2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
28 #include "classifier.h"
30 #include "discovery.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "multipath.h"
44 #include "ofp-print.h"
46 #include "ofproto-sflow.h"
48 #include "openflow/nicira-ext.h"
49 #include "openflow/openflow.h"
50 #include "openvswitch/datapath-protocol.h"
54 #include "poll-loop.h"
58 #include "stream-ssl.h"
62 #include "unaligned.h"
67 VLOG_DEFINE_THIS_MODULE(ofproto);
69 COVERAGE_DEFINE(facet_changed_rule);
70 COVERAGE_DEFINE(facet_revalidate);
71 COVERAGE_DEFINE(odp_overflow);
72 COVERAGE_DEFINE(ofproto_agg_request);
73 COVERAGE_DEFINE(ofproto_costly_flags);
74 COVERAGE_DEFINE(ofproto_ctlr_action);
75 COVERAGE_DEFINE(ofproto_del_rule);
76 COVERAGE_DEFINE(ofproto_error);
77 COVERAGE_DEFINE(ofproto_expiration);
78 COVERAGE_DEFINE(ofproto_expired);
79 COVERAGE_DEFINE(ofproto_flows_req);
80 COVERAGE_DEFINE(ofproto_flush);
81 COVERAGE_DEFINE(ofproto_invalidated);
82 COVERAGE_DEFINE(ofproto_no_packet_in);
83 COVERAGE_DEFINE(ofproto_ofconn_stuck);
84 COVERAGE_DEFINE(ofproto_ofp2odp);
85 COVERAGE_DEFINE(ofproto_packet_in);
86 COVERAGE_DEFINE(ofproto_packet_out);
87 COVERAGE_DEFINE(ofproto_queue_req);
88 COVERAGE_DEFINE(ofproto_recv_openflow);
89 COVERAGE_DEFINE(ofproto_reinit_ports);
90 COVERAGE_DEFINE(ofproto_unexpected_rule);
91 COVERAGE_DEFINE(ofproto_uninstallable);
92 COVERAGE_DEFINE(ofproto_update_port);
94 #include "sflow_api.h"
99 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
100 struct netdev *netdev;
101 struct ofp_phy_port opp; /* In host byte order. */
105 static void ofport_free(struct ofport *);
106 static void hton_ofp_phy_port(struct ofp_phy_port *);
108 struct action_xlate_ctx {
109 /* action_xlate_ctx_init() initializes these members. */
112 struct ofproto *ofproto;
114 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
115 * this flow when actions change header fields. */
118 /* The packet corresponding to 'flow', or a null pointer if we are
119 * revalidating without a packet to refer to. */
120 const struct ofpbuf *packet;
122 /* If nonnull, called just before executing a resubmit action.
124 * This is normally null so the client has to set it manually after
125 * calling action_xlate_ctx_init(). */
126 void (*resubmit_hook)(struct action_xlate_ctx *, const struct rule *);
128 /* If true, the speciality of 'flow' should be checked before executing
129 * its actions. If special_cb returns false on 'flow' rendered
130 * uninstallable and no actions will be executed. */
133 /* xlate_actions() initializes and uses these members. The client might want
134 * to look at them after it returns. */
136 struct ofpbuf *odp_actions; /* Datapath actions. */
137 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
138 bool may_set_up_flow; /* True ordinarily; false if the actions must
139 * be reassessed for every packet. */
140 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
142 /* xlate_actions() initializes and uses these members, but the client has no
143 * reason to look at them. */
145 int recurse; /* Recursion level, via xlate_table_action. */
146 int last_pop_priority; /* Offset in 'odp_actions' just past most
147 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
150 static void action_xlate_ctx_init(struct action_xlate_ctx *,
151 struct ofproto *, const struct flow *,
152 const struct ofpbuf *);
153 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
154 const union ofp_action *in, size_t n_in);
156 /* An OpenFlow flow. */
158 long long int used; /* Time last used; time created if not used. */
159 long long int created; /* Creation time. */
163 * - Do include packets and bytes from facets that have been deleted or
164 * whose own statistics have been folded into the rule.
166 * - Do include packets and bytes sent "by hand" that were accounted to
167 * the rule without any facet being involved (this is a rare corner
168 * case in rule_execute()).
170 * - Do not include packet or bytes that can be obtained from any facet's
171 * packet_count or byte_count member or that can be obtained from the
172 * datapath by, e.g., dpif_flow_get() for any facet.
174 uint64_t packet_count; /* Number of packets received. */
175 uint64_t byte_count; /* Number of bytes received. */
177 ovs_be64 flow_cookie; /* Controller-issued identifier. */
179 struct cls_rule cr; /* In owning ofproto's classifier. */
180 uint16_t idle_timeout; /* In seconds from time of last use. */
181 uint16_t hard_timeout; /* In seconds from time of creation. */
182 bool send_flow_removed; /* Send a flow removed message? */
183 int n_actions; /* Number of elements in actions[]. */
184 union ofp_action *actions; /* OpenFlow actions. */
185 struct list facets; /* List of "struct facet"s. */
188 static struct rule *rule_from_cls_rule(const struct cls_rule *);
189 static bool rule_is_hidden(const struct rule *);
191 static struct rule *rule_create(const struct cls_rule *,
192 const union ofp_action *, size_t n_actions,
193 uint16_t idle_timeout, uint16_t hard_timeout,
194 ovs_be64 flow_cookie, bool send_flow_removed);
195 static void rule_destroy(struct ofproto *, struct rule *);
196 static void rule_free(struct rule *);
198 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
199 static void rule_insert(struct ofproto *, struct rule *);
200 static void rule_remove(struct ofproto *, struct rule *);
202 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
203 static void rule_get_stats(const struct rule *, uint64_t *packets,
206 /* An exact-match instantiation of an OpenFlow flow. */
208 long long int used; /* Time last used; time created if not used. */
212 * - Do include packets and bytes sent "by hand", e.g. with
215 * - Do include packets and bytes that were obtained from the datapath
216 * when a flow was deleted (e.g. dpif_flow_del()) or when its
217 * statistics were reset (e.g. dpif_flow_put() with
218 * DPIF_FP_ZERO_STATS).
220 * - Do not include any packets or bytes that can currently be obtained
221 * from the datapath by, e.g., dpif_flow_get().
223 uint64_t packet_count; /* Number of packets received. */
224 uint64_t byte_count; /* Number of bytes received. */
226 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
227 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
229 /* Number of bytes passed to account_cb. This may include bytes that can
230 * currently obtained from the datapath (thus, it can be greater than
232 uint64_t accounted_bytes;
234 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
235 struct list list_node; /* In owning rule's 'facets' list. */
236 struct rule *rule; /* Owning rule. */
237 struct flow flow; /* Exact-match flow. */
238 bool installed; /* Installed in datapath? */
239 bool may_install; /* True ordinarily; false if actions must
240 * be reassessed for every packet. */
241 size_t actions_len; /* Number of bytes in actions[]. */
242 struct nlattr *actions; /* Datapath actions. */
243 tag_type tags; /* Tags (set only by hooks). */
244 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
247 static struct facet *facet_create(struct ofproto *, struct rule *,
249 const struct ofpbuf *packet);
250 static void facet_remove(struct ofproto *, struct facet *);
251 static void facet_free(struct facet *);
253 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
254 static bool facet_revalidate(struct ofproto *, struct facet *);
256 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
257 static void facet_uninstall(struct ofproto *, struct facet *);
258 static void facet_flush_stats(struct ofproto *, struct facet *);
260 static void facet_make_actions(struct ofproto *, struct facet *,
261 const struct ofpbuf *packet);
262 static void facet_update_stats(struct ofproto *, struct facet *,
263 const struct dpif_flow_stats *);
265 /* ofproto supports two kinds of OpenFlow connections:
267 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
268 * maintains persistent connections to these controllers and by default
269 * sends them asynchronous messages such as packet-ins.
271 * - "Service" connections, e.g. from ovs-ofctl. When these connections
272 * drop, it is the other side's responsibility to reconnect them if
273 * necessary. ofproto does not send them asynchronous messages by default.
275 * Currently, active (tcp, ssl, unix) connections are always "primary"
276 * connections and passive (ptcp, pssl, punix) connections are always "service"
277 * connections. There is no inherent reason for this, but it reflects the
281 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
282 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
285 /* A listener for incoming OpenFlow "service" connections. */
287 struct hmap_node node; /* In struct ofproto's "services" hmap. */
288 struct pvconn *pvconn; /* OpenFlow connection listener. */
290 /* These are not used by ofservice directly. They are settings for
291 * accepted "struct ofconn"s from the pvconn. */
292 int probe_interval; /* Max idle time before probing, in seconds. */
293 int rate_limit; /* Max packet-in rate in packets per second. */
294 int burst_limit; /* Limit on accumulating packet credits. */
297 static struct ofservice *ofservice_lookup(struct ofproto *,
299 static int ofservice_create(struct ofproto *,
300 const struct ofproto_controller *);
301 static void ofservice_reconfigure(struct ofservice *,
302 const struct ofproto_controller *);
303 static void ofservice_destroy(struct ofproto *, struct ofservice *);
305 /* An OpenFlow connection. */
307 struct ofproto *ofproto; /* The ofproto that owns this connection. */
308 struct list node; /* In struct ofproto's "all_conns" list. */
309 struct rconn *rconn; /* OpenFlow connection. */
310 enum ofconn_type type; /* Type. */
311 enum nx_flow_format flow_format; /* Currently selected flow format. */
313 /* OFPT_PACKET_IN related data. */
314 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
315 #define N_SCHEDULERS 2
316 struct pinsched *schedulers[N_SCHEDULERS];
317 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
318 int miss_send_len; /* Bytes to send of buffered packets. */
320 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
321 * requests, and the maximum number before we stop reading OpenFlow
323 #define OFCONN_REPLY_MAX 100
324 struct rconn_packet_counter *reply_counter;
326 /* type == OFCONN_PRIMARY only. */
327 enum nx_role role; /* Role. */
328 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
329 struct discovery *discovery; /* Controller discovery object, if enabled. */
330 struct status_category *ss; /* Switch status category. */
331 enum ofproto_band band; /* In-band or out-of-band? */
335 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
337 static void ofconn_destroy(struct ofconn *);
338 static void ofconn_run(struct ofconn *);
339 static void ofconn_wait(struct ofconn *);
340 static bool ofconn_receives_async_msgs(const struct ofconn *);
341 static char *ofconn_make_name(const struct ofproto *, const char *target);
342 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
344 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
345 struct rconn_packet_counter *counter);
347 static void send_packet_in(struct ofproto *, struct dpif_upcall *,
348 const struct flow *, bool clone);
349 static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn);
353 uint64_t datapath_id; /* Datapath ID. */
354 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
355 char *mfr_desc; /* Manufacturer. */
356 char *hw_desc; /* Hardware. */
357 char *sw_desc; /* Software version. */
358 char *serial_desc; /* Serial number. */
359 char *dp_desc; /* Datapath description. */
363 struct netdev_monitor *netdev_monitor;
364 struct hmap ports; /* Contains "struct ofport"s. */
365 struct shash port_by_name;
369 struct switch_status *switch_status;
370 struct fail_open *fail_open;
371 struct netflow *netflow;
372 struct ofproto_sflow *sflow;
374 /* In-band control. */
375 struct in_band *in_band;
376 long long int next_in_band_update;
377 struct sockaddr_in *extra_in_band_remotes;
378 size_t n_extra_remotes;
382 struct classifier cls;
383 long long int next_expiration;
387 bool need_revalidate;
388 struct tag_set revalidate_set;
390 /* OpenFlow connections. */
391 struct hmap controllers; /* Controller "struct ofconn"s. */
392 struct list all_conns; /* Contains "struct ofconn"s. */
393 enum ofproto_fail_mode fail_mode;
395 /* OpenFlow listeners. */
396 struct hmap services; /* Contains "struct ofservice"s. */
397 struct pvconn **snoops;
400 /* Hooks for ovs-vswitchd. */
401 const struct ofhooks *ofhooks;
404 /* Used by default ofhooks. */
405 struct mac_learning *ml;
408 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
409 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
411 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
413 static const struct ofhooks default_ofhooks;
415 static uint64_t pick_datapath_id(const struct ofproto *);
416 static uint64_t pick_fallback_dpid(void);
418 static int ofproto_expire(struct ofproto *);
420 static void handle_upcall(struct ofproto *, struct dpif_upcall *);
422 static void handle_openflow(struct ofconn *, struct ofpbuf *);
424 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
425 static void update_port(struct ofproto *, const char *devname);
426 static int init_ports(struct ofproto *);
427 static void reinit_ports(struct ofproto *);
429 static void ofproto_unixctl_init(void);
432 ofproto_create(const char *datapath, const char *datapath_type,
433 const struct ofhooks *ofhooks, void *aux,
434 struct ofproto **ofprotop)
442 ofproto_unixctl_init();
444 /* Connect to datapath and start listening for messages. */
445 error = dpif_open(datapath, datapath_type, &dpif);
447 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
450 error = dpif_recv_set_mask(dpif,
451 ((1u << DPIF_UC_MISS) |
452 (1u << DPIF_UC_ACTION) |
453 (1u << DPIF_UC_SAMPLE)));
455 VLOG_ERR("failed to listen on datapath %s: %s",
456 datapath, strerror(error));
460 dpif_flow_flush(dpif);
461 dpif_recv_purge(dpif);
463 /* Initialize settings. */
464 p = xzalloc(sizeof *p);
465 p->fallback_dpid = pick_fallback_dpid();
466 p->datapath_id = p->fallback_dpid;
467 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
468 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
469 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
470 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
471 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
473 /* Initialize datapath. */
475 p->netdev_monitor = netdev_monitor_create();
476 hmap_init(&p->ports);
477 shash_init(&p->port_by_name);
478 p->max_ports = dpif_get_max_ports(dpif);
480 /* Initialize submodules. */
481 p->switch_status = switch_status_create(p);
486 /* Initialize in-band control. */
488 p->in_band_queue = -1;
490 /* Initialize flow table. */
491 classifier_init(&p->cls);
492 p->next_expiration = time_msec() + 1000;
494 /* Initialize facet table. */
495 hmap_init(&p->facets);
496 p->need_revalidate = false;
497 tag_set_init(&p->revalidate_set);
499 /* Initialize OpenFlow connections. */
500 list_init(&p->all_conns);
501 hmap_init(&p->controllers);
502 hmap_init(&p->services);
506 /* Initialize hooks. */
508 p->ofhooks = ofhooks;
512 p->ofhooks = &default_ofhooks;
514 p->ml = mac_learning_create();
517 /* Pick final datapath ID. */
518 p->datapath_id = pick_datapath_id(p);
519 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
521 shash_add_once(&all_ofprotos, dpif_name(p->dpif), p);
528 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
530 uint64_t old_dpid = p->datapath_id;
531 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
532 if (p->datapath_id != old_dpid) {
533 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
535 /* Force all active connections to reconnect, since there is no way to
536 * notify a controller that the datapath ID has changed. */
537 ofproto_reconnect_controllers(p);
542 is_discovery_controller(const struct ofproto_controller *c)
544 return !strcmp(c->target, "discover");
548 is_in_band_controller(const struct ofproto_controller *c)
550 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
553 /* Creates a new controller in 'ofproto'. Some of the settings are initially
554 * drawn from 'c', but update_controller() needs to be called later to finish
555 * the new ofconn's configuration. */
557 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
559 struct discovery *discovery;
560 struct ofconn *ofconn;
562 if (is_discovery_controller(c)) {
563 int error = discovery_create(c->accept_re, c->update_resolv_conf,
564 ofproto->dpif, ofproto->switch_status,
573 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
574 ofconn->pktbuf = pktbuf_create();
575 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
577 ofconn->discovery = discovery;
579 char *name = ofconn_make_name(ofproto, c->target);
580 rconn_connect(ofconn->rconn, c->target, name);
583 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
584 hash_string(c->target, 0));
587 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
588 * target or turn discovery on or off (these are done by creating new ofconns
589 * and deleting old ones), but it can update the rest of an ofconn's
592 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
596 ofconn->band = (is_in_band_controller(c)
597 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
599 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
601 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
602 rconn_set_probe_interval(ofconn->rconn, probe_interval);
604 if (ofconn->discovery) {
605 discovery_set_update_resolv_conf(ofconn->discovery,
606 c->update_resolv_conf);
607 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
610 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
614 ofconn_get_target(const struct ofconn *ofconn)
616 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
619 static struct ofconn *
620 find_controller_by_target(struct ofproto *ofproto, const char *target)
622 struct ofconn *ofconn;
624 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
625 hash_string(target, 0), &ofproto->controllers) {
626 if (!strcmp(ofconn_get_target(ofconn), target)) {
634 update_in_band_remotes(struct ofproto *ofproto)
636 const struct ofconn *ofconn;
637 struct sockaddr_in *addrs;
638 size_t max_addrs, n_addrs;
642 /* Allocate enough memory for as many remotes as we could possibly have. */
643 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
644 addrs = xmalloc(max_addrs * sizeof *addrs);
647 /* Add all the remotes. */
649 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
650 struct sockaddr_in *sin = &addrs[n_addrs];
652 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
656 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
657 if (sin->sin_addr.s_addr) {
658 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
661 if (ofconn->discovery) {
665 for (i = 0; i < ofproto->n_extra_remotes; i++) {
666 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
669 /* Create or update or destroy in-band.
671 * Ordinarily we only enable in-band if there's at least one remote
672 * address, but discovery needs the in-band rules for DHCP to be installed
673 * even before we know any remote addresses. */
674 if (n_addrs || discovery) {
675 if (!ofproto->in_band) {
676 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
679 if (ofproto->in_band) {
680 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
682 in_band_set_queue(ofproto->in_band, ofproto->in_band_queue);
683 ofproto->next_in_band_update = time_msec() + 1000;
685 in_band_destroy(ofproto->in_band);
686 ofproto->in_band = NULL;
694 update_fail_open(struct ofproto *p)
696 struct ofconn *ofconn;
698 if (!hmap_is_empty(&p->controllers)
699 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
700 struct rconn **rconns;
704 p->fail_open = fail_open_create(p, p->switch_status);
708 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
709 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
710 rconns[n++] = ofconn->rconn;
713 fail_open_set_controllers(p->fail_open, rconns, n);
714 /* p->fail_open takes ownership of 'rconns'. */
716 fail_open_destroy(p->fail_open);
722 ofproto_set_controllers(struct ofproto *p,
723 const struct ofproto_controller *controllers,
724 size_t n_controllers)
726 struct shash new_controllers;
727 struct ofconn *ofconn, *next_ofconn;
728 struct ofservice *ofservice, *next_ofservice;
732 /* Create newly configured controllers and services.
733 * Create a name to ofproto_controller mapping in 'new_controllers'. */
734 shash_init(&new_controllers);
735 for (i = 0; i < n_controllers; i++) {
736 const struct ofproto_controller *c = &controllers[i];
738 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
739 if (!find_controller_by_target(p, c->target)) {
740 add_controller(p, c);
742 } else if (!pvconn_verify_name(c->target)) {
743 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
747 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
748 dpif_name(p->dpif), c->target);
752 shash_add_once(&new_controllers, c->target, &controllers[i]);
755 /* Delete controllers that are no longer configured.
756 * Update configuration of all now-existing controllers. */
758 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
759 struct ofproto_controller *c;
761 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
763 ofconn_destroy(ofconn);
765 update_controller(ofconn, c);
772 /* Delete services that are no longer configured.
773 * Update configuration of all now-existing services. */
774 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
775 struct ofproto_controller *c;
777 c = shash_find_data(&new_controllers,
778 pvconn_get_name(ofservice->pvconn));
780 ofservice_destroy(p, ofservice);
782 ofservice_reconfigure(ofservice, c);
786 shash_destroy(&new_controllers);
788 update_in_band_remotes(p);
791 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
792 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
793 struct ofconn, hmap_node);
794 ofconn->ss = switch_status_register(p->switch_status, "remote",
795 rconn_status_cb, ofconn->rconn);
800 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
802 p->fail_mode = fail_mode;
806 /* Drops the connections between 'ofproto' and all of its controllers, forcing
807 * them to reconnect. */
809 ofproto_reconnect_controllers(struct ofproto *ofproto)
811 struct ofconn *ofconn;
813 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
814 rconn_reconnect(ofconn->rconn);
819 any_extras_changed(const struct ofproto *ofproto,
820 const struct sockaddr_in *extras, size_t n)
824 if (n != ofproto->n_extra_remotes) {
828 for (i = 0; i < n; i++) {
829 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
830 const struct sockaddr_in *new = &extras[i];
832 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
833 old->sin_port != new->sin_port) {
841 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
842 * in-band control should guarantee access, in the same way that in-band
843 * control guarantees access to OpenFlow controllers. */
845 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
846 const struct sockaddr_in *extras, size_t n)
848 if (!any_extras_changed(ofproto, extras, n)) {
852 free(ofproto->extra_in_band_remotes);
853 ofproto->n_extra_remotes = n;
854 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
856 update_in_band_remotes(ofproto);
859 /* Sets the OpenFlow queue used by flows set up by in-band control on
860 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
861 * flows will use the default queue. */
863 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
865 if (queue_id != ofproto->in_band_queue) {
866 ofproto->in_band_queue = queue_id;
867 update_in_band_remotes(ofproto);
872 ofproto_set_desc(struct ofproto *p,
873 const char *mfr_desc, const char *hw_desc,
874 const char *sw_desc, const char *serial_desc,
877 struct ofp_desc_stats *ods;
880 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
881 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
882 sizeof ods->mfr_desc);
885 p->mfr_desc = xstrdup(mfr_desc);
888 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
889 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
890 sizeof ods->hw_desc);
893 p->hw_desc = xstrdup(hw_desc);
896 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
897 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
898 sizeof ods->sw_desc);
901 p->sw_desc = xstrdup(sw_desc);
904 if (strlen(serial_desc) >= sizeof ods->serial_num) {
905 VLOG_WARN("truncating serial_desc, must be less than %zu "
907 sizeof ods->serial_num);
909 free(p->serial_desc);
910 p->serial_desc = xstrdup(serial_desc);
913 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
914 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
915 sizeof ods->dp_desc);
918 p->dp_desc = xstrdup(dp_desc);
923 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
924 const struct svec *svec)
926 struct pvconn **pvconns = *pvconnsp;
927 size_t n_pvconns = *n_pvconnsp;
931 for (i = 0; i < n_pvconns; i++) {
932 pvconn_close(pvconns[i]);
936 pvconns = xmalloc(svec->n * sizeof *pvconns);
938 for (i = 0; i < svec->n; i++) {
939 const char *name = svec->names[i];
940 struct pvconn *pvconn;
943 error = pvconn_open(name, &pvconn);
945 pvconns[n_pvconns++] = pvconn;
947 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
955 *n_pvconnsp = n_pvconns;
961 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
963 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
967 ofproto_set_netflow(struct ofproto *ofproto,
968 const struct netflow_options *nf_options)
970 if (nf_options && nf_options->collectors.n) {
971 if (!ofproto->netflow) {
972 ofproto->netflow = netflow_create();
974 return netflow_set_options(ofproto->netflow, nf_options);
976 netflow_destroy(ofproto->netflow);
977 ofproto->netflow = NULL;
983 ofproto_set_sflow(struct ofproto *ofproto,
984 const struct ofproto_sflow_options *oso)
986 struct ofproto_sflow *os = ofproto->sflow;
989 struct ofport *ofport;
991 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
992 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
993 ofproto_sflow_add_port(os, ofport->odp_port,
994 netdev_get_name(ofport->netdev));
997 ofproto_sflow_set_options(os, oso);
999 ofproto_sflow_destroy(os);
1000 ofproto->sflow = NULL;
1005 ofproto_get_datapath_id(const struct ofproto *ofproto)
1007 return ofproto->datapath_id;
1011 ofproto_has_primary_controller(const struct ofproto *ofproto)
1013 return !hmap_is_empty(&ofproto->controllers);
1016 enum ofproto_fail_mode
1017 ofproto_get_fail_mode(const struct ofproto *p)
1019 return p->fail_mode;
1023 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
1027 for (i = 0; i < ofproto->n_snoops; i++) {
1028 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
1033 ofproto_destroy(struct ofproto *p)
1035 struct ofservice *ofservice, *next_ofservice;
1036 struct ofconn *ofconn, *next_ofconn;
1037 struct ofport *ofport, *next_ofport;
1044 shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif));
1046 /* Destroy fail-open and in-band early, since they touch the classifier. */
1047 fail_open_destroy(p->fail_open);
1048 p->fail_open = NULL;
1050 in_band_destroy(p->in_band);
1052 free(p->extra_in_band_remotes);
1054 ofproto_flush_flows(p);
1055 classifier_destroy(&p->cls);
1056 hmap_destroy(&p->facets);
1058 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1059 ofconn_destroy(ofconn);
1061 hmap_destroy(&p->controllers);
1063 dpif_close(p->dpif);
1064 netdev_monitor_destroy(p->netdev_monitor);
1065 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1066 hmap_remove(&p->ports, &ofport->hmap_node);
1067 ofport_free(ofport);
1069 shash_destroy(&p->port_by_name);
1071 switch_status_destroy(p->switch_status);
1072 netflow_destroy(p->netflow);
1073 ofproto_sflow_destroy(p->sflow);
1075 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
1076 ofservice_destroy(p, ofservice);
1078 hmap_destroy(&p->services);
1080 for (i = 0; i < p->n_snoops; i++) {
1081 pvconn_close(p->snoops[i]);
1085 mac_learning_destroy(p->ml);
1090 free(p->serial_desc);
1093 hmap_destroy(&p->ports);
1099 ofproto_run(struct ofproto *p)
1101 int error = ofproto_run1(p);
1103 error = ofproto_run2(p, false);
1109 process_port_change(struct ofproto *ofproto, int error, char *devname)
1111 if (error == ENOBUFS) {
1112 reinit_ports(ofproto);
1113 } else if (!error) {
1114 update_port(ofproto, devname);
1119 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1120 * means that 'ofconn' is more interesting for monitoring than a lower return
1123 snoop_preference(const struct ofconn *ofconn)
1125 switch (ofconn->role) {
1126 case NX_ROLE_MASTER:
1133 /* Shouldn't happen. */
1138 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1139 * Connects this vconn to a controller. */
1141 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1143 struct ofconn *ofconn, *best;
1145 /* Pick a controller for monitoring. */
1147 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1148 if (ofconn->type == OFCONN_PRIMARY
1149 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1155 rconn_add_monitor(best->rconn, vconn);
1157 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1163 ofproto_run1(struct ofproto *p)
1165 struct ofconn *ofconn, *next_ofconn;
1166 struct ofservice *ofservice;
1171 if (shash_is_empty(&p->port_by_name)) {
1175 for (i = 0; i < 50; i++) {
1176 struct dpif_upcall packet;
1178 error = dpif_recv(p->dpif, &packet);
1180 if (error == ENODEV) {
1181 /* Someone destroyed the datapath behind our back. The caller
1182 * better destroy us and give up, because we're just going to
1183 * spin from here on out. */
1184 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1185 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1186 dpif_name(p->dpif));
1192 handle_upcall(p, &packet);
1195 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1196 process_port_change(p, error, devname);
1198 while ((error = netdev_monitor_poll(p->netdev_monitor,
1199 &devname)) != EAGAIN) {
1200 process_port_change(p, error, devname);
1204 if (time_msec() >= p->next_in_band_update) {
1205 update_in_band_remotes(p);
1207 in_band_run(p->in_band);
1210 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1214 /* Fail-open maintenance. Do this after processing the ofconns since
1215 * fail-open checks the status of the controller rconn. */
1217 fail_open_run(p->fail_open);
1220 HMAP_FOR_EACH (ofservice, node, &p->services) {
1221 struct vconn *vconn;
1224 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1226 struct rconn *rconn;
1229 rconn = rconn_create(ofservice->probe_interval, 0);
1230 name = ofconn_make_name(p, vconn_get_name(vconn));
1231 rconn_connect_unreliably(rconn, vconn, name);
1234 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1235 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1236 ofservice->burst_limit);
1237 } else if (retval != EAGAIN) {
1238 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1242 for (i = 0; i < p->n_snoops; i++) {
1243 struct vconn *vconn;
1246 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1248 add_snooper(p, vconn);
1249 } else if (retval != EAGAIN) {
1250 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1254 if (time_msec() >= p->next_expiration) {
1255 int delay = ofproto_expire(p);
1256 p->next_expiration = time_msec() + delay;
1257 COVERAGE_INC(ofproto_expiration);
1261 netflow_run(p->netflow);
1264 ofproto_sflow_run(p->sflow);
1271 ofproto_run2(struct ofproto *p, bool revalidate_all)
1273 /* Figure out what we need to revalidate now, if anything. */
1274 struct tag_set revalidate_set = p->revalidate_set;
1275 if (p->need_revalidate) {
1276 revalidate_all = true;
1279 /* Clear the revalidation flags. */
1280 tag_set_init(&p->revalidate_set);
1281 p->need_revalidate = false;
1283 /* Now revalidate if there's anything to do. */
1284 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1285 struct facet *facet, *next;
1287 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1289 || tag_set_intersects(&revalidate_set, facet->tags)) {
1290 facet_revalidate(p, facet);
1299 ofproto_wait(struct ofproto *p)
1301 struct ofservice *ofservice;
1302 struct ofconn *ofconn;
1305 dpif_recv_wait(p->dpif);
1306 dpif_port_poll_wait(p->dpif);
1307 netdev_monitor_poll_wait(p->netdev_monitor);
1308 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1309 ofconn_wait(ofconn);
1312 poll_timer_wait_until(p->next_in_band_update);
1313 in_band_wait(p->in_band);
1316 fail_open_wait(p->fail_open);
1319 ofproto_sflow_wait(p->sflow);
1321 if (!tag_set_is_empty(&p->revalidate_set)) {
1322 poll_immediate_wake();
1324 if (p->need_revalidate) {
1325 /* Shouldn't happen, but if it does just go around again. */
1326 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1327 poll_immediate_wake();
1328 } else if (p->next_expiration != LLONG_MAX) {
1329 poll_timer_wait_until(p->next_expiration);
1331 HMAP_FOR_EACH (ofservice, node, &p->services) {
1332 pvconn_wait(ofservice->pvconn);
1334 for (i = 0; i < p->n_snoops; i++) {
1335 pvconn_wait(p->snoops[i]);
1340 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1342 tag_set_add(&ofproto->revalidate_set, tag);
1346 ofproto_get_revalidate_set(struct ofproto *ofproto)
1348 return &ofproto->revalidate_set;
1352 ofproto_is_alive(const struct ofproto *p)
1354 return !hmap_is_empty(&p->controllers);
1358 ofproto_get_ofproto_controller_info(const struct ofproto * ofproto,
1361 const struct ofconn *ofconn;
1365 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
1366 const struct rconn *rconn = ofconn->rconn;
1367 const int last_error = rconn_get_last_error(rconn);
1368 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
1370 shash_add(info, rconn_get_target(rconn), cinfo);
1372 cinfo->is_connected = rconn_is_connected(rconn);
1373 cinfo->role = ofconn->role;
1378 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
1379 cinfo->pairs.values[cinfo->pairs.n++] =
1380 xstrdup(ovs_retval_to_string(last_error));
1383 cinfo->pairs.keys[cinfo->pairs.n] = "state";
1384 cinfo->pairs.values[cinfo->pairs.n++] =
1385 xstrdup(rconn_get_state(rconn));
1387 cinfo->pairs.keys[cinfo->pairs.n] = "time_in_state";
1388 cinfo->pairs.values[cinfo->pairs.n++] =
1389 xasprintf("%u", rconn_get_state_elapsed(rconn));
1394 ofproto_free_ofproto_controller_info(struct shash *info)
1396 struct shash_node *node;
1398 SHASH_FOR_EACH (node, info) {
1399 struct ofproto_controller_info *cinfo = node->data;
1400 while (cinfo->pairs.n) {
1401 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
1405 shash_destroy(info);
1408 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1410 * This is almost the same as calling dpif_port_del() directly on the
1411 * datapath, but it also makes 'ofproto' close its open netdev for the port
1412 * (if any). This makes it possible to create a new netdev of a different
1413 * type under the same name, which otherwise the netdev library would refuse
1414 * to do because of the conflict. (The netdev would eventually get closed on
1415 * the next trip through ofproto_run(), but this interface is more direct.)
1417 * Returns 0 if successful, otherwise a positive errno. */
1419 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1421 struct ofport *ofport = get_port(ofproto, odp_port);
1422 const char *name = ofport ? ofport->opp.name : "<unknown>";
1425 error = dpif_port_del(ofproto->dpif, odp_port);
1427 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1428 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1429 } else if (ofport) {
1430 /* 'name' is ofport->opp.name and update_port() is going to destroy
1431 * 'ofport'. Just in case update_port() refers to 'name' after it
1432 * destroys 'ofport', make a copy of it around the update_port()
1434 char *devname = xstrdup(name);
1435 update_port(ofproto, devname);
1441 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1442 * true if 'odp_port' exists and should be included, false otherwise. */
1444 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1446 struct ofport *ofport = get_port(ofproto, odp_port);
1447 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1451 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1452 const union ofp_action *actions, size_t n_actions,
1453 const struct ofpbuf *packet)
1455 struct action_xlate_ctx ctx;
1456 struct ofpbuf *odp_actions;
1458 action_xlate_ctx_init(&ctx, p, flow, packet);
1459 /* Always xlate packets originated in this function. */
1460 ctx.check_special = false;
1461 odp_actions = xlate_actions(&ctx, actions, n_actions);
1463 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1465 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet);
1467 ofpbuf_delete(odp_actions);
1472 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1473 * performs the 'n_actions' actions in 'actions'. The new flow will not
1476 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1477 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1478 * controllers; otherwise, it will be hidden.
1480 * The caller retains ownership of 'cls_rule' and 'actions'. */
1482 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1483 const union ofp_action *actions, size_t n_actions)
1486 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1487 rule_insert(p, rule);
1491 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1495 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1498 rule_remove(ofproto, rule);
1503 ofproto_flush_flows(struct ofproto *ofproto)
1505 struct facet *facet, *next_facet;
1506 struct rule *rule, *next_rule;
1507 struct cls_cursor cursor;
1509 COVERAGE_INC(ofproto_flush);
1511 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1512 /* Mark the facet as not installed so that facet_remove() doesn't
1513 * bother trying to uninstall it. There is no point in uninstalling it
1514 * individually since we are about to blow away all the facets with
1515 * dpif_flow_flush(). */
1516 facet->installed = false;
1517 facet_remove(ofproto, facet);
1520 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1521 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1522 rule_remove(ofproto, rule);
1525 dpif_flow_flush(ofproto->dpif);
1526 if (ofproto->in_band) {
1527 in_band_flushed(ofproto->in_band);
1529 if (ofproto->fail_open) {
1530 fail_open_flushed(ofproto->fail_open);
1535 reinit_ports(struct ofproto *p)
1537 struct dpif_port_dump dump;
1538 struct shash_node *node;
1539 struct shash devnames;
1540 struct ofport *ofport;
1541 struct dpif_port dpif_port;
1543 COVERAGE_INC(ofproto_reinit_ports);
1545 shash_init(&devnames);
1546 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1547 shash_add_once (&devnames, ofport->opp.name, NULL);
1549 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1550 shash_add_once (&devnames, dpif_port.name, NULL);
1553 SHASH_FOR_EACH (node, &devnames) {
1554 update_port(p, node->name);
1556 shash_destroy(&devnames);
1559 static struct ofport *
1560 make_ofport(const struct dpif_port *dpif_port)
1562 struct netdev_options netdev_options;
1563 enum netdev_flags flags;
1564 struct ofport *ofport;
1565 struct netdev *netdev;
1568 memset(&netdev_options, 0, sizeof netdev_options);
1569 netdev_options.name = dpif_port->name;
1570 netdev_options.type = dpif_port->type;
1571 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1573 error = netdev_open(&netdev_options, &netdev);
1575 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1576 "cannot be opened (%s)",
1577 dpif_port->name, dpif_port->port_no,
1578 dpif_port->name, strerror(error));
1582 ofport = xzalloc(sizeof *ofport);
1583 ofport->netdev = netdev;
1584 ofport->odp_port = dpif_port->port_no;
1585 ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no);
1586 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1587 ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name);
1589 netdev_get_flags(netdev, &flags);
1590 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1592 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1594 netdev_get_features(netdev,
1595 &ofport->opp.curr, &ofport->opp.advertised,
1596 &ofport->opp.supported, &ofport->opp.peer);
1601 ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port)
1603 if (get_port(p, dpif_port->port_no)) {
1604 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1605 dpif_port->port_no);
1607 } else if (shash_find(&p->port_by_name, dpif_port->name)) {
1608 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1617 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1619 const struct ofp_phy_port *a = &a_->opp;
1620 const struct ofp_phy_port *b = &b_->opp;
1622 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1623 return (a->port_no == b->port_no
1624 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1625 && !strcmp(a->name, b->name)
1626 && a->state == b->state
1627 && a->config == b->config
1628 && a->curr == b->curr
1629 && a->advertised == b->advertised
1630 && a->supported == b->supported
1631 && a->peer == b->peer);
1635 send_port_status(struct ofproto *p, const struct ofport *ofport,
1638 /* XXX Should limit the number of queued port status change messages. */
1639 struct ofconn *ofconn;
1640 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1641 struct ofp_port_status *ops;
1644 /* Primary controllers, even slaves, should always get port status
1645 updates. Otherwise obey ofconn_receives_async_msgs(). */
1646 if (ofconn->type != OFCONN_PRIMARY
1647 && !ofconn_receives_async_msgs(ofconn)) {
1651 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1652 ops->reason = reason;
1653 ops->desc = ofport->opp;
1654 hton_ofp_phy_port(&ops->desc);
1655 queue_tx(b, ofconn, NULL);
1660 ofport_install(struct ofproto *p, struct ofport *ofport)
1662 const char *netdev_name = ofport->opp.name;
1664 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1665 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1666 shash_add(&p->port_by_name, netdev_name, ofport);
1668 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1673 ofport_remove(struct ofproto *p, struct ofport *ofport)
1675 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1676 hmap_remove(&p->ports, &ofport->hmap_node);
1677 shash_delete(&p->port_by_name,
1678 shash_find(&p->port_by_name, ofport->opp.name));
1680 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1685 ofport_free(struct ofport *ofport)
1688 netdev_close(ofport->netdev);
1693 static struct ofport *
1694 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1696 struct ofport *port;
1698 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1699 hash_int(odp_port, 0), &ofproto->ports) {
1700 if (port->odp_port == odp_port) {
1708 update_port(struct ofproto *p, const char *devname)
1710 struct dpif_port dpif_port;
1711 struct ofport *old_ofport;
1712 struct ofport *new_ofport;
1715 COVERAGE_INC(ofproto_update_port);
1717 /* Query the datapath for port information. */
1718 error = dpif_port_query_by_name(p->dpif, devname, &dpif_port);
1720 /* Find the old ofport. */
1721 old_ofport = shash_find_data(&p->port_by_name, devname);
1724 /* There's no port named 'devname' but there might be a port with
1725 * the same port number. This could happen if a port is deleted
1726 * and then a new one added in its place very quickly, or if a port
1727 * is renamed. In the former case we want to send an OFPPR_DELETE
1728 * and an OFPPR_ADD, and in the latter case we want to send a
1729 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1730 * the old port's ifindex against the new port, or perhaps less
1731 * reliably but more portably by comparing the old port's MAC
1732 * against the new port's MAC. However, this code isn't that smart
1733 * and always sends an OFPPR_MODIFY (XXX). */
1734 old_ofport = get_port(p, dpif_port.port_no);
1736 } else if (error != ENOENT && error != ENODEV) {
1737 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1738 "%s", strerror(error));
1742 /* Create a new ofport. */
1743 new_ofport = !error ? make_ofport(&dpif_port) : NULL;
1745 /* Eliminate a few pathological cases. */
1746 if (!old_ofport && !new_ofport) {
1748 } else if (old_ofport && new_ofport) {
1749 /* Most of the 'config' bits are OpenFlow soft state, but
1750 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1751 * OpenFlow bits from old_ofport. (make_ofport() only sets
1752 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1753 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1755 if (ofport_equal(old_ofport, new_ofport)) {
1756 /* False alarm--no change. */
1757 ofport_free(new_ofport);
1762 /* Now deal with the normal cases. */
1764 ofport_remove(p, old_ofport);
1767 ofport_install(p, new_ofport);
1769 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1770 (!old_ofport ? OFPPR_ADD
1771 : !new_ofport ? OFPPR_DELETE
1773 ofport_free(old_ofport);
1776 dpif_port_destroy(&dpif_port);
1780 init_ports(struct ofproto *p)
1782 struct dpif_port_dump dump;
1783 struct dpif_port dpif_port;
1785 DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) {
1786 if (!ofport_conflicts(p, &dpif_port)) {
1787 struct ofport *ofport = make_ofport(&dpif_port);
1789 ofport_install(p, ofport);
1797 static struct ofconn *
1798 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1800 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1801 ofconn->ofproto = p;
1802 list_push_back(&p->all_conns, &ofconn->node);
1803 ofconn->rconn = rconn;
1804 ofconn->type = type;
1805 ofconn->flow_format = NXFF_OPENFLOW10;
1806 ofconn->role = NX_ROLE_OTHER;
1807 ofconn->packet_in_counter = rconn_packet_counter_create ();
1808 ofconn->pktbuf = NULL;
1809 ofconn->miss_send_len = 0;
1810 ofconn->reply_counter = rconn_packet_counter_create ();
1815 ofconn_destroy(struct ofconn *ofconn)
1817 if (ofconn->type == OFCONN_PRIMARY) {
1818 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1820 discovery_destroy(ofconn->discovery);
1822 list_remove(&ofconn->node);
1823 switch_status_unregister(ofconn->ss);
1824 rconn_destroy(ofconn->rconn);
1825 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1826 rconn_packet_counter_destroy(ofconn->reply_counter);
1827 pktbuf_destroy(ofconn->pktbuf);
1832 ofconn_run(struct ofconn *ofconn)
1834 struct ofproto *p = ofconn->ofproto;
1838 if (ofconn->discovery) {
1839 char *controller_name;
1840 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1841 discovery_question_connectivity(ofconn->discovery);
1843 if (discovery_run(ofconn->discovery, &controller_name)) {
1844 if (controller_name) {
1845 char *ofconn_name = ofconn_make_name(p, controller_name);
1846 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1849 rconn_disconnect(ofconn->rconn);
1854 for (i = 0; i < N_SCHEDULERS; i++) {
1855 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1858 rconn_run(ofconn->rconn);
1860 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1861 /* Limit the number of iterations to prevent other tasks from
1863 for (iteration = 0; iteration < 50; iteration++) {
1864 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1869 fail_open_maybe_recover(p->fail_open);
1871 handle_openflow(ofconn, of_msg);
1872 ofpbuf_delete(of_msg);
1876 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1877 ofconn_destroy(ofconn);
1882 ofconn_wait(struct ofconn *ofconn)
1886 if (ofconn->discovery) {
1887 discovery_wait(ofconn->discovery);
1889 for (i = 0; i < N_SCHEDULERS; i++) {
1890 pinsched_wait(ofconn->schedulers[i]);
1892 rconn_run_wait(ofconn->rconn);
1893 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1894 rconn_recv_wait(ofconn->rconn);
1896 COVERAGE_INC(ofproto_ofconn_stuck);
1900 /* Returns true if 'ofconn' should receive asynchronous messages. */
1902 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1904 if (ofconn->type == OFCONN_PRIMARY) {
1905 /* Primary controllers always get asynchronous messages unless they
1906 * have configured themselves as "slaves". */
1907 return ofconn->role != NX_ROLE_SLAVE;
1909 /* Service connections don't get asynchronous messages unless they have
1910 * explicitly asked for them by setting a nonzero miss send length. */
1911 return ofconn->miss_send_len > 0;
1915 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1916 * and 'target', suitable for use in log messages for identifying the
1919 * The name is dynamically allocated. The caller should free it (with free())
1920 * when it is no longer needed. */
1922 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1924 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1928 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1932 for (i = 0; i < N_SCHEDULERS; i++) {
1933 struct pinsched **s = &ofconn->schedulers[i];
1937 *s = pinsched_create(rate, burst,
1938 ofconn->ofproto->switch_status);
1940 pinsched_set_limits(*s, rate, burst);
1943 pinsched_destroy(*s);
1950 ofservice_reconfigure(struct ofservice *ofservice,
1951 const struct ofproto_controller *c)
1953 ofservice->probe_interval = c->probe_interval;
1954 ofservice->rate_limit = c->rate_limit;
1955 ofservice->burst_limit = c->burst_limit;
1958 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1959 * positive errno value. */
1961 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1963 struct ofservice *ofservice;
1964 struct pvconn *pvconn;
1967 error = pvconn_open(c->target, &pvconn);
1972 ofservice = xzalloc(sizeof *ofservice);
1973 hmap_insert(&ofproto->services, &ofservice->node,
1974 hash_string(c->target, 0));
1975 ofservice->pvconn = pvconn;
1977 ofservice_reconfigure(ofservice, c);
1983 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1985 hmap_remove(&ofproto->services, &ofservice->node);
1986 pvconn_close(ofservice->pvconn);
1990 /* Finds and returns the ofservice within 'ofproto' that has the given
1991 * 'target', or a null pointer if none exists. */
1992 static struct ofservice *
1993 ofservice_lookup(struct ofproto *ofproto, const char *target)
1995 struct ofservice *ofservice;
1997 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1998 &ofproto->services) {
1999 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2006 /* Returns true if 'rule' should be hidden from the controller.
2008 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2009 * (e.g. by in-band control) and are intentionally hidden from the
2012 rule_is_hidden(const struct rule *rule)
2014 return rule->cr.priority > UINT16_MAX;
2017 /* Creates and returns a new rule initialized as specified.
2019 * The caller is responsible for inserting the rule into the classifier (with
2020 * rule_insert()). */
2021 static struct rule *
2022 rule_create(const struct cls_rule *cls_rule,
2023 const union ofp_action *actions, size_t n_actions,
2024 uint16_t idle_timeout, uint16_t hard_timeout,
2025 ovs_be64 flow_cookie, bool send_flow_removed)
2027 struct rule *rule = xzalloc(sizeof *rule);
2028 rule->cr = *cls_rule;
2029 rule->idle_timeout = idle_timeout;
2030 rule->hard_timeout = hard_timeout;
2031 rule->flow_cookie = flow_cookie;
2032 rule->used = rule->created = time_msec();
2033 rule->send_flow_removed = send_flow_removed;
2034 list_init(&rule->facets);
2035 if (n_actions > 0) {
2036 rule->n_actions = n_actions;
2037 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2043 static struct rule *
2044 rule_from_cls_rule(const struct cls_rule *cls_rule)
2046 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2050 rule_free(struct rule *rule)
2052 free(rule->actions);
2056 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2057 * destroying any that no longer has a rule (which is probably all of them).
2059 * The caller must have already removed 'rule' from the classifier. */
2061 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2063 struct facet *facet, *next_facet;
2064 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2065 facet_revalidate(ofproto, facet);
2070 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2071 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2074 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2076 const union ofp_action *oa;
2077 struct actions_iterator i;
2079 if (out_port == htons(OFPP_NONE)) {
2082 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2083 oa = actions_next(&i)) {
2084 if (action_outputs_to_port(oa, out_port)) {
2091 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2092 * 'packet', which arrived on 'in_port'.
2094 * Takes ownership of 'packet'. */
2096 execute_odp_actions(struct ofproto *ofproto, const struct flow *flow,
2097 const struct nlattr *odp_actions, size_t actions_len,
2098 struct ofpbuf *packet)
2100 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2101 && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
2102 /* As an optimization, avoid a round-trip from userspace to kernel to
2103 * userspace. This also avoids possibly filling up kernel packet
2104 * buffers along the way. */
2105 struct dpif_upcall upcall;
2107 upcall.type = DPIF_UC_ACTION;
2108 upcall.packet = packet;
2111 upcall.userdata = nl_attr_get_u64(odp_actions);
2112 upcall.sample_pool = 0;
2113 upcall.actions = NULL;
2114 upcall.actions_len = 0;
2116 send_packet_in(ofproto, &upcall, flow, false);
2122 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2123 ofpbuf_delete(packet);
2128 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2129 * statistics appropriately. 'packet' must have at least sizeof(struct
2130 * ofp_packet_in) bytes of headroom.
2132 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2133 * applying flow_extract() to 'packet' would yield the same flow as
2136 * 'facet' must have accurately composed ODP actions; that is, it must not be
2137 * in need of revalidation.
2139 * Takes ownership of 'packet'. */
2141 facet_execute(struct ofproto *ofproto, struct facet *facet,
2142 struct ofpbuf *packet)
2144 struct dpif_flow_stats stats;
2146 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2148 flow_extract_stats(&facet->flow, packet, &stats);
2149 if (execute_odp_actions(ofproto, &facet->flow,
2150 facet->actions, facet->actions_len, packet)) {
2151 facet_update_stats(ofproto, facet, &stats);
2152 facet->used = time_msec();
2153 netflow_flow_update_time(ofproto->netflow,
2154 &facet->nf_flow, facet->used);
2158 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2159 * statistics (or the statistics for one of its facets) appropriately.
2160 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2162 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2163 * with statistics for 'packet' either way.
2165 * Takes ownership of 'packet'. */
2167 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2168 struct ofpbuf *packet)
2170 struct action_xlate_ctx ctx;
2171 struct ofpbuf *odp_actions;
2172 struct facet *facet;
2176 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2178 flow_extract(packet, 0, in_port, &flow);
2180 /* First look for a related facet. If we find one, account it to that. */
2181 facet = facet_lookup_valid(ofproto, &flow);
2182 if (facet && facet->rule == rule) {
2183 facet_execute(ofproto, facet, packet);
2187 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2188 * create a new facet for it and use that. */
2189 if (rule_lookup(ofproto, &flow) == rule) {
2190 facet = facet_create(ofproto, rule, &flow, packet);
2191 facet_execute(ofproto, facet, packet);
2192 facet_install(ofproto, facet, true);
2196 /* We can't account anything to a facet. If we were to try, then that
2197 * facet would have a non-matching rule, busting our invariants. */
2198 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2199 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2200 size = packet->size;
2201 if (execute_odp_actions(ofproto, &flow, odp_actions->data,
2202 odp_actions->size, packet)) {
2203 rule->used = time_msec();
2204 rule->packet_count++;
2205 rule->byte_count += size;
2207 ofpbuf_delete(odp_actions);
2210 /* Inserts 'rule' into 'p''s flow table. */
2212 rule_insert(struct ofproto *p, struct rule *rule)
2214 struct rule *displaced_rule;
2216 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2217 if (displaced_rule) {
2218 rule_destroy(p, displaced_rule);
2220 p->need_revalidate = true;
2223 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2224 * 'flow' and an example 'packet' within that flow.
2226 * The caller must already have determined that no facet with an identical
2227 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2228 * 'ofproto''s classifier table. */
2229 static struct facet *
2230 facet_create(struct ofproto *ofproto, struct rule *rule,
2231 const struct flow *flow, const struct ofpbuf *packet)
2233 struct facet *facet;
2235 facet = xzalloc(sizeof *facet);
2236 facet->used = time_msec();
2237 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2238 list_push_back(&rule->facets, &facet->list_node);
2240 facet->flow = *flow;
2241 netflow_flow_init(&facet->nf_flow);
2242 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2244 facet_make_actions(ofproto, facet, packet);
2250 facet_free(struct facet *facet)
2252 free(facet->actions);
2256 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2258 * - Removes 'rule' from the classifier.
2260 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2261 * destroys them), via rule_destroy().
2264 rule_remove(struct ofproto *ofproto, struct rule *rule)
2266 COVERAGE_INC(ofproto_del_rule);
2267 ofproto->need_revalidate = true;
2268 classifier_remove(&ofproto->cls, &rule->cr);
2269 rule_destroy(ofproto, rule);
2272 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2274 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2275 * rule's statistics, via facet_uninstall().
2277 * - Removes 'facet' from its rule and from ofproto->facets.
2280 facet_remove(struct ofproto *ofproto, struct facet *facet)
2282 facet_uninstall(ofproto, facet);
2283 facet_flush_stats(ofproto, facet);
2284 hmap_remove(&ofproto->facets, &facet->hmap_node);
2285 list_remove(&facet->list_node);
2289 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2291 facet_make_actions(struct ofproto *p, struct facet *facet,
2292 const struct ofpbuf *packet)
2294 const struct rule *rule = facet->rule;
2295 struct ofpbuf *odp_actions;
2296 struct action_xlate_ctx ctx;
2298 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2299 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2300 facet->tags = ctx.tags;
2301 facet->may_install = ctx.may_set_up_flow;
2302 facet->nf_flow.output_iface = ctx.nf_output_iface;
2304 if (facet->actions_len != odp_actions->size
2305 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2306 free(facet->actions);
2307 facet->actions_len = odp_actions->size;
2308 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2311 ofpbuf_delete(odp_actions);
2315 facet_put__(struct ofproto *ofproto, struct facet *facet,
2316 const struct nlattr *actions, size_t actions_len,
2317 struct dpif_flow_stats *stats)
2319 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2320 enum dpif_flow_put_flags flags;
2323 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
2325 flags |= DPIF_FP_ZERO_STATS;
2326 facet->dp_packet_count = 0;
2327 facet->dp_byte_count = 0;
2330 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2331 odp_flow_key_from_flow(&key, &facet->flow);
2332 assert(key.base == keybuf);
2334 return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
2335 actions, actions_len, stats);
2338 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2339 * 'zero_stats' is true, clears any existing statistics from the datapath for
2342 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2344 struct dpif_flow_stats stats;
2346 if (facet->may_install
2347 && !facet_put__(p, facet, facet->actions, facet->actions_len,
2348 zero_stats ? &stats : NULL)) {
2349 facet->installed = true;
2353 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2354 * to the accounting hook function in the ofhooks structure. */
2356 facet_account(struct ofproto *ofproto,
2357 struct facet *facet, uint64_t extra_bytes)
2359 uint64_t total_bytes = facet->byte_count + extra_bytes;
2361 if (ofproto->ofhooks->account_flow_cb
2362 && total_bytes > facet->accounted_bytes)
2364 ofproto->ofhooks->account_flow_cb(
2365 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2366 total_bytes - facet->accounted_bytes, ofproto->aux);
2367 facet->accounted_bytes = total_bytes;
2371 /* If 'rule' is installed in the datapath, uninstalls it. */
2373 facet_uninstall(struct ofproto *p, struct facet *facet)
2375 if (facet->installed) {
2376 uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
2377 struct dpif_flow_stats stats;
2380 ofpbuf_use_stack(&key, keybuf, sizeof keybuf);
2381 odp_flow_key_from_flow(&key, &facet->flow);
2382 assert(key.base == keybuf);
2384 if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
2385 facet_update_stats(p, facet, &stats);
2387 facet->installed = false;
2388 facet->dp_packet_count = 0;
2389 facet->dp_byte_count = 0;
2393 /* Returns true if the only action for 'facet' is to send to the controller.
2394 * (We don't report NetFlow expiration messages for such facets because they
2395 * are just part of the control logic for the network, not real traffic). */
2397 facet_is_controller_flow(struct facet *facet)
2400 && facet->rule->n_actions == 1
2401 && action_outputs_to_port(&facet->rule->actions[0],
2402 htons(OFPP_CONTROLLER)));
2405 /* Folds all of 'facet''s statistics into its rule. Also updates the
2406 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2407 * 'facet''s statistics in the datapath should have been zeroed and folded into
2408 * its packet and byte counts before this function is called. */
2410 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2412 assert(!facet->dp_byte_count);
2413 assert(!facet->dp_packet_count);
2415 facet_account(ofproto, facet, 0);
2417 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2418 struct ofexpired expired;
2419 expired.flow = facet->flow;
2420 expired.packet_count = facet->packet_count;
2421 expired.byte_count = facet->byte_count;
2422 expired.used = facet->used;
2423 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2426 facet->rule->packet_count += facet->packet_count;
2427 facet->rule->byte_count += facet->byte_count;
2429 /* Reset counters to prevent double counting if 'facet' ever gets
2431 facet->packet_count = 0;
2432 facet->byte_count = 0;
2433 facet->accounted_bytes = 0;
2435 netflow_flow_clear(&facet->nf_flow);
2438 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2439 * Returns it if found, otherwise a null pointer.
2441 * The returned facet might need revalidation; use facet_lookup_valid()
2442 * instead if that is important. */
2443 static struct facet *
2444 facet_find(struct ofproto *ofproto, const struct flow *flow)
2446 struct facet *facet;
2448 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2450 if (flow_equal(flow, &facet->flow)) {
2458 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2459 * Returns it if found, otherwise a null pointer.
2461 * The returned facet is guaranteed to be valid. */
2462 static struct facet *
2463 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2465 struct facet *facet = facet_find(ofproto, flow);
2467 /* The facet we found might not be valid, since we could be in need of
2468 * revalidation. If it is not valid, don't return it. */
2470 && ofproto->need_revalidate
2471 && !facet_revalidate(ofproto, facet)) {
2472 COVERAGE_INC(ofproto_invalidated);
2479 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2481 * - If the rule found is different from 'facet''s current rule, moves
2482 * 'facet' to the new rule and recompiles its actions.
2484 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2485 * where it is and recompiles its actions anyway.
2487 * - If there is none, destroys 'facet'.
2489 * Returns true if 'facet' still exists, false if it has been destroyed. */
2491 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2493 struct action_xlate_ctx ctx;
2494 struct ofpbuf *odp_actions;
2495 struct rule *new_rule;
2496 bool actions_changed;
2498 COVERAGE_INC(facet_revalidate);
2500 /* Determine the new rule. */
2501 new_rule = rule_lookup(ofproto, &facet->flow);
2503 /* No new rule, so delete the facet. */
2504 facet_remove(ofproto, facet);
2508 /* Calculate new ODP actions.
2510 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2511 * emit a NetFlow expiration and, if so, we need to have the old state
2512 * around to properly compose it. */
2513 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2514 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2515 actions_changed = (facet->actions_len != odp_actions->size
2516 || memcmp(facet->actions, odp_actions->data,
2517 facet->actions_len));
2519 /* If the ODP actions changed or the installability changed, then we need
2520 * to talk to the datapath. */
2521 if (actions_changed || ctx.may_set_up_flow != facet->installed) {
2522 if (ctx.may_set_up_flow) {
2523 struct dpif_flow_stats stats;
2525 facet_put__(ofproto, facet,
2526 odp_actions->data, odp_actions->size, &stats);
2527 facet_update_stats(ofproto, facet, &stats);
2529 facet_uninstall(ofproto, facet);
2532 /* The datapath flow is gone or has zeroed stats, so push stats out of
2533 * 'facet' into 'rule'. */
2534 facet_flush_stats(ofproto, facet);
2537 /* Update 'facet' now that we've taken care of all the old state. */
2538 facet->tags = ctx.tags;
2539 facet->nf_flow.output_iface = ctx.nf_output_iface;
2540 facet->may_install = ctx.may_set_up_flow;
2541 if (actions_changed) {
2542 free(facet->actions);
2543 facet->actions_len = odp_actions->size;
2544 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2546 if (facet->rule != new_rule) {
2547 COVERAGE_INC(facet_changed_rule);
2548 list_remove(&facet->list_node);
2549 list_push_back(&new_rule->facets, &facet->list_node);
2550 facet->rule = new_rule;
2551 facet->used = new_rule->created;
2554 ofpbuf_delete(odp_actions);
2560 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2561 struct rconn_packet_counter *counter)
2563 update_openflow_length(msg);
2564 if (rconn_send(ofconn->rconn, msg, counter)) {
2570 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2573 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
2575 COVERAGE_INC(ofproto_error);
2576 queue_tx(buf, ofconn, ofconn->reply_counter);
2581 hton_ofp_phy_port(struct ofp_phy_port *opp)
2583 opp->port_no = htons(opp->port_no);
2584 opp->config = htonl(opp->config);
2585 opp->state = htonl(opp->state);
2586 opp->curr = htonl(opp->curr);
2587 opp->advertised = htonl(opp->advertised);
2588 opp->supported = htonl(opp->supported);
2589 opp->peer = htonl(opp->peer);
2593 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2595 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2600 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2602 struct ofp_switch_features *osf;
2604 struct ofport *port;
2606 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2607 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2608 osf->n_buffers = htonl(pktbuf_capacity());
2610 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2611 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2612 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2613 (1u << OFPAT_SET_VLAN_VID) |
2614 (1u << OFPAT_SET_VLAN_PCP) |
2615 (1u << OFPAT_STRIP_VLAN) |
2616 (1u << OFPAT_SET_DL_SRC) |
2617 (1u << OFPAT_SET_DL_DST) |
2618 (1u << OFPAT_SET_NW_SRC) |
2619 (1u << OFPAT_SET_NW_DST) |
2620 (1u << OFPAT_SET_NW_TOS) |
2621 (1u << OFPAT_SET_TP_SRC) |
2622 (1u << OFPAT_SET_TP_DST) |
2623 (1u << OFPAT_ENQUEUE));
2625 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2626 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2629 queue_tx(buf, ofconn, ofconn->reply_counter);
2634 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2637 struct ofp_switch_config *osc;
2641 /* Figure out flags. */
2642 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2643 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2646 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2647 osc->flags = htons(flags);
2648 osc->miss_send_len = htons(ofconn->miss_send_len);
2649 queue_tx(buf, ofconn, ofconn->reply_counter);
2655 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2657 uint16_t flags = ntohs(osc->flags);
2659 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2660 switch (flags & OFPC_FRAG_MASK) {
2661 case OFPC_FRAG_NORMAL:
2662 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2664 case OFPC_FRAG_DROP:
2665 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2668 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2674 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2679 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2680 * flow translation. */
2681 #define MAX_RESUBMIT_RECURSION 16
2683 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2684 struct action_xlate_ctx *ctx);
2687 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2689 const struct ofport *ofport = get_port(ctx->ofproto, port);
2692 if (ofport->opp.config & OFPPC_NO_FWD) {
2693 /* Forwarding disabled on port. */
2698 * We don't have an ofport record for this port, but it doesn't hurt to
2699 * allow forwarding to it anyhow. Maybe such a port will appear later
2700 * and we're pre-populating the flow table.
2704 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port);
2705 ctx->nf_output_iface = port;
2708 static struct rule *
2709 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2711 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2715 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2717 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2718 uint16_t old_in_port;
2721 /* Look up a flow with 'in_port' as the input port. Then restore the
2722 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2723 * have surprising behavior). */
2724 old_in_port = ctx->flow.in_port;
2725 ctx->flow.in_port = in_port;
2726 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2727 ctx->flow.in_port = old_in_port;
2729 if (ctx->resubmit_hook) {
2730 ctx->resubmit_hook(ctx, rule);
2735 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2739 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2741 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2742 MAX_RESUBMIT_RECURSION);
2747 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2748 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2750 struct ofport *ofport;
2752 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2753 uint16_t odp_port = ofport->odp_port;
2754 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2755 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
2758 *nf_output_iface = NF_OUT_FLOOD;
2762 xlate_output_action__(struct action_xlate_ctx *ctx,
2763 uint16_t port, uint16_t max_len)
2766 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2768 ctx->nf_output_iface = NF_OUT_DROP;
2772 add_output_action(ctx, ctx->flow.in_port);
2775 xlate_table_action(ctx, ctx->flow.in_port);
2778 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2779 ctx->odp_actions, &ctx->tags,
2780 &ctx->nf_output_iface,
2781 ctx->ofproto->aux)) {
2782 COVERAGE_INC(ofproto_uninstallable);
2783 ctx->may_set_up_flow = false;
2787 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2788 &ctx->nf_output_iface, ctx->odp_actions);
2791 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2792 &ctx->nf_output_iface, ctx->odp_actions);
2794 case OFPP_CONTROLLER:
2795 nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
2798 add_output_action(ctx, ODPP_LOCAL);
2801 odp_port = ofp_port_to_odp_port(port);
2802 if (odp_port != ctx->flow.in_port) {
2803 add_output_action(ctx, odp_port);
2808 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2809 ctx->nf_output_iface = NF_OUT_FLOOD;
2810 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2811 ctx->nf_output_iface = prev_nf_output_iface;
2812 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2813 ctx->nf_output_iface != NF_OUT_FLOOD) {
2814 ctx->nf_output_iface = NF_OUT_MULTI;
2819 xlate_output_action(struct action_xlate_ctx *ctx,
2820 const struct ofp_action_output *oao)
2822 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2825 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2826 * optimization, because we're going to add another action that sets the
2827 * priority immediately after, or because there are no actions following the
2830 remove_pop_action(struct action_xlate_ctx *ctx)
2832 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2833 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2834 ctx->last_pop_priority = -1;
2839 add_pop_action(struct action_xlate_ctx *ctx)
2841 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2842 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
2843 ctx->last_pop_priority = ctx->odp_actions->size;
2848 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2849 const struct ofp_action_enqueue *oae)
2851 uint16_t ofp_port, odp_port;
2855 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2858 /* Fall back to ordinary output action. */
2859 xlate_output_action__(ctx, ntohs(oae->port), 0);
2863 /* Figure out ODP output port. */
2864 ofp_port = ntohs(oae->port);
2865 if (ofp_port != OFPP_IN_PORT) {
2866 odp_port = ofp_port_to_odp_port(ofp_port);
2868 odp_port = ctx->flow.in_port;
2871 /* Add ODP actions. */
2872 remove_pop_action(ctx);
2873 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2874 add_output_action(ctx, odp_port);
2875 add_pop_action(ctx);
2877 /* Update NetFlow output port. */
2878 if (ctx->nf_output_iface == NF_OUT_DROP) {
2879 ctx->nf_output_iface = odp_port;
2880 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2881 ctx->nf_output_iface = NF_OUT_MULTI;
2886 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2887 const struct nx_action_set_queue *nasq)
2892 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2895 /* Couldn't translate queue to a priority, so ignore. A warning
2896 * has already been logged. */
2900 remove_pop_action(ctx);
2901 nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
2905 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2907 ovs_be16 tci = ctx->flow.vlan_tci;
2908 if (!(tci & htons(VLAN_CFI))) {
2909 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
2911 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
2912 tci & ~htons(VLAN_CFI));
2916 struct xlate_reg_state {
2922 save_reg_state(const struct action_xlate_ctx *ctx,
2923 struct xlate_reg_state *state)
2925 state->vlan_tci = ctx->flow.vlan_tci;
2926 state->tun_id = ctx->flow.tun_id;
2930 update_reg_state(struct action_xlate_ctx *ctx,
2931 const struct xlate_reg_state *state)
2933 if (ctx->flow.vlan_tci != state->vlan_tci) {
2934 xlate_set_dl_tci(ctx);
2936 if (ctx->flow.tun_id != state->tun_id) {
2937 nl_msg_put_be64(ctx->odp_actions,
2938 ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
2943 xlate_nicira_action(struct action_xlate_ctx *ctx,
2944 const struct nx_action_header *nah)
2946 const struct nx_action_resubmit *nar;
2947 const struct nx_action_set_tunnel *nast;
2948 const struct nx_action_set_queue *nasq;
2949 const struct nx_action_multipath *nam;
2950 enum nx_action_subtype subtype = ntohs(nah->subtype);
2951 struct xlate_reg_state state;
2954 assert(nah->vendor == htonl(NX_VENDOR_ID));
2956 case NXAST_RESUBMIT:
2957 nar = (const struct nx_action_resubmit *) nah;
2958 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2961 case NXAST_SET_TUNNEL:
2962 nast = (const struct nx_action_set_tunnel *) nah;
2963 tun_id = htonll(ntohl(nast->tun_id));
2964 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
2965 ctx->flow.tun_id = tun_id;
2968 case NXAST_DROP_SPOOFED_ARP:
2969 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2970 nl_msg_put_flag(ctx->odp_actions,
2971 ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
2975 case NXAST_SET_QUEUE:
2976 nasq = (const struct nx_action_set_queue *) nah;
2977 xlate_set_queue_action(ctx, nasq);
2980 case NXAST_POP_QUEUE:
2981 add_pop_action(ctx);
2984 case NXAST_REG_MOVE:
2985 save_reg_state(ctx, &state);
2986 nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
2988 update_reg_state(ctx, &state);
2991 case NXAST_REG_LOAD:
2992 save_reg_state(ctx, &state);
2993 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2995 update_reg_state(ctx, &state);
2999 /* Nothing to do. */
3002 case NXAST_SET_TUNNEL64:
3003 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
3004 nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
3005 ctx->flow.tun_id = tun_id;
3008 case NXAST_MULTIPATH:
3009 nam = (const struct nx_action_multipath *) nah;
3010 multipath_execute(nam, &ctx->flow);
3013 /* If you add a new action here that modifies flow data, don't forget to
3014 * update the flow key in ctx->flow at the same time. */
3016 case NXAST_SNAT__OBSOLETE:
3018 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
3024 do_xlate_actions(const union ofp_action *in, size_t n_in,
3025 struct action_xlate_ctx *ctx)
3027 struct actions_iterator iter;
3028 const union ofp_action *ia;
3029 const struct ofport *port;
3031 port = get_port(ctx->ofproto, ctx->flow.in_port);
3032 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
3033 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
3034 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
3035 /* Drop this flow. */
3039 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
3040 enum ofp_action_type type = ntohs(ia->type);
3041 const struct ofp_action_dl_addr *oada;
3045 xlate_output_action(ctx, &ia->output);
3048 case OFPAT_SET_VLAN_VID:
3049 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
3050 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
3051 xlate_set_dl_tci(ctx);
3054 case OFPAT_SET_VLAN_PCP:
3055 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
3056 ctx->flow.vlan_tci |= htons(
3057 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
3058 xlate_set_dl_tci(ctx);
3061 case OFPAT_STRIP_VLAN:
3062 ctx->flow.vlan_tci = htons(0);
3063 xlate_set_dl_tci(ctx);
3066 case OFPAT_SET_DL_SRC:
3067 oada = ((struct ofp_action_dl_addr *) ia);
3068 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
3069 oada->dl_addr, ETH_ADDR_LEN);
3070 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
3073 case OFPAT_SET_DL_DST:
3074 oada = ((struct ofp_action_dl_addr *) ia);
3075 nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
3076 oada->dl_addr, ETH_ADDR_LEN);
3077 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3080 case OFPAT_SET_NW_SRC:
3081 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
3082 ia->nw_addr.nw_addr);
3083 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3086 case OFPAT_SET_NW_DST:
3087 nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
3088 ia->nw_addr.nw_addr);
3089 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3092 case OFPAT_SET_NW_TOS:
3093 nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
3095 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3098 case OFPAT_SET_TP_SRC:
3099 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
3100 ia->tp_port.tp_port);
3101 ctx->flow.tp_src = ia->tp_port.tp_port;
3104 case OFPAT_SET_TP_DST:
3105 nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
3106 ia->tp_port.tp_port);
3107 ctx->flow.tp_dst = ia->tp_port.tp_port;
3111 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3115 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3119 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3126 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3127 struct ofproto *ofproto, const struct flow *flow,
3128 const struct ofpbuf *packet)
3130 ctx->ofproto = ofproto;
3132 ctx->packet = packet;
3133 ctx->resubmit_hook = NULL;
3134 ctx->check_special = true;
3137 static struct ofpbuf *
3138 xlate_actions(struct action_xlate_ctx *ctx,
3139 const union ofp_action *in, size_t n_in)
3141 COVERAGE_INC(ofproto_ofp2odp);
3143 ctx->odp_actions = ofpbuf_new(512);
3145 ctx->may_set_up_flow = true;
3146 ctx->nf_output_iface = NF_OUT_DROP;
3148 ctx->last_pop_priority = -1;
3150 if (!ctx->check_special
3151 || !ctx->ofproto->ofhooks->special_cb
3152 || ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet,
3153 ctx->ofproto->aux)) {
3154 do_xlate_actions(in, n_in, ctx);
3156 ctx->may_set_up_flow = false;
3159 remove_pop_action(ctx);
3161 /* Check with in-band control to see if we're allowed to set up this
3163 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3164 ctx->odp_actions->data, ctx->odp_actions->size)) {
3165 ctx->may_set_up_flow = false;
3168 return ctx->odp_actions;
3171 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3172 * error message code (composed with ofp_mkerr()) for the caller to propagate
3173 * upward. Otherwise, returns 0.
3175 * The log message mentions 'msg_type'. */
3177 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3179 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3180 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3181 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3184 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3191 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3193 struct ofproto *p = ofconn->ofproto;
3194 struct ofp_packet_out *opo;
3195 struct ofpbuf payload, *buffer;
3196 union ofp_action *ofp_actions;
3197 struct action_xlate_ctx ctx;
3198 struct ofpbuf *odp_actions;
3199 struct ofpbuf request;
3201 size_t n_ofp_actions;
3205 COVERAGE_INC(ofproto_packet_out);
3207 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3212 /* Get ofp_packet_out. */
3213 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3214 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3217 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3218 &ofp_actions, &n_ofp_actions);
3224 if (opo->buffer_id != htonl(UINT32_MAX)) {
3225 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3227 if (error || !buffer) {
3236 /* Extract flow, check actions. */
3237 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3239 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3245 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3246 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3247 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3248 ofpbuf_delete(odp_actions);
3251 ofpbuf_delete(buffer);
3256 update_port_config(struct ofproto *p, struct ofport *port,
3257 uint32_t config, uint32_t mask)
3259 mask &= config ^ port->opp.config;
3260 if (mask & OFPPC_PORT_DOWN) {
3261 if (config & OFPPC_PORT_DOWN) {
3262 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3264 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3267 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3268 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3269 if (mask & REVALIDATE_BITS) {
3270 COVERAGE_INC(ofproto_costly_flags);
3271 port->opp.config ^= mask & REVALIDATE_BITS;
3272 p->need_revalidate = true;
3274 #undef REVALIDATE_BITS
3275 if (mask & OFPPC_NO_PACKET_IN) {
3276 port->opp.config ^= OFPPC_NO_PACKET_IN;
3281 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3283 struct ofproto *p = ofconn->ofproto;
3284 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3285 struct ofport *port;
3288 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3293 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3295 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3296 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3297 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3299 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3300 if (opm->advertise) {
3301 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3307 static struct ofpbuf *
3308 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3310 struct ofp_stats_reply *osr;
3313 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3314 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3316 osr->flags = htons(0);
3320 static struct ofpbuf *
3321 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3323 const struct ofp_stats_request *osr
3324 = (const struct ofp_stats_request *) request;
3325 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3329 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3330 struct ofpbuf **msgp)
3332 struct ofpbuf *msg = *msgp;
3333 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3334 if (nbytes + msg->size > UINT16_MAX) {
3335 struct ofp_stats_reply *reply = msg->data;
3336 reply->flags = htons(OFPSF_REPLY_MORE);
3337 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3338 queue_tx(msg, ofconn, ofconn->reply_counter);
3340 return ofpbuf_put_uninit(*msgp, nbytes);
3343 static struct ofpbuf *
3344 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3346 struct nicira_stats_msg *nsm;
3349 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3350 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3351 nsm->type = htons(OFPST_VENDOR);
3352 nsm->flags = htons(0);
3353 nsm->vendor = htonl(NX_VENDOR_ID);
3354 nsm->subtype = subtype;
3358 static struct ofpbuf *
3359 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3361 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3365 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3366 struct ofpbuf **msgp)
3368 struct ofpbuf *msg = *msgp;
3369 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3370 if (nbytes + msg->size > UINT16_MAX) {
3371 struct nicira_stats_msg *reply = msg->data;
3372 reply->flags = htons(OFPSF_REPLY_MORE);
3373 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3374 queue_tx(msg, ofconn, ofconn->reply_counter);
3376 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3380 handle_desc_stats_request(struct ofconn *ofconn,
3381 const struct ofp_header *request)
3383 struct ofproto *p = ofconn->ofproto;
3384 struct ofp_desc_stats *ods;
3387 msg = start_ofp_stats_reply(request, sizeof *ods);
3388 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3389 memset(ods, 0, sizeof *ods);
3390 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3391 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3392 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3393 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3394 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3395 queue_tx(msg, ofconn, ofconn->reply_counter);
3401 handle_table_stats_request(struct ofconn *ofconn,
3402 const struct ofp_header *request)
3404 struct ofproto *p = ofconn->ofproto;
3405 struct ofp_table_stats *ots;
3408 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3410 /* Classifier table. */
3411 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3412 memset(ots, 0, sizeof *ots);
3413 strcpy(ots->name, "classifier");
3414 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3415 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3416 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3417 ots->active_count = htonl(classifier_count(&p->cls));
3418 put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */
3419 put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */
3421 queue_tx(msg, ofconn, ofconn->reply_counter);
3426 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3427 struct ofpbuf **msgp)
3429 struct netdev_stats stats;
3430 struct ofp_port_stats *ops;
3432 /* Intentionally ignore return value, since errors will set
3433 * 'stats' to all-1s, which is correct for OpenFlow, and
3434 * netdev_get_stats() will log errors. */
3435 netdev_get_stats(port->netdev, &stats);
3437 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3438 ops->port_no = htons(port->opp.port_no);
3439 memset(ops->pad, 0, sizeof ops->pad);
3440 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
3441 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
3442 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
3443 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
3444 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
3445 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
3446 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
3447 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
3448 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
3449 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
3450 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
3451 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
3455 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3457 struct ofproto *p = ofconn->ofproto;
3458 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3459 struct ofp_port_stats *ops;
3461 struct ofport *port;
3463 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3464 if (psr->port_no != htons(OFPP_NONE)) {
3465 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3467 append_port_stat(port, ofconn, &msg);
3470 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3471 append_port_stat(port, ofconn, &msg);
3475 queue_tx(msg, ofconn, ofconn->reply_counter);
3480 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3482 long long int msecs = time_msec() - start;
3483 *sec = htonl(msecs / 1000);
3484 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3488 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3489 ovs_be16 out_port, struct ofpbuf **replyp)
3491 struct ofp_flow_stats *ofs;
3492 uint64_t packet_count, byte_count;
3494 size_t act_len, len;
3496 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3500 act_len = sizeof *rule->actions * rule->n_actions;
3501 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3503 rule_get_stats(rule, &packet_count, &byte_count);
3505 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3506 ofs->length = htons(len);
3509 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3510 rule->flow_cookie, &cookie);
3511 put_32aligned_be64(&ofs->cookie, cookie);
3512 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3513 ofs->priority = htons(rule->cr.priority);
3514 ofs->idle_timeout = htons(rule->idle_timeout);
3515 ofs->hard_timeout = htons(rule->hard_timeout);
3516 memset(ofs->pad2, 0, sizeof ofs->pad2);
3517 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
3518 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
3519 if (rule->n_actions > 0) {
3520 memcpy(ofs->actions, rule->actions, act_len);
3525 is_valid_table(uint8_t table_id)
3527 return table_id == 0 || table_id == 0xff;
3531 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3533 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3534 struct ofpbuf *reply;
3536 COVERAGE_INC(ofproto_flows_req);
3537 reply = start_ofp_stats_reply(oh, 1024);
3538 if (is_valid_table(fsr->table_id)) {
3539 struct cls_cursor cursor;
3540 struct cls_rule target;
3543 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3545 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3546 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3547 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3550 queue_tx(reply, ofconn, ofconn->reply_counter);
3556 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3557 ovs_be16 out_port, struct ofpbuf **replyp)
3559 struct nx_flow_stats *nfs;
3560 uint64_t packet_count, byte_count;
3561 size_t act_len, start_len;
3562 struct ofpbuf *reply;
3564 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3568 rule_get_stats(rule, &packet_count, &byte_count);
3570 act_len = sizeof *rule->actions * rule->n_actions;
3572 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3573 start_len = (*replyp)->size;
3576 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3579 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3580 nfs->cookie = rule->flow_cookie;
3581 nfs->priority = htons(rule->cr.priority);
3582 nfs->idle_timeout = htons(rule->idle_timeout);
3583 nfs->hard_timeout = htons(rule->hard_timeout);
3584 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3585 memset(nfs->pad2, 0, sizeof nfs->pad2);
3586 nfs->packet_count = htonll(packet_count);
3587 nfs->byte_count = htonll(byte_count);
3588 if (rule->n_actions > 0) {
3589 ofpbuf_put(reply, rule->actions, act_len);
3591 nfs->length = htons(reply->size - start_len);
3595 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3597 struct nx_flow_stats_request *nfsr;
3598 struct cls_rule target;
3599 struct ofpbuf *reply;
3603 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3605 /* Dissect the message. */
3606 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3607 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3612 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3615 COVERAGE_INC(ofproto_flows_req);
3616 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3617 if (is_valid_table(nfsr->table_id)) {
3618 struct cls_cursor cursor;
3621 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3622 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3623 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3626 queue_tx(reply, ofconn, ofconn->reply_counter);
3632 flow_stats_ds(struct rule *rule, struct ds *results)
3634 uint64_t packet_count, byte_count;
3635 size_t act_len = sizeof *rule->actions * rule->n_actions;
3637 rule_get_stats(rule, &packet_count, &byte_count);
3639 ds_put_format(results, "duration=%llds, ",
3640 (time_msec() - rule->created) / 1000);
3641 ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0);
3642 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3643 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3644 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3645 cls_rule_format(&rule->cr, results);
3646 ds_put_char(results, ',');
3648 ofp_print_actions(results, &rule->actions->header, act_len);
3650 ds_put_cstr(results, "drop");
3652 ds_put_cstr(results, "\n");
3655 /* Adds a pretty-printed description of all flows to 'results', including
3656 * those marked hidden by secchan (e.g., by in-band control). */
3658 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3660 struct cls_cursor cursor;
3663 cls_cursor_init(&cursor, &p->cls, NULL);
3664 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3665 flow_stats_ds(rule, results);
3670 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3671 ovs_be16 out_port, uint8_t table_id,
3672 struct ofp_aggregate_stats_reply *oasr)
3674 uint64_t total_packets = 0;
3675 uint64_t total_bytes = 0;
3678 COVERAGE_INC(ofproto_agg_request);
3680 if (is_valid_table(table_id)) {
3681 struct cls_cursor cursor;
3684 cls_cursor_init(&cursor, &ofproto->cls, target);
3685 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3686 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3687 uint64_t packet_count;
3688 uint64_t byte_count;
3690 rule_get_stats(rule, &packet_count, &byte_count);
3692 total_packets += packet_count;
3693 total_bytes += byte_count;
3699 oasr->flow_count = htonl(n_flows);
3700 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
3701 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
3702 memset(oasr->pad, 0, sizeof oasr->pad);
3706 handle_aggregate_stats_request(struct ofconn *ofconn,
3707 const struct ofp_header *oh)
3709 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3710 struct ofp_aggregate_stats_reply *reply;
3711 struct cls_rule target;
3714 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3717 msg = start_ofp_stats_reply(oh, sizeof *reply);
3718 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3719 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3720 request->table_id, reply);
3721 queue_tx(msg, ofconn, ofconn->reply_counter);
3726 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3728 struct nx_aggregate_stats_request *request;
3729 struct ofp_aggregate_stats_reply *reply;
3730 struct cls_rule target;
3735 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3737 /* Dissect the message. */
3738 request = ofpbuf_pull(&b, sizeof *request);
3739 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3744 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3748 COVERAGE_INC(ofproto_flows_req);
3749 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3750 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3751 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3752 request->table_id, reply);
3753 queue_tx(buf, ofconn, ofconn->reply_counter);
3758 struct queue_stats_cbdata {
3759 struct ofconn *ofconn;
3760 struct ofport *ofport;
3765 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3766 const struct netdev_queue_stats *stats)
3768 struct ofp_queue_stats *reply;
3770 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3771 reply->port_no = htons(cbdata->ofport->opp.port_no);
3772 memset(reply->pad, 0, sizeof reply->pad);
3773 reply->queue_id = htonl(queue_id);
3774 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
3775 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
3776 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
3780 handle_queue_stats_dump_cb(uint32_t queue_id,
3781 struct netdev_queue_stats *stats,
3784 struct queue_stats_cbdata *cbdata = cbdata_;
3786 put_queue_stats(cbdata, queue_id, stats);
3790 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3791 struct queue_stats_cbdata *cbdata)
3793 cbdata->ofport = port;
3794 if (queue_id == OFPQ_ALL) {
3795 netdev_dump_queue_stats(port->netdev,
3796 handle_queue_stats_dump_cb, cbdata);
3798 struct netdev_queue_stats stats;
3800 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3801 put_queue_stats(cbdata, queue_id, &stats);
3807 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3809 struct ofproto *ofproto = ofconn->ofproto;
3810 const struct ofp_queue_stats_request *qsr;
3811 struct queue_stats_cbdata cbdata;
3812 struct ofport *port;
3813 unsigned int port_no;
3816 qsr = ofputil_stats_body(oh);
3818 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3821 COVERAGE_INC(ofproto_queue_req);
3823 cbdata.ofconn = ofconn;
3824 cbdata.msg = start_ofp_stats_reply(oh, 128);
3826 port_no = ntohs(qsr->port_no);
3827 queue_id = ntohl(qsr->queue_id);
3828 if (port_no == OFPP_ALL) {
3829 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3830 handle_queue_stats_for_port(port, queue_id, &cbdata);
3832 } else if (port_no < ofproto->max_ports) {
3833 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3835 handle_queue_stats_for_port(port, queue_id, &cbdata);
3838 ofpbuf_delete(cbdata.msg);
3839 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3841 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3847 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3850 if (used > facet->used) {
3852 if (used > facet->rule->used) {
3853 facet->rule->used = used;
3855 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3859 /* Folds the statistics from 'stats' into the counters in 'facet'.
3861 * Because of the meaning of a facet's counters, it only makes sense to do this
3862 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3863 * packet that was sent by hand or if it represents statistics that have been
3864 * cleared out of the datapath. */
3866 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3867 const struct dpif_flow_stats *stats)
3869 if (stats->n_packets) {
3870 facet_update_time(ofproto, facet, stats->used);
3871 facet->packet_count += stats->n_packets;
3872 facet->byte_count += stats->n_bytes;
3873 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3877 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3878 * in which no matching flow already exists in the flow table.
3880 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3881 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3882 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3884 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3887 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3889 struct ofproto *p = ofconn->ofproto;
3890 struct ofpbuf *packet;
3895 if (fm->flags & OFPFF_CHECK_OVERLAP
3896 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3897 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3901 if (fm->buffer_id != UINT32_MAX) {
3902 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3906 in_port = UINT16_MAX;
3909 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3910 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3911 fm->flags & OFPFF_SEND_FLOW_REM);
3912 rule_insert(p, rule);
3914 rule_execute(p, rule, in_port, packet);
3919 static struct rule *
3920 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3922 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3926 send_buffered_packet(struct ofconn *ofconn,
3927 struct rule *rule, uint32_t buffer_id)
3929 struct ofpbuf *packet;
3933 if (buffer_id == UINT32_MAX) {
3937 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3942 rule_execute(ofconn->ofproto, rule, in_port, packet);
3947 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3949 struct modify_flows_cbdata {
3950 struct ofproto *ofproto;
3951 const struct flow_mod *fm;
3955 static int modify_flow(struct ofproto *, const struct flow_mod *,
3958 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3959 * encoded by ofp_mkerr() on failure.
3961 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3964 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3966 struct ofproto *p = ofconn->ofproto;
3967 struct rule *match = NULL;
3968 struct cls_cursor cursor;
3971 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3972 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3973 if (!rule_is_hidden(rule)) {
3975 modify_flow(p, fm, rule);
3980 /* This credits the packet to whichever flow happened to match last.
3981 * That's weird. Maybe we should do a lookup for the flow that
3982 * actually matches the packet? Who knows. */
3983 send_buffered_packet(ofconn, match, fm->buffer_id);
3986 return add_flow(ofconn, fm);
3990 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3991 * code as encoded by ofp_mkerr() on failure.
3993 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3996 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3998 struct ofproto *p = ofconn->ofproto;
3999 struct rule *rule = find_flow_strict(p, fm);
4000 if (rule && !rule_is_hidden(rule)) {
4001 modify_flow(p, fm, rule);
4002 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4004 return add_flow(ofconn, fm);
4008 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4009 * been identified as a flow in 'p''s flow table to be modified, by changing
4010 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4011 * ofp_action[] structures). */
4013 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4015 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4017 rule->flow_cookie = fm->cookie;
4019 /* If the actions are the same, do nothing. */
4020 if (fm->n_actions == rule->n_actions
4022 || !memcmp(fm->actions, rule->actions, actions_len))) {
4026 /* Replace actions. */
4027 free(rule->actions);
4028 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4029 rule->n_actions = fm->n_actions;
4031 p->need_revalidate = true;
4036 /* OFPFC_DELETE implementation. */
4038 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4040 /* Implements OFPFC_DELETE. */
4042 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4044 struct rule *rule, *next_rule;
4045 struct cls_cursor cursor;
4047 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4048 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4049 delete_flow(p, rule, htons(fm->out_port));
4053 /* Implements OFPFC_DELETE_STRICT. */
4055 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4057 struct rule *rule = find_flow_strict(p, fm);
4059 delete_flow(p, rule, htons(fm->out_port));
4063 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4064 * been identified as a flow to delete from 'p''s flow table, by deleting the
4065 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4068 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4069 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4070 * specified 'out_port'. */
4072 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4074 if (rule_is_hidden(rule)) {
4078 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4082 rule_send_removed(p, rule, OFPRR_DELETE);
4083 rule_remove(p, rule);
4087 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4089 struct ofproto *p = ofconn->ofproto;
4093 error = reject_slave_controller(ofconn, "flow_mod");
4098 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4103 /* We do not support the emergency flow cache. It will hopefully get
4104 * dropped from OpenFlow in the near future. */
4105 if (fm.flags & OFPFF_EMERG) {
4106 /* There isn't a good fit for an error code, so just state that the
4107 * flow table is full. */
4108 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4111 error = validate_actions(fm.actions, fm.n_actions,
4112 &fm.cr.flow, p->max_ports);
4117 switch (fm.command) {
4119 return add_flow(ofconn, &fm);
4122 return modify_flows_loose(ofconn, &fm);
4124 case OFPFC_MODIFY_STRICT:
4125 return modify_flow_strict(ofconn, &fm);
4128 delete_flows_loose(p, &fm);
4131 case OFPFC_DELETE_STRICT:
4132 delete_flow_strict(p, &fm);
4136 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4141 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4143 const struct nxt_tun_id_cookie *msg
4144 = (const struct nxt_tun_id_cookie *) oh;
4146 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4151 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4153 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4154 struct nx_role_request *reply;
4158 if (ofconn->type != OFCONN_PRIMARY) {
4159 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4161 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4164 role = ntohl(nrr->role);
4165 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4166 && role != NX_ROLE_SLAVE) {
4167 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4169 /* There's no good error code for this. */
4170 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4173 if (role == NX_ROLE_MASTER) {
4174 struct ofconn *other;
4176 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4177 if (other->role == NX_ROLE_MASTER) {
4178 other->role = NX_ROLE_SLAVE;
4182 ofconn->role = role;
4184 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4185 reply->role = htonl(role);
4186 queue_tx(buf, ofconn, ofconn->reply_counter);
4192 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4194 const struct nxt_set_flow_format *msg
4195 = (const struct nxt_set_flow_format *) oh;
4198 format = ntohl(msg->format);
4199 if (format == NXFF_OPENFLOW10
4200 || format == NXFF_TUN_ID_FROM_COOKIE
4201 || format == NXFF_NXM) {
4202 ofconn->flow_format = format;
4205 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4210 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4212 struct ofp_header *ob;
4215 /* Currently, everything executes synchronously, so we can just
4216 * immediately send the barrier reply. */
4217 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4218 queue_tx(buf, ofconn, ofconn->reply_counter);
4223 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4225 const struct ofp_header *oh = msg->data;
4226 const struct ofputil_msg_type *type;
4229 error = ofputil_decode_msg_type(oh, &type);
4234 switch (ofputil_msg_type_code(type)) {
4235 /* OpenFlow requests. */
4236 case OFPUTIL_OFPT_ECHO_REQUEST:
4237 return handle_echo_request(ofconn, oh);
4239 case OFPUTIL_OFPT_FEATURES_REQUEST:
4240 return handle_features_request(ofconn, oh);
4242 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4243 return handle_get_config_request(ofconn, oh);
4245 case OFPUTIL_OFPT_SET_CONFIG:
4246 return handle_set_config(ofconn, msg->data);
4248 case OFPUTIL_OFPT_PACKET_OUT:
4249 return handle_packet_out(ofconn, oh);
4251 case OFPUTIL_OFPT_PORT_MOD:
4252 return handle_port_mod(ofconn, oh);
4254 case OFPUTIL_OFPT_FLOW_MOD:
4255 return handle_flow_mod(ofconn, oh);
4257 case OFPUTIL_OFPT_BARRIER_REQUEST:
4258 return handle_barrier_request(ofconn, oh);
4260 /* OpenFlow replies. */
4261 case OFPUTIL_OFPT_ECHO_REPLY:
4264 /* Nicira extension requests. */
4265 case OFPUTIL_NXT_STATUS_REQUEST:
4266 return switch_status_handle_request(
4267 ofconn->ofproto->switch_status, ofconn->rconn, oh);
4269 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4270 return handle_tun_id_from_cookie(ofconn, oh);
4272 case OFPUTIL_NXT_ROLE_REQUEST:
4273 return handle_role_request(ofconn, oh);
4275 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4276 return handle_nxt_set_flow_format(ofconn, oh);
4278 case OFPUTIL_NXT_FLOW_MOD:
4279 return handle_flow_mod(ofconn, oh);
4281 /* OpenFlow statistics requests. */
4282 case OFPUTIL_OFPST_DESC_REQUEST:
4283 return handle_desc_stats_request(ofconn, oh);
4285 case OFPUTIL_OFPST_FLOW_REQUEST:
4286 return handle_flow_stats_request(ofconn, oh);
4288 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4289 return handle_aggregate_stats_request(ofconn, oh);
4291 case OFPUTIL_OFPST_TABLE_REQUEST:
4292 return handle_table_stats_request(ofconn, oh);
4294 case OFPUTIL_OFPST_PORT_REQUEST:
4295 return handle_port_stats_request(ofconn, oh);
4297 case OFPUTIL_OFPST_QUEUE_REQUEST:
4298 return handle_queue_stats_request(ofconn, oh);
4300 /* Nicira extension statistics requests. */
4301 case OFPUTIL_NXST_FLOW_REQUEST:
4302 return handle_nxst_flow(ofconn, oh);
4304 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4305 return handle_nxst_aggregate(ofconn, oh);
4307 case OFPUTIL_INVALID:
4308 case OFPUTIL_OFPT_HELLO:
4309 case OFPUTIL_OFPT_ERROR:
4310 case OFPUTIL_OFPT_FEATURES_REPLY:
4311 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4312 case OFPUTIL_OFPT_PACKET_IN:
4313 case OFPUTIL_OFPT_FLOW_REMOVED:
4314 case OFPUTIL_OFPT_PORT_STATUS:
4315 case OFPUTIL_OFPT_BARRIER_REPLY:
4316 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4317 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4318 case OFPUTIL_OFPST_DESC_REPLY:
4319 case OFPUTIL_OFPST_FLOW_REPLY:
4320 case OFPUTIL_OFPST_QUEUE_REPLY:
4321 case OFPUTIL_OFPST_PORT_REPLY:
4322 case OFPUTIL_OFPST_TABLE_REPLY:
4323 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4324 case OFPUTIL_NXT_STATUS_REPLY:
4325 case OFPUTIL_NXT_ROLE_REPLY:
4326 case OFPUTIL_NXT_FLOW_REMOVED:
4327 case OFPUTIL_NXST_FLOW_REPLY:
4328 case OFPUTIL_NXST_AGGREGATE_REPLY:
4330 if (VLOG_IS_WARN_ENABLED()) {
4331 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4332 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4335 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4336 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4338 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4344 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4346 int error = handle_openflow__(ofconn, ofp_msg);
4348 send_error_oh(ofconn, ofp_msg->data, error);
4350 COVERAGE_INC(ofproto_recv_openflow);
4354 handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4356 struct facet *facet;
4359 /* Obtain in_port and tun_id, at least. */
4360 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4362 /* Set header pointers in 'flow'. */
4363 flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
4365 if (p->ofhooks->special_cb
4366 && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) {
4367 ofpbuf_delete(upcall->packet);
4371 /* Check with in-band control to see if this packet should be sent
4372 * to the local port regardless of the flow table. */
4373 if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) {
4374 struct ofpbuf odp_actions;
4376 ofpbuf_init(&odp_actions, 32);
4377 nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, ODPP_LOCAL);
4378 dpif_execute(p->dpif, odp_actions.data, odp_actions.size,
4380 ofpbuf_uninit(&odp_actions);
4383 facet = facet_lookup_valid(p, &flow);
4385 struct rule *rule = rule_lookup(p, &flow);
4387 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4388 struct ofport *port = get_port(p, flow.in_port);
4390 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4391 COVERAGE_INC(ofproto_no_packet_in);
4392 /* XXX install 'drop' flow entry */
4393 ofpbuf_delete(upcall->packet);
4397 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4401 COVERAGE_INC(ofproto_packet_in);
4402 send_packet_in(p, upcall, &flow, false);
4406 facet = facet_create(p, rule, &flow, upcall->packet);
4407 } else if (!facet->may_install) {
4408 /* The facet is not installable, that is, we need to process every
4409 * packet, so process the current packet's actions into 'facet'. */
4410 facet_make_actions(p, facet, upcall->packet);
4413 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4415 * Extra-special case for fail-open mode.
4417 * We are in fail-open mode and the packet matched the fail-open rule,
4418 * but we are connected to a controller too. We should send the packet
4419 * up to the controller in the hope that it will try to set up a flow
4420 * and thereby allow us to exit fail-open.
4422 * See the top-level comment in fail-open.c for more information.
4424 send_packet_in(p, upcall, &flow, true);
4427 facet_execute(p, facet, upcall->packet);
4428 facet_install(p, facet, false);
4432 handle_upcall(struct ofproto *p, struct dpif_upcall *upcall)
4436 switch (upcall->type) {
4437 case DPIF_UC_ACTION:
4438 COVERAGE_INC(ofproto_ctlr_action);
4439 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4440 send_packet_in(p, upcall, &flow, false);
4443 case DPIF_UC_SAMPLE:
4445 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
4446 ofproto_sflow_received(p->sflow, upcall, &flow);
4448 ofpbuf_delete(upcall->packet);
4452 handle_miss_upcall(p, upcall);
4455 case DPIF_N_UC_TYPES:
4457 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4462 /* Flow expiration. */
4464 static int ofproto_dp_max_idle(const struct ofproto *);
4465 static void ofproto_update_stats(struct ofproto *);
4466 static void rule_expire(struct ofproto *, struct rule *);
4467 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4469 /* This function is called periodically by ofproto_run(). Its job is to
4470 * collect updates for the flows that have been installed into the datapath,
4471 * most importantly when they last were used, and then use that information to
4472 * expire flows that have not been used recently.
4474 * Returns the number of milliseconds after which it should be called again. */
4476 ofproto_expire(struct ofproto *ofproto)
4478 struct rule *rule, *next_rule;
4479 struct cls_cursor cursor;
4482 /* Update stats for each flow in the datapath. */
4483 ofproto_update_stats(ofproto);
4485 /* Expire facets that have been idle too long. */
4486 dp_max_idle = ofproto_dp_max_idle(ofproto);
4487 ofproto_expire_facets(ofproto, dp_max_idle);
4489 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4490 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4491 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4492 rule_expire(ofproto, rule);
4495 /* Let the hook know that we're at a stable point: all outstanding data
4496 * in existing flows has been accounted to the account_cb. Thus, the
4497 * hook can now reasonably do operations that depend on having accurate
4498 * flow volume accounting (currently, that's just bond rebalancing). */
4499 if (ofproto->ofhooks->account_checkpoint_cb) {
4500 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4503 return MIN(dp_max_idle, 1000);
4506 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4509 ofproto_update_stats(struct ofproto *p)
4511 const struct dpif_flow_stats *stats;
4512 struct dpif_flow_dump dump;
4513 const struct nlattr *key;
4516 dpif_flow_dump_start(&dump, p->dpif);
4517 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4518 struct facet *facet;
4521 if (odp_flow_key_to_flow(key, key_len, &flow)) {
4525 odp_flow_key_format(key, key_len, &s);
4526 VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
4532 facet = facet_find(p, &flow);
4534 if (facet && facet->installed) {
4536 if (stats->n_packets >= facet->dp_packet_count) {
4537 facet->packet_count += stats->n_packets - facet->dp_packet_count;
4539 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4542 if (stats->n_bytes >= facet->dp_byte_count) {
4543 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
4545 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4548 facet->dp_packet_count = stats->n_packets;
4549 facet->dp_byte_count = stats->n_bytes;
4551 facet_update_time(p, facet, stats->used);
4552 facet_account(p, facet, stats->n_bytes);
4554 /* There's a flow in the datapath that we know nothing about.
4556 COVERAGE_INC(ofproto_unexpected_rule);
4557 dpif_flow_del(p->dpif, key, key_len, NULL);
4560 dpif_flow_dump_done(&dump);
4563 /* Calculates and returns the number of milliseconds of idle time after which
4564 * facets should expire from the datapath and we should fold their statistics
4565 * into their parent rules in userspace. */
4567 ofproto_dp_max_idle(const struct ofproto *ofproto)
4570 * Idle time histogram.
4572 * Most of the time a switch has a relatively small number of facets. When
4573 * this is the case we might as well keep statistics for all of them in
4574 * userspace and to cache them in the kernel datapath for performance as
4577 * As the number of facets increases, the memory required to maintain
4578 * statistics about them in userspace and in the kernel becomes
4579 * significant. However, with a large number of facets it is likely that
4580 * only a few of them are "heavy hitters" that consume a large amount of
4581 * bandwidth. At this point, only heavy hitters are worth caching in the
4582 * kernel and maintaining in userspaces; other facets we can discard.
4584 * The technique used to compute the idle time is to build a histogram with
4585 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4586 * that is installed in the kernel gets dropped in the appropriate bucket.
4587 * After the histogram has been built, we compute the cutoff so that only
4588 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4589 * cached. At least the most-recently-used bucket of facets is kept, so
4590 * actually an arbitrary number of facets can be kept in any given
4591 * expiration run (though the next run will delete most of those unless
4592 * they receive additional data).
4594 * This requires a second pass through the facets, in addition to the pass
4595 * made by ofproto_update_stats(), because the former function never looks
4596 * at uninstallable facets.
4598 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4599 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4600 int buckets[N_BUCKETS] = { 0 };
4601 struct facet *facet;
4606 total = hmap_count(&ofproto->facets);
4607 if (total <= 1000) {
4608 return N_BUCKETS * BUCKET_WIDTH;
4611 /* Build histogram. */
4613 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4614 long long int idle = now - facet->used;
4615 int bucket = (idle <= 0 ? 0
4616 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4617 : (unsigned int) idle / BUCKET_WIDTH);
4621 /* Find the first bucket whose flows should be expired. */
4622 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4623 if (buckets[bucket]) {
4626 subtotal += buckets[bucket++];
4627 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4632 if (VLOG_IS_DBG_ENABLED()) {
4636 ds_put_cstr(&s, "keep");
4637 for (i = 0; i < N_BUCKETS; i++) {
4639 ds_put_cstr(&s, ", drop");
4642 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4645 VLOG_INFO("%s: %s (msec:count)",
4646 dpif_name(ofproto->dpif), ds_cstr(&s));
4650 return bucket * BUCKET_WIDTH;
4654 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4656 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4657 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4658 struct ofexpired expired;
4660 if (facet->installed) {
4661 struct dpif_flow_stats stats;
4663 facet_put__(ofproto, facet, facet->actions, facet->actions_len,
4665 facet_update_stats(ofproto, facet, &stats);
4668 expired.flow = facet->flow;
4669 expired.packet_count = facet->packet_count;
4670 expired.byte_count = facet->byte_count;
4671 expired.used = facet->used;
4672 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4677 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4679 long long int cutoff = time_msec() - dp_max_idle;
4680 struct facet *facet, *next_facet;
4682 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4683 facet_active_timeout(ofproto, facet);
4684 if (facet->used < cutoff) {
4685 facet_remove(ofproto, facet);
4690 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4691 * then delete it entirely. */
4693 rule_expire(struct ofproto *ofproto, struct rule *rule)
4695 struct facet *facet, *next_facet;
4699 /* Has 'rule' expired? */
4701 if (rule->hard_timeout
4702 && now > rule->created + rule->hard_timeout * 1000) {
4703 reason = OFPRR_HARD_TIMEOUT;
4704 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4705 && now >rule->used + rule->idle_timeout * 1000) {
4706 reason = OFPRR_IDLE_TIMEOUT;
4711 COVERAGE_INC(ofproto_expired);
4713 /* Update stats. (This is a no-op if the rule expired due to an idle
4714 * timeout, because that only happens when the rule has no facets left.) */
4715 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4716 facet_remove(ofproto, facet);
4719 /* Get rid of the rule. */
4720 if (!rule_is_hidden(rule)) {
4721 rule_send_removed(ofproto, rule, reason);
4723 rule_remove(ofproto, rule);
4726 static struct ofpbuf *
4727 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4730 struct ofp_flow_removed *ofr;
4733 ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
4734 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4735 rule->flow_cookie, &ofr->cookie);
4736 ofr->priority = htons(rule->cr.priority);
4737 ofr->reason = reason;
4738 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4739 ofr->idle_timeout = htons(rule->idle_timeout);
4740 ofr->packet_count = htonll(rule->packet_count);
4741 ofr->byte_count = htonll(rule->byte_count);
4746 static struct ofpbuf *
4747 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4749 struct nx_flow_removed *nfr;
4753 make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
4754 match_len = nx_put_match(buf, &rule->cr);
4757 nfr->cookie = rule->flow_cookie;
4758 nfr->priority = htons(rule->cr.priority);
4759 nfr->reason = reason;
4760 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4761 nfr->idle_timeout = htons(rule->idle_timeout);
4762 nfr->match_len = htons(match_len);
4763 nfr->packet_count = htonll(rule->packet_count);
4764 nfr->byte_count = htonll(rule->byte_count);
4770 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4772 struct ofconn *ofconn;
4774 if (!rule->send_flow_removed) {
4778 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4781 if (!rconn_is_connected(ofconn->rconn)
4782 || !ofconn_receives_async_msgs(ofconn)) {
4786 msg = (ofconn->flow_format == NXFF_NXM
4787 ? compose_nx_flow_removed(rule, reason)
4788 : compose_ofp_flow_removed(ofconn, rule, reason));
4790 /* Account flow expirations under ofconn->reply_counter, the counter
4791 * for replies to OpenFlow requests. That works because preventing
4792 * OpenFlow requests from being processed also prevents new flows from
4793 * being added (and expiring). (It also prevents processing OpenFlow
4794 * requests that would not add new flows, so it is imperfect.) */
4795 queue_tx(msg, ofconn, ofconn->reply_counter);
4799 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4800 * The returned statistics include statistics for all of 'rule''s facets. */
4802 rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes)
4805 struct facet *facet;
4807 /* Start from historical data for 'rule' itself that are no longer tracked
4808 * in facets. This counts, for example, facets that have expired. */
4809 p = rule->packet_count;
4810 b = rule->byte_count;
4812 /* Add any statistics that are tracked by facets. This includes
4813 * statistical data recently updated by ofproto_update_stats() as well as
4814 * stats for packets that were executed "by hand" via dpif_execute(). */
4815 LIST_FOR_EACH (facet, list_node, &rule->facets) {
4816 p += facet->packet_count;
4817 b += facet->byte_count;
4824 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4826 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
4828 struct ofconn *ofconn = ofconn_;
4830 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
4831 ofconn->packet_in_counter, 100);
4834 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4835 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4836 * scheduler for sending.
4838 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4839 * Otherwise, ownership is transferred to this function. */
4841 schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall,
4842 const struct flow *flow, bool clone)
4844 enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) };
4845 struct ofproto *ofproto = ofconn->ofproto;
4846 struct ofp_packet_in *opi;
4847 int total_len, send_len;
4848 struct ofpbuf *packet;
4852 /* Get OpenFlow buffer_id. */
4853 if (upcall->type == DPIF_UC_ACTION) {
4854 buffer_id = UINT32_MAX;
4855 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4856 buffer_id = pktbuf_get_null();
4857 } else if (!ofconn->pktbuf) {
4858 buffer_id = UINT32_MAX;
4860 buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port);
4863 /* Figure out how much of the packet to send. */
4864 total_len = send_len = upcall->packet->size;
4865 if (buffer_id != UINT32_MAX) {
4866 send_len = MIN(send_len, ofconn->miss_send_len);
4868 if (upcall->type == DPIF_UC_ACTION) {
4869 send_len = MIN(send_len, upcall->userdata);
4872 /* Copy or steal buffer for OFPT_PACKET_IN. */
4874 packet = ofpbuf_clone_data_with_headroom(upcall->packet->data,
4875 send_len, OPI_SIZE);
4877 packet = upcall->packet;
4878 packet->size = send_len;
4881 /* Add OFPT_PACKET_IN. */
4882 opi = ofpbuf_push_zeros(packet, OPI_SIZE);
4883 opi->header.version = OFP_VERSION;
4884 opi->header.type = OFPT_PACKET_IN;
4885 opi->total_len = htons(total_len);
4886 opi->in_port = htons(odp_port_to_ofp_port(flow->in_port));
4887 opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
4888 opi->buffer_id = htonl(buffer_id);
4889 update_openflow_length(packet);
4891 /* Hand over to packet scheduler. It might immediately call into
4892 * do_send_packet_in() or it might buffer it for a while (until a later
4893 * call to pinsched_run()). */
4894 idx = upcall->type == DPIF_UC_MISS ? 0 : 1;
4895 pinsched_send(ofconn->schedulers[idx], flow->in_port,
4896 packet, do_send_packet_in, ofconn);
4899 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4900 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4901 * their individual configurations.
4903 * Takes ownership of 'packet'. */
4905 send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall,
4906 const struct flow *flow, bool clone)
4908 struct ofconn *ofconn, *prev;
4911 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4912 if (ofconn_receives_async_msgs(ofconn)) {
4914 schedule_packet_in(prev, upcall, flow, true);
4920 schedule_packet_in(prev, upcall, flow, clone);
4921 } else if (!clone) {
4922 ofpbuf_delete(upcall->packet);
4927 pick_datapath_id(const struct ofproto *ofproto)
4929 const struct ofport *port;
4931 port = get_port(ofproto, ODPP_LOCAL);
4933 uint8_t ea[ETH_ADDR_LEN];
4936 error = netdev_get_etheraddr(port->netdev, ea);
4938 return eth_addr_to_uint64(ea);
4940 VLOG_WARN("could not get MAC address for %s (%s)",
4941 netdev_get_name(port->netdev), strerror(error));
4943 return ofproto->fallback_dpid;
4947 pick_fallback_dpid(void)
4949 uint8_t ea[ETH_ADDR_LEN];
4950 eth_addr_nicira_random(ea);
4951 return eth_addr_to_uint64(ea);
4955 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4956 void *aux OVS_UNUSED)
4958 const struct shash_node *node;
4962 SHASH_FOR_EACH (node, &all_ofprotos) {
4963 ds_put_format(&results, "%s\n", node->name);
4965 unixctl_command_reply(conn, 200, ds_cstr(&results));
4966 ds_destroy(&results);
4969 struct ofproto_trace {
4970 struct action_xlate_ctx ctx;
4976 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4978 ds_put_char_multiple(result, '\t', level);
4980 ds_put_cstr(result, "No match\n");
4984 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4985 ntohll(rule->flow_cookie));
4986 cls_rule_format(&rule->cr, result);
4987 ds_put_char(result, '\n');
4989 ds_put_char_multiple(result, '\t', level);
4990 ds_put_cstr(result, "OpenFlow ");
4991 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4992 rule->n_actions * sizeof *rule->actions);
4993 ds_put_char(result, '\n');
4997 trace_format_flow(struct ds *result, int level, const char *title,
4998 struct ofproto_trace *trace)
5000 ds_put_char_multiple(result, '\t', level);
5001 ds_put_format(result, "%s: ", title);
5002 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
5003 ds_put_cstr(result, "unchanged");
5005 flow_format(result, &trace->ctx.flow);
5006 trace->flow = trace->ctx.flow;
5008 ds_put_char(result, '\n');
5012 trace_resubmit(struct action_xlate_ctx *ctx, const struct rule *rule)
5014 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
5015 struct ds *result = trace->result;
5017 ds_put_char(result, '\n');
5018 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
5019 trace_format_rule(result, ctx->recurse + 1, rule);
5023 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
5024 void *aux OVS_UNUSED)
5026 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5027 char *args = xstrdup(args_);
5028 char *save_ptr = NULL;
5029 struct ofproto *ofproto;
5030 struct ofpbuf packet;
5038 ofpbuf_init(&packet, strlen(args) / 2);
5041 dpname = strtok_r(args, " ", &save_ptr);
5042 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5043 in_port_s = strtok_r(NULL, " ", &save_ptr);
5044 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5045 if (!dpname || !in_port_s || !packet_s) {
5046 unixctl_command_reply(conn, 501, "Bad command syntax");
5050 ofproto = shash_find_data(&all_ofprotos, dpname);
5052 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5057 tun_id = htonll(strtoull(tun_id_s, NULL, 10));
5058 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5060 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5061 packet_s += strspn(packet_s, " ");
5062 if (*packet_s != '\0') {
5063 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5066 if (packet.size < ETH_HEADER_LEN) {
5067 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5071 ds_put_cstr(&result, "Packet: ");
5072 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5073 ds_put_cstr(&result, s);
5076 flow_extract(&packet, tun_id, in_port, &flow);
5077 ds_put_cstr(&result, "Flow: ");
5078 flow_format(&result, &flow);
5079 ds_put_char(&result, '\n');
5081 rule = rule_lookup(ofproto, &flow);
5082 trace_format_rule(&result, 0, rule);
5084 struct ofproto_trace trace;
5085 struct ofpbuf *odp_actions;
5087 trace.result = &result;
5089 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5090 trace.ctx.resubmit_hook = trace_resubmit;
5091 odp_actions = xlate_actions(&trace.ctx,
5092 rule->actions, rule->n_actions);
5094 ds_put_char(&result, '\n');
5095 trace_format_flow(&result, 0, "Final flow", &trace);
5096 ds_put_cstr(&result, "Datapath actions: ");
5097 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5098 ofpbuf_delete(odp_actions);
5101 unixctl_command_reply(conn, 200, ds_cstr(&result));
5104 ds_destroy(&result);
5105 ofpbuf_uninit(&packet);
5110 ofproto_unixctl_init(void)
5112 static bool registered;
5118 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5119 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5123 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5124 struct ofpbuf *odp_actions, tag_type *tags,
5125 uint16_t *nf_output_iface, void *ofproto_)
5127 struct ofproto *ofproto = ofproto_;
5130 /* Drop frames for reserved multicast addresses. */
5131 if (eth_addr_is_reserved(flow->dl_dst)) {
5135 /* Learn source MAC (but don't try to learn from revalidation). */
5136 if (packet != NULL) {
5137 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5139 GRAT_ARP_LOCK_NONE);
5141 /* The log messages here could actually be useful in debugging,
5142 * so keep the rate limit relatively high. */
5143 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5144 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5145 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5146 ofproto_revalidate(ofproto, rev_tag);
5150 /* Determine output port. */
5151 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5154 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5155 nf_output_iface, odp_actions);
5156 } else if (out_port != flow->in_port) {
5157 nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port);
5158 *nf_output_iface = out_port;
5166 static const struct ofhooks default_ofhooks = {
5167 default_normal_ofhook_cb,