2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
28 #include "classifier.h"
30 #include "discovery.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "multipath.h"
44 #include "ofp-print.h"
46 #include "ofproto-sflow.h"
48 #include "openflow/nicira-ext.h"
49 #include "openflow/openflow.h"
50 #include "openvswitch/datapath-protocol.h"
54 #include "poll-loop.h"
58 #include "stream-ssl.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto);
68 COVERAGE_DEFINE(facet_changed_rule);
69 COVERAGE_DEFINE(facet_revalidate);
70 COVERAGE_DEFINE(odp_overflow);
71 COVERAGE_DEFINE(ofproto_agg_request);
72 COVERAGE_DEFINE(ofproto_costly_flags);
73 COVERAGE_DEFINE(ofproto_ctlr_action);
74 COVERAGE_DEFINE(ofproto_del_rule);
75 COVERAGE_DEFINE(ofproto_error);
76 COVERAGE_DEFINE(ofproto_expiration);
77 COVERAGE_DEFINE(ofproto_expired);
78 COVERAGE_DEFINE(ofproto_flows_req);
79 COVERAGE_DEFINE(ofproto_flush);
80 COVERAGE_DEFINE(ofproto_invalidated);
81 COVERAGE_DEFINE(ofproto_no_packet_in);
82 COVERAGE_DEFINE(ofproto_ofconn_stuck);
83 COVERAGE_DEFINE(ofproto_ofp2odp);
84 COVERAGE_DEFINE(ofproto_packet_in);
85 COVERAGE_DEFINE(ofproto_packet_out);
86 COVERAGE_DEFINE(ofproto_queue_req);
87 COVERAGE_DEFINE(ofproto_recv_openflow);
88 COVERAGE_DEFINE(ofproto_reinit_ports);
89 COVERAGE_DEFINE(ofproto_unexpected_rule);
90 COVERAGE_DEFINE(ofproto_uninstallable);
91 COVERAGE_DEFINE(ofproto_update_port);
93 #include "sflow_api.h"
98 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
99 struct netdev *netdev;
100 struct ofp_phy_port opp; /* In host byte order. */
104 static void ofport_free(struct ofport *);
105 static void hton_ofp_phy_port(struct ofp_phy_port *);
107 struct action_xlate_ctx {
108 /* action_xlate_ctx_init() initializes these members. */
111 struct ofproto *ofproto;
113 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
114 * this flow when actions change header fields. */
117 /* The packet corresponding to 'flow', or a null pointer if we are
118 * revalidating without a packet to refer to. */
119 const struct ofpbuf *packet;
121 /* If nonnull, called just before executing a resubmit action.
123 * This is normally null so the client has to set it manually after
124 * calling action_xlate_ctx_init(). */
125 void (*resubmit_hook)(struct action_xlate_ctx *, const struct rule *);
127 /* xlate_actions() initializes and uses these members. The client might want
128 * to look at them after it returns. */
130 struct ofpbuf *odp_actions; /* Datapath actions. */
131 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
132 bool may_set_up_flow; /* True ordinarily; false if the actions must
133 * be reassessed for every packet. */
134 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
136 /* xlate_actions() initializes and uses these members, but the client has no
137 * reason to look at them. */
139 int recurse; /* Recursion level, via xlate_table_action. */
140 int last_pop_priority; /* Offset in 'odp_actions' just past most
141 * recently added ODPAT_SET_PRIORITY. */
144 static void action_xlate_ctx_init(struct action_xlate_ctx *,
145 struct ofproto *, const struct flow *,
146 const struct ofpbuf *);
147 static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
148 const union ofp_action *in, size_t n_in);
150 /* An OpenFlow flow. */
152 long long int used; /* Time last used; time created if not used. */
153 long long int created; /* Creation time. */
157 * - Do include packets and bytes from facets that have been deleted or
158 * whose own statistics have been folded into the rule.
160 * - Do include packets and bytes sent "by hand" that were accounted to
161 * the rule without any facet being involved (this is a rare corner
162 * case in rule_execute()).
164 * - Do not include packet or bytes that can be obtained from any facet's
165 * packet_count or byte_count member or that can be obtained from the
166 * datapath by, e.g., dpif_flow_get() for any facet.
168 uint64_t packet_count; /* Number of packets received. */
169 uint64_t byte_count; /* Number of bytes received. */
171 ovs_be64 flow_cookie; /* Controller-issued identifier. */
173 struct cls_rule cr; /* In owning ofproto's classifier. */
174 uint16_t idle_timeout; /* In seconds from time of last use. */
175 uint16_t hard_timeout; /* In seconds from time of creation. */
176 bool send_flow_removed; /* Send a flow removed message? */
177 int n_actions; /* Number of elements in actions[]. */
178 union ofp_action *actions; /* OpenFlow actions. */
179 struct list facets; /* List of "struct facet"s. */
182 static struct rule *rule_from_cls_rule(const struct cls_rule *);
183 static bool rule_is_hidden(const struct rule *);
185 static struct rule *rule_create(const struct cls_rule *,
186 const union ofp_action *, size_t n_actions,
187 uint16_t idle_timeout, uint16_t hard_timeout,
188 ovs_be64 flow_cookie, bool send_flow_removed);
189 static void rule_destroy(struct ofproto *, struct rule *);
190 static void rule_free(struct rule *);
192 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
193 static void rule_insert(struct ofproto *, struct rule *);
194 static void rule_remove(struct ofproto *, struct rule *);
196 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
198 /* An exact-match instantiation of an OpenFlow flow. */
200 long long int used; /* Time last used; time created if not used. */
204 * - Do include packets and bytes sent "by hand", e.g. with
207 * - Do include packets and bytes that were obtained from the datapath
208 * when a flow was deleted (e.g. dpif_flow_del()) or when its
209 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
211 * - Do not include any packets or bytes that can currently be obtained
212 * from the datapath by, e.g., dpif_flow_get().
214 uint64_t packet_count; /* Number of packets received. */
215 uint64_t byte_count; /* Number of bytes received. */
217 /* Number of bytes passed to account_cb. This may include bytes that can
218 * currently obtained from the datapath (thus, it can be greater than
220 uint64_t accounted_bytes;
222 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
223 struct list list_node; /* In owning rule's 'facets' list. */
224 struct rule *rule; /* Owning rule. */
225 struct flow flow; /* Exact-match flow. */
226 bool installed; /* Installed in datapath? */
227 bool may_install; /* True ordinarily; false if actions must
228 * be reassessed for every packet. */
229 size_t actions_len; /* Number of bytes in actions[]. */
230 struct nlattr *actions; /* Datapath actions. */
231 tag_type tags; /* Tags (set only by hooks). */
232 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
235 static struct facet *facet_create(struct ofproto *, struct rule *,
237 const struct ofpbuf *packet);
238 static void facet_remove(struct ofproto *, struct facet *);
239 static void facet_free(struct facet *);
241 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
242 static bool facet_revalidate(struct ofproto *, struct facet *);
244 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
245 static void facet_uninstall(struct ofproto *, struct facet *);
246 static void facet_flush_stats(struct ofproto *, struct facet *);
248 static void facet_make_actions(struct ofproto *, struct facet *,
249 const struct ofpbuf *packet);
250 static void facet_update_stats(struct ofproto *, struct facet *,
251 const struct odp_flow_stats *);
253 /* ofproto supports two kinds of OpenFlow connections:
255 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
256 * maintains persistent connections to these controllers and by default
257 * sends them asynchronous messages such as packet-ins.
259 * - "Service" connections, e.g. from ovs-ofctl. When these connections
260 * drop, it is the other side's responsibility to reconnect them if
261 * necessary. ofproto does not send them asynchronous messages by default.
263 * Currently, active (tcp, ssl, unix) connections are always "primary"
264 * connections and passive (ptcp, pssl, punix) connections are always "service"
265 * connections. There is no inherent reason for this, but it reflects the
269 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
270 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
273 /* A listener for incoming OpenFlow "service" connections. */
275 struct hmap_node node; /* In struct ofproto's "services" hmap. */
276 struct pvconn *pvconn; /* OpenFlow connection listener. */
278 /* These are not used by ofservice directly. They are settings for
279 * accepted "struct ofconn"s from the pvconn. */
280 int probe_interval; /* Max idle time before probing, in seconds. */
281 int rate_limit; /* Max packet-in rate in packets per second. */
282 int burst_limit; /* Limit on accumulating packet credits. */
285 static struct ofservice *ofservice_lookup(struct ofproto *,
287 static int ofservice_create(struct ofproto *,
288 const struct ofproto_controller *);
289 static void ofservice_reconfigure(struct ofservice *,
290 const struct ofproto_controller *);
291 static void ofservice_destroy(struct ofproto *, struct ofservice *);
293 /* An OpenFlow connection. */
295 struct ofproto *ofproto; /* The ofproto that owns this connection. */
296 struct list node; /* In struct ofproto's "all_conns" list. */
297 struct rconn *rconn; /* OpenFlow connection. */
298 enum ofconn_type type; /* Type. */
299 enum nx_flow_format flow_format; /* Currently selected flow format. */
301 /* OFPT_PACKET_IN related data. */
302 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
303 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
304 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
305 int miss_send_len; /* Bytes to send of buffered packets. */
307 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
308 * requests, and the maximum number before we stop reading OpenFlow
310 #define OFCONN_REPLY_MAX 100
311 struct rconn_packet_counter *reply_counter;
313 /* type == OFCONN_PRIMARY only. */
314 enum nx_role role; /* Role. */
315 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
316 struct discovery *discovery; /* Controller discovery object, if enabled. */
317 struct status_category *ss; /* Switch status category. */
318 enum ofproto_band band; /* In-band or out-of-band? */
321 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
322 * "schedulers" array. Their values are 0 and 1, and their meanings and values
323 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
324 * case anything ever changes, check their values here. */
325 #define N_SCHEDULERS 2
326 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
327 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
328 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
329 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
331 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
333 static void ofconn_destroy(struct ofconn *);
334 static void ofconn_run(struct ofconn *);
335 static void ofconn_wait(struct ofconn *);
336 static bool ofconn_receives_async_msgs(const struct ofconn *);
337 static char *ofconn_make_name(const struct ofproto *, const char *target);
338 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
340 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
341 struct rconn_packet_counter *counter);
343 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
344 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
348 uint64_t datapath_id; /* Datapath ID. */
349 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
350 char *mfr_desc; /* Manufacturer. */
351 char *hw_desc; /* Hardware. */
352 char *sw_desc; /* Software version. */
353 char *serial_desc; /* Serial number. */
354 char *dp_desc; /* Datapath description. */
358 struct netdev_monitor *netdev_monitor;
359 struct hmap ports; /* Contains "struct ofport"s. */
360 struct shash port_by_name;
364 struct switch_status *switch_status;
365 struct fail_open *fail_open;
366 struct netflow *netflow;
367 struct ofproto_sflow *sflow;
369 /* In-band control. */
370 struct in_band *in_band;
371 long long int next_in_band_update;
372 struct sockaddr_in *extra_in_band_remotes;
373 size_t n_extra_remotes;
377 struct classifier cls;
378 long long int next_expiration;
382 bool need_revalidate;
383 struct tag_set revalidate_set;
385 /* OpenFlow connections. */
386 struct hmap controllers; /* Controller "struct ofconn"s. */
387 struct list all_conns; /* Contains "struct ofconn"s. */
388 enum ofproto_fail_mode fail_mode;
390 /* OpenFlow listeners. */
391 struct hmap services; /* Contains "struct ofservice"s. */
392 struct pvconn **snoops;
395 /* Hooks for ovs-vswitchd. */
396 const struct ofhooks *ofhooks;
399 /* Used by default ofhooks. */
400 struct mac_learning *ml;
403 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
404 static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos);
406 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
408 static const struct ofhooks default_ofhooks;
410 static uint64_t pick_datapath_id(const struct ofproto *);
411 static uint64_t pick_fallback_dpid(void);
413 static int ofproto_expire(struct ofproto *);
415 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
417 static void handle_openflow(struct ofconn *, struct ofpbuf *);
419 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
420 static void update_port(struct ofproto *, const char *devname);
421 static int init_ports(struct ofproto *);
422 static void reinit_ports(struct ofproto *);
424 static void ofproto_unixctl_init(void);
427 ofproto_create(const char *datapath, const char *datapath_type,
428 const struct ofhooks *ofhooks, void *aux,
429 struct ofproto **ofprotop)
431 struct odp_stats stats;
438 ofproto_unixctl_init();
440 /* Connect to datapath and start listening for messages. */
441 error = dpif_open(datapath, datapath_type, &dpif);
443 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
446 error = dpif_get_dp_stats(dpif, &stats);
448 VLOG_ERR("failed to obtain stats for datapath %s: %s",
449 datapath, strerror(error));
453 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
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 = stats.max_ports;
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++) {
1178 error = dpif_recv(p->dpif, &buf);
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_odp_msg(p, buf);
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);
1357 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1359 * This is almost the same as calling dpif_port_del() directly on the
1360 * datapath, but it also makes 'ofproto' close its open netdev for the port
1361 * (if any). This makes it possible to create a new netdev of a different
1362 * type under the same name, which otherwise the netdev library would refuse
1363 * to do because of the conflict. (The netdev would eventually get closed on
1364 * the next trip through ofproto_run(), but this interface is more direct.)
1366 * Returns 0 if successful, otherwise a positive errno. */
1368 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1370 struct ofport *ofport = get_port(ofproto, odp_port);
1371 const char *name = ofport ? ofport->opp.name : "<unknown>";
1374 error = dpif_port_del(ofproto->dpif, odp_port);
1376 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1377 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1378 } else if (ofport) {
1379 /* 'name' is ofport->opp.name and update_port() is going to destroy
1380 * 'ofport'. Just in case update_port() refers to 'name' after it
1381 * destroys 'ofport', make a copy of it around the update_port()
1383 char *devname = xstrdup(name);
1384 update_port(ofproto, devname);
1390 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1391 * true if 'odp_port' exists and should be included, false otherwise. */
1393 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1395 struct ofport *ofport = get_port(ofproto, odp_port);
1396 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1400 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1401 const union ofp_action *actions, size_t n_actions,
1402 const struct ofpbuf *packet)
1404 struct action_xlate_ctx ctx;
1405 struct ofpbuf *odp_actions;
1407 action_xlate_ctx_init(&ctx, p, flow, packet);
1408 odp_actions = xlate_actions(&ctx, actions, n_actions);
1410 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1412 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet);
1414 ofpbuf_delete(odp_actions);
1419 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1420 * performs the 'n_actions' actions in 'actions'. The new flow will not
1423 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1424 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1425 * controllers; otherwise, it will be hidden.
1427 * The caller retains ownership of 'cls_rule' and 'actions'. */
1429 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1430 const union ofp_action *actions, size_t n_actions)
1433 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1434 rule_insert(p, rule);
1438 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1442 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1445 rule_remove(ofproto, rule);
1450 ofproto_flush_flows(struct ofproto *ofproto)
1452 struct facet *facet, *next_facet;
1453 struct rule *rule, *next_rule;
1454 struct cls_cursor cursor;
1456 COVERAGE_INC(ofproto_flush);
1458 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1459 /* Mark the facet as not installed so that facet_remove() doesn't
1460 * bother trying to uninstall it. There is no point in uninstalling it
1461 * individually since we are about to blow away all the facets with
1462 * dpif_flow_flush(). */
1463 facet->installed = false;
1464 facet_remove(ofproto, facet);
1467 cls_cursor_init(&cursor, &ofproto->cls, NULL);
1468 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1469 rule_remove(ofproto, rule);
1472 dpif_flow_flush(ofproto->dpif);
1473 if (ofproto->in_band) {
1474 in_band_flushed(ofproto->in_band);
1476 if (ofproto->fail_open) {
1477 fail_open_flushed(ofproto->fail_open);
1482 reinit_ports(struct ofproto *p)
1484 struct svec devnames;
1485 struct ofport *ofport;
1486 struct odp_port *odp_ports;
1490 COVERAGE_INC(ofproto_reinit_ports);
1492 svec_init(&devnames);
1493 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1494 svec_add (&devnames, ofport->opp.name);
1496 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1497 for (i = 0; i < n_odp_ports; i++) {
1498 svec_add (&devnames, odp_ports[i].devname);
1502 svec_sort_unique(&devnames);
1503 for (i = 0; i < devnames.n; i++) {
1504 update_port(p, devnames.names[i]);
1506 svec_destroy(&devnames);
1509 static struct ofport *
1510 make_ofport(const struct odp_port *odp_port)
1512 struct netdev_options netdev_options;
1513 enum netdev_flags flags;
1514 struct ofport *ofport;
1515 struct netdev *netdev;
1518 memset(&netdev_options, 0, sizeof netdev_options);
1519 netdev_options.name = odp_port->devname;
1520 netdev_options.type = odp_port->type;
1521 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1523 error = netdev_open(&netdev_options, &netdev);
1525 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1526 "cannot be opened (%s)",
1527 odp_port->devname, odp_port->port,
1528 odp_port->devname, strerror(error));
1532 ofport = xmalloc(sizeof *ofport);
1533 ofport->netdev = netdev;
1534 ofport->odp_port = odp_port->port;
1535 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1536 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1537 memcpy(ofport->opp.name, odp_port->devname,
1538 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1539 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1541 netdev_get_flags(netdev, &flags);
1542 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1544 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1546 netdev_get_features(netdev,
1547 &ofport->opp.curr, &ofport->opp.advertised,
1548 &ofport->opp.supported, &ofport->opp.peer);
1553 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1555 if (get_port(p, odp_port->port)) {
1556 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1559 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1560 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1569 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1571 const struct ofp_phy_port *a = &a_->opp;
1572 const struct ofp_phy_port *b = &b_->opp;
1574 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1575 return (a->port_no == b->port_no
1576 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1577 && !strcmp(a->name, b->name)
1578 && a->state == b->state
1579 && a->config == b->config
1580 && a->curr == b->curr
1581 && a->advertised == b->advertised
1582 && a->supported == b->supported
1583 && a->peer == b->peer);
1587 send_port_status(struct ofproto *p, const struct ofport *ofport,
1590 /* XXX Should limit the number of queued port status change messages. */
1591 struct ofconn *ofconn;
1592 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1593 struct ofp_port_status *ops;
1596 /* Primary controllers, even slaves, should always get port status
1597 updates. Otherwise obey ofconn_receives_async_msgs(). */
1598 if (ofconn->type != OFCONN_PRIMARY
1599 && !ofconn_receives_async_msgs(ofconn)) {
1603 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1604 ops->reason = reason;
1605 ops->desc = ofport->opp;
1606 hton_ofp_phy_port(&ops->desc);
1607 queue_tx(b, ofconn, NULL);
1612 ofport_install(struct ofproto *p, struct ofport *ofport)
1614 const char *netdev_name = ofport->opp.name;
1616 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1617 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1618 shash_add(&p->port_by_name, netdev_name, ofport);
1620 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1625 ofport_remove(struct ofproto *p, struct ofport *ofport)
1627 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1628 hmap_remove(&p->ports, &ofport->hmap_node);
1629 shash_delete(&p->port_by_name,
1630 shash_find(&p->port_by_name, ofport->opp.name));
1632 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1637 ofport_free(struct ofport *ofport)
1640 netdev_close(ofport->netdev);
1645 static struct ofport *
1646 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1648 struct ofport *port;
1650 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1651 hash_int(odp_port, 0), &ofproto->ports) {
1652 if (port->odp_port == odp_port) {
1660 update_port(struct ofproto *p, const char *devname)
1662 struct odp_port odp_port;
1663 struct ofport *old_ofport;
1664 struct ofport *new_ofport;
1667 COVERAGE_INC(ofproto_update_port);
1669 /* Query the datapath for port information. */
1670 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1672 /* Find the old ofport. */
1673 old_ofport = shash_find_data(&p->port_by_name, devname);
1676 /* There's no port named 'devname' but there might be a port with
1677 * the same port number. This could happen if a port is deleted
1678 * and then a new one added in its place very quickly, or if a port
1679 * is renamed. In the former case we want to send an OFPPR_DELETE
1680 * and an OFPPR_ADD, and in the latter case we want to send a
1681 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1682 * the old port's ifindex against the new port, or perhaps less
1683 * reliably but more portably by comparing the old port's MAC
1684 * against the new port's MAC. However, this code isn't that smart
1685 * and always sends an OFPPR_MODIFY (XXX). */
1686 old_ofport = get_port(p, odp_port.port);
1688 } else if (error != ENOENT && error != ENODEV) {
1689 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1690 "%s", strerror(error));
1694 /* Create a new ofport. */
1695 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1697 /* Eliminate a few pathological cases. */
1698 if (!old_ofport && !new_ofport) {
1700 } else if (old_ofport && new_ofport) {
1701 /* Most of the 'config' bits are OpenFlow soft state, but
1702 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1703 * OpenFlow bits from old_ofport. (make_ofport() only sets
1704 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1705 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1707 if (ofport_equal(old_ofport, new_ofport)) {
1708 /* False alarm--no change. */
1709 ofport_free(new_ofport);
1714 /* Now deal with the normal cases. */
1716 ofport_remove(p, old_ofport);
1719 ofport_install(p, new_ofport);
1721 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1722 (!old_ofport ? OFPPR_ADD
1723 : !new_ofport ? OFPPR_DELETE
1725 ofport_free(old_ofport);
1729 init_ports(struct ofproto *p)
1731 struct odp_port *ports;
1736 error = dpif_port_list(p->dpif, &ports, &n_ports);
1741 for (i = 0; i < n_ports; i++) {
1742 const struct odp_port *odp_port = &ports[i];
1743 if (!ofport_conflicts(p, odp_port)) {
1744 struct ofport *ofport = make_ofport(odp_port);
1746 ofport_install(p, ofport);
1754 static struct ofconn *
1755 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1757 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1758 ofconn->ofproto = p;
1759 list_push_back(&p->all_conns, &ofconn->node);
1760 ofconn->rconn = rconn;
1761 ofconn->type = type;
1762 ofconn->flow_format = NXFF_OPENFLOW10;
1763 ofconn->role = NX_ROLE_OTHER;
1764 ofconn->packet_in_counter = rconn_packet_counter_create ();
1765 ofconn->pktbuf = NULL;
1766 ofconn->miss_send_len = 0;
1767 ofconn->reply_counter = rconn_packet_counter_create ();
1772 ofconn_destroy(struct ofconn *ofconn)
1774 if (ofconn->type == OFCONN_PRIMARY) {
1775 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1777 discovery_destroy(ofconn->discovery);
1779 list_remove(&ofconn->node);
1780 switch_status_unregister(ofconn->ss);
1781 rconn_destroy(ofconn->rconn);
1782 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1783 rconn_packet_counter_destroy(ofconn->reply_counter);
1784 pktbuf_destroy(ofconn->pktbuf);
1789 ofconn_run(struct ofconn *ofconn)
1791 struct ofproto *p = ofconn->ofproto;
1795 if (ofconn->discovery) {
1796 char *controller_name;
1797 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1798 discovery_question_connectivity(ofconn->discovery);
1800 if (discovery_run(ofconn->discovery, &controller_name)) {
1801 if (controller_name) {
1802 char *ofconn_name = ofconn_make_name(p, controller_name);
1803 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1806 rconn_disconnect(ofconn->rconn);
1811 for (i = 0; i < N_SCHEDULERS; i++) {
1812 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1815 rconn_run(ofconn->rconn);
1817 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1818 /* Limit the number of iterations to prevent other tasks from
1820 for (iteration = 0; iteration < 50; iteration++) {
1821 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1826 fail_open_maybe_recover(p->fail_open);
1828 handle_openflow(ofconn, of_msg);
1829 ofpbuf_delete(of_msg);
1833 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1834 ofconn_destroy(ofconn);
1839 ofconn_wait(struct ofconn *ofconn)
1843 if (ofconn->discovery) {
1844 discovery_wait(ofconn->discovery);
1846 for (i = 0; i < N_SCHEDULERS; i++) {
1847 pinsched_wait(ofconn->schedulers[i]);
1849 rconn_run_wait(ofconn->rconn);
1850 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1851 rconn_recv_wait(ofconn->rconn);
1853 COVERAGE_INC(ofproto_ofconn_stuck);
1857 /* Returns true if 'ofconn' should receive asynchronous messages. */
1859 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1861 if (ofconn->type == OFCONN_PRIMARY) {
1862 /* Primary controllers always get asynchronous messages unless they
1863 * have configured themselves as "slaves". */
1864 return ofconn->role != NX_ROLE_SLAVE;
1866 /* Service connections don't get asynchronous messages unless they have
1867 * explicitly asked for them by setting a nonzero miss send length. */
1868 return ofconn->miss_send_len > 0;
1872 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1873 * and 'target', suitable for use in log messages for identifying the
1876 * The name is dynamically allocated. The caller should free it (with free())
1877 * when it is no longer needed. */
1879 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1881 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1885 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1889 for (i = 0; i < N_SCHEDULERS; i++) {
1890 struct pinsched **s = &ofconn->schedulers[i];
1894 *s = pinsched_create(rate, burst,
1895 ofconn->ofproto->switch_status);
1897 pinsched_set_limits(*s, rate, burst);
1900 pinsched_destroy(*s);
1907 ofservice_reconfigure(struct ofservice *ofservice,
1908 const struct ofproto_controller *c)
1910 ofservice->probe_interval = c->probe_interval;
1911 ofservice->rate_limit = c->rate_limit;
1912 ofservice->burst_limit = c->burst_limit;
1915 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1916 * positive errno value. */
1918 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1920 struct ofservice *ofservice;
1921 struct pvconn *pvconn;
1924 error = pvconn_open(c->target, &pvconn);
1929 ofservice = xzalloc(sizeof *ofservice);
1930 hmap_insert(&ofproto->services, &ofservice->node,
1931 hash_string(c->target, 0));
1932 ofservice->pvconn = pvconn;
1934 ofservice_reconfigure(ofservice, c);
1940 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1942 hmap_remove(&ofproto->services, &ofservice->node);
1943 pvconn_close(ofservice->pvconn);
1947 /* Finds and returns the ofservice within 'ofproto' that has the given
1948 * 'target', or a null pointer if none exists. */
1949 static struct ofservice *
1950 ofservice_lookup(struct ofproto *ofproto, const char *target)
1952 struct ofservice *ofservice;
1954 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1955 &ofproto->services) {
1956 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1963 /* Returns true if 'rule' should be hidden from the controller.
1965 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1966 * (e.g. by in-band control) and are intentionally hidden from the
1969 rule_is_hidden(const struct rule *rule)
1971 return rule->cr.priority > UINT16_MAX;
1974 /* Creates and returns a new rule initialized as specified.
1976 * The caller is responsible for inserting the rule into the classifier (with
1977 * rule_insert()). */
1978 static struct rule *
1979 rule_create(const struct cls_rule *cls_rule,
1980 const union ofp_action *actions, size_t n_actions,
1981 uint16_t idle_timeout, uint16_t hard_timeout,
1982 ovs_be64 flow_cookie, bool send_flow_removed)
1984 struct rule *rule = xzalloc(sizeof *rule);
1985 rule->cr = *cls_rule;
1986 rule->idle_timeout = idle_timeout;
1987 rule->hard_timeout = hard_timeout;
1988 rule->flow_cookie = flow_cookie;
1989 rule->used = rule->created = time_msec();
1990 rule->send_flow_removed = send_flow_removed;
1991 list_init(&rule->facets);
1992 if (n_actions > 0) {
1993 rule->n_actions = n_actions;
1994 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
2000 static struct rule *
2001 rule_from_cls_rule(const struct cls_rule *cls_rule)
2003 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
2007 rule_free(struct rule *rule)
2009 free(rule->actions);
2013 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2014 * destroying any that no longer has a rule (which is probably all of them).
2016 * The caller must have already removed 'rule' from the classifier. */
2018 rule_destroy(struct ofproto *ofproto, struct rule *rule)
2020 struct facet *facet, *next_facet;
2021 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
2022 facet_revalidate(ofproto, facet);
2027 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2028 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2031 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
2033 const union ofp_action *oa;
2034 struct actions_iterator i;
2036 if (out_port == htons(OFPP_NONE)) {
2039 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
2040 oa = actions_next(&i)) {
2041 if (action_outputs_to_port(oa, out_port)) {
2048 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2049 * 'packet', which arrived on 'in_port'.
2051 * Takes ownership of 'packet'. */
2053 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
2054 const struct nlattr *odp_actions, size_t actions_len,
2055 struct ofpbuf *packet)
2057 if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
2058 && odp_actions->nla_type == ODPAT_CONTROLLER) {
2059 /* As an optimization, avoid a round-trip from userspace to kernel to
2060 * userspace. This also avoids possibly filling up kernel packet
2061 * buffers along the way. */
2062 struct odp_msg *msg;
2064 msg = ofpbuf_push_uninit(packet, sizeof *msg);
2065 msg->type = _ODPL_ACTION_NR;
2066 msg->length = sizeof(struct odp_msg) + packet->size;
2067 msg->port = in_port;
2068 msg->arg = nl_attr_get_u64(odp_actions);
2070 send_packet_in(ofproto, packet);
2076 error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
2077 ofpbuf_delete(packet);
2082 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2083 * statistics appropriately. 'packet' must have at least sizeof(struct
2084 * ofp_packet_in) bytes of headroom.
2086 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2087 * applying flow_extract() to 'packet' would yield the same flow as
2090 * 'facet' must have accurately composed ODP actions; that is, it must not be
2091 * in need of revalidation.
2093 * Takes ownership of 'packet'. */
2095 facet_execute(struct ofproto *ofproto, struct facet *facet,
2096 struct ofpbuf *packet)
2098 struct odp_flow_stats stats;
2100 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2102 flow_extract_stats(&facet->flow, packet, &stats);
2103 if (execute_odp_actions(ofproto, facet->flow.in_port,
2104 facet->actions, facet->actions_len, packet)) {
2105 facet_update_stats(ofproto, facet, &stats);
2106 facet->used = time_msec();
2107 netflow_flow_update_time(ofproto->netflow,
2108 &facet->nf_flow, facet->used);
2112 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2113 * statistics (or the statistics for one of its facets) appropriately.
2114 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2116 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2117 * with statistics for 'packet' either way.
2119 * Takes ownership of 'packet'. */
2121 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2122 struct ofpbuf *packet)
2124 struct action_xlate_ctx ctx;
2125 struct ofpbuf *odp_actions;
2126 struct facet *facet;
2130 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2132 flow_extract(packet, 0, in_port, &flow);
2134 /* First look for a related facet. If we find one, account it to that. */
2135 facet = facet_lookup_valid(ofproto, &flow);
2136 if (facet && facet->rule == rule) {
2137 facet_execute(ofproto, facet, packet);
2141 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2142 * create a new facet for it and use that. */
2143 if (rule_lookup(ofproto, &flow) == rule) {
2144 facet = facet_create(ofproto, rule, &flow, packet);
2145 facet_execute(ofproto, facet, packet);
2146 facet_install(ofproto, facet, true);
2150 /* We can't account anything to a facet. If we were to try, then that
2151 * facet would have a non-matching rule, busting our invariants. */
2152 action_xlate_ctx_init(&ctx, ofproto, &flow, packet);
2153 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2154 size = packet->size;
2155 if (execute_odp_actions(ofproto, in_port, odp_actions->data,
2156 odp_actions->size, packet)) {
2157 rule->used = time_msec();
2158 rule->packet_count++;
2159 rule->byte_count += size;
2161 ofpbuf_delete(odp_actions);
2164 /* Inserts 'rule' into 'p''s flow table. */
2166 rule_insert(struct ofproto *p, struct rule *rule)
2168 struct rule *displaced_rule;
2170 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2171 if (displaced_rule) {
2172 rule_destroy(p, displaced_rule);
2174 p->need_revalidate = true;
2177 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2178 * 'flow' and an example 'packet' within that flow.
2180 * The caller must already have determined that no facet with an identical
2181 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2182 * 'ofproto''s classifier table. */
2183 static struct facet *
2184 facet_create(struct ofproto *ofproto, struct rule *rule,
2185 const struct flow *flow, const struct ofpbuf *packet)
2187 struct facet *facet;
2189 facet = xzalloc(sizeof *facet);
2190 facet->used = time_msec();
2191 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2192 list_push_back(&rule->facets, &facet->list_node);
2194 facet->flow = *flow;
2195 netflow_flow_init(&facet->nf_flow);
2196 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2198 facet_make_actions(ofproto, facet, packet);
2204 facet_free(struct facet *facet)
2206 free(facet->actions);
2210 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2212 * - Removes 'rule' from the classifier.
2214 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2215 * destroys them), via rule_destroy().
2218 rule_remove(struct ofproto *ofproto, struct rule *rule)
2220 COVERAGE_INC(ofproto_del_rule);
2221 ofproto->need_revalidate = true;
2222 classifier_remove(&ofproto->cls, &rule->cr);
2223 rule_destroy(ofproto, rule);
2226 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2228 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2229 * rule's statistics, via facet_uninstall().
2231 * - Removes 'facet' from its rule and from ofproto->facets.
2234 facet_remove(struct ofproto *ofproto, struct facet *facet)
2236 facet_uninstall(ofproto, facet);
2237 facet_flush_stats(ofproto, facet);
2238 hmap_remove(&ofproto->facets, &facet->hmap_node);
2239 list_remove(&facet->list_node);
2243 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2245 facet_make_actions(struct ofproto *p, struct facet *facet,
2246 const struct ofpbuf *packet)
2248 const struct rule *rule = facet->rule;
2249 struct ofpbuf *odp_actions;
2250 struct action_xlate_ctx ctx;
2252 action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
2253 odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions);
2254 facet->tags = ctx.tags;
2255 facet->may_install = ctx.may_set_up_flow;
2256 facet->nf_flow.output_iface = ctx.nf_output_iface;
2258 if (facet->actions_len != odp_actions->size
2259 || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
2260 free(facet->actions);
2261 facet->actions_len = odp_actions->size;
2262 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2265 ofpbuf_delete(odp_actions);
2269 facet_put__(struct ofproto *ofproto, struct facet *facet, int flags,
2270 struct odp_flow_put *put)
2272 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2273 odp_flow_key_from_flow(&put->flow.key, &facet->flow);
2274 put->flow.actions = facet->actions;
2275 put->flow.actions_len = facet->actions_len;
2276 put->flow.flags = 0;
2278 return dpif_flow_put(ofproto->dpif, put);
2281 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2282 * 'zero_stats' is true, clears any existing statistics from the datapath for
2285 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2287 if (facet->may_install) {
2288 struct odp_flow_put put;
2291 flags = ODPPF_CREATE | ODPPF_MODIFY;
2293 flags |= ODPPF_ZERO_STATS;
2295 if (!facet_put__(p, facet, flags, &put)) {
2296 facet->installed = true;
2301 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2302 * to the accounting hook function in the ofhooks structure. */
2304 facet_account(struct ofproto *ofproto,
2305 struct facet *facet, uint64_t extra_bytes)
2307 uint64_t total_bytes = facet->byte_count + extra_bytes;
2309 if (ofproto->ofhooks->account_flow_cb
2310 && total_bytes > facet->accounted_bytes)
2312 ofproto->ofhooks->account_flow_cb(
2313 &facet->flow, facet->tags, facet->actions, facet->actions_len,
2314 total_bytes - facet->accounted_bytes, ofproto->aux);
2315 facet->accounted_bytes = total_bytes;
2319 /* If 'rule' is installed in the datapath, uninstalls it. */
2321 facet_uninstall(struct ofproto *p, struct facet *facet)
2323 if (facet->installed) {
2324 struct odp_flow odp_flow;
2326 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
2327 odp_flow.actions = NULL;
2328 odp_flow.actions_len = 0;
2330 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2331 facet_update_stats(p, facet, &odp_flow.stats);
2333 facet->installed = false;
2337 /* Returns true if the only action for 'facet' is to send to the controller.
2338 * (We don't report NetFlow expiration messages for such facets because they
2339 * are just part of the control logic for the network, not real traffic). */
2341 facet_is_controller_flow(struct facet *facet)
2344 && facet->rule->n_actions == 1
2345 && action_outputs_to_port(&facet->rule->actions[0],
2346 htons(OFPP_CONTROLLER)));
2349 /* Folds all of 'facet''s statistics into its rule. Also updates the
2350 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2352 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2354 facet_account(ofproto, facet, 0);
2356 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2357 struct ofexpired expired;
2358 expired.flow = facet->flow;
2359 expired.packet_count = facet->packet_count;
2360 expired.byte_count = facet->byte_count;
2361 expired.used = facet->used;
2362 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2365 facet->rule->packet_count += facet->packet_count;
2366 facet->rule->byte_count += facet->byte_count;
2368 /* Reset counters to prevent double counting if 'facet' ever gets
2370 facet->packet_count = 0;
2371 facet->byte_count = 0;
2372 facet->accounted_bytes = 0;
2374 netflow_flow_clear(&facet->nf_flow);
2377 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2378 * Returns it if found, otherwise a null pointer.
2380 * The returned facet might need revalidation; use facet_lookup_valid()
2381 * instead if that is important. */
2382 static struct facet *
2383 facet_find(struct ofproto *ofproto, const struct flow *flow)
2385 struct facet *facet;
2387 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2389 if (flow_equal(flow, &facet->flow)) {
2397 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2398 * Returns it if found, otherwise a null pointer.
2400 * The returned facet is guaranteed to be valid. */
2401 static struct facet *
2402 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2404 struct facet *facet = facet_find(ofproto, flow);
2406 /* The facet we found might not be valid, since we could be in need of
2407 * revalidation. If it is not valid, don't return it. */
2409 && ofproto->need_revalidate
2410 && !facet_revalidate(ofproto, facet)) {
2411 COVERAGE_INC(ofproto_invalidated);
2418 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2420 * - If the rule found is different from 'facet''s current rule, moves
2421 * 'facet' to the new rule and recompiles its actions.
2423 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2424 * where it is and recompiles its actions anyway.
2426 * - If there is none, destroys 'facet'.
2428 * Returns true if 'facet' still exists, false if it has been destroyed. */
2430 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2432 struct action_xlate_ctx ctx;
2433 struct ofpbuf *odp_actions;
2434 struct rule *new_rule;
2435 bool actions_changed;
2437 COVERAGE_INC(facet_revalidate);
2439 /* Determine the new rule. */
2440 new_rule = rule_lookup(ofproto, &facet->flow);
2442 /* No new rule, so delete the facet. */
2443 facet_remove(ofproto, facet);
2447 /* Calculate new ODP actions.
2449 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2450 * emit a NetFlow expiration and, if so, we need to have the old state
2451 * around to properly compose it. */
2452 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
2453 odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions);
2454 actions_changed = (facet->actions_len != odp_actions->size
2455 || memcmp(facet->actions, odp_actions->data,
2456 facet->actions_len));
2458 /* If the ODP actions changed or the installability changed, then we need
2459 * to talk to the datapath. */
2460 if (actions_changed || facet->may_install != facet->installed) {
2461 if (facet->may_install) {
2462 struct odp_flow_put put;
2464 memset(&put.flow.stats, 0, sizeof put.flow.stats);
2465 odp_flow_key_from_flow(&put.flow.key, &facet->flow);
2466 put.flow.actions = odp_actions->data;
2467 put.flow.actions_len = odp_actions->size;
2469 put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS;
2470 dpif_flow_put(ofproto->dpif, &put);
2472 facet_update_stats(ofproto, facet, &put.flow.stats);
2474 facet_uninstall(ofproto, facet);
2477 /* The datapath flow is gone or has zeroed stats, so push stats out of
2478 * 'facet' into 'rule'. */
2479 facet_flush_stats(ofproto, facet);
2482 /* Update 'facet' now that we've taken care of all the old state. */
2483 facet->tags = ctx.tags;
2484 facet->nf_flow.output_iface = ctx.nf_output_iface;
2485 facet->may_install = ctx.may_set_up_flow;
2486 if (actions_changed) {
2487 free(facet->actions);
2488 facet->actions_len = odp_actions->size;
2489 facet->actions = xmemdup(odp_actions->data, odp_actions->size);
2491 if (facet->rule != new_rule) {
2492 COVERAGE_INC(facet_changed_rule);
2493 list_remove(&facet->list_node);
2494 list_push_back(&new_rule->facets, &facet->list_node);
2495 facet->rule = new_rule;
2496 facet->used = new_rule->created;
2499 ofpbuf_delete(odp_actions);
2505 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2506 struct rconn_packet_counter *counter)
2508 update_openflow_length(msg);
2509 if (rconn_send(ofconn->rconn, msg, counter)) {
2515 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2518 struct ofpbuf *buf = make_ofp_error_msg(error, oh);
2520 COVERAGE_INC(ofproto_error);
2521 queue_tx(buf, ofconn, ofconn->reply_counter);
2526 hton_ofp_phy_port(struct ofp_phy_port *opp)
2528 opp->port_no = htons(opp->port_no);
2529 opp->config = htonl(opp->config);
2530 opp->state = htonl(opp->state);
2531 opp->curr = htonl(opp->curr);
2532 opp->advertised = htonl(opp->advertised);
2533 opp->supported = htonl(opp->supported);
2534 opp->peer = htonl(opp->peer);
2538 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2540 queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter);
2545 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2547 struct ofp_switch_features *osf;
2549 struct ofport *port;
2551 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2552 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2553 osf->n_buffers = htonl(pktbuf_capacity());
2555 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2556 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2557 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2558 (1u << OFPAT_SET_VLAN_VID) |
2559 (1u << OFPAT_SET_VLAN_PCP) |
2560 (1u << OFPAT_STRIP_VLAN) |
2561 (1u << OFPAT_SET_DL_SRC) |
2562 (1u << OFPAT_SET_DL_DST) |
2563 (1u << OFPAT_SET_NW_SRC) |
2564 (1u << OFPAT_SET_NW_DST) |
2565 (1u << OFPAT_SET_NW_TOS) |
2566 (1u << OFPAT_SET_TP_SRC) |
2567 (1u << OFPAT_SET_TP_DST) |
2568 (1u << OFPAT_ENQUEUE));
2570 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2571 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2574 queue_tx(buf, ofconn, ofconn->reply_counter);
2579 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2582 struct ofp_switch_config *osc;
2586 /* Figure out flags. */
2587 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2588 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2591 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2592 osc->flags = htons(flags);
2593 osc->miss_send_len = htons(ofconn->miss_send_len);
2594 queue_tx(buf, ofconn, ofconn->reply_counter);
2600 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2602 uint16_t flags = ntohs(osc->flags);
2604 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2605 switch (flags & OFPC_FRAG_MASK) {
2606 case OFPC_FRAG_NORMAL:
2607 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2609 case OFPC_FRAG_DROP:
2610 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2613 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2619 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2624 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2625 * flow translation. */
2626 #define MAX_RESUBMIT_RECURSION 8
2628 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2629 struct action_xlate_ctx *ctx);
2632 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2634 const struct ofport *ofport = get_port(ctx->ofproto, port);
2637 if (ofport->opp.config & OFPPC_NO_FWD) {
2638 /* Forwarding disabled on port. */
2643 * We don't have an ofport record for this port, but it doesn't hurt to
2644 * allow forwarding to it anyhow. Maybe such a port will appear later
2645 * and we're pre-populating the flow table.
2649 nl_msg_put_u32(ctx->odp_actions, ODPAT_OUTPUT, port);
2650 ctx->nf_output_iface = port;
2653 static struct rule *
2654 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2656 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2660 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2662 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2663 uint16_t old_in_port;
2666 /* Look up a flow with 'in_port' as the input port. Then restore the
2667 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2668 * have surprising behavior). */
2669 old_in_port = ctx->flow.in_port;
2670 ctx->flow.in_port = in_port;
2671 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2672 ctx->flow.in_port = old_in_port;
2674 if (ctx->resubmit_hook) {
2675 ctx->resubmit_hook(ctx, rule);
2680 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2684 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2686 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2687 MAX_RESUBMIT_RECURSION);
2692 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2693 uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
2695 struct ofport *ofport;
2697 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2698 uint16_t odp_port = ofport->odp_port;
2699 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2700 nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, odp_port);
2703 *nf_output_iface = NF_OUT_FLOOD;
2707 xlate_output_action__(struct action_xlate_ctx *ctx,
2708 uint16_t port, uint16_t max_len)
2711 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2713 ctx->nf_output_iface = NF_OUT_DROP;
2717 add_output_action(ctx, ctx->flow.in_port);
2720 xlate_table_action(ctx, ctx->flow.in_port);
2723 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2724 ctx->odp_actions, &ctx->tags,
2725 &ctx->nf_output_iface,
2726 ctx->ofproto->aux)) {
2727 COVERAGE_INC(ofproto_uninstallable);
2728 ctx->may_set_up_flow = false;
2732 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2733 &ctx->nf_output_iface, ctx->odp_actions);
2736 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2737 &ctx->nf_output_iface, ctx->odp_actions);
2739 case OFPP_CONTROLLER:
2740 nl_msg_put_u64(ctx->odp_actions, ODPAT_CONTROLLER, max_len);
2743 add_output_action(ctx, ODPP_LOCAL);
2746 odp_port = ofp_port_to_odp_port(port);
2747 if (odp_port != ctx->flow.in_port) {
2748 add_output_action(ctx, odp_port);
2753 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2754 ctx->nf_output_iface = NF_OUT_FLOOD;
2755 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2756 ctx->nf_output_iface = prev_nf_output_iface;
2757 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2758 ctx->nf_output_iface != NF_OUT_FLOOD) {
2759 ctx->nf_output_iface = NF_OUT_MULTI;
2764 xlate_output_action(struct action_xlate_ctx *ctx,
2765 const struct ofp_action_output *oao)
2767 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2770 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2771 * optimization, because we're going to add another action that sets the
2772 * priority immediately after, or because there are no actions following the
2775 remove_pop_action(struct action_xlate_ctx *ctx)
2777 if (ctx->odp_actions->size == ctx->last_pop_priority) {
2778 ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
2779 ctx->last_pop_priority = -1;
2784 add_pop_action(struct action_xlate_ctx *ctx)
2786 if (ctx->odp_actions->size != ctx->last_pop_priority) {
2787 nl_msg_put_flag(ctx->odp_actions, ODPAT_POP_PRIORITY);
2788 ctx->last_pop_priority = ctx->odp_actions->size;
2793 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2794 const struct ofp_action_enqueue *oae)
2796 uint16_t ofp_port, odp_port;
2800 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2803 /* Fall back to ordinary output action. */
2804 xlate_output_action__(ctx, ntohs(oae->port), 0);
2808 /* Figure out ODP output port. */
2809 ofp_port = ntohs(oae->port);
2810 if (ofp_port != OFPP_IN_PORT) {
2811 odp_port = ofp_port_to_odp_port(ofp_port);
2813 odp_port = ctx->flow.in_port;
2816 /* Add ODP actions. */
2817 remove_pop_action(ctx);
2818 nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority);
2819 add_output_action(ctx, odp_port);
2820 add_pop_action(ctx);
2822 /* Update NetFlow output port. */
2823 if (ctx->nf_output_iface == NF_OUT_DROP) {
2824 ctx->nf_output_iface = odp_port;
2825 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2826 ctx->nf_output_iface = NF_OUT_MULTI;
2831 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2832 const struct nx_action_set_queue *nasq)
2837 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2840 /* Couldn't translate queue to a priority, so ignore. A warning
2841 * has already been logged. */
2845 remove_pop_action(ctx);
2846 nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority);
2850 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2852 ovs_be16 tci = ctx->flow.vlan_tci;
2853 if (!(tci & htons(VLAN_CFI))) {
2854 nl_msg_put_flag(ctx->odp_actions, ODPAT_STRIP_VLAN);
2856 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_DL_TCI,
2857 tci & ~htons(VLAN_CFI));
2862 xlate_reg_move_action(struct action_xlate_ctx *ctx,
2863 const struct nx_action_reg_move *narm)
2865 ovs_be16 old_tci = ctx->flow.vlan_tci;
2867 nxm_execute_reg_move(narm, &ctx->flow);
2869 if (ctx->flow.vlan_tci != old_tci) {
2870 xlate_set_dl_tci(ctx);
2875 xlate_nicira_action(struct action_xlate_ctx *ctx,
2876 const struct nx_action_header *nah)
2878 const struct nx_action_resubmit *nar;
2879 const struct nx_action_set_tunnel *nast;
2880 const struct nx_action_set_queue *nasq;
2881 const struct nx_action_multipath *nam;
2882 enum nx_action_subtype subtype = ntohs(nah->subtype);
2885 assert(nah->vendor == htonl(NX_VENDOR_ID));
2887 case NXAST_RESUBMIT:
2888 nar = (const struct nx_action_resubmit *) nah;
2889 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2892 case NXAST_SET_TUNNEL:
2893 nast = (const struct nx_action_set_tunnel *) nah;
2894 tun_id = htonll(ntohl(nast->tun_id));
2895 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id);
2896 ctx->flow.tun_id = tun_id;
2899 case NXAST_DROP_SPOOFED_ARP:
2900 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2901 nl_msg_put_flag(ctx->odp_actions, ODPAT_DROP_SPOOFED_ARP);
2905 case NXAST_SET_QUEUE:
2906 nasq = (const struct nx_action_set_queue *) nah;
2907 xlate_set_queue_action(ctx, nasq);
2910 case NXAST_POP_QUEUE:
2911 add_pop_action(ctx);
2914 case NXAST_REG_MOVE:
2915 xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah);
2918 case NXAST_REG_LOAD:
2919 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2923 /* Nothing to do. */
2926 case NXAST_SET_TUNNEL64:
2927 tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
2928 nl_msg_put_be64(ctx->odp_actions, ODPAT_SET_TUNNEL, tun_id);
2929 ctx->flow.tun_id = tun_id;
2932 case NXAST_MULTIPATH:
2933 nam = (const struct nx_action_multipath *) nah;
2934 multipath_execute(nam, &ctx->flow);
2937 /* If you add a new action here that modifies flow data, don't forget to
2938 * update the flow key in ctx->flow at the same time. */
2940 case NXAST_SNAT__OBSOLETE:
2942 VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
2948 do_xlate_actions(const union ofp_action *in, size_t n_in,
2949 struct action_xlate_ctx *ctx)
2951 struct actions_iterator iter;
2952 const union ofp_action *ia;
2953 const struct ofport *port;
2955 port = get_port(ctx->ofproto, ctx->flow.in_port);
2956 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2957 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2958 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2959 /* Drop this flow. */
2963 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2964 enum ofp_action_type type = ntohs(ia->type);
2965 const struct ofp_action_dl_addr *oada;
2969 xlate_output_action(ctx, &ia->output);
2972 case OFPAT_SET_VLAN_VID:
2973 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
2974 ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
2975 xlate_set_dl_tci(ctx);
2978 case OFPAT_SET_VLAN_PCP:
2979 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
2980 ctx->flow.vlan_tci |= htons(
2981 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
2982 xlate_set_dl_tci(ctx);
2985 case OFPAT_STRIP_VLAN:
2986 ctx->flow.vlan_tci = htons(0);
2987 xlate_set_dl_tci(ctx);
2990 case OFPAT_SET_DL_SRC:
2991 oada = ((struct ofp_action_dl_addr *) ia);
2992 nl_msg_put_unspec(ctx->odp_actions, ODPAT_SET_DL_SRC,
2993 oada->dl_addr, ETH_ADDR_LEN);
2994 memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
2997 case OFPAT_SET_DL_DST:
2998 oada = ((struct ofp_action_dl_addr *) ia);
2999 nl_msg_put_unspec(ctx->odp_actions, ODPAT_SET_DL_DST,
3000 oada->dl_addr, ETH_ADDR_LEN);
3001 memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
3004 case OFPAT_SET_NW_SRC:
3005 nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_SRC,
3006 ia->nw_addr.nw_addr);
3007 ctx->flow.nw_src = ia->nw_addr.nw_addr;
3010 case OFPAT_SET_NW_DST:
3011 nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_DST,
3012 ia->nw_addr.nw_addr);
3013 ctx->flow.nw_dst = ia->nw_addr.nw_addr;
3016 case OFPAT_SET_NW_TOS:
3017 nl_msg_put_u8(ctx->odp_actions, ODPAT_SET_NW_TOS,
3019 ctx->flow.nw_tos = ia->nw_tos.nw_tos;
3022 case OFPAT_SET_TP_SRC:
3023 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_SRC,
3024 ia->tp_port.tp_port);
3025 ctx->flow.tp_src = ia->tp_port.tp_port;
3028 case OFPAT_SET_TP_DST:
3029 nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_DST,
3030 ia->tp_port.tp_port);
3031 ctx->flow.tp_dst = ia->tp_port.tp_port;
3035 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
3039 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
3043 VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
3050 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
3051 struct ofproto *ofproto, const struct flow *flow,
3052 const struct ofpbuf *packet)
3054 ctx->ofproto = ofproto;
3056 ctx->packet = packet;
3057 ctx->resubmit_hook = NULL;
3060 static struct ofpbuf *
3061 xlate_actions(struct action_xlate_ctx *ctx,
3062 const union ofp_action *in, size_t n_in)
3064 COVERAGE_INC(ofproto_ofp2odp);
3066 ctx->odp_actions = ofpbuf_new(512);
3068 ctx->may_set_up_flow = true;
3069 ctx->nf_output_iface = NF_OUT_DROP;
3071 ctx->last_pop_priority = -1;
3072 do_xlate_actions(in, n_in, ctx);
3073 remove_pop_action(ctx);
3075 /* Check with in-band control to see if we're allowed to set up this
3077 if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow,
3078 ctx->odp_actions->data, ctx->odp_actions->size)) {
3079 ctx->may_set_up_flow = false;
3082 return ctx->odp_actions;
3085 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3086 * error message code (composed with ofp_mkerr()) for the caller to propagate
3087 * upward. Otherwise, returns 0.
3089 * The log message mentions 'msg_type'. */
3091 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3093 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3094 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3095 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3098 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3105 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3107 struct ofproto *p = ofconn->ofproto;
3108 struct ofp_packet_out *opo;
3109 struct ofpbuf payload, *buffer;
3110 union ofp_action *ofp_actions;
3111 struct action_xlate_ctx ctx;
3112 struct ofpbuf *odp_actions;
3113 struct ofpbuf request;
3115 size_t n_ofp_actions;
3119 COVERAGE_INC(ofproto_packet_out);
3121 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3126 /* Get ofp_packet_out. */
3127 ofpbuf_use_const(&request, oh, ntohs(oh->length));
3128 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
3131 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3132 &ofp_actions, &n_ofp_actions);
3138 if (opo->buffer_id != htonl(UINT32_MAX)) {
3139 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3141 if (error || !buffer) {
3150 /* Extract flow, check actions. */
3151 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3153 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3159 action_xlate_ctx_init(&ctx, p, &flow, &payload);
3160 odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
3161 dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload);
3162 ofpbuf_delete(odp_actions);
3165 ofpbuf_delete(buffer);
3170 update_port_config(struct ofproto *p, struct ofport *port,
3171 uint32_t config, uint32_t mask)
3173 mask &= config ^ port->opp.config;
3174 if (mask & OFPPC_PORT_DOWN) {
3175 if (config & OFPPC_PORT_DOWN) {
3176 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3178 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3181 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3182 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3183 if (mask & REVALIDATE_BITS) {
3184 COVERAGE_INC(ofproto_costly_flags);
3185 port->opp.config ^= mask & REVALIDATE_BITS;
3186 p->need_revalidate = true;
3188 #undef REVALIDATE_BITS
3189 if (mask & OFPPC_NO_PACKET_IN) {
3190 port->opp.config ^= OFPPC_NO_PACKET_IN;
3195 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3197 struct ofproto *p = ofconn->ofproto;
3198 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
3199 struct ofport *port;
3202 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3207 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3209 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3210 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3211 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3213 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3214 if (opm->advertise) {
3215 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3221 static struct ofpbuf *
3222 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3224 struct ofp_stats_reply *osr;
3227 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3228 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3230 osr->flags = htons(0);
3234 static struct ofpbuf *
3235 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
3237 const struct ofp_stats_request *osr
3238 = (const struct ofp_stats_request *) request;
3239 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
3243 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3244 struct ofpbuf **msgp)
3246 struct ofpbuf *msg = *msgp;
3247 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3248 if (nbytes + msg->size > UINT16_MAX) {
3249 struct ofp_stats_reply *reply = msg->data;
3250 reply->flags = htons(OFPSF_REPLY_MORE);
3251 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3252 queue_tx(msg, ofconn, ofconn->reply_counter);
3254 return ofpbuf_put_uninit(*msgp, nbytes);
3257 static struct ofpbuf *
3258 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3260 struct nicira_stats_msg *nsm;
3263 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3264 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3265 nsm->type = htons(OFPST_VENDOR);
3266 nsm->flags = htons(0);
3267 nsm->vendor = htonl(NX_VENDOR_ID);
3268 nsm->subtype = subtype;
3272 static struct ofpbuf *
3273 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3275 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3279 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3280 struct ofpbuf **msgp)
3282 struct ofpbuf *msg = *msgp;
3283 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3284 if (nbytes + msg->size > UINT16_MAX) {
3285 struct nicira_stats_msg *reply = msg->data;
3286 reply->flags = htons(OFPSF_REPLY_MORE);
3287 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3288 queue_tx(msg, ofconn, ofconn->reply_counter);
3290 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3294 handle_desc_stats_request(struct ofconn *ofconn,
3295 const struct ofp_header *request)
3297 struct ofproto *p = ofconn->ofproto;
3298 struct ofp_desc_stats *ods;
3301 msg = start_ofp_stats_reply(request, sizeof *ods);
3302 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3303 memset(ods, 0, sizeof *ods);
3304 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3305 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3306 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3307 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3308 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3309 queue_tx(msg, ofconn, ofconn->reply_counter);
3315 handle_table_stats_request(struct ofconn *ofconn,
3316 const struct ofp_header *request)
3318 struct ofproto *p = ofconn->ofproto;
3319 struct ofp_table_stats *ots;
3322 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3324 /* Classifier table. */
3325 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3326 memset(ots, 0, sizeof *ots);
3327 strcpy(ots->name, "classifier");
3328 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3329 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3330 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3331 ots->active_count = htonl(classifier_count(&p->cls));
3332 ots->lookup_count = htonll(0); /* XXX */
3333 ots->matched_count = htonll(0); /* XXX */
3335 queue_tx(msg, ofconn, ofconn->reply_counter);
3340 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3341 struct ofpbuf **msgp)
3343 struct netdev_stats stats;
3344 struct ofp_port_stats *ops;
3346 /* Intentionally ignore return value, since errors will set
3347 * 'stats' to all-1s, which is correct for OpenFlow, and
3348 * netdev_get_stats() will log errors. */
3349 netdev_get_stats(port->netdev, &stats);
3351 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3352 ops->port_no = htons(port->opp.port_no);
3353 memset(ops->pad, 0, sizeof ops->pad);
3354 ops->rx_packets = htonll(stats.rx_packets);
3355 ops->tx_packets = htonll(stats.tx_packets);
3356 ops->rx_bytes = htonll(stats.rx_bytes);
3357 ops->tx_bytes = htonll(stats.tx_bytes);
3358 ops->rx_dropped = htonll(stats.rx_dropped);
3359 ops->tx_dropped = htonll(stats.tx_dropped);
3360 ops->rx_errors = htonll(stats.rx_errors);
3361 ops->tx_errors = htonll(stats.tx_errors);
3362 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3363 ops->rx_over_err = htonll(stats.rx_over_errors);
3364 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3365 ops->collisions = htonll(stats.collisions);
3369 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3371 struct ofproto *p = ofconn->ofproto;
3372 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
3373 struct ofp_port_stats *ops;
3375 struct ofport *port;
3377 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
3378 if (psr->port_no != htons(OFPP_NONE)) {
3379 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3381 append_port_stat(port, ofconn, &msg);
3384 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3385 append_port_stat(port, ofconn, &msg);
3389 queue_tx(msg, ofconn, ofconn->reply_counter);
3393 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3394 * '*packet_countp' and '*byte_countp'. The returned statistics include
3395 * statistics for all of 'rule''s facets. */
3397 query_stats(struct ofproto *p, struct rule *rule,
3398 uint64_t *packet_countp, uint64_t *byte_countp)
3400 uint64_t packet_count, byte_count;
3401 struct facet *facet;
3402 struct odp_flow *odp_flows;
3405 /* Start from historical data for 'rule' itself that are no longer tracked
3406 * by the datapath. This counts, for example, facets that have expired. */
3407 packet_count = rule->packet_count;
3408 byte_count = rule->byte_count;
3410 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3412 * Also, add any statistics that are not tracked by the datapath for each
3413 * facet. This includes, for example, statistics for packets that were
3414 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3416 odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows);
3418 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3419 struct odp_flow *odp_flow = &odp_flows[n_odp_flows++];
3420 odp_flow_key_from_flow(&odp_flow->key, &facet->flow);
3421 packet_count += facet->packet_count;
3422 byte_count += facet->byte_count;
3425 /* Fetch up-to-date statistics from the datapath and add them in. */
3426 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3429 for (i = 0; i < n_odp_flows; i++) {
3430 struct odp_flow *odp_flow = &odp_flows[i];
3431 packet_count += odp_flow->stats.n_packets;
3432 byte_count += odp_flow->stats.n_bytes;
3437 /* Return the stats to the caller. */
3438 *packet_countp = packet_count;
3439 *byte_countp = byte_count;
3443 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3445 long long int msecs = time_msec() - start;
3446 *sec = htonl(msecs / 1000);
3447 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3451 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
3452 ovs_be16 out_port, struct ofpbuf **replyp)
3454 struct ofp_flow_stats *ofs;
3455 uint64_t packet_count, byte_count;
3456 size_t act_len, len;
3458 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3462 act_len = sizeof *rule->actions * rule->n_actions;
3463 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3465 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3467 ofs = append_ofp_stats_reply(len, ofconn, replyp);
3468 ofs->length = htons(len);
3471 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match,
3472 rule->flow_cookie, &ofs->cookie);
3473 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3474 ofs->priority = htons(rule->cr.priority);
3475 ofs->idle_timeout = htons(rule->idle_timeout);
3476 ofs->hard_timeout = htons(rule->hard_timeout);
3477 memset(ofs->pad2, 0, sizeof ofs->pad2);
3478 ofs->packet_count = htonll(packet_count);
3479 ofs->byte_count = htonll(byte_count);
3480 if (rule->n_actions > 0) {
3481 memcpy(ofs->actions, rule->actions, act_len);
3486 is_valid_table(uint8_t table_id)
3488 return table_id == 0 || table_id == 0xff;
3492 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3494 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
3495 struct ofpbuf *reply;
3497 COVERAGE_INC(ofproto_flows_req);
3498 reply = start_ofp_stats_reply(oh, 1024);
3499 if (is_valid_table(fsr->table_id)) {
3500 struct cls_cursor cursor;
3501 struct cls_rule target;
3504 ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0,
3506 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3507 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3508 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
3511 queue_tx(reply, ofconn, ofconn->reply_counter);
3517 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
3518 ovs_be16 out_port, struct ofpbuf **replyp)
3520 struct nx_flow_stats *nfs;
3521 uint64_t packet_count, byte_count;
3522 size_t act_len, start_len;
3523 struct ofpbuf *reply;
3525 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
3529 query_stats(ofconn->ofproto, rule, &packet_count, &byte_count);
3531 act_len = sizeof *rule->actions * rule->n_actions;
3533 start_len = (*replyp)->size;
3534 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
3537 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
3540 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3541 nfs->cookie = rule->flow_cookie;
3542 nfs->priority = htons(rule->cr.priority);
3543 nfs->idle_timeout = htons(rule->idle_timeout);
3544 nfs->hard_timeout = htons(rule->hard_timeout);
3545 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
3546 memset(nfs->pad2, 0, sizeof nfs->pad2);
3547 nfs->packet_count = htonll(packet_count);
3548 nfs->byte_count = htonll(byte_count);
3549 if (rule->n_actions > 0) {
3550 ofpbuf_put(reply, rule->actions, act_len);
3552 nfs->length = htons(reply->size - start_len);
3556 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
3558 struct nx_flow_stats_request *nfsr;
3559 struct cls_rule target;
3560 struct ofpbuf *reply;
3564 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3566 /* Dissect the message. */
3567 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
3568 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
3573 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3576 COVERAGE_INC(ofproto_flows_req);
3577 reply = start_nxstats_reply(&nfsr->nsm, 1024);
3578 if (is_valid_table(nfsr->table_id)) {
3579 struct cls_cursor cursor;
3582 cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target);
3583 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3584 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
3587 queue_tx(reply, ofconn, ofconn->reply_counter);
3593 flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results)
3595 uint64_t packet_count, byte_count;
3596 size_t act_len = sizeof *rule->actions * rule->n_actions;
3598 query_stats(ofproto, rule, &packet_count, &byte_count);
3600 ds_put_format(results, "duration=%llds, ",
3601 (time_msec() - rule->created) / 1000);
3602 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3603 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3604 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3605 cls_rule_format(&rule->cr, results);
3607 ofp_print_actions(results, &rule->actions->header, act_len);
3609 ds_put_cstr(results, "drop");
3611 ds_put_cstr(results, "\n");
3614 /* Adds a pretty-printed description of all flows to 'results', including
3615 * those marked hidden by secchan (e.g., by in-band control). */
3617 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3619 struct cls_cursor cursor;
3622 cls_cursor_init(&cursor, &p->cls, NULL);
3623 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3624 flow_stats_ds(p, rule, results);
3629 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3630 ovs_be16 out_port, uint8_t table_id,
3631 struct ofp_aggregate_stats_reply *oasr)
3633 uint64_t total_packets = 0;
3634 uint64_t total_bytes = 0;
3637 COVERAGE_INC(ofproto_agg_request);
3639 if (is_valid_table(table_id)) {
3640 struct cls_cursor cursor;
3643 cls_cursor_init(&cursor, &ofproto->cls, target);
3644 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3645 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
3646 uint64_t packet_count;
3647 uint64_t byte_count;
3649 query_stats(ofproto, rule, &packet_count, &byte_count);
3651 total_packets += packet_count;
3652 total_bytes += byte_count;
3658 oasr->flow_count = htonl(n_flows);
3659 oasr->packet_count = htonll(total_packets);
3660 oasr->byte_count = htonll(total_bytes);
3661 memset(oasr->pad, 0, sizeof oasr->pad);
3665 handle_aggregate_stats_request(struct ofconn *ofconn,
3666 const struct ofp_header *oh)
3668 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
3669 struct ofp_aggregate_stats_reply *reply;
3670 struct cls_rule target;
3673 ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0,
3676 msg = start_ofp_stats_reply(oh, sizeof *reply);
3677 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3678 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3679 request->table_id, reply);
3680 queue_tx(msg, ofconn, ofconn->reply_counter);
3685 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
3687 struct nx_aggregate_stats_request *request;
3688 struct ofp_aggregate_stats_reply *reply;
3689 struct cls_rule target;
3694 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3696 /* Dissect the message. */
3697 request = ofpbuf_pull(&b, sizeof *request);
3698 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
3703 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3707 COVERAGE_INC(ofproto_flows_req);
3708 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3709 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3710 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3711 request->table_id, reply);
3712 queue_tx(buf, ofconn, ofconn->reply_counter);
3717 struct queue_stats_cbdata {
3718 struct ofconn *ofconn;
3719 struct ofport *ofport;
3724 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3725 const struct netdev_queue_stats *stats)
3727 struct ofp_queue_stats *reply;
3729 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3730 reply->port_no = htons(cbdata->ofport->opp.port_no);
3731 memset(reply->pad, 0, sizeof reply->pad);
3732 reply->queue_id = htonl(queue_id);
3733 reply->tx_bytes = htonll(stats->tx_bytes);
3734 reply->tx_packets = htonll(stats->tx_packets);
3735 reply->tx_errors = htonll(stats->tx_errors);
3739 handle_queue_stats_dump_cb(uint32_t queue_id,
3740 struct netdev_queue_stats *stats,
3743 struct queue_stats_cbdata *cbdata = cbdata_;
3745 put_queue_stats(cbdata, queue_id, stats);
3749 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3750 struct queue_stats_cbdata *cbdata)
3752 cbdata->ofport = port;
3753 if (queue_id == OFPQ_ALL) {
3754 netdev_dump_queue_stats(port->netdev,
3755 handle_queue_stats_dump_cb, cbdata);
3757 struct netdev_queue_stats stats;
3759 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3760 put_queue_stats(cbdata, queue_id, &stats);
3766 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
3768 struct ofproto *ofproto = ofconn->ofproto;
3769 const struct ofp_queue_stats_request *qsr;
3770 struct queue_stats_cbdata cbdata;
3771 struct ofport *port;
3772 unsigned int port_no;
3775 qsr = ofputil_stats_body(oh);
3777 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3780 COVERAGE_INC(ofproto_queue_req);
3782 cbdata.ofconn = ofconn;
3783 cbdata.msg = start_ofp_stats_reply(oh, 128);
3785 port_no = ntohs(qsr->port_no);
3786 queue_id = ntohl(qsr->queue_id);
3787 if (port_no == OFPP_ALL) {
3788 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3789 handle_queue_stats_for_port(port, queue_id, &cbdata);
3791 } else if (port_no < ofproto->max_ports) {
3792 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3794 handle_queue_stats_for_port(port, queue_id, &cbdata);
3797 ofpbuf_delete(cbdata.msg);
3798 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3800 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3805 static long long int
3806 msec_from_nsec(uint64_t sec, uint32_t nsec)
3808 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3812 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3813 const struct odp_flow_stats *stats)
3815 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3816 if (used > facet->used) {
3818 if (used > facet->rule->used) {
3819 facet->rule->used = used;
3821 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3825 /* Folds the statistics from 'stats' into the counters in 'facet'.
3827 * Because of the meaning of a facet's counters, it only makes sense to do this
3828 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3829 * packet that was sent by hand or if it represents statistics that have been
3830 * cleared out of the datapath. */
3832 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3833 const struct odp_flow_stats *stats)
3835 if (stats->n_packets) {
3836 facet_update_time(ofproto, facet, stats);
3837 facet->packet_count += stats->n_packets;
3838 facet->byte_count += stats->n_bytes;
3839 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3843 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3844 * in which no matching flow already exists in the flow table.
3846 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3847 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3848 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3850 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3853 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3855 struct ofproto *p = ofconn->ofproto;
3856 struct ofpbuf *packet;
3861 if (fm->flags & OFPFF_CHECK_OVERLAP
3862 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3863 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3867 if (fm->buffer_id != UINT32_MAX) {
3868 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3872 in_port = UINT16_MAX;
3875 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3876 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3877 fm->flags & OFPFF_SEND_FLOW_REM);
3878 rule_insert(p, rule);
3880 rule_execute(p, rule, in_port, packet);
3885 static struct rule *
3886 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3888 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3892 send_buffered_packet(struct ofconn *ofconn,
3893 struct rule *rule, uint32_t buffer_id)
3895 struct ofpbuf *packet;
3899 if (buffer_id == UINT32_MAX) {
3903 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3908 rule_execute(ofconn->ofproto, rule, in_port, packet);
3913 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3915 struct modify_flows_cbdata {
3916 struct ofproto *ofproto;
3917 const struct flow_mod *fm;
3921 static int modify_flow(struct ofproto *, const struct flow_mod *,
3924 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3925 * encoded by ofp_mkerr() on failure.
3927 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3930 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3932 struct ofproto *p = ofconn->ofproto;
3933 struct rule *match = NULL;
3934 struct cls_cursor cursor;
3937 cls_cursor_init(&cursor, &p->cls, &fm->cr);
3938 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3939 if (!rule_is_hidden(rule)) {
3941 modify_flow(p, fm, rule);
3946 /* This credits the packet to whichever flow happened to match last.
3947 * That's weird. Maybe we should do a lookup for the flow that
3948 * actually matches the packet? Who knows. */
3949 send_buffered_packet(ofconn, match, fm->buffer_id);
3952 return add_flow(ofconn, fm);
3956 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3957 * code as encoded by ofp_mkerr() on failure.
3959 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3962 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
3964 struct ofproto *p = ofconn->ofproto;
3965 struct rule *rule = find_flow_strict(p, fm);
3966 if (rule && !rule_is_hidden(rule)) {
3967 modify_flow(p, fm, rule);
3968 return send_buffered_packet(ofconn, rule, fm->buffer_id);
3970 return add_flow(ofconn, fm);
3974 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3975 * been identified as a flow in 'p''s flow table to be modified, by changing
3976 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3977 * ofp_action[] structures). */
3979 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
3981 size_t actions_len = fm->n_actions * sizeof *rule->actions;
3983 rule->flow_cookie = fm->cookie;
3985 /* If the actions are the same, do nothing. */
3986 if (fm->n_actions == rule->n_actions
3988 || !memcmp(fm->actions, rule->actions, actions_len))) {
3992 /* Replace actions. */
3993 free(rule->actions);
3994 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
3995 rule->n_actions = fm->n_actions;
3997 p->need_revalidate = true;
4002 /* OFPFC_DELETE implementation. */
4004 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4006 /* Implements OFPFC_DELETE. */
4008 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4010 struct rule *rule, *next_rule;
4011 struct cls_cursor cursor;
4013 cls_cursor_init(&cursor, &p->cls, &fm->cr);
4014 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4015 delete_flow(p, rule, htons(fm->out_port));
4019 /* Implements OFPFC_DELETE_STRICT. */
4021 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4023 struct rule *rule = find_flow_strict(p, fm);
4025 delete_flow(p, rule, htons(fm->out_port));
4029 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4030 * been identified as a flow to delete from 'p''s flow table, by deleting the
4031 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4034 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4035 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4036 * specified 'out_port'. */
4038 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4040 if (rule_is_hidden(rule)) {
4044 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4048 rule_send_removed(p, rule, OFPRR_DELETE);
4049 rule_remove(p, rule);
4053 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4055 struct ofproto *p = ofconn->ofproto;
4059 error = reject_slave_controller(ofconn, "flow_mod");
4064 error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format);
4069 /* We do not support the emergency flow cache. It will hopefully get
4070 * dropped from OpenFlow in the near future. */
4071 if (fm.flags & OFPFF_EMERG) {
4072 /* There isn't a good fit for an error code, so just state that the
4073 * flow table is full. */
4074 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4077 error = validate_actions(fm.actions, fm.n_actions,
4078 &fm.cr.flow, p->max_ports);
4083 switch (fm.command) {
4085 return add_flow(ofconn, &fm);
4088 return modify_flows_loose(ofconn, &fm);
4090 case OFPFC_MODIFY_STRICT:
4091 return modify_flow_strict(ofconn, &fm);
4094 delete_flows_loose(p, &fm);
4097 case OFPFC_DELETE_STRICT:
4098 delete_flow_strict(p, &fm);
4102 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4107 handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh)
4109 const struct nxt_tun_id_cookie *msg
4110 = (const struct nxt_tun_id_cookie *) oh;
4112 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4117 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4119 struct nx_role_request *nrr = (struct nx_role_request *) oh;
4120 struct nx_role_request *reply;
4124 if (ofconn->type != OFCONN_PRIMARY) {
4125 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4127 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4130 role = ntohl(nrr->role);
4131 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4132 && role != NX_ROLE_SLAVE) {
4133 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4135 /* There's no good error code for this. */
4136 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4139 if (role == NX_ROLE_MASTER) {
4140 struct ofconn *other;
4142 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4143 if (other->role == NX_ROLE_MASTER) {
4144 other->role = NX_ROLE_SLAVE;
4148 ofconn->role = role;
4150 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
4151 reply->role = htonl(role);
4152 queue_tx(buf, ofconn, ofconn->reply_counter);
4158 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4160 const struct nxt_set_flow_format *msg
4161 = (const struct nxt_set_flow_format *) oh;
4164 format = ntohl(msg->format);
4165 if (format == NXFF_OPENFLOW10
4166 || format == NXFF_TUN_ID_FROM_COOKIE
4167 || format == NXFF_NXM) {
4168 ofconn->flow_format = format;
4171 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4176 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4178 struct ofp_header *ob;
4181 /* Currently, everything executes synchronously, so we can just
4182 * immediately send the barrier reply. */
4183 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4184 queue_tx(buf, ofconn, ofconn->reply_counter);
4189 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4191 const struct ofp_header *oh = msg->data;
4192 const struct ofputil_msg_type *type;
4195 error = ofputil_decode_msg_type(oh, &type);
4200 switch (ofputil_msg_type_code(type)) {
4201 /* OpenFlow requests. */
4202 case OFPUTIL_OFPT_ECHO_REQUEST:
4203 return handle_echo_request(ofconn, oh);
4205 case OFPUTIL_OFPT_FEATURES_REQUEST:
4206 return handle_features_request(ofconn, oh);
4208 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
4209 return handle_get_config_request(ofconn, oh);
4211 case OFPUTIL_OFPT_SET_CONFIG:
4212 return handle_set_config(ofconn, msg->data);
4214 case OFPUTIL_OFPT_PACKET_OUT:
4215 return handle_packet_out(ofconn, oh);
4217 case OFPUTIL_OFPT_PORT_MOD:
4218 return handle_port_mod(ofconn, oh);
4220 case OFPUTIL_OFPT_FLOW_MOD:
4221 return handle_flow_mod(ofconn, oh);
4223 case OFPUTIL_OFPT_BARRIER_REQUEST:
4224 return handle_barrier_request(ofconn, oh);
4226 /* OpenFlow replies. */
4227 case OFPUTIL_OFPT_ECHO_REPLY:
4230 /* Nicira extension requests. */
4231 case OFPUTIL_NXT_STATUS_REQUEST:
4232 return switch_status_handle_request(
4233 ofconn->ofproto->switch_status, ofconn->rconn, oh);
4235 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE:
4236 return handle_tun_id_from_cookie(ofconn, oh);
4238 case OFPUTIL_NXT_ROLE_REQUEST:
4239 return handle_role_request(ofconn, oh);
4241 case OFPUTIL_NXT_SET_FLOW_FORMAT:
4242 return handle_nxt_set_flow_format(ofconn, oh);
4244 case OFPUTIL_NXT_FLOW_MOD:
4245 return handle_flow_mod(ofconn, oh);
4247 /* OpenFlow statistics requests. */
4248 case OFPUTIL_OFPST_DESC_REQUEST:
4249 return handle_desc_stats_request(ofconn, oh);
4251 case OFPUTIL_OFPST_FLOW_REQUEST:
4252 return handle_flow_stats_request(ofconn, oh);
4254 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
4255 return handle_aggregate_stats_request(ofconn, oh);
4257 case OFPUTIL_OFPST_TABLE_REQUEST:
4258 return handle_table_stats_request(ofconn, oh);
4260 case OFPUTIL_OFPST_PORT_REQUEST:
4261 return handle_port_stats_request(ofconn, oh);
4263 case OFPUTIL_OFPST_QUEUE_REQUEST:
4264 return handle_queue_stats_request(ofconn, oh);
4266 /* Nicira extension statistics requests. */
4267 case OFPUTIL_NXST_FLOW_REQUEST:
4268 return handle_nxst_flow(ofconn, oh);
4270 case OFPUTIL_NXST_AGGREGATE_REQUEST:
4271 return handle_nxst_aggregate(ofconn, oh);
4273 case OFPUTIL_INVALID:
4274 case OFPUTIL_OFPT_HELLO:
4275 case OFPUTIL_OFPT_ERROR:
4276 case OFPUTIL_OFPT_FEATURES_REPLY:
4277 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
4278 case OFPUTIL_OFPT_PACKET_IN:
4279 case OFPUTIL_OFPT_FLOW_REMOVED:
4280 case OFPUTIL_OFPT_PORT_STATUS:
4281 case OFPUTIL_OFPT_BARRIER_REPLY:
4282 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
4283 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
4284 case OFPUTIL_OFPST_DESC_REPLY:
4285 case OFPUTIL_OFPST_FLOW_REPLY:
4286 case OFPUTIL_OFPST_QUEUE_REPLY:
4287 case OFPUTIL_OFPST_PORT_REPLY:
4288 case OFPUTIL_OFPST_TABLE_REPLY:
4289 case OFPUTIL_OFPST_AGGREGATE_REPLY:
4290 case OFPUTIL_NXT_STATUS_REPLY:
4291 case OFPUTIL_NXT_ROLE_REPLY:
4292 case OFPUTIL_NXT_FLOW_REMOVED:
4293 case OFPUTIL_NXST_FLOW_REPLY:
4294 case OFPUTIL_NXST_AGGREGATE_REPLY:
4296 if (VLOG_IS_WARN_ENABLED()) {
4297 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4298 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4301 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
4302 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
4304 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4310 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4312 int error = handle_openflow__(ofconn, ofp_msg);
4314 send_error_oh(ofconn, ofp_msg->data, error);
4316 COVERAGE_INC(ofproto_recv_openflow);
4320 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4322 struct odp_msg *msg = packet->data;
4323 struct ofpbuf payload;
4324 struct facet *facet;
4327 ofpbuf_use_const(&payload, msg + 1, msg->length - sizeof *msg);
4328 flow_extract(&payload, msg->arg, msg->port, &flow);
4330 packet->l2 = payload.l2;
4331 packet->l3 = payload.l3;
4332 packet->l4 = payload.l4;
4333 packet->l7 = payload.l7;
4335 /* Check with in-band control to see if this packet should be sent
4336 * to the local port regardless of the flow table. */
4337 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4338 struct ofpbuf odp_actions;
4340 ofpbuf_init(&odp_actions, 32);
4341 nl_msg_put_u32(&odp_actions, ODPAT_OUTPUT, ODPP_LOCAL);
4342 dpif_execute(p->dpif, odp_actions.data, odp_actions.size, &payload);
4343 ofpbuf_uninit(&odp_actions);
4346 facet = facet_lookup_valid(p, &flow);
4348 struct rule *rule = rule_lookup(p, &flow);
4350 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4351 struct ofport *port = get_port(p, msg->port);
4353 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4354 COVERAGE_INC(ofproto_no_packet_in);
4355 /* XXX install 'drop' flow entry */
4356 ofpbuf_delete(packet);
4360 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4364 COVERAGE_INC(ofproto_packet_in);
4365 send_packet_in(p, packet);
4369 facet = facet_create(p, rule, &flow, packet);
4370 } else if (!facet->may_install) {
4371 /* The facet is not installable, that is, we need to process every
4372 * packet, so process the current packet's actions into 'facet'. */
4373 facet_make_actions(p, facet, packet);
4376 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4378 * Extra-special case for fail-open mode.
4380 * We are in fail-open mode and the packet matched the fail-open rule,
4381 * but we are connected to a controller too. We should send the packet
4382 * up to the controller in the hope that it will try to set up a flow
4383 * and thereby allow us to exit fail-open.
4385 * See the top-level comment in fail-open.c for more information.
4387 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4388 DPIF_RECV_MSG_PADDING));
4391 ofpbuf_pull(packet, sizeof *msg);
4392 facet_execute(p, facet, packet);
4393 facet_install(p, facet, false);
4397 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4399 struct odp_msg *msg = packet->data;
4401 switch (msg->type) {
4402 case _ODPL_ACTION_NR:
4403 COVERAGE_INC(ofproto_ctlr_action);
4404 send_packet_in(p, packet);
4407 case _ODPL_SFLOW_NR:
4409 ofproto_sflow_received(p->sflow, msg);
4411 ofpbuf_delete(packet);
4415 handle_odp_miss_msg(p, packet);
4419 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4425 /* Flow expiration. */
4427 static int ofproto_dp_max_idle(const struct ofproto *);
4428 static void ofproto_update_used(struct ofproto *);
4429 static void rule_expire(struct ofproto *, struct rule *);
4430 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4432 /* This function is called periodically by ofproto_run(). Its job is to
4433 * collect updates for the flows that have been installed into the datapath,
4434 * most importantly when they last were used, and then use that information to
4435 * expire flows that have not been used recently.
4437 * Returns the number of milliseconds after which it should be called again. */
4439 ofproto_expire(struct ofproto *ofproto)
4441 struct rule *rule, *next_rule;
4442 struct cls_cursor cursor;
4445 /* Update 'used' for each flow in the datapath. */
4446 ofproto_update_used(ofproto);
4448 /* Expire facets that have been idle too long. */
4449 dp_max_idle = ofproto_dp_max_idle(ofproto);
4450 ofproto_expire_facets(ofproto, dp_max_idle);
4452 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4453 cls_cursor_init(&cursor, &ofproto->cls, NULL);
4454 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
4455 rule_expire(ofproto, rule);
4458 /* Let the hook know that we're at a stable point: all outstanding data
4459 * in existing flows has been accounted to the account_cb. Thus, the
4460 * hook can now reasonably do operations that depend on having accurate
4461 * flow volume accounting (currently, that's just bond rebalancing). */
4462 if (ofproto->ofhooks->account_checkpoint_cb) {
4463 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4466 return MIN(dp_max_idle, 1000);
4469 /* Update 'used' member of installed facets. */
4471 ofproto_update_used(struct ofproto *p)
4473 struct odp_flow *flows;
4478 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4483 for (i = 0; i < n_flows; i++) {
4484 struct odp_flow *f = &flows[i];
4485 struct facet *facet;
4488 odp_flow_key_to_flow(&f->key, &flow);
4489 facet = facet_find(p, &flow);
4491 if (facet && facet->installed) {
4492 facet_update_time(p, facet, &f->stats);
4493 facet_account(p, facet, f->stats.n_bytes);
4495 /* There's a flow in the datapath that we know nothing about.
4497 COVERAGE_INC(ofproto_unexpected_rule);
4498 dpif_flow_del(p->dpif, f);
4505 /* Calculates and returns the number of milliseconds of idle time after which
4506 * facets should expire from the datapath and we should fold their statistics
4507 * into their parent rules in userspace. */
4509 ofproto_dp_max_idle(const struct ofproto *ofproto)
4512 * Idle time histogram.
4514 * Most of the time a switch has a relatively small number of facets. When
4515 * this is the case we might as well keep statistics for all of them in
4516 * userspace and to cache them in the kernel datapath for performance as
4519 * As the number of facets increases, the memory required to maintain
4520 * statistics about them in userspace and in the kernel becomes
4521 * significant. However, with a large number of facets it is likely that
4522 * only a few of them are "heavy hitters" that consume a large amount of
4523 * bandwidth. At this point, only heavy hitters are worth caching in the
4524 * kernel and maintaining in userspaces; other facets we can discard.
4526 * The technique used to compute the idle time is to build a histogram with
4527 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4528 * that is installed in the kernel gets dropped in the appropriate bucket.
4529 * After the histogram has been built, we compute the cutoff so that only
4530 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4531 * cached. At least the most-recently-used bucket of facets is kept, so
4532 * actually an arbitrary number of facets can be kept in any given
4533 * expiration run (though the next run will delete most of those unless
4534 * they receive additional data).
4536 * This requires a second pass through the facets, in addition to the pass
4537 * made by ofproto_update_used(), because the former function never looks
4538 * at uninstallable facets.
4540 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4541 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4542 int buckets[N_BUCKETS] = { 0 };
4543 struct facet *facet;
4548 total = hmap_count(&ofproto->facets);
4549 if (total <= 1000) {
4550 return N_BUCKETS * BUCKET_WIDTH;
4553 /* Build histogram. */
4555 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4556 long long int idle = now - facet->used;
4557 int bucket = (idle <= 0 ? 0
4558 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4559 : (unsigned int) idle / BUCKET_WIDTH);
4563 /* Find the first bucket whose flows should be expired. */
4564 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4565 if (buckets[bucket]) {
4568 subtotal += buckets[bucket++];
4569 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4574 if (VLOG_IS_DBG_ENABLED()) {
4578 ds_put_cstr(&s, "keep");
4579 for (i = 0; i < N_BUCKETS; i++) {
4581 ds_put_cstr(&s, ", drop");
4584 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4587 VLOG_INFO("%s: %s (msec:count)",
4588 dpif_name(ofproto->dpif), ds_cstr(&s));
4592 return bucket * BUCKET_WIDTH;
4596 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4598 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4599 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4600 struct ofexpired expired;
4601 struct odp_flow odp_flow;
4603 /* Get updated flow stats.
4605 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4606 * updated TCP flags and (2) the dpif_flow_list_all() in
4607 * ofproto_update_used() zeroed TCP flags. */
4608 memset(&odp_flow, 0, sizeof odp_flow);
4609 if (facet->installed) {
4610 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
4611 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4612 dpif_flow_get(ofproto->dpif, &odp_flow);
4614 if (odp_flow.stats.n_packets) {
4615 facet_update_time(ofproto, facet, &odp_flow.stats);
4616 netflow_flow_update_flags(&facet->nf_flow,
4617 odp_flow.stats.tcp_flags);
4621 expired.flow = facet->flow;
4622 expired.packet_count = facet->packet_count +
4623 odp_flow.stats.n_packets;
4624 expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes;
4625 expired.used = facet->used;
4627 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4632 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4634 long long int cutoff = time_msec() - dp_max_idle;
4635 struct facet *facet, *next_facet;
4637 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4638 facet_active_timeout(ofproto, facet);
4639 if (facet->used < cutoff) {
4640 facet_remove(ofproto, facet);
4645 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4646 * then delete it entirely. */
4648 rule_expire(struct ofproto *ofproto, struct rule *rule)
4650 struct facet *facet, *next_facet;
4654 /* Has 'rule' expired? */
4656 if (rule->hard_timeout
4657 && now > rule->created + rule->hard_timeout * 1000) {
4658 reason = OFPRR_HARD_TIMEOUT;
4659 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4660 && now >rule->used + rule->idle_timeout * 1000) {
4661 reason = OFPRR_IDLE_TIMEOUT;
4666 COVERAGE_INC(ofproto_expired);
4668 /* Update stats. (This is a no-op if the rule expired due to an idle
4669 * timeout, because that only happens when the rule has no facets left.) */
4670 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4671 facet_remove(ofproto, facet);
4674 /* Get rid of the rule. */
4675 if (!rule_is_hidden(rule)) {
4676 rule_send_removed(ofproto, rule, reason);
4678 rule_remove(ofproto, rule);
4681 static struct ofpbuf *
4682 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4685 struct ofp_flow_removed *ofr;
4688 ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf);
4689 ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match,
4690 rule->flow_cookie, &ofr->cookie);
4691 ofr->priority = htons(rule->cr.priority);
4692 ofr->reason = reason;
4693 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4694 ofr->idle_timeout = htons(rule->idle_timeout);
4695 ofr->packet_count = htonll(rule->packet_count);
4696 ofr->byte_count = htonll(rule->byte_count);
4701 static struct ofpbuf *
4702 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4704 struct nx_flow_removed *nfr;
4708 make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf);
4709 match_len = nx_put_match(buf, &rule->cr);
4712 nfr->cookie = rule->flow_cookie;
4713 nfr->priority = htons(rule->cr.priority);
4714 nfr->reason = reason;
4715 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4716 nfr->idle_timeout = htons(rule->idle_timeout);
4717 nfr->match_len = htons(match_len);
4718 nfr->packet_count = htonll(rule->packet_count);
4719 nfr->byte_count = htonll(rule->byte_count);
4725 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4727 struct ofconn *ofconn;
4729 if (!rule->send_flow_removed) {
4733 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4736 if (!rconn_is_connected(ofconn->rconn)
4737 || !ofconn_receives_async_msgs(ofconn)) {
4741 msg = (ofconn->flow_format == NXFF_NXM
4742 ? compose_nx_flow_removed(rule, reason)
4743 : compose_ofp_flow_removed(ofconn, rule, reason));
4745 /* Account flow expirations under ofconn->reply_counter, the counter
4746 * for replies to OpenFlow requests. That works because preventing
4747 * OpenFlow requests from being processed also prevents new flows from
4748 * being added (and expiring). (It also prevents processing OpenFlow
4749 * requests that would not add new flows, so it is imperfect.) */
4750 queue_tx(msg, ofconn, ofconn->reply_counter);
4754 /* pinsched callback for sending 'packet' on 'ofconn'. */
4756 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4758 struct ofconn *ofconn = ofconn_;
4760 rconn_send_with_limit(ofconn->rconn, packet,
4761 ofconn->packet_in_counter, 100);
4764 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4765 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4766 * packet scheduler for sending.
4768 * 'max_len' specifies the maximum number of bytes of the packet to send on
4769 * 'ofconn' (INT_MAX specifies no limit).
4771 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4772 * ownership is transferred to this function. */
4774 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4777 struct ofproto *ofproto = ofconn->ofproto;
4778 struct ofp_packet_in *opi = packet->data;
4779 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4780 int send_len, trim_size;
4784 if (opi->reason == OFPR_ACTION) {
4785 buffer_id = UINT32_MAX;
4786 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4787 buffer_id = pktbuf_get_null();
4788 } else if (!ofconn->pktbuf) {
4789 buffer_id = UINT32_MAX;
4791 struct ofpbuf payload;
4793 ofpbuf_use_const(&payload, opi->data,
4794 packet->size - offsetof(struct ofp_packet_in, data));
4795 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4798 /* Figure out how much of the packet to send. */
4799 send_len = ntohs(opi->total_len);
4800 if (buffer_id != UINT32_MAX) {
4801 send_len = MIN(send_len, ofconn->miss_send_len);
4803 send_len = MIN(send_len, max_len);
4805 /* Adjust packet length and clone if necessary. */
4806 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4808 packet = ofpbuf_clone_data(packet->data, trim_size);
4811 packet->size = trim_size;
4814 /* Update packet headers. */
4815 opi->buffer_id = htonl(buffer_id);
4816 update_openflow_length(packet);
4818 /* Hand over to packet scheduler. It might immediately call into
4819 * do_send_packet_in() or it might buffer it for a while (until a later
4820 * call to pinsched_run()). */
4821 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4822 packet, do_send_packet_in, ofconn);
4825 /* Replace struct odp_msg header in 'packet' by equivalent struct
4826 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4827 * returned by dpif_recv()).
4829 * The conversion is not complete: the caller still needs to trim any unneeded
4830 * payload off the end of the buffer, set the length in the OpenFlow header,
4831 * and set buffer_id. Those require us to know the controller settings and so
4832 * must be done on a per-controller basis.
4834 * Returns the maximum number of bytes of the packet that should be sent to
4835 * the controller (INT_MAX if no limit). */
4837 do_convert_to_packet_in(struct ofpbuf *packet)
4839 struct odp_msg *msg = packet->data;
4840 struct ofp_packet_in *opi;
4846 /* Extract relevant header fields */
4847 if (msg->type == _ODPL_ACTION_NR) {
4848 reason = OFPR_ACTION;
4851 reason = OFPR_NO_MATCH;
4854 total_len = msg->length - sizeof *msg;
4855 in_port = odp_port_to_ofp_port(msg->port);
4857 /* Repurpose packet buffer by overwriting header. */
4858 ofpbuf_pull(packet, sizeof(struct odp_msg));
4859 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4860 opi->header.version = OFP_VERSION;
4861 opi->header.type = OFPT_PACKET_IN;
4862 opi->total_len = htons(total_len);
4863 opi->in_port = htons(in_port);
4864 opi->reason = reason;
4869 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4870 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4871 * as necessary according to their individual configurations.
4873 * 'packet' must have sufficient headroom to convert it into a struct
4874 * ofp_packet_in (e.g. as returned by dpif_recv()).
4876 * Takes ownership of 'packet'. */
4878 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4880 struct ofconn *ofconn, *prev;
4883 max_len = do_convert_to_packet_in(packet);
4886 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4887 if (ofconn_receives_async_msgs(ofconn)) {
4889 schedule_packet_in(prev, packet, max_len, true);
4895 schedule_packet_in(prev, packet, max_len, false);
4897 ofpbuf_delete(packet);
4902 pick_datapath_id(const struct ofproto *ofproto)
4904 const struct ofport *port;
4906 port = get_port(ofproto, ODPP_LOCAL);
4908 uint8_t ea[ETH_ADDR_LEN];
4911 error = netdev_get_etheraddr(port->netdev, ea);
4913 return eth_addr_to_uint64(ea);
4915 VLOG_WARN("could not get MAC address for %s (%s)",
4916 netdev_get_name(port->netdev), strerror(error));
4918 return ofproto->fallback_dpid;
4922 pick_fallback_dpid(void)
4924 uint8_t ea[ETH_ADDR_LEN];
4925 eth_addr_nicira_random(ea);
4926 return eth_addr_to_uint64(ea);
4930 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
4931 void *aux OVS_UNUSED)
4933 const struct shash_node *node;
4937 SHASH_FOR_EACH (node, &all_ofprotos) {
4938 ds_put_format(&results, "%s\n", node->name);
4940 unixctl_command_reply(conn, 200, ds_cstr(&results));
4941 ds_destroy(&results);
4944 struct ofproto_trace {
4945 struct action_xlate_ctx ctx;
4951 trace_format_rule(struct ds *result, int level, const struct rule *rule)
4953 ds_put_char_multiple(result, '\t', level);
4955 ds_put_cstr(result, "No match\n");
4959 ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
4960 ntohll(rule->flow_cookie));
4961 cls_rule_format(&rule->cr, result);
4962 ds_put_char(result, '\n');
4964 ds_put_char_multiple(result, '\t', level);
4965 ds_put_cstr(result, "OpenFlow ");
4966 ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
4967 rule->n_actions * sizeof *rule->actions);
4968 ds_put_char(result, '\n');
4972 trace_format_flow(struct ds *result, int level, const char *title,
4973 struct ofproto_trace *trace)
4975 ds_put_char_multiple(result, '\t', level);
4976 ds_put_format(result, "%s: ", title);
4977 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
4978 ds_put_cstr(result, "unchanged");
4980 flow_format(result, &trace->ctx.flow);
4981 trace->flow = trace->ctx.flow;
4983 ds_put_char(result, '\n');
4987 trace_resubmit(struct action_xlate_ctx *ctx, const struct rule *rule)
4989 struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
4990 struct ds *result = trace->result;
4992 ds_put_char(result, '\n');
4993 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
4994 trace_format_rule(result, ctx->recurse + 1, rule);
4998 ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
4999 void *aux OVS_UNUSED)
5001 char *dpname, *in_port_s, *tun_id_s, *packet_s;
5002 char *args = xstrdup(args_);
5003 char *save_ptr = NULL;
5004 struct ofproto *ofproto;
5005 struct ofpbuf packet;
5013 ofpbuf_init(&packet, strlen(args) / 2);
5016 dpname = strtok_r(args, " ", &save_ptr);
5017 tun_id_s = strtok_r(NULL, " ", &save_ptr);
5018 in_port_s = strtok_r(NULL, " ", &save_ptr);
5019 packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
5020 if (!dpname || !in_port_s || !packet_s) {
5021 unixctl_command_reply(conn, 501, "Bad command syntax");
5025 ofproto = shash_find_data(&all_ofprotos, dpname);
5027 unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
5032 tun_id = ntohl(strtoul(tun_id_s, NULL, 10));
5033 in_port = ofp_port_to_odp_port(atoi(in_port_s));
5035 packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
5036 packet_s += strspn(packet_s, " ");
5037 if (*packet_s != '\0') {
5038 unixctl_command_reply(conn, 501, "Trailing garbage in command");
5041 if (packet.size < ETH_HEADER_LEN) {
5042 unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
5046 ds_put_cstr(&result, "Packet: ");
5047 s = ofp_packet_to_string(packet.data, packet.size, packet.size);
5048 ds_put_cstr(&result, s);
5051 flow_extract(&packet, tun_id, in_port, &flow);
5052 ds_put_cstr(&result, "Flow: ");
5053 flow_format(&result, &flow);
5054 ds_put_char(&result, '\n');
5056 rule = rule_lookup(ofproto, &flow);
5057 trace_format_rule(&result, 0, rule);
5059 struct ofproto_trace trace;
5060 struct ofpbuf *odp_actions;
5062 trace.result = &result;
5064 action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
5065 trace.ctx.resubmit_hook = trace_resubmit;
5066 odp_actions = xlate_actions(&trace.ctx,
5067 rule->actions, rule->n_actions);
5069 ds_put_char(&result, '\n');
5070 trace_format_flow(&result, 0, "Final flow", &trace);
5071 ds_put_cstr(&result, "Datapath actions: ");
5072 format_odp_actions(&result, odp_actions->data, odp_actions->size);
5073 ofpbuf_delete(odp_actions);
5076 unixctl_command_reply(conn, 200, ds_cstr(&result));
5079 ds_destroy(&result);
5080 ofpbuf_uninit(&packet);
5085 ofproto_unixctl_init(void)
5087 static bool registered;
5093 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);
5094 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
5098 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5099 struct ofpbuf *odp_actions, tag_type *tags,
5100 uint16_t *nf_output_iface, void *ofproto_)
5102 struct ofproto *ofproto = ofproto_;
5105 /* Drop frames for reserved multicast addresses. */
5106 if (eth_addr_is_reserved(flow->dl_dst)) {
5110 /* Learn source MAC (but don't try to learn from revalidation). */
5111 if (packet != NULL) {
5112 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5114 GRAT_ARP_LOCK_NONE);
5116 /* The log messages here could actually be useful in debugging,
5117 * so keep the rate limit relatively high. */
5118 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5119 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5120 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5121 ofproto_revalidate(ofproto, rev_tag);
5125 /* Determine output port. */
5126 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5129 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5130 nf_output_iface, odp_actions);
5131 } else if (out_port != flow->in_port) {
5132 nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, out_port);
5133 *nf_output_iface = out_port;
5141 static const struct ofhooks default_ofhooks = {
5142 default_normal_ofhook_cb,