ofproto_sflow_destroy(p->sflow);
1089 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1090 ofservice_destroy(p, ofservice);
1092 hmap_destroy(&p->services);
1094 for (i = 0; i < p->n_snoops; i++) {
1095 pvconn_close(p->snoops[i]);
1099 mac_learning_destroy(p->ml);
1104 free(p->serial_desc);
1107 hmap_destroy(&p->ports);
1113 ofproto_run(struct ofproto *p)
1115 int error = ofproto_run1(p);
1117 error = ofproto_run2(p, false);
1123 process_port_change(struct ofproto *ofproto, int error, char *devname)
1125 if (error == ENOBUFS) {
1126 reinit_ports(ofproto);
1127 } else if (!error) {
1128 update_port(ofproto, devname);
1133 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1134 * means that 'ofconn' is more interesting for monitoring than a lower return
1137 snoop_preference(const struct ofconn *ofconn)
1139 switch (ofconn->role) {
1140 case NX_ROLE_MASTER:
1147 /* Shouldn't happen. */
1152 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1153 * Connects this vconn to a controller. */
1155 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1157 struct ofconn *ofconn, *best;
1159 /* Pick a controller for monitoring. */
1161 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1162 if (ofconn->type == OFCONN_PRIMARY
1163 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1169 rconn_add_monitor(best->rconn, vconn);
1171 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1177 ofproto_run1(struct ofproto *p)
1179 struct ofconn *ofconn, *next_ofconn;
1180 struct ofservice *ofservice;
1181 struct ofport *ofport;
1186 if (shash_is_empty(&p->port_by_name)) {
1190 for (i = 0; i < 50; i++) {
1191 struct dpif_upcall packet;
1193 error = dpif_recv(p->dpif, &packet);
1195 if (error == ENODEV) {
1196 /* Someone destroyed the datapath behind our back. The caller
1197 * better destroy us and give up, because we're just going to
1198 * spin from here on out. */
1199 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1200 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1201 dpif_name(p->dpif));
1207 handle_upcall(p, &packet);
1210 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1211 process_port_change(p, error, devname);
1213 while ((error = netdev_monitor_poll(p->netdev_monitor,
1214 &devname)) != EAGAIN) {
1215 process_port_change(p, error, devname);
1218 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1219 ofport_run(p, ofport);
1223 if (time_msec() >= p->next_in_band_update) {
1224 update_in_band_remotes(p);
1226 in_band_run(p->in_band);
1229 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1233 /* Fail-open maintenance. Do this after processing the ofconns since
1234 * fail-open checks the status of the controller rconn. */
1236 fail_open_run(p->fail_open);
1239 HMAP_FOR_EACH (ofservice, node, &p->services) {
1240 struct vconn *vconn;
1243 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1245 struct rconn *rconn;
1248 rconn = rconn_create(ofservice->probe_interval, 0);
1249 name = ofconn_make_name(p, vconn_get_name(vconn));
1250 rconn_connect_unreliably(rconn, vconn, name);
1253 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1254 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1255 ofservice->burst_limit);
1256 } else if (retval != EAGAIN) {
1257 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1261 for (i = 0; i < p->n_snoops; i++) {
1262 struct vconn *vconn;
1265 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1267 add_snooper(p, vconn);
1268 } else if (retval != EAGAIN) {
1269 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1273 if (time_msec() >= p->next_expiration) {
1274 int delay = ofproto_expire(p);
1275 p->next_expiration = time_msec() + delay;
1276 COVERAGE_INC(ofproto_expiration);
1280 netflow_run(p->netflow);
1283 ofproto_sflow_run(p->sflow);
1290 ofproto_run2(struct ofproto *p, bool revalidate_all)
1292 /* Figure out what we need to revalidate now, if anything. */
1293 struct tag_set revalidate_set = p->revalidate_set;
1294 if (p->need_revalidate) {
1295 revalidate_all = true;
1298 /* Clear the revalidation flags. */
1299 tag_set_init(&p->revalidate_set);
1300 p->need_revalidate = false;
1302 /* Now revalidate if there's anything to do. */
1303 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1304 struct facet *facet, *next;
1306 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1308 || tag_set_intersects(&revalidate_set, facet->tags)) {
1309 facet_revalidate(p, facet);
1318 ofproto_wait(struct ofproto *p)
1320 struct ofservice *ofservice;
1321 struct ofconn *ofconn;
1322 struct ofport *ofport;
1325 dpif_recv_wait(p->dpif);
1326 dpif_port_poll_wait(p->dpif);
1327 netdev_monitor_poll_wait(p->netdev_monitor);
1328 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1329 ofport_wait(ofport);
1331 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1332 ofconn_wait(ofconn);
1335 poll_timer_wait_until(p->next_in_band_update);
1336 in_band_wait(p->in_band);
1339 fail_open_wait(p->fail_open);
1342 ofproto_sflow_wait(p->sflow);
1344 if (!tag_set_is_empty(&p->revalidate_set)) {
1345 poll_immediate_wake();
1347 if (p->need_revalidate) {
1348 /* Shouldn't happen, but if it does just go around again. */
1349 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1350 poll_immediate_wake();
1351 } else if (p->next_expiration != LLONG_MAX) {
1352 poll_timer_wait_until(p->next_expiration);
1354 HMAP_FOR_EACH (ofservice, node, &p->services) {
1355 pvconn_wait(ofservice->pvconn);
1357 for (i = 0; i < p->n_snoops; i++) {
1358 pvconn_wait(p->snoops[i]);
1363 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1365 tag_set_add(&ofproto->revalidate_set, tag);
1369 ofproto_get_revalidate_set(struct ofproto *ofproto)
1371 return &ofproto->revalidate_set;
1375 ofproto_is_alive(const struct ofproto *p)
1377 return !hmap_is_empty(&p->controllers);
1381 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1384 const struct ofconn *ofconn;
1388 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1389 const struct rconn *rconn = ofconn->rconn;
1390 time_t now = time_now();
1391 time_t last_connection = rconn_get_last_connection(rconn);
1392 time_t last_disconnect = rconn_get_last_disconnect(rconn);
1393 const int last_error = rconn_get_last_error(rconn);
1394 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1396 shash_add(info, rconn_get_target(rconn), cinfo);
1398 cinfo->is_connected = rconn_is_connected(rconn);
1399 cinfo->role = ofconn->role;
1404 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1405 cinfo->pairs.values[cinfo->pairs.n++] =
1406 xstrdup(ovs_retval_to_string(last_error));
1409 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1410 cinfo->pairs.values[cinfo->pairs.n++] =
1411 xstrdup(rconn_get_state(rconn));
1413 if (last_connection != TIME_MIN) {
1414 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
1415 cinfo->pairs.values[cinfo->pairs.n++]
1416 = xasprintf("%ld", (long int) (now - last_connection));
1419 if (last_disconnect != TIME_MIN) {
1420 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
1421 cinfo->pairs.values[cinfo->pairs.n++]
1422 = xasprintf("%ld", (long int) (now - last_disconnect));
1428 ofproto_free_ofproto_controller_info(struct shash *info)
1430 struct shash_node *node;
1432 SHASH_FOR_EACH (node, info) {
1433 struct ofproto_controller_info *cinfo = node->data;
1434 while (cinfo->pairs.n) {
1435 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1439 shash_destroy(info);
1442 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1444 * This is almost the same as calling dpif_port_del() directly on the
1445 * datapath, but it also makes 'ofproto' close its open netdev for the port
1446 * (if any). This makes it possible to create a new netdev of a different
1447 * type under the same name, which otherwise the netdev library would refuse
1448 * to do because of the conflict. (The netdev would eventually get closed on
1449 * the next trip through ofproto_run(), but this interface is more direct.)
1451 * Returns 0 if successful, otherwise a positive errno. */
1453 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1455 struct ofport *ofport = get_port(ofproto, odp_port);
1456 const char *name = ofport ? ofport->opp.name : "<unknown>";
1459 error = dpif_port_del(ofproto->dpif, odp_port);
1461 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1462 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1463 } else if (ofport) {
1464 /* 'name' is ofport->opp.name and update_port() is going to destroy
1465 * 'ofport'. Just in case update_port() refers to 'name' after it
1466 * destroys 'ofport', make a copy of it around the update_port()
1468 char *devname = xstrdup(name);
1469 update_port(ofproto, devname);
1475 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1476 * true if 'odp_port' exists and should be included, false otherwise. */
1478 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1480 struct ofport *ofport = get_port(ofproto, odp_port);
1481 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1484 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1485 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1486 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1488 * Returns 0 if successful, otherwise a positive errno value. */
1490 ofproto_send_packet(struct ofproto *ofproto,
1491 uint32_t port_no, uint16_t vlan_tci,
1492 const struct ofpbuf *packet)
1494 struct ofpbuf odp_actions;
1497 ofpbuf_init(&odp_actions, 32);
1498 if (vlan_tci != 0) {
1499 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
1500 ntohs(vlan_tci & ~VLAN_CFI));
1502 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1503 error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
1505 ofpbuf_uninit(&odp_actions);
1508 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
1509 dpif_name(ofproto->dpif), port_no, strerror(error));
1514 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1515 * performs the 'n_actions' actions in 'actions'. The new flow will not
1518 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1519 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1520 * controllers; otherwise, it will be hidden.
1522 * The caller retains ownership of 'cls_rule' and 'actions'. */
1524 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1525 const union ofp_action *actions, size_t n_actions)
1528 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1529 rule_insert(p, rule);
1533 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1537 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1540 rule_remove(ofproto, rule);
1545 ofproto_flush_flows(struct ofproto *ofproto)
1547 struct facet *facet, *next_facet;
1548 struct rule *rule, *next_rule;
1549 struct cls_cursor cursor;
1551 COVERAGE_INC(ofproto_flush);
1553 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1554 /* Mark the facet as not installed so that facet_remove() doesn't
1555 * bother trying to uninstall it. There is no point in uninstalling it
1556 * individually since we are about to blow away all the facets with
1557 * dpif_flow_flush(). */
1558 facet->installed = false;
1559 facet->dp_packet_count = 0;
1560 facet->dp_byte_count = 0;
1561 facet_remove(ofproto, facet);
1564 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1565 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1566 rule_remove(ofproto, rule);
1569 dpif_flow_flush(ofproto->dpif);
1570 if (ofproto->in_band) {
1571 in_band_flushed(ofproto->in_band);
1573 if (ofproto->fail_open) {
1574 fail_open_flushed(ofproto->fail_open);
1579 reinit_ports(struct ofproto *p)
1581 struct dpif_port_dump dump;
1582 struct shash_node *node;
1583 struct shash devnames;
1584 struct ofport *ofport;
1585 struct dpif_port dpif_port;
1587 COVERAGE_INC(ofproto_reinit_ports);
1589 shash_init(&devnames);
1590 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1591 shash_add_once (&devnames, ofport->opp.name, NULL);
1593 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1594 shash_add_once (&devnames, dpif_port.name, NULL);
1597 SHASH_FOR_EACH (node, &devnames) {
1598 update_port(p, node->name);
1600 shash_destroy(&devnames);
1603 static struct ofport *
1604 make_ofport(const struct dpif_port *dpif_port)
1606 struct netdev_options netdev_options;
1607 enum netdev_flags flags;
1608 struct ofport *ofport;
1609 struct netdev *netdev;
1612 memset(&netdev_options, 0, sizeof netdev_options);
1613 netdev_options.name = dpif_port->name;
1614 netdev_options.type = dpif_port->type;
1615 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1617 error = netdev_open(&netdev_options, &netdev);
1619 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1620 "cannot be opened (%s)",
1621 dpif_port->name, dpif_port->port_no,
1622 dpif_port->name, strerror(error));
1626 ofport = xzalloc(sizeof *ofport);
1627 ofport->netdev = netdev;
1628 ofport->odp_port = dpif_port->port_no;
1629 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1630 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1631 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1633 netdev_get_flags(netdev, &flags);
1634 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1636 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1638 netdev_get_features(netdev,
1639 &ofport->opp.curr, &ofport->opp.advertised,
1640 &ofport->opp.supported, &ofport->opp.peer);
1645 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1647 if (get_port(p, dpif_port->port_no)) {
1648 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1649 dpif_port->port_no);
1651 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1652 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1661 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1663 const struct ofp_phy_port *a = &a_->opp;
1664 const struct ofp_phy_port *b = &b_->opp;
1666 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1667 return (a->port_no == b->port_no
1668 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1669 && !strcmp(a->name, b->name)
1670 && a->state == b->state
1671 && a->config == b->config
1672 && a->curr == b->curr
1673 && a->advertised == b->advertised
1674 && a->supported == b->supported
1675 && a->peer == b->peer);
1679 send_port_status(struct ofproto *p, const struct ofport *ofport,
1682 /* XXX Should limit the number of queued port status change messages. */
1683 struct ofconn *ofconn;
1684 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1685 struct ofp_port_status *ops;
1688 /* Primary controllers, even slaves, should always get port status
1689 updates. Otherwise obey ofconn_receives_async_msgs(). */
1690 if (ofconn->type != OFCONN_PRIMARY
1691 && !ofconn_receives_async_msgs(ofconn)) {
1695 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1696 ops->reason = reason;
1697 ops->desc = ofport->opp;
1698 hton_ofp_phy_port(&ops->desc);
1699 queue_tx(b, ofconn, NULL);
1704 ofport_install(struct ofproto *p, struct ofport *ofport)
1706 const char *netdev_name = ofport->opp.name;
1708 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1709 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1710 shash_add(&p->port_by_name, netdev_name, ofport);
1712 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1717 ofport_remove(struct ofproto *p, struct ofport *ofport)
1719 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1720 hmap_remove(&p->ports, &ofport->hmap_node);
1721 shash_delete(&p->port_by_name,
1722 shash_find(&p->port_by_name, ofport->opp.name));
1724 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1729 ofport_run(struct ofproto *ofproto, struct ofport *ofport)
1732 struct ofpbuf *packet = cfm_run(ofport->cfm);
1734 ofproto_send_packet(ofproto, ofport->odp_port, 0, packet);
1735 ofpbuf_delete(packet);
1741 ofport_wait(struct ofport *ofport)
1744 cfm_wait(ofport->cfm);
1749 ofport_free(struct ofport *ofport)
1752 cfm_destroy(ofport->cfm);
1753 netdev_close(ofport->netdev);
1758 static struct ofport *
1759 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1761 struct ofport *port;
1763 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1764 hash_int(odp_port, 0), &ofproto->ports) {
1765 if (port->odp_port == odp_port) {
1773 update_port(struct ofproto *p, const char *devname)
1775 struct dpif_port dpif_port;
1776 struct ofport *old_ofport;
1777 struct ofport *new_ofport;
1780 COVERAGE_INC(ofproto_update_port);
1782 /* Query the datapath for port information. */
1783 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1785 /* Find the old ofport. */
1786 old_ofport = shash_find_data(&p->port_by_name, devname);
1789 /* There's no port named 'devname' but there might be a port with
1790 * the same port number. This could happen if a port is deleted
1791 * and then a new one added in its place very quickly, or if a port
1792 * is renamed. In the former case we want to send an OFPPR_DELETE
1793 * and an OFPPR_ADD, and in the latter case we want to send a
1794 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1795 * the old port's ifindex against the new port, or perhaps less
1796 * reliably but more portably by comparing the old port's MAC
1797 * against the new port's MAC. However, this code isn't that smart
1798 * and always sends an OFPPR_MODIFY (XXX). */
1799 old_ofport = get_port(p, dpif_port.port_no);
1801 } else if (error != ENOENT && error != ENODEV) {
1802 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1803 "%s", strerror(error));
1807 /* Create a new ofport. */
1808 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1810 /* Eliminate a few pathological cases. */
1811 if (!old_ofport && !new_ofport) {
1813 } else if (old_ofport && new_ofport) {
1814 /* Most of the 'config' bits are OpenFlow soft state, but
1815 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1816 * OpenFlow bits from old_ofport. (make_ofport() only sets
1817 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1818 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1820 if (ofport_equal(old_ofport, new_ofport)) {
1821 /* False alarm--no change. */
1822 ofport_free(new_ofport);
1827 /* Now deal with the normal cases. */
1829 ofport_remove(p, old_ofport);
1832 ofport_install(p, new_ofport);
1834 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1835 (!old_ofport ? OFPPR_ADD
1836 : !new_ofport ? OFPPR_DELETE
1838 ofport_free(old_ofport);
1841 dpif_port_destroy(&dpif_port);
1845 init_ports(struct ofproto *p)
1847 struct dpif_port_dump dump;
1848 struct dpif_port dpif_port;
1850 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1851 if (!ofport_conflicts(p, &dpif_port)) {
1852 struct ofport *ofport = make_ofport(&dpif_port);
1854 ofport_install(p, ofport);
1862 static struct ofconn *
1863 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1865 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1866 ofconn->ofproto = p;
1867 list_push_back(&p->all_conns, &ofconn->node);
1868 ofconn->rconn = rconn;
1869 ofconn->type = type;
1870 ofconn->flow_format = NXFF_OPENFLOW10;
1871 ofconn->role = NX_ROLE_OTHER;
1872 ofconn->packet_in_counter = rconn_packet_counter_create ();
1873 ofconn->pktbuf = NULL;
1874 ofconn->miss_send_len = 0;
1875 ofconn->reply_counter = rconn_packet_counter_create ();
1880 ofconn_destroy(struct ofconn *ofconn)
1882 if (ofconn->type == OFCONN_PRIMARY) {
1883 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1886 list_remove(&ofconn->node);
1887 rconn_destroy(ofconn->rconn);
1888 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1889 rconn_packet_counter_destroy(ofconn->reply_counter);
1890 pktbuf_destroy(ofconn->pktbuf);
1895 ofconn_run(struct ofconn *ofconn)
1897 struct ofproto *p = ofconn->ofproto;
1901 for (i = 0; i < N_SCHEDULERS; i++) {
1902 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1905 rconn_run(ofconn->rconn);
1907 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1908 /* Limit the number of iterations to prevent other tasks from
1910 for (iteration = 0; iteration < 50; iteration++) {
1911 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1916 fail_open_maybe_recover(p->fail_open);
1918 handle_openflow(ofconn, of_msg);
1919 ofpbuf_delete(of_msg);
1923 if (!rconn_is_alive(ofconn->rconn)) {
1924 ofconn_destroy(ofconn);
1929 ofconn_wait(struct ofconn *ofconn)
1933 for (i = 0; i < N_SCHEDULERS; i++) {
1934 pinsched_wait(ofconn->schedulers[i]);
1936 rconn_run_wait(ofconn->rconn);
1937 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1938 rconn_recv_wait(ofconn->rconn);
1940 COVERAGE_INC(ofproto_ofconn_stuck);
1944 /* Returns true if 'ofconn' should receive asynchronous messages. */
1946 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1948 if (ofconn->type == OFCONN_PRIMARY) {
1949 /* Primary controllers always get asynchronous messages unless they
1950 * have configured themselves as "slaves". */
1951 return ofconn->role != NX_ROLE_SLAVE;
1953 /* Service connections don't get asynchronous messages unless they have
1954 * explicitly asked for them by setting a nonzero miss send length. */
1955 return ofconn->miss_send_len > 0;
1959 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1960 * and 'target', suitable for use in log messages for identifying the
1963 * The name is dynamically allocated. The caller should free it (with free())
1964 * when it is no longer needed. */
1966 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1968 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1972 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1976 for (i = 0; i < N_SCHEDULERS; i++) {
1977 struct pinsched **s = &ofconn->schedulers[i];
1981 *s = pinsched_create(rate, burst);
1983 pinsched_set_limits(*s, rate, burst);
1986 pinsched_destroy(*s);
1993 ofservice_reconfigure(struct ofservice *ofservice,
1994 const struct ofproto_controller *c)
1996 ofservice->probe_interval = c->probe_interval;
1997 ofservice->rate_limit = c->rate_limit;
1998 ofservice->burst_limit = c->burst_limit;
2001 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
2002 * positive errno value. */
2004 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
2006 struct ofservice *ofservice;
2007 struct pvconn *pvconn;
2010 error = pvconn_open(c->target, &pvconn);
2015 ofservice = xzalloc(sizeof *ofservice);
2016 hmap_insert(&ofproto->services, &ofservice->node,
2017 hash_string(c->target, 0));
2018 ofservice->pvconn = pvconn;
2020 ofservice_reconfigure(ofservice, c);
2026 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
2028 hmap_remove(&ofproto->services, &ofservice->node);
2029 pvconn_close(ofservice->pvconn);
2033 /* Finds and returns the ofservice within 'ofproto' that has the given
2034 * 'target', or a null pointer if none exists. */
2035 static struct ofservice *
2036 ofservice_lookup(struct ofproto *ofproto, const char *target)
2038 struct ofservice *ofservice;
2040 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
2041 &ofproto->services) {
2042 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2049 /* Returns true if 'rule' should be hidden from the controller.
2051 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2052 * (e.g. by in-band control) and are intentionally hidden from the
2055 rule_is_hidden(const struct rule *rule)
2057 return rule->cr.priority > UINT16_MAX;
2060 /* Creates and returns a new rule initialized as specified.
2062 * The caller is responsible for inserting the rule into the classifier (with
2063 * rule_insert()). */
2064 static struct rule *
2065 rule_create(const struct cls_rule *cls_rule,
2066 const union ofp_action *actions, size_t n_actions,
2067 uint16_t idle_timeout, uint16_t hard_timeout,
2068 ovs_be64 flow_cookie, bool send_flow_removed)
2070 struct rule *rule = xzalloc(sizeof *rule);
2071 rule->cr = *cls_rule;
2072 rule->idle_timeout = idle_timeout;
2073 rule->hard_timeout = hard_timeout;
2074 rule->flow_cookie = flow_cookie;
2075 rule->used = rule->created = time_msec();
2076 rule->send_flow_removed = send_flow_removed;
2077 list_init(&rule->facets);
2078 if (n_actions > 0) {
2079 rule->n_actions = n_actions;
2080 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2086 static struct rule *
2087 rule_from_cls_rule(const struct cls_rule *cls_rule)
2089 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2093 rule_free(struct rule *rule)
2095 free(rule->actions);
2099 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2100 * destroying any that no longer has a rule (which is probably all of them).
2102 * The caller must have already removed 'rule' from the classifier. */
2104 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2106 struct facet *facet, *next_facet;
2107 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2108 facet_revalidate(ofproto, facet);
2113 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2114 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2117 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2119 const union ofp_action *oa;
2120 struct actions_iterator i;
2122 if (out_port == htons(OFPP_NONE)) {
2125 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2126 oa = actions_next(&i)) {
2127 if (action_outputs_to_port(oa, out_port)) {
2134 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2135 * 'packet', which arrived on 'in_port'.
2137 * Takes ownership of 'packet'. */
2139 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2140 const struct nlattr *odp_actions, size_t actions_len,
2141 struct ofpbuf *packet)
2143 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2144 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2145 /* As an optimization, avoid a round-trip from userspace to kernel to
2146 * userspace. This also avoids possibly filling up kernel packet
2147 * buffers along the way. */
2148 struct dpif_upcall upcall;
2150 upcall.type = DPIF_UC_ACTION;
2151 upcall.packet = packet;
2154 upcall.userdata = nl_attr_get_u64(odp_actions);
2155 upcall.sample_pool = 0;
2156 upcall.actions = NULL;
2157 upcall.actions_len = 0;
2159 send_packet_in(ofproto, &upcall, flow, false);
2165 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2166 ofpbuf_delete(packet);
2171 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2172 * statistics appropriately. 'packet' must have at least sizeof(struct
2173 * ofp_packet_in) bytes of headroom.
2175 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2176 * applying flow_extract() to 'packet' would yield the same flow as
2179 * 'facet' must have accurately composed ODP actions; that is, it must not be
2180 * in need of revalidation.
2182 * Takes ownership of 'packet'. */
2184 facet_execute(struct ofproto *ofproto, struct facet *facet,
2185 struct ofpbuf *packet)
2187 struct dpif_flow_stats stats;
2189 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2191 flow_extract_stats(&facet->flow, packet, &stats);
2192 stats.used = time_msec();
2193 if (execute_odp_actions(ofproto, &facet->flow,
2194 facet->actions, facet->actions_len, packet)) {
2195 facet_update_stats(ofproto, facet, &stats);
2199 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2200 * statistics (or the statistics for one of its facets) appropriately.
2201 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2203 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2204 * with statistics for 'packet' either way.
2206 * Takes ownership of 'packet'. */
2208 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2209 struct ofpbuf *packet)
2211 struct action_xlate_ctx ctx;
2212 struct ofpbuf *odp_actions;
2213 struct facet *facet;
2217 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2219 flow_extract(packet, 0, in_port, &flow);
2221 /* First look for a related facet. If we find one, account it to that. */
2222 facet = facet_lookup_valid(ofproto, &flow);
2223 if (facet && facet->rule == rule) {
2224 facet_execute(ofproto, facet, packet);
2228 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2229 * create a new facet for it and use that. */
2230 if (rule_lookup(ofproto, &flow) == rule) {
2231 facet = facet_create(ofproto, rule, &flow, packet);
2232 facet_execute(ofproto, facet, packet);
2233 facet_install(ofproto, facet, true);
2237 /* We can't account anything to a facet. If we were to try, then that
2238 * facet would have a non-matching rule, busting our invariants. */
2239 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2240 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2241 size = packet->size;
2242 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2243 odp_actions->size, packet)) {
2244 rule->used = time_msec();
2245 rule->packet_count++;
2246 rule->byte_count += size;
2247 flow_push_stats(ofproto, rule, &flow, 1, size, rule->used);
2249 ofpbuf_delete(odp_actions);
2252 /* Inserts 'rule' into 'p''s flow table. */
2254 rule_insert(struct ofproto *p, struct rule *rule)
2256 struct rule *displaced_rule;
2258 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2259 if (displaced_rule) {
2260 rule_destroy(p, displaced_rule);
2262 p->need_revalidate = true;
2265 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2266 * 'flow' and an example 'packet' within that flow.
2268 * The caller must already have determined that no facet with an identical
2269 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2270 * 'ofproto''s classifier table. */
2271 static struct facet *
2272 facet_create(struct ofproto *ofproto, struct rule *rule,
2273 const struct flow *flow, const struct ofpbuf *packet)
2275 struct facet *facet;
2277 facet = xzalloc(sizeof *facet);
2278 facet->used = time_msec();
2279 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2280 list_push_back(&rule->facets, &facet->list_node);
2282 facet->flow = *flow;
2283 netflow_flow_init(&facet->nf_flow);
2284 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2286 facet_make_actions(ofproto, facet, packet);
2292 facet_free(struct facet *facet)
2294 free(facet->actions);
2298 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2300 * - Removes 'rule' from the classifier.
2302 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2303 * destroys them), via rule_destroy().
2306 rule_remove(struct ofproto *ofproto, struct rule *rule)
2308 COVERAGE_INC(ofproto_del_rule);
2309 ofproto->need_revalidate = true;
2310 classifier_remove(&ofproto->cls, &rule->cr);
2311 rule_destroy(ofproto, rule);
2314 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2316 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2317 * rule's statistics, via facet_uninstall().
2319 * - Removes 'facet' from its rule and from ofproto->facets.
2322 facet_remove(struct ofproto *ofproto, struct facet *facet)
2324 facet_uninstall(ofproto, facet);
2325 facet_flush_stats(ofproto, facet);
2326 hmap_remove(&ofproto->facets, &facet->hmap_node);
2327 list_remove(&facet->list_node);
2331 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2333 facet_make_actions(struct ofproto *p, struct facet *facet,
2334 const struct ofpbuf *packet)
2336 const struct rule *rule = facet->rule;
2337 struct ofpbuf *odp_actions;
2338 struct action_xlate_ctx ctx;
2340 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2341 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2342 facet->tags = ctx.tags;
2343 facet->may_install = ctx.may_set_up_flow;
2344 facet->nf_flow.output_iface = ctx.nf_output_iface;
2346 if (facet->actions_len != odp_actions->size
2347 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2348 free(facet->actions);
2349 facet->actions_len = odp_actions->size;
2350 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2353 ofpbuf_delete(odp_actions);
2357 facet_put__(struct ofproto *ofproto, struct facet *facet,
2358 const struct nlattr *actions, size_t actions_len,
2359 struct dpif_flow_stats *stats)
2361 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2362 enum dpif_flow_put_flags flags;
2365 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2367 flags |= DPIF_FP_ZERO_STATS;
2368 facet->dp_packet_count = 0;
2369 facet->dp_byte_count = 0;
2372 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2373 odp_flow_key_from_flow(&key, &facet->flow);
2374 assert(key.base == keybuf);
2376 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2377 actions, actions_len, stats);
2380 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2381 * 'zero_stats' is true, clears any existing statistics from the datapath for
2384 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2386 struct dpif_flow_stats stats;
2388 if (facet->may_install
2389 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2390 zero_stats ? &stats : NULL)) {
2391 facet->installed = true;
2395 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2396 * to the accounting hook function in the ofhooks structure. */
2398 facet_account(struct ofproto *ofproto,
2399 struct facet *facet, uint64_t extra_bytes)
2401 uint64_t total_bytes = facet->byte_count + extra_bytes;
2403 if (ofproto->ofhooks->account_flow_cb
2404 && total_bytes > facet->accounted_bytes)
2406 ofproto->ofhooks->account_flow_cb(
2407 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2408 total_bytes - facet->accounted_bytes, ofproto->aux);
2409 facet->accounted_bytes = total_bytes;
2413 /* If 'rule' is installed in the datapath, uninstalls it. */
2415 facet_uninstall(struct ofproto *p, struct facet *facet)
2417 if (facet->installed) {
2418 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2419 struct dpif_flow_stats stats;
2422 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2423 odp_flow_key_from_flow(&key, &facet->flow);
2424 assert(key.base == keybuf);
2426 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2427 facet_update_stats(p, facet, &stats);
2429 facet->installed = false;
2430 facet->dp_packet_count = 0;
2431 facet->dp_byte_count = 0;
2433 assert(facet->dp_packet_count == 0);
2434 assert(facet->dp_byte_count == 0);
2438 /* Returns true if the only action for 'facet' is to send to the controller.
2439 * (We don't report NetFlow expiration messages for such facets because they
2440 * are just part of the control logic for the network, not real traffic). */
2442 facet_is_controller_flow(struct facet *facet)
2445 && facet->rule->n_actions == 1
2446 && action_outputs_to_port(&facet->rule->actions[0],
2447 htons(OFPP_CONTROLLER)));
2450 /* Folds all of 'facet''s statistics into its rule. Also updates the
2451 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2452 * 'facet''s statistics in the datapath should have been zeroed and folded into
2453 * its packet and byte counts before this function is called. */
2455 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2457 assert(!facet->dp_byte_count);
2458 assert(!facet->dp_packet_count);
2460 facet_push_stats(ofproto, facet);
2461 facet_account(ofproto, facet, 0);
2463 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2464 struct ofexpired expired;
2465 expired.flow = facet->flow;
2466 expired.packet_count = facet->packet_count;
2467 expired.byte_count = facet->byte_count;
2468 expired.used = facet->used;
2469 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2472 facet->rule->packet_count += facet->packet_count;
2473 facet->rule->byte_count += facet->byte_count;
2475 /* Reset counters to prevent double counting if 'facet' ever gets
2477 facet->packet_count = 0;
2478 facet->byte_count = 0;
2479 facet->rs_packet_count = 0;
2480 facet->rs_byte_count = 0;
2481 facet->accounted_bytes = 0;
2483 netflow_flow_clear(&facet->nf_flow);
2486 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2487 * Returns it if found, otherwise a null pointer.
2489 * The returned facet might need revalidation; use facet_lookup_valid()
2490 * instead if that is important. */
2491 static struct facet *
2492 facet_find(struct ofproto *ofproto, const struct flow *flow)
2494 struct facet *facet;
2496 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2498 if (flow_equal(flow, &facet->flow)) {
2506 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2507 * Returns it if found, otherwise a null pointer.
2509 * The returned facet is guaranteed to be valid. */
2510 static struct facet *
2511 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2513 struct facet *facet = facet_find(ofproto, flow);
2515 /* The facet we found might not be valid, since we could be in need of
2516 * revalidation. If it is not valid, don't return it. */
2518 && ofproto->need_revalidate
2519 && !facet_revalidate(ofproto, facet)) {
2520 COVERAGE_INC(ofproto_invalidated);
2527 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2529 * - If the rule found is different from 'facet''s current rule, moves
2530 * 'facet' to the new rule and recompiles its actions.
2532 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2533 * where it is and recompiles its actions anyway.
2535 * - If there is none, destroys 'facet'.
2537 * Returns true if 'facet' still exists, false if it has been destroyed. */
2539 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2541 struct action_xlate_ctx ctx;
2542 struct ofpbuf *odp_actions;
2543 struct rule *new_rule;
2544 bool actions_changed;
2546 COVERAGE_INC(facet_revalidate);
2548 /* Determine the new rule. */
2549 new_rule = rule_lookup(ofproto, &facet->flow);
2551 /* No new rule, so delete the facet. */
2552 facet_remove(ofproto, facet);
2556 /* Calculate new ODP actions.
2558 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2559 * emit a NetFlow expiration and, if so, we need to have the old state
2560 * around to properly compose it. */
2561 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2562 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2563 actions_changed = (facet->actions_len != odp_actions->size
2564 || memcmp(facet->actions, odp_actions->data,
2565 facet->actions_len));
2567 /* If the ODP actions changed or the installability changed, then we need
2568 * to talk to the datapath. */
2569 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2570 if (ctx.may_set_up_flow) {
2571 struct dpif_flow_stats stats;
2573 facet_put__(ofproto, facet,
2574 odp_actions->data, odp_actions->size, &stats);
2575 facet_update_stats(ofproto, facet, &stats);
2577 facet_uninstall(ofproto, facet);
2580 /* The datapath flow is gone or has zeroed stats, so push stats out of
2581 * 'facet' into 'rule'. */
2582 facet_flush_stats(ofproto, facet);
2585 /* Update 'facet' now that we've taken care of all the old state. */
2586 facet->tags = ctx.tags;
2587 facet->nf_flow.output_iface = ctx.nf_output_iface;
2588 facet->may_install = ctx.may_set_up_flow;
2589 if (actions_changed) {
2590 free(facet->actions);
2591 facet->actions_len = odp_actions->size;
2592 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2594 if (facet->rule != new_rule) {
2595 COVERAGE_INC(facet_changed_rule);
2596 list_remove(&facet->list_node);
2597 list_push_back(&new_rule->facets, &facet->list_node);
2598 facet->rule = new_rule;
2599 facet->used = new_rule->created;
2600 facet->rs_used = facet->used;
2603 ofpbuf_delete(odp_actions);
2609 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2610 struct rconn_packet_counter *counter)
2612 update_openflow_length(msg);
2613 if (rconn_send(ofconn->rconn, msg, counter)) {
2619 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2622 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2624 COVERAGE_INC(ofproto_error);
2625 queue_tx(buf, ofconn, ofconn->reply_counter);
2630 hton_ofp_phy_port(struct ofp_phy_port *opp)
2632 opp->port_no = htons(opp->port_no);
2633 opp->config = htonl(opp->config);
2634 opp->state = htonl(opp->state);
2635 opp->curr = htonl(opp->curr);
2636 opp->advertised = htonl(opp->advertised);
2637 opp->supported = htonl(opp->supported);
2638 opp->peer = htonl(opp->peer);
2642 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2644 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2649 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2651 struct ofp_switch_features *osf;
2653 struct ofport *port;
2655 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2656 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2657 osf->n_buffers = htonl(pktbuf_capacity());
2659 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2660 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2661 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2662 (1u << OFPAT_SET_VLAN_VID) |
2663 (1u << OFPAT_SET_VLAN_PCP) |
2664 (1u << OFPAT_STRIP_VLAN) |
2665 (1u << OFPAT_SET_DL_SRC) |
2666 (1u << OFPAT_SET_DL_DST) |
2667 (1u << OFPAT_SET_NW_SRC) |
2668 (1u << OFPAT_SET_NW_DST) |
2669 (1u << OFPAT_SET_NW_TOS) |
2670 (1u << OFPAT_SET_TP_SRC) |
2671 (1u << OFPAT_SET_TP_DST) |
2672 (1u << OFPAT_ENQUEUE));
2674 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2675 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2678 queue_tx(buf, ofconn, ofconn->reply_counter);
2683 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2686 struct ofp_switch_config *osc;
2690 /* Figure out flags. */
2691 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2692 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2695 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2696 osc->flags = htons(flags);
2697 osc->miss_send_len = htons(ofconn->miss_send_len);
2698 queue_tx(buf, ofconn, ofconn->reply_counter);
2704 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2706 uint16_t flags = ntohs(osc->flags);
2708 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2709 switch (flags & OFPC_FRAG_MASK) {
2710 case OFPC_FRAG_NORMAL:
2711 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2713 case OFPC_FRAG_DROP:
2714 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2717 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2723 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2728 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2729 struct action_xlate_ctx *ctx);
2732 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2734 const struct ofport *ofport = get_port(ctx->ofproto, port);
2737 if (ofport->opp.config & OFPPC_NO_FWD) {
2738 /* Forwarding disabled on port. */
2743 * We don't have an ofport record for this port, but it doesn't hurt to
2744 * allow forwarding to it anyhow. Maybe such a port will appear later
2745 * and we're pre-populating the flow table.
2749 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2750 ctx->nf_output_iface = port;
2753 static struct rule *
2754 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2756 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2760 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2762 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2763 uint16_t old_in_port;
2766 /* Look up a flow with 'in_port' as the input port. Then restore the
2767 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2768 * have surprising behavior). */
2769 old_in_port = ctx->flow.in_port;
2770 ctx->flow.in_port = in_port;
2771 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2772 ctx->flow.in_port = old_in_port;
2774 if (ctx->resubmit_hook) {
2775 ctx->resubmit_hook(ctx, rule);
2780 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2784 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2786 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2787 MAX_RESUBMIT_RECURSION);
2792 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2793 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2795 struct ofport *ofport;
2797 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2798 uint16_t odp_port = ofport->odp_port;
2799 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2800 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2803 *nf_output_iface = NF_OUT_FLOOD;
2807 xlate_output_action__(struct action_xlate_ctx *ctx,
2808 uint16_t port, uint16_t max_len)
2811 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2813 ctx->nf_output_iface = NF_OUT_DROP;
2817 add_output_action(ctx, ctx->flow.in_port);
2820 xlate_table_action(ctx, ctx->flow.in_port);
2823 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2824 ctx->odp_actions, &ctx->tags,
2825 &ctx->nf_output_iface,
2826 ctx->ofproto->aux)) {
2827 COVERAGE_INC(ofproto_uninstallable);
2828 ctx->may_set_up_flow = false;
2832 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2833 &ctx->nf_output_iface, ctx->odp_actions);
2836 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2837 &ctx->nf_output_iface, ctx->odp_actions);
2839 case OFPP_CONTROLLER:
2840 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2843 add_output_action(ctx, ODPP_LOCAL);
2846 odp_port = ofp_port_to_odp_port(port);
2847 if (odp_port != ctx->flow.in_port) {
2848 add_output_action(ctx, odp_port);
2853 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2854 ctx->nf_output_iface = NF_OUT_FLOOD;
2855 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2856 ctx->nf_output_iface = prev_nf_output_iface;
2857 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2858 ctx->nf_output_iface != NF_OUT_FLOOD) {
2859 ctx->nf_output_iface = NF_OUT_MULTI;
2864 xlate_output_action(struct action_xlate_ctx *ctx,
2865 const struct ofp_action_output *oao)
2867 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2870 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2871 * optimization, because we're going to add another action that sets the
2872 * priority immediately after, or because there are no actions following the
2875 remove_pop_action(struct action_xlate_ctx *ctx)
2877 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2878 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2879 ctx->last_pop_priority = -1;
2884 add_pop_action(struct action_xlate_ctx *ctx)
2886 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2887 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2888 ctx->last_pop_priority = ctx->odp_actions->size;
2893 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2894 const struct ofp_action_enqueue *oae)
2896 uint16_t ofp_port, odp_port;
2900 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2903 /* Fall back to ordinary output action. */
2904 xlate_output_action__(ctx, ntohs(oae->port), 0);
2908 /* Figure out ODP output port. */
2909 ofp_port = ntohs(oae->port);
2910 if (ofp_port != OFPP_IN_PORT) {
2911 odp_port = ofp_port_to_odp_port(ofp_port);
2913 odp_port = ctx->flow.in_port;
2916 /* Add ODP actions. */
2917 remove_pop_action(ctx);
2918 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2919 add_output_action(ctx, odp_port);
2920 add_pop_action(ctx);
2922 /* Update NetFlow output port. */
2923 if (ctx->nf_output_iface == NF_OUT_DROP) {
2924 ctx->nf_output_iface = odp_port;
2925 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2926 ctx->nf_output_iface = NF_OUT_MULTI;
2931 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2932 const struct nx_action_set_queue *nasq)
2937 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2940 /* Couldn't translate queue to a priority, so ignore. A warning
2941 * has already been logged. */
2945 remove_pop_action(ctx);
2946 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2950 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2952 ovs_be16 tci = ctx->flow.vlan_tci;
2953 if (!(tci & htons(VLAN_CFI))) {
2954 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2956 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2957 tci & ~htons(VLAN_CFI));
2961 struct xlate_reg_state {
2967 save_reg_state(const struct action_xlate_ctx *ctx,
2968 struct xlate_reg_state *state)
2970 state->vlan_tci = ctx->flow.vlan_tci;
2971 state->tun_id = ctx->flow.tun_id;
2975 update_reg_state(struct action_xlate_ctx *ctx,
2976 const struct xlate_reg_state *state)
2978 if (ctx->flow.vlan_tci != state->vlan_tci) {
2979 xlate_set_dl_tci(ctx);
2981 if (ctx->flow.tun_id != state->tun_id) {
2982 nl_msg_put_be64(ctx->odp_actions,
2983 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2988 xlate_nicira_action(struct action_xlate_ctx *ctx,
2989 const struct nx_action_header *nah)
2991 const struct nx_action_resubmit *nar;
2992 const struct nx_action_set_tunnel *nast;
2993 const struct nx_action_set_queue *nasq;
2994 const struct nx_action_multipath *nam;
2995 enum nx_action_subtype subtype = ntohs(nah->subtype);
2996 struct xlate_reg_state state;
2999 assert(nah->vendor == htonl(NX_VENDOR_ID));
3001 case NXAST_RESUBMIT:
3002 nar = (const struct nx_action_resubmit *) nah;
3003 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
3006 case NXAST_SET_TUNNEL:
3007 nast = (const struct nx_action_set_tunnel *) nah;
3008 tun_id = htonll(ntohl(nast->tun_id));
3009 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3010 ctx->flow.tun_id = tun_id;
3013 case NXAST_DROP_SPOOFED_ARP:
3014 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
3015 nl_msg_put_flag(ctx->odp_actions,
3016 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
3020 case NXAST_SET_QUEUE:
3021 nasq = (const struct nx_action_set_queue *) nah;
3022 xlate_set_queue_action(ctx, nasq);
3025 case NXAST_POP_QUEUE:
3026 add_pop_action(ctx);
3029 case NXAST_REG_MOVE:
3030 save_reg_state(ctx, &state);
3031 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
3033 update_reg_state(ctx, &state);
3036 case NXAST_REG_LOAD:
3037 save_reg_state(ctx, &state);
3038 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
3040 update_reg_state(ctx, &state);
3044 /* Nothing to do. */
3047 case NXAST_SET_TUNNEL64:
3048 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
3049 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3050 ctx->flow.tun_id = tun_id;
3053 case NXAST_MULTIPATH:
3054 nam = (const struct nx_action_multipath *) nah;
3055 multipath_execute(nam, &ctx->flow);
3058 /* If you add a new action here that modifies flow data, don't forget to
3059 * update the flow key in ctx->flow at the same time. */
3061 case NXAST_SNAT__OBSOLETE:
3063 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3069 do_xlate_actions(const union ofp_action *in, size_t n_in,
3070 struct action_xlate_ctx *ctx)
3072 struct actions_iterator iter;
3073 const union ofp_action *ia;
3074 const struct ofport *port;
3076 port = get_port(ctx->ofproto, ctx->flow.in_port);
3077 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3078 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3079 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3080 /* Drop this flow. */
3084 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3085 enum ofp_action_type type = ntohs(ia->type);
3086 const struct ofp_action_dl_addr *oada;
3090 xlate_output_action(ctx, &ia->output);
3093 case OFPAT_SET_VLAN_VID:
3094 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3095 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3096 xlate_set_dl_tci(ctx);
3099 case OFPAT_SET_VLAN_PCP:
3100 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3101 ctx->flow.vlan_tci |= htons(
3102 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3103 xlate_set_dl_tci(ctx);
3106 case OFPAT_STRIP_VLAN:
3107 ctx->flow.vlan_tci = htons(0);
3108 xlate_set_dl_tci(ctx);
3111 case OFPAT_SET_DL_SRC:
3112 oada = ((struct ofp_action_dl_addr *) ia);
3113 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3114 oada->dl_addr, ETH_ADDR_LEN);
3115 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3118 case OFPAT_SET_DL_DST:
3119 oada = ((struct ofp_action_dl_addr *) ia);
3120 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3121 oada->dl_addr, ETH_ADDR_LEN);
3122 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3125 case OFPAT_SET_NW_SRC:
3126 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3127 ia->nw_addr.nw_addr);
3128 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3131 case OFPAT_SET_NW_DST:
3132 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3133 ia->nw_addr.nw_addr);
3134 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3137 case OFPAT_SET_NW_TOS:
3138 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3140 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3143 case OFPAT_SET_TP_SRC:
3144 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3145 ia->tp_port.tp_port);
3146 ctx->flow.tp_src = ia->tp_port.tp_port;
3149 case OFPAT_SET_TP_DST:
3150 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3151 ia->tp_port.tp_port);
3152 ctx->flow.tp_dst = ia->tp_port.tp_port;
3156 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3160 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3164 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3171 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3172 struct ofproto *ofproto, const struct flow *flow,
3173 const struct ofpbuf *packet)
3175 ctx->ofproto = ofproto;
3177 ctx->packet = packet;
3178 ctx->resubmit_hook = NULL;
3179 ctx->check_special = true;
3183 ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow,
3184 const struct ofpbuf *packet)
3186 struct ofport *ofport;
3188 ofport = get_port(ofproto, flow->in_port);
3189 if (ofport && ofport->cfm) {
3190 cfm_process_heartbeat(ofport->cfm, packet);
3194 static struct ofpbuf *
3195 xlate_actions(struct action_xlate_ctx *ctx,
3196 const union ofp_action *in, size_t n_in)
3198 COVERAGE_INC(ofproto_ofp2odp);
3200 ctx->odp_actions = ofpbuf_new(512);
3202 ctx->may_set_up_flow = true;
3203 ctx->nf_output_iface = NF_OUT_DROP;
3205 ctx->last_pop_priority = -1;
3207 if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) {
3209 ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet);
3211 ctx->may_set_up_flow = false;
3212 } else if (ctx->check_special
3213 && ctx->ofproto->ofhooks->special_cb
3214 && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3215 ctx->ofproto->aux)) {
3216 ctx->may_set_up_flow = false;
3218 do_xlate_actions(in, n_in, ctx);
3221 remove_pop_action(ctx);
3223 /* Check with in-band control to see if we're allowed to set up this
3225 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3226 ctx->odp_actions->data, ctx->odp_actions->size)) {
3227 ctx->may_set_up_flow = false;
3230 return ctx->odp_actions;
3233 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3234 * error message code (composed with ofp_mkerr()) for the caller to propagate
3235 * upward. Otherwise, returns 0.
3237 * The log message mentions 'msg_type'. */
3239 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3241 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3242 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3243 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3246 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3253 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3255 struct ofproto *p = ofconn->ofproto;
3256 struct ofp_packet_out *opo;
3257 struct ofpbuf payload, *buffer;
3258 union ofp_action *ofp_actions;
3259 struct action_xlate_ctx ctx;
3260 struct ofpbuf *odp_actions;
3261 struct ofpbuf request;
3263 size_t n_ofp_actions;
3267 COVERAGE_INC(ofproto_packet_out);
3269 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3274 /* Get ofp_packet_out. */
3275 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3276 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3279 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3280 &ofp_actions, &n_ofp_actions);
3286 if (opo->buffer_id != htonl(UINT32_MAX)) {
3287 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3289 if (error || !buffer) {
3298 /* Extract flow, check actions. */
3299 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3301 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3307 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3308 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3309 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3310 ofpbuf_delete(odp_actions);
3313 ofpbuf_delete(buffer);
3318 update_port_config(struct ofproto *p, struct ofport *port,
3319 uint32_t config, uint32_t mask)
3321 mask &= config ^ port->opp.config;
3322 if (mask & OFPPC_PORT_DOWN) {
3323 if (config & OFPPC_PORT_DOWN) {
3324 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3326 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3329 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3330 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3331 if (mask & REVALIDATE_BITS) {
3332 COVERAGE_INC(ofproto_costly_flags);
3333 port->opp.config ^= mask & REVALIDATE_BITS;
3334 p->need_revalidate = true;
3336 #undef REVALIDATE_BITS
3337 if (mask & OFPPC_NO_PACKET_IN) {
3338 port->opp.config ^= OFPPC_NO_PACKET_IN;
3343 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3345 struct ofproto *p = ofconn->ofproto;
3346 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3347 struct ofport *port;
3350 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3355 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3357 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3358 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3359 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3361 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3362 if (opm->advertise) {
3363 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3369 static struct ofpbuf *
3370 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3372 struct ofp_stats_reply *osr;
3375 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3376 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3378 osr->flags = htons(0);
3382 static struct ofpbuf *
3383 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3385 const struct ofp_stats_request *osr
3386 = (const struct ofp_stats_request *) request;
3387 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3391 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3392 struct ofpbuf **msgp)
3394 struct ofpbuf *msg = *msgp;
3395 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3396 if (nbytes + msg->size > UINT16_MAX) {
3397 struct ofp_stats_reply *reply = msg->data;
3398 reply->flags = htons(OFPSF_REPLY_MORE);
3399 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3400 queue_tx(msg, ofconn, ofconn->reply_counter);
3402 return ofpbuf_put_uninit(*msgp, nbytes);
3405 static struct ofpbuf *
3406 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3408 struct nicira_stats_msg *nsm;
3411 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3412 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3413 nsm->type = htons(OFPST_VENDOR);
3414 nsm->flags = htons(0);
3415 nsm->vendor = htonl(NX_VENDOR_ID);
3416 nsm->subtype = subtype;
3420 static struct ofpbuf *
3421 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3423 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3427 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3428 struct ofpbuf **msgp)
3430 struct ofpbuf *msg = *msgp;
3431 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3432 if (nbytes + msg->size > UINT16_MAX) {
3433 struct nicira_stats_msg *reply = msg->data;
3434 reply->flags = htons(OFPSF_REPLY_MORE);
3435 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3436 queue_tx(msg, ofconn, ofconn->reply_counter);
3438 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3442 handle_desc_stats_request(struct ofconn *ofconn,
3443 const struct ofp_header *request)
3445 struct ofproto *p = ofconn->ofproto;
3446 struct ofp_desc_stats *ods;
3449 msg = start_ofp_stats_reply(request, sizeof *ods);
3450 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3451 memset(ods, 0, sizeof *ods);
3452 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3453 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3454 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3455 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3456 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3457 queue_tx(msg, ofconn, ofconn->reply_counter);
3463 handle_table_stats_request(struct ofconn *ofconn,
3464 const struct ofp_header *request)
3466 struct ofproto *p = ofconn->ofproto;
3467 struct ofp_table_stats *ots;
3470 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3472 /* Classifier table. */
3473 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3474 memset(ots, 0, sizeof *ots);
3475 strcpy(ots->name, "classifier");
3476 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3477 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3478 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3479 ots->active_count = htonl(classifier_count(&p->cls));
3480 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3481 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3483 queue_tx(msg, ofconn, ofconn->reply_counter);
3488 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3489 struct ofpbuf **msgp)
3491 struct netdev_stats stats;
3492 struct ofp_port_stats *ops;
3494 /* Intentionally ignore return value, since errors will set
3495 * 'stats' to all-1s, which is correct for OpenFlow, and
3496 * netdev_get_stats() will log errors. */
3497 netdev_get_stats(port->netdev, &stats);
3499 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3500 ops->port_no = htons(port->opp.port_no);
3501 memset(ops->pad, 0, sizeof ops->pad);
3502 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3503 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3504 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3505 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3506 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3507 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3508 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3509 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3510 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3511 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3512 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3513 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3517 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3519 struct ofproto *p = ofconn->ofproto;
3520 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3521 struct ofp_port_stats *ops;
3523 struct ofport *port;
3525 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3526 if (psr->port_no != htons(OFPP_NONE)) {
3527 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3529 append_port_stat(port, ofconn, &msg);
3532 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3533 append_port_stat(port, ofconn, &msg);
3537 queue_tx(msg, ofconn, ofconn->reply_counter);
3542 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3544 long long int msecs = time_msec() - start;
3545 *sec = htonl(msecs / 1000);
3546 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3550 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3551 ovs_be16 out_port, struct ofpbuf **replyp)
3553 struct ofp_flow_stats *ofs;
3554 uint64_t packet_count, byte_count;
3556 size_t act_len, len;
3558 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3562 act_len = sizeof *rule->actions * rule->n_actions;
3563 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3565 rule_get_stats(rule, &packet_count, &byte_count);
3567 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3568 ofs->length = htons(len);
3571 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3572 rule->flow_cookie, &cookie);
3573 put_32aligned_be64(&ofs->cookie, cookie);
3574 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3575 ofs->priority = htons(rule->cr.priority);
3576 ofs->idle_timeout = htons(rule->idle_timeout);
3577 ofs->hard_timeout = htons(rule->hard_timeout);
3578 memset(ofs->pad2, 0, sizeof ofs->pad2);
3579 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3580 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3581 if (rule->n_actions > 0) {
3582 memcpy(ofs->actions, rule->actions, act_len);
3587 is_valid_table(uint8_t table_id)
3589 if (table_id == 0 || table_id == 0xff) {
3592 /* It would probably be better to reply with an error but there doesn't
3593 * seem to be any appropriate value, so that might just be
3595 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
3602 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3604 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3605 struct ofpbuf *reply;
3607 COVERAGE_INC(ofproto_flows_req);
3608 reply = start_ofp_stats_reply(oh, 1024);
3609 if (is_valid_table(fsr->table_id)) {
3610 struct cls_cursor cursor;
3611 struct cls_rule target;
3614 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3616 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3617 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3618 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3621 queue_tx(reply, ofconn, ofconn->reply_counter);
3627 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3628 ovs_be16 out_port, struct ofpbuf **replyp)
3630 struct nx_flow_stats *nfs;
3631 uint64_t packet_count, byte_count;
3632 size_t act_len, start_len;
3633 struct ofpbuf *reply;
3635 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3639 rule_get_stats(rule, &packet_count, &byte_count);
3641 act_len = sizeof *rule->actions * rule->n_actions;
3643 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3644 start_len = (*replyp)->size;
3647 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3650 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3651 nfs->cookie = rule->flow_cookie;
3652 nfs->priority = htons(rule->cr.priority);
3653 nfs->idle_timeout = htons(rule->idle_timeout);
3654 nfs->hard_timeout = htons(rule->hard_timeout);
3655 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3656 memset(nfs->pad2, 0, sizeof nfs->pad2);
3657 nfs->packet_count = htonll(packet_count);
3658 nfs->byte_count = htonll(byte_count);
3659 if (rule->n_actions > 0) {
3660 ofpbuf_put(reply, rule->actions, act_len);
3662 nfs->length = htons(reply->size - start_len);
3666 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3668 struct nx_flow_stats_request *nfsr;
3669 struct cls_rule target;
3670 struct ofpbuf *reply;
3674 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3676 /* Dissect the message. */
3677 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3678 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3683 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3686 COVERAGE_INC(ofproto_flows_req);
3687 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3688 if (is_valid_table(nfsr->table_id)) {
3689 struct cls_cursor cursor;
3692 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3693 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3694 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3697 queue_tx(reply, ofconn, ofconn->reply_counter);
3703 flow_stats_ds(struct rule *rule, struct ds *results)
3705 uint64_t packet_count, byte_count;
3706 size_t act_len = sizeof *rule->actions * rule->n_actions;
3708 rule_get_stats(rule, &packet_count, &byte_count);
3710 ds_put_format(results, "duration=%llds, ",
3711 (time_msec() - rule->created) / 1000);
3712 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3713 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3714 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3715 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3716 cls_rule_format(&rule->cr, results);
3717 ds_put_char(results, ',');
3719 ofp_print_actions(results, &rule->actions->header, act_len);
3721 ds_put_cstr(results, "drop");
3723 ds_put_cstr(results, "\n");
3726 /* Adds a pretty-printed description of all flows to 'results', including
3727 * hidden flows (e.g., set up by in-band control). */
3729 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3731 struct cls_cursor cursor;
3734 cls_cursor_init(&cursor, &p->cls, NULL);
3735 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3736 flow_stats_ds(rule, results);
3741 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3742 ovs_be16 out_port, uint8_t table_id,
3743 struct ofp_aggregate_stats_reply *oasr)
3745 uint64_t total_packets = 0;
3746 uint64_t total_bytes = 0;
3749 COVERAGE_INC(ofproto_agg_request);
3751 if (is_valid_table(table_id)) {
3752 struct cls_cursor cursor;
3755 cls_cursor_init(&cursor, &ofproto->cls, target);
3756 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3757 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3758 uint64_t packet_count;
3759 uint64_t byte_count;
3761 rule_get_stats(rule, &packet_count, &byte_count);
3763 total_packets += packet_count;
3764 total_bytes += byte_count;
3770 oasr->flow_count = htonl(n_flows);
3771 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3772 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3773 memset(oasr->pad, 0, sizeof oasr->pad);
3777 handle_aggregate_stats_request(struct ofconn *ofconn,
3778 const struct ofp_header *oh)
3780 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3781 struct ofp_aggregate_stats_reply *reply;
3782 struct cls_rule target;
3785 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3788 msg = start_ofp_stats_reply(oh, sizeof *reply);
3789 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3790 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3791 request->table_id, reply);
3792 queue_tx(msg, ofconn, ofconn->reply_counter);
3797 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3799 struct nx_aggregate_stats_request *request;
3800 struct ofp_aggregate_stats_reply *reply;
3801 struct cls_rule target;
3806 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3808 /* Dissect the message. */
3809 request = ofpbuf_pull(&b, sizeof *request);
3810 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3815 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3819 COVERAGE_INC(ofproto_flows_req);
3820 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3821 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3822 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3823 request->table_id, reply);
3824 queue_tx(buf, ofconn, ofconn->reply_counter);
3829 struct queue_stats_cbdata {
3830 struct ofconn *ofconn;
3831 struct ofport *ofport;
3836 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3837 const struct netdev_queue_stats *stats)
3839 struct ofp_queue_stats *reply;
3841 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3842 reply->port_no = htons(cbdata->ofport->opp.port_no);
3843 memset(reply->pad, 0, sizeof reply->pad);
3844 reply->queue_id = htonl(queue_id);
3845 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3846 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3847 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3851 handle_queue_stats_dump_cb(uint32_t queue_id,
3852 struct netdev_queue_stats *stats,
3855 struct queue_stats_cbdata *cbdata = cbdata_;
3857 put_queue_stats(cbdata, queue_id, stats);
3861 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3862 struct queue_stats_cbdata *cbdata)
3864 cbdata->ofport = port;
3865 if (queue_id == OFPQ_ALL) {
3866 netdev_dump_queue_stats(port->netdev,
3867 handle_queue_stats_dump_cb, cbdata);
3869 struct netdev_queue_stats stats;
3871 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3872 put_queue_stats(cbdata, queue_id, &stats);
3878 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3880 struct ofproto *ofproto = ofconn->ofproto;
3881 const struct ofp_queue_stats_request *qsr;
3882 struct queue_stats_cbdata cbdata;
3883 struct ofport *port;
3884 unsigned int port_no;
3887 qsr = ofputil_stats_body(oh);
3889 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3892 COVERAGE_INC(ofproto_queue_req);
3894 cbdata.ofconn = ofconn;
3895 cbdata.msg = start_ofp_stats_reply(oh, 128);
3897 port_no = ntohs(qsr->port_no);
3898 queue_id = ntohl(qsr->queue_id);
3899 if (port_no == OFPP_ALL) {
3900 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3901 handle_queue_stats_for_port(port, queue_id, &cbdata);
3903 } else if (port_no < ofproto->max_ports) {
3904 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3906 handle_queue_stats_for_port(port, queue_id, &cbdata);
3909 ofpbuf_delete(cbdata.msg);
3910 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3912 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3917 /* Updates 'facet''s used time. Caller is responsible for calling
3918 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3920 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3923 if (used > facet->used) {
3925 if (used > facet->rule->used) {
3926 facet->rule->used = used;
3928 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3932 /* Folds the statistics from 'stats' into the counters in 'facet'.
3934 * Because of the meaning of a facet's counters, it only makes sense to do this
3935 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3936 * packet that was sent by hand or if it represents statistics that have been
3937 * cleared out of the datapath. */
3939 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3940 const struct dpif_flow_stats *stats)
3942 if (stats->n_packets || stats->used > facet->used) {
3943 facet_update_time(ofproto, facet, stats->used);
3944 facet->packet_count += stats->n_packets;
3945 facet->byte_count += stats->n_bytes;
3946 facet_push_stats(ofproto, facet);
3947 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3952 facet_push_stats(struct ofproto *ofproto, struct facet *facet)
3954 uint64_t rs_packets, rs_bytes;
3956 assert(facet->packet_count >= facet->rs_packet_count);
3957 assert(facet->byte_count >= facet->rs_byte_count);
3958 assert(facet->used >= facet->rs_used);
3960 rs_packets = facet->packet_count - facet->rs_packet_count;
3961 rs_bytes = facet->byte_count - facet->rs_byte_count;
3963 if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
3964 facet->rs_packet_count = facet->packet_count;
3965 facet->rs_byte_count = facet->byte_count;
3966 facet->rs_used = facet->used;
3968 flow_push_stats(ofproto, facet->rule, &facet->flow,
3969 rs_packets, rs_bytes, facet->used);
3973 struct ofproto_push {
3974 struct action_xlate_ctx ctx;
3981 push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
3983 struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
3986 rule->packet_count += push->packets;
3987 rule->byte_count += push->bytes;
3988 rule->used = MAX(push->used, rule->used);
3992 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3993 * 'rule''s actions. */
3995 flow_push_stats(struct ofproto *ofproto, const struct rule *rule,
3996 struct flow *flow, uint64_t packets, uint64_t bytes,
3999 struct ofproto_push push;
4001 push.packets = packets;
4005 action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
4006 push.ctx.resubmit_hook = push_resubmit;
4007 ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions));
4010 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4011 * in which no matching flow already exists in the flow table.
4013 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4014 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
4015 * OpenFlow error code as encoded by ofp_mkerr() on failure.
4017 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4020 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
4022 struct ofproto *p = ofconn->ofproto;
4023 struct ofpbuf *packet;
4028 if (fm->flags & OFPFF_CHECK_OVERLAP
4029 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
4030 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
4034 if (fm->buffer_id != UINT32_MAX) {
4035 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
4039 in_port = UINT16_MAX;
4042 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
4043 fm->idle_timeout, fm->hard_timeout, fm->cookie,
4044 fm->flags & OFPFF_SEND_FLOW_REM);
4045 rule_insert(p, rule);
4047 rule_execute(p, rule, in_port, packet);
4052 static struct rule *
4053 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
4055 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
4059 send_buffered_packet(struct ofconn *ofconn,
4060 struct rule *rule, uint32_t buffer_id)
4062 struct ofpbuf *packet;
4066 if (buffer_id == UINT32_MAX) {
4070 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
4075 rule_execute(ofconn->ofproto, rule, in_port, packet);
4080 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4082 struct modify_flows_cbdata {
4083 struct ofproto *ofproto;
4084 const struct flow_mod *fm;
4088 static int modify_flow(struct ofproto *, const struct flow_mod *,
4091 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4092 * encoded by ofp_mkerr() on failure.
4094 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4097 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
4099 struct ofproto *p = ofconn->ofproto;
4100 struct rule *match = NULL;
4101 struct cls_cursor cursor;
4104 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4105 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4106 if (!rule_is_hidden(rule)) {
4108 modify_flow(p, fm, rule);
4113 /* This credits the packet to whichever flow happened to match last.
4114 * That's weird. Maybe we should do a lookup for the flow that
4115 * actually matches the packet? Who knows. */
4116 send_buffered_packet(ofconn, match, fm->buffer_id);
4119 return add_flow(ofconn, fm);
4123 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4124 * code as encoded by ofp_mkerr() on failure.
4126 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4129 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4131 struct ofproto *p = ofconn->ofproto;
4132 struct rule *rule = find_flow_strict(p, fm);
4133 if (rule && !rule_is_hidden(rule)) {
4134 modify_flow(p, fm, rule);
4135 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4137 return add_flow(ofconn, fm);
4141 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4142 * been identified as a flow in 'p''s flow table to be modified, by changing
4143 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4144 * ofp_action[] structures). */
4146 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4148 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4150 rule->flow_cookie = fm->cookie;
4152 /* If the actions are the same, do nothing. */
4153 if (fm->n_actions == rule->n_actions
4155 || !memcmp(fm->actions, rule->actions, actions_len))) {
4159 /* Replace actions. */
4160 free(rule->actions);
4161 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4162 rule->n_actions = fm->n_actions;
4164 p->need_revalidate = true;
4169 /* OFPFC_DELETE implementation. */
4171 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4173 /* Implements OFPFC_DELETE. */
4175 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4177 struct rule *rule, *next_rule;
4178 struct cls_cursor cursor;
4180 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4181 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4182 delete_flow(p, rule, htons(fm->out_port));
4186 /* Implements OFPFC_DELETE_STRICT. */
4188 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4190 struct rule *rule = find_flow_strict(p, fm);
4192 delete_flow(p, rule, htons(fm->out_port));
4196 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4197 * been identified as a flow to delete from 'p''s flow table, by deleting the
4198 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4201 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4202 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4203 * specified 'out_port'. */
4205 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4207 if (rule_is_hidden(rule)) {
4211 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4215 rule_send_removed(p, rule, OFPRR_DELETE);
4216 rule_remove(p, rule);
4220 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4222 struct ofproto *p = ofconn->ofproto;
4226 error = reject_slave_controller(ofconn, "flow_mod");
4231 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4236 /* We do not support the emergency flow cache. It will hopefully get
4237 * dropped from OpenFlow in the near future. */
4238 if (fm.flags & OFPFF_EMERG) {
4239 /* There isn't a good fit for an error code, so just state that the
4240 * flow table is full. */
4241 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4244 error = validate_actions(fm.actions, fm.n_actions,
4245 &fm.cr.flow, p->max_ports);
4250 switch (fm.command) {
4252 return add_flow(ofconn, &fm);
4255 return modify_flows_loose(ofconn, &fm);
4257 case OFPFC_MODIFY_STRICT:
4258 return modify_flow_strict(ofconn, &fm);
4261 delete_flows_loose(p, &fm);
4264 case OFPFC_DELETE_STRICT:
4265 delete_flow_strict(p, &fm);
4269 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4274 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4276 const struct nxt_tun_id_cookie *msg
4277 = (const struct nxt_tun_id_cookie *) oh;
4279 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4284 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4286 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4287 struct nx_role_request *reply;
4291 if (ofconn->type != OFCONN_PRIMARY) {
4292 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4294 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4297 role = ntohl(nrr->role);
4298 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4299 && role != NX_ROLE_SLAVE) {
4300 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4302 /* There's no good error code for this. */
4303 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4306 if (role == NX_ROLE_MASTER) {
4307 struct ofconn *other;
4309 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4310 if (other->role == NX_ROLE_MASTER) {
4311 other->role = NX_ROLE_SLAVE;
4315 ofconn->role = role;
4317 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4318 reply->role = htonl(role);
4319 queue_tx(buf, ofconn, ofconn->reply_counter);
4325 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4327 const struct nxt_set_flow_format *msg
4328 = (const struct nxt_set_flow_format *) oh;
4331 format = ntohl(msg->format);
4332 if (format == NXFF_OPENFLOW10
4333 || format == NXFF_TUN_ID_FROM_COOKIE
4334 || format == NXFF_NXM) {
4335 ofconn->flow_format = format;
4338 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4343 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4345 struct ofp_header *ob;
4348 /* Currently, everything executes synchronously, so we can just
4349 * immediately send the barrier reply. */
4350 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4351 queue_tx(buf, ofconn, ofconn->reply_counter);
4356 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4358 const struct ofp_header *oh = msg->data;
4359 const struct ofputil_msg_type *type;
4362 error = ofputil_decode_msg_type(oh, &type);
4367 switch (ofputil_msg_type_code(type)) {
4368 /* OpenFlow requests. */
4369 case OFPUTIL_OFPT_ECHO_REQUEST:
4370 return handle_echo_request(ofconn, oh);
4372 case OFPUTIL_OFPT_FEATURES_REQUEST:
4373 return handle_features_request(ofconn, oh);
4375 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4376 return handle_get_config_request(ofconn, oh);
4378 case OFPUTIL_OFPT_SET_CONFIG:
4379 return handle_set_config(ofconn, msg->data);
4381 case OFPUTIL_OFPT_PACKET_OUT:
4382 return handle_packet_out(ofconn, oh);
4384 case OFPUTIL_OFPT_PORT_MOD:
4385 return handle_port_mod(ofconn, oh);
4387 case OFPUTIL_OFPT_FLOW_MOD:
4388 return handle_flow_mod(ofconn, oh);
4390 case OFPUTIL_OFPT_BARRIER_REQUEST:
4391 return handle_barrier_request(ofconn, oh);
4393 /* OpenFlow replies. */
4394 case OFPUTIL_OFPT_ECHO_REPLY:
4397 /* Nicira extension requests. */
4398 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4399 return handle_tun_id_from_cookie(ofconn, oh);
4401 case OFPUTIL_NXT_ROLE_REQUEST:
4402 return handle_role_request(ofconn, oh);
4404 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4405 return handle_nxt_set_flow_format(ofconn, oh);
4407 case OFPUTIL_NXT_FLOW_MOD:
4408 return handle_flow_mod(ofconn, oh);
4410 /* OpenFlow statistics requests. */
4411 case OFPUTIL_OFPST_DESC_REQUEST:
4412 return handle_desc_stats_request(ofconn, oh);
4414 case OFPUTIL_OFPST_FLOW_REQUEST:
4415 return handle_flow_stats_request(ofconn, oh);
4417 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4418 return handle_aggregate_stats_request(ofconn, oh);
4420 case OFPUTIL_OFPST_TABLE_REQUEST:
4421 return handle_table_stats_request(ofconn, oh);
4423 case OFPUTIL_OFPST_PORT_REQUEST:
4424 return handle_port_stats_request(ofconn, oh);
4426 case OFPUTIL_OFPST_QUEUE_REQUEST:
4427 return handle_queue_stats_request(ofconn, oh);
4429 /* Nicira extension statistics requests. */
4430 case OFPUTIL_NXST_FLOW_REQUEST:
4431 return handle_nxst_flow(ofconn, oh);
4433 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4434 return handle_nxst_aggregate(ofconn, oh);
4436 case OFPUTIL_INVALID:
4437 case OFPUTIL_OFPT_HELLO:
4438 case OFPUTIL_OFPT_ERROR:
4439 case OFPUTIL_OFPT_FEATURES_REPLY:
4440 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4441 case OFPUTIL_OFPT_PACKET_IN:
4442 case OFPUTIL_OFPT_FLOW_REMOVED:
4443 case OFPUTIL_OFPT_PORT_STATUS:
4444 case OFPUTIL_OFPT_BARRIER_REPLY:
4445 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4446 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4447 case OFPUTIL_OFPST_DESC_REPLY:
4448 case OFPUTIL_OFPST_FLOW_REPLY:
4449 case OFPUTIL_OFPST_QUEUE_REPLY:
4450 case OFPUTIL_OFPST_PORT_REPLY:
4451 case OFPUTIL_OFPST_TABLE_REPLY:
4452 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4453 case OFPUTIL_NXT_ROLE_REPLY:
4454 case OFPUTIL_NXT_FLOW_REMOVED:
4455 case OFPUTIL_NXST_FLOW_REPLY:
4456 case OFPUTIL_NXST_AGGREGATE_REPLY:
4458 if (VLOG_IS_WARN_ENABLED()) {
4459 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4460 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4463 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4464 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4466 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4472 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4474 int error = handle_openflow__(ofconn, ofp_msg);
4476 send_error_oh(ofconn, ofp_msg->data, error);
4478 COVERAGE_INC(ofproto_recv_openflow);
4482 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4484 struct facet *facet;
4487 /* Obtain in_port and tun_id, at least. */
4488 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4490 /* Set header pointers in 'flow'. */
4491 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4493 if (cfm_should_process_flow(&flow)) {
4494 ofproto_process_cfm(p, &flow, upcall->packet);
4495 ofpbuf_delete(upcall->packet);
4497 } else if (p->ofhooks->special_cb
4498 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4499 ofpbuf_delete(upcall->packet);
4503 /* Check with in-band control to see if this packet should be sent
4504 * to the local port regardless of the flow table. */
4505 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4506 ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet);
4509 facet = facet_lookup_valid(p, &flow);
4511 struct rule *rule = rule_lookup(p, &flow);
4513 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4514 struct ofport *port = get_port(p, flow.in_port);
4516 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4517 COVERAGE_INC(ofproto_no_packet_in);
4518 /* XXX install 'drop' flow entry */
4519 ofpbuf_delete(upcall->packet);
4523 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4527 COVERAGE_INC(ofproto_packet_in);
4528 send_packet_in(p, upcall, &flow, false);
4532 facet = facet_create(p, rule, &flow, upcall->packet);
4533 } else if (!facet->may_install) {
4534 /* The facet is not installable, that is, we need to process every
4535 * packet, so process the current packet's actions into 'facet'. */
4536 facet_make_actions(p, facet, upcall->packet);
4539 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4541 * Extra-special case for fail-open mode.
4543 * We are in fail-open mode and the packet matched the fail-open rule,
4544 * but we are connected to a controller too. We should send the packet
4545 * up to the controller in the hope that it will try to set up a flow
4546 * and thereby allow us to exit fail-open.
4548 * See the top-level comment in fail-open.c for more information.
4550 send_packet_in(p, upcall, &flow, true);
4553 facet_execute(p, facet, upcall->packet);
4554 facet_install(p, facet, false);
4558 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4562 switch (upcall->type) {
4563 case DPIF_UC_ACTION:
4564 COVERAGE_INC(ofproto_ctlr_action);
4565 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4566 send_packet_in(p, upcall, &flow, false);
4569 case DPIF_UC_SAMPLE:
4571 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4572 ofproto_sflow_received(p->sflow, upcall, &flow);
4574 ofpbuf_delete(upcall->packet);
4578 handle_miss_upcall(p, upcall);
4581 case DPIF_N_UC_TYPES:
4583 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4588 /* Flow expiration. */
4590 static int ofproto_dp_max_idle(const struct ofproto *);
4591 static void ofproto_update_stats(struct ofproto *);
4592 static void rule_expire(struct ofproto *, struct rule *);
4593 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4595 /* This function is called periodically by ofproto_run(). Its job is to
4596 * collect updates for the flows that have been installed into the datapath,
4597 * most importantly when they last were used, and then use that information to
4598 * expire flows that have not been used recently.
4600 * Returns the number of milliseconds after which it should be called again. */
4602 ofproto_expire(struct ofproto *ofproto)
4604 struct rule *rule, *next_rule;
4605 struct cls_cursor cursor;
4608 /* Update stats for each flow in the datapath. */
4609 ofproto_update_stats(ofproto);
4611 /* Expire facets that have been idle too long. */
4612 dp_max_idle = ofproto_dp_max_idle(ofproto);
4613 ofproto_expire_facets(ofproto, dp_max_idle);
4615 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4616 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4617 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4618 rule_expire(ofproto, rule);
4621 /* Let the hook know that we're at a stable point: all outstanding data
4622 * in existing flows has been accounted to the account_cb. Thus, the
4623 * hook can now reasonably do operations that depend on having accurate
4624 * flow volume accounting (currently, that's just bond rebalancing). */
4625 if (ofproto->ofhooks->account_checkpoint_cb) {
4626 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4629 return MIN(dp_max_idle, 1000);
4632 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4634 * This function also pushes statistics updates to rules which each facet
4635 * resubmits into. Generally these statistics will be accurate. However, if a
4636 * facet changes the rule it resubmits into at some time in between
4637 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4638 * old rule will be incorrectly attributed to the new rule. This could be
4639 * avoided by calling ofproto_update_stats() whenever rules are created or
4640 * deleted. However, the performance impact of making so many calls to the
4641 * datapath do not justify the benefit of having perfectly accurate statistics.
4644 ofproto_update_stats(struct ofproto *p)
4646 const struct dpif_flow_stats *stats;
4647 struct dpif_flow_dump dump;
4648 const struct nlattr *key;
4651 dpif_flow_dump_start(&dump, p->dpif);
4652 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4653 struct facet *facet;
4656 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4660 odp_flow_key_format(key, key_len, &s);
4661 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4667 facet = facet_find(p, &flow);
4669 if (facet && facet->installed) {
4671 if (stats->n_packets >= facet->dp_packet_count) {
4672 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4674 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4677 if (stats->n_bytes >= facet->dp_byte_count) {
4678 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4680 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4683 facet->dp_packet_count = stats->n_packets;
4684 facet->dp_byte_count = stats->n_bytes;
4686 facet_update_time(p, facet, stats->used);
4687 facet_account(p, facet, stats->n_bytes);
4688 facet_push_stats(p, facet);
4690 /* There's a flow in the datapath that we know nothing about.
4692 COVERAGE_INC(ofproto_unexpected_rule);
4693 dpif_flow_del(p->dpif, key, key_len, NULL);
4696 dpif_flow_dump_done(&dump);
4699 /* Calculates and returns the number of milliseconds of idle time after which
4700 * facets should expire from the datapath and we should fold their statistics
4701 * into their parent rules in userspace. */
4703 ofproto_dp_max_idle(const struct ofproto *ofproto)
4706 * Idle time histogram.
4708 * Most of the time a switch has a relatively small number of facets. When
4709 * this is the case we might as well keep statistics for all of them in
4710 * userspace and to cache them in the kernel datapath for performance as
4713 * As the number of facets increases, the memory required to maintain
4714 * statistics about them in userspace and in the kernel becomes
4715 * significant. However, with a large number of facets it is likely that
4716 * only a few of them are "heavy hitters" that consume a large amount of
4717 * bandwidth. At this point, only heavy hitters are worth caching in the
4718 * kernel and maintaining in userspaces; other facets we can discard.
4720 * The technique used to compute the idle time is to build a histogram with
4721 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4722 * that is installed in the kernel gets dropped in the appropriate bucket.
4723 * After the histogram has been built, we compute the cutoff so that only
4724 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4725 * cached. At least the most-recently-used bucket of facets is kept, so
4726 * actually an arbitrary number of facets can be kept in any given
4727 * expiration run (though the next run will delete most of those unless
4728 * they receive additional data).
4730 * This requires a second pass through the facets, in addition to the pass
4731 * made by ofproto_update_stats(), because the former function never looks
4732 * at uninstallable facets.
4734 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4735 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4736 int buckets[N_BUCKETS] = { 0 };
4737 struct facet *facet;
4742 total = hmap_count(&ofproto->facets);
4743 if (total <= 1000) {
4744 return N_BUCKETS * BUCKET_WIDTH;
4747 /* Build histogram. */
4749 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4750 long long int idle = now - facet->used;
4751 int bucket = (idle <= 0 ? 0
4752 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4753 : (unsigned int) idle / BUCKET_WIDTH);
4757 /* Find the first bucket whose flows should be expired. */
4758 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4759 if (buckets[bucket]) {
4762 subtotal += buckets[bucket++];
4763 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4768 if (VLOG_IS_DBG_ENABLED()) {
4772 ds_put_cstr(&s, "keep");
4773 for (i = 0; i < N_BUCKETS; i++) {
4775 ds_put_cstr(&s, ", drop");
4778 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4781 VLOG_INFO("%s: %s (msec:count)",
4782 dpif_name(ofproto->dpif), ds_cstr(&s));
4786 return bucket * BUCKET_WIDTH;
4790 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4792 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4793 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4794 struct ofexpired expired;
4796 if (facet->installed) {
4797 struct dpif_flow_stats stats;
4799 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4801 facet_update_stats(ofproto, facet, &stats);
4804 expired.flow = facet->flow;
4805 expired.packet_count = facet->packet_count;
4806 expired.byte_count = facet->byte_count;
4807 expired.used = facet->used;
4808 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4813 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4815 long long int cutoff = time_msec() - dp_max_idle;
4816 struct facet *facet, *next_facet;
4818 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4819 facet_active_timeout(ofproto, facet);
4820 if (facet->used < cutoff) {
4821 facet_remove(ofproto, facet);
4826 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4827 * then delete it entirely. */
4829 rule_expire(struct ofproto *ofproto, struct rule *rule)
4831 struct facet *facet, *next_facet;
4835 /* Has 'rule' expired? */
4837 if (rule->hard_timeout
4838 && now > rule->created + rule->hard_timeout * 1000) {
4839 reason = OFPRR_HARD_TIMEOUT;
4840 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4841 && now >rule->used + rule->idle_timeout * 1000) {
4842 reason = OFPRR_IDLE_TIMEOUT;
4847 COVERAGE_INC(ofproto_expired);
4849 /* Update stats. (This is a no-op if the rule expired due to an idle
4850 * timeout, because that only happens when the rule has no facets left.) */
4851 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4852 facet_remove(ofproto, facet);
4855 /* Get rid of the rule. */
4856 if (!rule_is_hidden(rule)) {
4857 rule_send_removed(ofproto, rule, reason);
4859 rule_remove(ofproto, rule);
4862 static struct ofpbuf *
4863 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4866 struct ofp_flow_removed *ofr;
4869 ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
4870 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4871 rule->flow_cookie, &ofr->cookie);
4872 ofr->priority = htons(rule->cr.priority);
4873 ofr->reason = reason;
4874 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4875 ofr->idle_timeout = htons(rule->idle_timeout);
4876 ofr->packet_count = htonll(rule->packet_count);
4877 ofr->byte_count = htonll(rule->byte_count);
4882 static struct ofpbuf *
4883 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4885 struct nx_flow_removed *nfr;
4889 make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
4890 match_len = nx_put_match(buf, &rule->cr);
4893 nfr->cookie = rule->flow_cookie;
4894 nfr->priority = htons(rule->cr.priority);
4895 nfr->reason = reason;
4896 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4897 nfr->idle_timeout = htons(rule->idle_timeout);
4898 nfr->match_len = htons(match_len);
4899 nfr->packet_count = htonll(rule->packet_count);
4900 nfr->byte_count = htonll(rule->byte_count);
4906 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4908 struct ofconn *ofconn;
4910 if (!rule->send_flow_removed) {
4914 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4917 if (!rconn_is_connected(ofconn->rconn)
4918 || !ofconn_receives_async_msgs(ofconn)) {
4922 msg = (ofconn->flow_format == NXFF_NXM
4923 ? compose_nx_flow_removed(rule, reason)
4924 : compose_ofp_flow_removed(ofconn, rule, reason));
4926 /* Account flow expirations under ofconn->reply_counter, the counter
4927 * for replies to OpenFlow requests. That works because preventing
4928 * OpenFlow requests from being processed also prevents new flows from
4929 * being added (and expiring). (It also prevents processing OpenFlow
4930 * requests that would not add new flows, so it is imperfect.) */
4931 queue_tx(msg, ofconn, ofconn->reply_counter);
4935 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4936 * The returned statistics include statistics for all of 'rule''s facets. */
4938 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4941 struct facet *facet;
4943 /* Start from historical data for 'rule' itself that are no longer tracked
4944 * in facets. This counts, for example, facets that have expired. */
4945 p = rule->packet_count;
4946 b = rule->byte_count;
4948 /* Add any statistics that are tracked by facets. This includes
4949 * statistical data recently updated by ofproto_update_stats() as well as
4950 * stats for packets that were executed "by hand" via dpif_execute(). */
4951 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4952 p += facet->packet_count;
4953 b += facet->byte_count;
4960 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4962 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4964 struct ofconn *ofconn = ofconn_;
4966 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4967 ofconn->packet_in_counter, 100);
4970 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4971 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4972 * scheduler for sending.
4974 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4975 * Otherwise, ownership is transferred to this function. */
4977 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4978 const struct flow *flow, bool clone)
4980 enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) };
4981 struct ofproto *ofproto = ofconn->ofproto;
4982 struct ofp_packet_in *opi;
4983 int total_len, send_len;
4984 struct ofpbuf *packet;
4988 /* Get OpenFlow buffer_id. */
4989 if (upcall->type == DPIF_UC_ACTION) {
4990 buffer_id = UINT32_MAX;
4991 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4992 buffer_id = pktbuf_get_null();
4993 } else if (!ofconn->pktbuf) {
4994 buffer_id = UINT32_MAX;
4996 buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port);
4999 /* Figure out how much of the packet to send. */
5000 total_len = send_len = upcall->packet->size;
5001 if (buffer_id != UINT32_MAX) {
5002 send_len = MIN(send_len, ofconn->miss_send_len);
5004 if (upcall->type == DPIF_UC_ACTION) {
5005 send_len = MIN(send_len, upcall->userdata);
5008 /* Copy or steal buffer for OFPT_PACKET_IN. */
5010 packet = ofpbuf_clone_data_with_headroom(upcall->packet->data,
5011 send_len, OPI_SIZE);
5013 packet = upcall->packet;
5014 packet->size = send_len;
5017 /* Add OFPT_PACKET_IN. */
5018 opi = ofpbuf_push_zeros(packet, OPI_SIZE);
5019 opi->header.version = OFP_VERSION;
5020 opi->header.type = OFPT_PACKET_IN;
5021 opi->total_len = htons(total_len);
5022 opi->in_port = htons(odp_port_to_ofp_port(flow->in_port));
5023 opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
5024 opi->buffer_id = htonl(buffer_id);
5025 update_openflow_length(packet);
5027 /* Hand over to packet scheduler. It might immediately call into
5028 * do_send_packet_in() or it might buffer it for a while (until a later
5029 * call to pinsched_run()). */
5030 idx = upcall->type == DPIF_UC_MISS ? 0 : 1;
5031 pinsched_send(ofconn->schedulers[idx], flow->in_port,
5032 packet, do_send_packet_in, ofconn);
5035 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
5036 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
5037 * their individual configurations.
5039 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
5040 * Otherwise, ownership is transferred to this function. */
5042 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
5043 const struct flow *flow, bool clone)
5045 struct ofconn *ofconn, *prev;
5048 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5049 if (ofconn_receives_async_msgs(ofconn)) {
5051 schedule_packet_in(prev, upcall, flow, true);
5057 schedule_packet_in(prev, upcall, flow, clone);
5058 } else if (!clone) {
5059 ofpbuf_delete(upcall->packet);
5064 pick_datapath_id(const struct ofproto *ofproto)
5066 const struct ofport *port;
5068 port = get_port(ofproto, ODPP_LOCAL);
5070 uint8_t ea[ETH_ADDR_LEN];
5073 error = netdev_get_etheraddr(port->netdev, ea);
5075 return eth_addr_to_uint64(ea);
5077 VLOG_WARN("could not get MAC address for %s (%s)",
5078 netdev_get_name(port->netdev), strerror(error));
5080 return ofproto->fallback_dpid;
5084 pick_fallback_dpid(void)
5086 uint8_t ea[ETH_ADDR_LEN];
5087 eth_addr_nicira_random(ea);
5088 return eth_addr_to_uint64(ea);
5092 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
5093 void *aux OVS_UNUSED)
5095 const struct shash_node *node;
5099 SHASH_FOR_EACH (node, &all_ofprotos) {
5100 ds_put_format(&results, "%s\n", node->name);
5102 unixctl_command_reply(conn, 200, ds_cstr(&results));
5103 ds_destroy(&results);
5106 struct ofproto_trace {
5107 struct action_xlate_ctx ctx;
5113 trace_format_rule(struct ds *result, int level, const struct rule *rule)
5115 ds_put_char_multiple(result, '\t', level);
5117 ds_put_cstr(result, "No match\n");
5121 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
5122 ntohll(rule->flow_cookie));
5123 cls_rule_format(&rule->cr, result);
5124 ds_put_char(result, '\n');
5126 ds_put_char_multiple(result, '\t', level);
5127 ds_put_cstr(result, "OpenFlow ");
5128 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
5129 rule->n_actions * sizeof *rule->actions);
5130 ds_put_char(result, '\n');
5134 trace_format_flow(struct ds *result, int level, const char *title,
5135 struct ofproto_trace *trace)
5137 ds_put_char_multiple(result, '\t', level);
5138 ds_put_format(result, "%s: ", title);
5139 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5140 ds_put_cstr(result, "unchanged");
5142 flow_format(result, &trace->ctx.flow);
5143 trace->flow = trace->ctx.flow;
5145 ds_put_char(result, '\n');
5149 trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule)
5151 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5152 struct ds *result = trace->result;
5154 ds_put_char(result, '\n');
5155 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5156 trace_format_rule(result, ctx->recurse + 1, rule);
5160 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5161 void *aux OVS_UNUSED)
5163 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5164 char *args = xstrdup(args_);
5165 char *save_ptr = NULL;
5166 struct ofproto *ofproto;
5167 struct ofpbuf packet;
5175 ofpbuf_init(&packet, strlen(args) / 2);
5178 dpname = strtok_r(args, " ", &save_ptr);
5179 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5180 in_port_s = strtok_r(NULL, " ", &save_ptr);
5181 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5182 if (!dpname || !in_port_s || !packet_s) {
5183 unixctl_command_reply(conn, 501, "Bad command syntax");
5187 ofproto = shash_find_data(&all_ofprotos, dpname);
5189 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5194 tun_id = htonll(strtoull(tun_id_s, NULL, 10));
5195 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5197 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5198 packet_s += strspn(packet_s, " ");
5199 if (*packet_s != '\0') {
5200 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5203 if (packet.size < ETH_HEADER_LEN) {
5204 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5208 ds_put_cstr(&result, "Packet: ");
5209 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5210 ds_put_cstr(&result, s);
5213 flow_extract(&packet, tun_id, in_port, &flow);
5214 ds_put_cstr(&result, "Flow: ");
5215 flow_format(&result, &flow);
5216 ds_put_char(&result, '\n');
5218 rule = rule_lookup(ofproto, &flow);
5219 trace_format_rule(&result, 0, rule);
5221 struct ofproto_trace trace;
5222 struct ofpbuf *odp_actions;
5224 trace.result = &result;
5226 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5227 trace.ctx.resubmit_hook = trace_resubmit;
5228 odp_actions = xlate_actions(&trace.ctx,
5229 rule->actions, rule->n_actions);
5231 ds_put_char(&result, '\n');
5232 trace_format_flow(&result, 0, "Final flow", &trace);
5233 ds_put_cstr(&result, "Datapath actions: ");
5234 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5235 ofpbuf_delete(odp_actions);
5238 unixctl_command_reply(conn, 200, ds_cstr(&result));
5241 ds_destroy(&result);
5242 ofpbuf_uninit(&packet);
5247 ofproto_unixctl_init(void)
5249 static bool registered;
5255 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5256 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5260 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5261 struct ofpbuf *odp_actions, tag_type *tags,
5262 uint16_t *nf_output_iface, void *ofproto_)
5264 struct ofproto *ofproto = ofproto_;
5265 struct mac_entry *dst_mac;
5267 /* Drop frames for reserved multicast addresses. */
5268 if (eth_addr_is_reserved(flow->dl_dst)) {
5272 /* Learn source MAC (but don't try to learn from revalidation). */
5274 && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) {
5275 struct mac_entry *src_mac;
5277 src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0);
5278 if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) {
5279 /* The log messages here could actually be useful in debugging,
5280 * so keep the rate limit relatively high. */
5281 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5282 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5283 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5285 ofproto_revalidate(ofproto,
5286 mac_learning_changed(ofproto->ml, src_mac));
5287 src_mac->port.i = flow->in_port;
5291 /* Determine output port. */
5292 dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags);
5294 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5295 nf_output_iface, odp_actions);
5297 int out_port = dst_mac->port.i;
5298 if (out_port != flow->in_port) {
5299 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5300 *nf_output_iface = out_port;
5309 static const struct ofhooks default_ofhooks = {
5310 default_normal_ofhook_cb,