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"
42 #include "ofp-print.h"
44 #include "ofproto-sflow.h"
46 #include "openflow/nicira-ext.h"
47 #include "openflow/openflow.h"
48 #include "openvswitch/datapath-protocol.h"
52 #include "poll-loop.h"
56 #include "stream-ssl.h"
64 VLOG_DEFINE_THIS_MODULE(ofproto);
66 #include "sflow_api.h"
69 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
70 struct netdev *netdev;
71 struct ofp_phy_port opp; /* In host byte order. */
75 static void ofport_free(struct ofport *);
76 static void hton_ofp_phy_port(struct ofp_phy_port *);
78 static int xlate_actions(const union ofp_action *in, size_t n_in,
79 const struct flow *, struct ofproto *,
80 const struct ofpbuf *packet,
81 struct odp_actions *out, tag_type *tags,
82 bool *may_set_up_flow, uint16_t *nf_output_iface);
84 /* An OpenFlow flow. */
86 long long int used; /* Time last used; time created if not used. */
87 long long int created; /* Creation time. */
91 * - Do include packets and bytes from facets that have been deleted or
92 * whose own statistics have been folded into the rule.
94 * - Do include packets and bytes sent "by hand" that were accounted to
95 * the rule without any facet being involved (this is a rare corner
96 * case in rule_execute()).
98 * - Do not include packet or bytes that can be obtained from any facet's
99 * packet_count or byte_count member or that can be obtained from the
100 * datapath by, e.g., dpif_flow_get() for any facet.
102 uint64_t packet_count; /* Number of packets received. */
103 uint64_t byte_count; /* Number of bytes received. */
105 ovs_be64 flow_cookie; /* Controller-issued identifier. */
107 struct cls_rule cr; /* In owning ofproto's classifier. */
108 uint16_t idle_timeout; /* In seconds from time of last use. */
109 uint16_t hard_timeout; /* In seconds from time of creation. */
110 bool send_flow_removed; /* Send a flow removed message? */
111 int n_actions; /* Number of elements in actions[]. */
112 union ofp_action *actions; /* OpenFlow actions. */
113 struct list facets; /* List of "struct facet"s. */
116 static struct rule *rule_from_cls_rule(const struct cls_rule *);
117 static bool rule_is_hidden(const struct rule *);
119 static struct rule *rule_create(const struct cls_rule *,
120 const union ofp_action *, size_t n_actions,
121 uint16_t idle_timeout, uint16_t hard_timeout,
122 ovs_be64 flow_cookie, bool send_flow_removed);
123 static void rule_destroy(struct ofproto *, struct rule *);
124 static void rule_free(struct rule *);
126 static struct rule *rule_lookup(struct ofproto *, const struct flow *);
127 static void rule_insert(struct ofproto *, struct rule *,
128 struct ofpbuf *packet, uint16_t in_port);
129 static void rule_remove(struct ofproto *, struct rule *);
131 static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason);
133 /* An exact-match instantiation of an OpenFlow flow. */
135 long long int used; /* Time last used; time created if not used. */
139 * - Do include packets and bytes sent "by hand", e.g. with
142 * - Do include packets and bytes that were obtained from the datapath
143 * when a flow was deleted (e.g. dpif_flow_del()) or when its
144 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
146 * - Do not include any packets or bytes that can currently be obtained
147 * from the datapath by, e.g., dpif_flow_get().
149 uint64_t packet_count; /* Number of packets received. */
150 uint64_t byte_count; /* Number of bytes received. */
152 /* Number of bytes passed to account_cb. This may include bytes that can
153 * currently obtained from the datapath (thus, it can be greater than
155 uint64_t accounted_bytes;
157 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
158 struct list list_node; /* In owning rule's 'facets' list. */
159 struct rule *rule; /* Owning rule. */
160 struct flow flow; /* Exact-match flow. */
161 bool installed; /* Installed in datapath? */
162 bool may_install; /* True ordinarily; false if actions must
163 * be reassessed for every packet. */
164 int n_actions; /* Number of elements in actions[]. */
165 union odp_action *actions; /* Datapath actions. */
166 tag_type tags; /* Tags (set only by hooks). */
167 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
170 static struct facet *facet_create(struct ofproto *, struct rule *,
172 const struct ofpbuf *packet);
173 static void facet_remove(struct ofproto *, struct facet *);
174 static void facet_free(struct facet *);
176 static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *);
177 static bool facet_revalidate(struct ofproto *, struct facet *);
179 static void facet_install(struct ofproto *, struct facet *, bool zero_stats);
180 static void facet_uninstall(struct ofproto *, struct facet *);
181 static void facet_flush_stats(struct ofproto *, struct facet *);
183 static bool facet_make_actions(struct ofproto *, struct facet *,
184 const struct ofpbuf *packet);
185 static void facet_update_stats(struct ofproto *, struct facet *,
186 const struct odp_flow_stats *);
188 /* ofproto supports two kinds of OpenFlow connections:
190 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
191 * maintains persistent connections to these controllers and by default
192 * sends them asynchronous messages such as packet-ins.
194 * - "Service" connections, e.g. from ovs-ofctl. When these connections
195 * drop, it is the other side's responsibility to reconnect them if
196 * necessary. ofproto does not send them asynchronous messages by default.
198 * Currently, active (tcp, ssl, unix) connections are always "primary"
199 * connections and passive (ptcp, pssl, punix) connections are always "service"
200 * connections. There is no inherent reason for this, but it reflects the
204 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
205 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
208 /* A listener for incoming OpenFlow "service" connections. */
210 struct hmap_node node; /* In struct ofproto's "services" hmap. */
211 struct pvconn *pvconn; /* OpenFlow connection listener. */
213 /* These are not used by ofservice directly. They are settings for
214 * accepted "struct ofconn"s from the pvconn. */
215 int probe_interval; /* Max idle time before probing, in seconds. */
216 int rate_limit; /* Max packet-in rate in packets per second. */
217 int burst_limit; /* Limit on accumulating packet credits. */
220 static struct ofservice *ofservice_lookup(struct ofproto *,
222 static int ofservice_create(struct ofproto *,
223 const struct ofproto_controller *);
224 static void ofservice_reconfigure(struct ofservice *,
225 const struct ofproto_controller *);
226 static void ofservice_destroy(struct ofproto *, struct ofservice *);
228 /* An OpenFlow connection. */
230 struct ofproto *ofproto; /* The ofproto that owns this connection. */
231 struct list node; /* In struct ofproto's "all_conns" list. */
232 struct rconn *rconn; /* OpenFlow connection. */
233 enum ofconn_type type; /* Type. */
234 int flow_format; /* One of NXFF_*. */
236 /* OFPT_PACKET_IN related data. */
237 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
238 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
239 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
240 int miss_send_len; /* Bytes to send of buffered packets. */
242 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
243 * requests, and the maximum number before we stop reading OpenFlow
245 #define OFCONN_REPLY_MAX 100
246 struct rconn_packet_counter *reply_counter;
248 /* type == OFCONN_PRIMARY only. */
249 enum nx_role role; /* Role. */
250 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
251 struct discovery *discovery; /* Controller discovery object, if enabled. */
252 struct status_category *ss; /* Switch status category. */
253 enum ofproto_band band; /* In-band or out-of-band? */
256 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
257 * "schedulers" array. Their values are 0 and 1, and their meanings and values
258 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
259 * case anything ever changes, check their values here. */
260 #define N_SCHEDULERS 2
261 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
262 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
263 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
264 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
266 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
268 static void ofconn_destroy(struct ofconn *);
269 static void ofconn_run(struct ofconn *);
270 static void ofconn_wait(struct ofconn *);
271 static bool ofconn_receives_async_msgs(const struct ofconn *);
272 static char *ofconn_make_name(const struct ofproto *, const char *target);
273 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
275 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
276 struct rconn_packet_counter *counter);
278 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
279 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
283 uint64_t datapath_id; /* Datapath ID. */
284 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
285 char *mfr_desc; /* Manufacturer. */
286 char *hw_desc; /* Hardware. */
287 char *sw_desc; /* Software version. */
288 char *serial_desc; /* Serial number. */
289 char *dp_desc; /* Datapath description. */
293 struct netdev_monitor *netdev_monitor;
294 struct hmap ports; /* Contains "struct ofport"s. */
295 struct shash port_by_name;
299 struct switch_status *switch_status;
300 struct fail_open *fail_open;
301 struct netflow *netflow;
302 struct ofproto_sflow *sflow;
304 /* In-band control. */
305 struct in_band *in_band;
306 long long int next_in_band_update;
307 struct sockaddr_in *extra_in_band_remotes;
308 size_t n_extra_remotes;
311 struct classifier cls;
312 long long int next_expiration;
316 bool need_revalidate;
317 struct tag_set revalidate_set;
319 /* OpenFlow connections. */
320 struct hmap controllers; /* Controller "struct ofconn"s. */
321 struct list all_conns; /* Contains "struct ofconn"s. */
322 enum ofproto_fail_mode fail_mode;
324 /* OpenFlow listeners. */
325 struct hmap services; /* Contains "struct ofservice"s. */
326 struct pvconn **snoops;
329 /* Hooks for ovs-vswitchd. */
330 const struct ofhooks *ofhooks;
333 /* Used by default ofhooks. */
334 struct mac_learning *ml;
337 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
339 static const struct ofhooks default_ofhooks;
341 static uint64_t pick_datapath_id(const struct ofproto *);
342 static uint64_t pick_fallback_dpid(void);
344 static int ofproto_expire(struct ofproto *);
346 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
348 static void handle_openflow(struct ofconn *, struct ofpbuf *);
350 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
351 static void update_port(struct ofproto *, const char *devname);
352 static int init_ports(struct ofproto *);
353 static void reinit_ports(struct ofproto *);
356 ofproto_create(const char *datapath, const char *datapath_type,
357 const struct ofhooks *ofhooks, void *aux,
358 struct ofproto **ofprotop)
360 struct odp_stats stats;
367 /* Connect to datapath and start listening for messages. */
368 error = dpif_open(datapath, datapath_type, &dpif);
370 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
373 error = dpif_get_dp_stats(dpif, &stats);
375 VLOG_ERR("failed to obtain stats for datapath %s: %s",
376 datapath, strerror(error));
380 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
382 VLOG_ERR("failed to listen on datapath %s: %s",
383 datapath, strerror(error));
387 dpif_flow_flush(dpif);
388 dpif_recv_purge(dpif);
390 /* Initialize settings. */
391 p = xzalloc(sizeof *p);
392 p->fallback_dpid = pick_fallback_dpid();
393 p->datapath_id = p->fallback_dpid;
394 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
395 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
396 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
397 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
398 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
400 /* Initialize datapath. */
402 p->netdev_monitor = netdev_monitor_create();
403 hmap_init(&p->ports);
404 shash_init(&p->port_by_name);
405 p->max_ports = stats.max_ports;
407 /* Initialize submodules. */
408 p->switch_status = switch_status_create(p);
414 /* Initialize flow table. */
415 classifier_init(&p->cls);
416 p->next_expiration = time_msec() + 1000;
418 /* Initialize facet table. */
419 hmap_init(&p->facets);
420 p->need_revalidate = false;
421 tag_set_init(&p->revalidate_set);
423 /* Initialize OpenFlow connections. */
424 list_init(&p->all_conns);
425 hmap_init(&p->controllers);
426 hmap_init(&p->services);
430 /* Initialize hooks. */
432 p->ofhooks = ofhooks;
436 p->ofhooks = &default_ofhooks;
438 p->ml = mac_learning_create();
441 /* Pick final datapath ID. */
442 p->datapath_id = pick_datapath_id(p);
443 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
450 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
452 uint64_t old_dpid = p->datapath_id;
453 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
454 if (p->datapath_id != old_dpid) {
455 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
457 /* Force all active connections to reconnect, since there is no way to
458 * notify a controller that the datapath ID has changed. */
459 ofproto_reconnect_controllers(p);
464 is_discovery_controller(const struct ofproto_controller *c)
466 return !strcmp(c->target, "discover");
470 is_in_band_controller(const struct ofproto_controller *c)
472 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
475 /* Creates a new controller in 'ofproto'. Some of the settings are initially
476 * drawn from 'c', but update_controller() needs to be called later to finish
477 * the new ofconn's configuration. */
479 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
481 struct discovery *discovery;
482 struct ofconn *ofconn;
484 if (is_discovery_controller(c)) {
485 int error = discovery_create(c->accept_re, c->update_resolv_conf,
486 ofproto->dpif, ofproto->switch_status,
495 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
496 ofconn->pktbuf = pktbuf_create();
497 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
499 ofconn->discovery = discovery;
501 char *name = ofconn_make_name(ofproto, c->target);
502 rconn_connect(ofconn->rconn, c->target, name);
505 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
506 hash_string(c->target, 0));
509 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
510 * target or turn discovery on or off (these are done by creating new ofconns
511 * and deleting old ones), but it can update the rest of an ofconn's
514 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
518 ofconn->band = (is_in_band_controller(c)
519 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
521 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
523 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
524 rconn_set_probe_interval(ofconn->rconn, probe_interval);
526 if (ofconn->discovery) {
527 discovery_set_update_resolv_conf(ofconn->discovery,
528 c->update_resolv_conf);
529 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
532 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
536 ofconn_get_target(const struct ofconn *ofconn)
538 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
541 static struct ofconn *
542 find_controller_by_target(struct ofproto *ofproto, const char *target)
544 struct ofconn *ofconn;
546 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
547 hash_string(target, 0), &ofproto->controllers) {
548 if (!strcmp(ofconn_get_target(ofconn), target)) {
556 update_in_band_remotes(struct ofproto *ofproto)
558 const struct ofconn *ofconn;
559 struct sockaddr_in *addrs;
560 size_t max_addrs, n_addrs;
564 /* Allocate enough memory for as many remotes as we could possibly have. */
565 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
566 addrs = xmalloc(max_addrs * sizeof *addrs);
569 /* Add all the remotes. */
571 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
572 struct sockaddr_in *sin = &addrs[n_addrs];
574 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
578 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
579 if (sin->sin_addr.s_addr) {
580 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
583 if (ofconn->discovery) {
587 for (i = 0; i < ofproto->n_extra_remotes; i++) {
588 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
591 /* Create or update or destroy in-band.
593 * Ordinarily we only enable in-band if there's at least one remote
594 * address, but discovery needs the in-band rules for DHCP to be installed
595 * even before we know any remote addresses. */
596 if (n_addrs || discovery) {
597 if (!ofproto->in_band) {
598 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
601 if (ofproto->in_band) {
602 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
604 ofproto->next_in_band_update = time_msec() + 1000;
606 in_band_destroy(ofproto->in_band);
607 ofproto->in_band = NULL;
615 update_fail_open(struct ofproto *p)
617 struct ofconn *ofconn;
619 if (!hmap_is_empty(&p->controllers)
620 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
621 struct rconn **rconns;
625 p->fail_open = fail_open_create(p, p->switch_status);
629 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
630 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
631 rconns[n++] = ofconn->rconn;
634 fail_open_set_controllers(p->fail_open, rconns, n);
635 /* p->fail_open takes ownership of 'rconns'. */
637 fail_open_destroy(p->fail_open);
643 ofproto_set_controllers(struct ofproto *p,
644 const struct ofproto_controller *controllers,
645 size_t n_controllers)
647 struct shash new_controllers;
648 struct ofconn *ofconn, *next_ofconn;
649 struct ofservice *ofservice, *next_ofservice;
653 /* Create newly configured controllers and services.
654 * Create a name to ofproto_controller mapping in 'new_controllers'. */
655 shash_init(&new_controllers);
656 for (i = 0; i < n_controllers; i++) {
657 const struct ofproto_controller *c = &controllers[i];
659 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
660 if (!find_controller_by_target(p, c->target)) {
661 add_controller(p, c);
663 } else if (!pvconn_verify_name(c->target)) {
664 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
668 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
669 dpif_name(p->dpif), c->target);
673 shash_add_once(&new_controllers, c->target, &controllers[i]);
676 /* Delete controllers that are no longer configured.
677 * Update configuration of all now-existing controllers. */
679 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
680 struct ofproto_controller *c;
682 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
684 ofconn_destroy(ofconn);
686 update_controller(ofconn, c);
693 /* Delete services that are no longer configured.
694 * Update configuration of all now-existing services. */
695 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
696 struct ofproto_controller *c;
698 c = shash_find_data(&new_controllers,
699 pvconn_get_name(ofservice->pvconn));
701 ofservice_destroy(p, ofservice);
703 ofservice_reconfigure(ofservice, c);
707 shash_destroy(&new_controllers);
709 update_in_band_remotes(p);
712 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
713 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
714 struct ofconn, hmap_node);
715 ofconn->ss = switch_status_register(p->switch_status, "remote",
716 rconn_status_cb, ofconn->rconn);
721 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
723 p->fail_mode = fail_mode;
727 /* Drops the connections between 'ofproto' and all of its controllers, forcing
728 * them to reconnect. */
730 ofproto_reconnect_controllers(struct ofproto *ofproto)
732 struct ofconn *ofconn;
734 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
735 rconn_reconnect(ofconn->rconn);
740 any_extras_changed(const struct ofproto *ofproto,
741 const struct sockaddr_in *extras, size_t n)
745 if (n != ofproto->n_extra_remotes) {
749 for (i = 0; i < n; i++) {
750 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
751 const struct sockaddr_in *new = &extras[i];
753 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
754 old->sin_port != new->sin_port) {
762 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
763 * in-band control should guarantee access, in the same way that in-band
764 * control guarantees access to OpenFlow controllers. */
766 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
767 const struct sockaddr_in *extras, size_t n)
769 if (!any_extras_changed(ofproto, extras, n)) {
773 free(ofproto->extra_in_band_remotes);
774 ofproto->n_extra_remotes = n;
775 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
777 update_in_band_remotes(ofproto);
781 ofproto_set_desc(struct ofproto *p,
782 const char *mfr_desc, const char *hw_desc,
783 const char *sw_desc, const char *serial_desc,
786 struct ofp_desc_stats *ods;
789 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
790 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
791 sizeof ods->mfr_desc);
794 p->mfr_desc = xstrdup(mfr_desc);
797 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
798 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
799 sizeof ods->hw_desc);
802 p->hw_desc = xstrdup(hw_desc);
805 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
806 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
807 sizeof ods->sw_desc);
810 p->sw_desc = xstrdup(sw_desc);
813 if (strlen(serial_desc) >= sizeof ods->serial_num) {
814 VLOG_WARN("truncating serial_desc, must be less than %zu "
816 sizeof ods->serial_num);
818 free(p->serial_desc);
819 p->serial_desc = xstrdup(serial_desc);
822 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
823 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
824 sizeof ods->dp_desc);
827 p->dp_desc = xstrdup(dp_desc);
832 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
833 const struct svec *svec)
835 struct pvconn **pvconns = *pvconnsp;
836 size_t n_pvconns = *n_pvconnsp;
840 for (i = 0; i < n_pvconns; i++) {
841 pvconn_close(pvconns[i]);
845 pvconns = xmalloc(svec->n * sizeof *pvconns);
847 for (i = 0; i < svec->n; i++) {
848 const char *name = svec->names[i];
849 struct pvconn *pvconn;
852 error = pvconn_open(name, &pvconn);
854 pvconns[n_pvconns++] = pvconn;
856 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
864 *n_pvconnsp = n_pvconns;
870 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
872 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
876 ofproto_set_netflow(struct ofproto *ofproto,
877 const struct netflow_options *nf_options)
879 if (nf_options && nf_options->collectors.n) {
880 if (!ofproto->netflow) {
881 ofproto->netflow = netflow_create();
883 return netflow_set_options(ofproto->netflow, nf_options);
885 netflow_destroy(ofproto->netflow);
886 ofproto->netflow = NULL;
892 ofproto_set_sflow(struct ofproto *ofproto,
893 const struct ofproto_sflow_options *oso)
895 struct ofproto_sflow *os = ofproto->sflow;
898 struct ofport *ofport;
900 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
901 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
902 ofproto_sflow_add_port(os, ofport->odp_port,
903 netdev_get_name(ofport->netdev));
906 ofproto_sflow_set_options(os, oso);
908 ofproto_sflow_destroy(os);
909 ofproto->sflow = NULL;
914 ofproto_get_datapath_id(const struct ofproto *ofproto)
916 return ofproto->datapath_id;
920 ofproto_has_primary_controller(const struct ofproto *ofproto)
922 return !hmap_is_empty(&ofproto->controllers);
925 enum ofproto_fail_mode
926 ofproto_get_fail_mode(const struct ofproto *p)
932 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
936 for (i = 0; i < ofproto->n_snoops; i++) {
937 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
942 ofproto_destroy(struct ofproto *p)
944 struct ofservice *ofservice, *next_ofservice;
945 struct ofconn *ofconn, *next_ofconn;
946 struct ofport *ofport, *next_ofport;
953 /* Destroy fail-open and in-band early, since they touch the classifier. */
954 fail_open_destroy(p->fail_open);
957 in_band_destroy(p->in_band);
959 free(p->extra_in_band_remotes);
961 ofproto_flush_flows(p);
962 classifier_destroy(&p->cls);
963 hmap_destroy(&p->facets);
965 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
966 ofconn_destroy(ofconn);
968 hmap_destroy(&p->controllers);
971 netdev_monitor_destroy(p->netdev_monitor);
972 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
973 hmap_remove(&p->ports, &ofport->hmap_node);
976 shash_destroy(&p->port_by_name);
978 switch_status_destroy(p->switch_status);
979 netflow_destroy(p->netflow);
980 ofproto_sflow_destroy(p->sflow);
982 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
983 ofservice_destroy(p, ofservice);
985 hmap_destroy(&p->services);
987 for (i = 0; i < p->n_snoops; i++) {
988 pvconn_close(p->snoops[i]);
992 mac_learning_destroy(p->ml);
997 free(p->serial_desc);
1000 hmap_destroy(&p->ports);
1006 ofproto_run(struct ofproto *p)
1008 int error = ofproto_run1(p);
1010 error = ofproto_run2(p, false);
1016 process_port_change(struct ofproto *ofproto, int error, char *devname)
1018 if (error == ENOBUFS) {
1019 reinit_ports(ofproto);
1020 } else if (!error) {
1021 update_port(ofproto, devname);
1026 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1027 * means that 'ofconn' is more interesting for monitoring than a lower return
1030 snoop_preference(const struct ofconn *ofconn)
1032 switch (ofconn->role) {
1033 case NX_ROLE_MASTER:
1040 /* Shouldn't happen. */
1045 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1046 * Connects this vconn to a controller. */
1048 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1050 struct ofconn *ofconn, *best;
1052 /* Pick a controller for monitoring. */
1054 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1055 if (ofconn->type == OFCONN_PRIMARY
1056 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1062 rconn_add_monitor(best->rconn, vconn);
1064 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1070 ofproto_run1(struct ofproto *p)
1072 struct ofconn *ofconn, *next_ofconn;
1073 struct ofservice *ofservice;
1078 if (shash_is_empty(&p->port_by_name)) {
1082 for (i = 0; i < 50; i++) {
1085 error = dpif_recv(p->dpif, &buf);
1087 if (error == ENODEV) {
1088 /* Someone destroyed the datapath behind our back. The caller
1089 * better destroy us and give up, because we're just going to
1090 * spin from here on out. */
1091 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1092 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1093 dpif_name(p->dpif));
1099 handle_odp_msg(p, buf);
1102 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1103 process_port_change(p, error, devname);
1105 while ((error = netdev_monitor_poll(p->netdev_monitor,
1106 &devname)) != EAGAIN) {
1107 process_port_change(p, error, devname);
1111 if (time_msec() >= p->next_in_band_update) {
1112 update_in_band_remotes(p);
1114 in_band_run(p->in_band);
1117 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1121 /* Fail-open maintenance. Do this after processing the ofconns since
1122 * fail-open checks the status of the controller rconn. */
1124 fail_open_run(p->fail_open);
1127 HMAP_FOR_EACH (ofservice, node, &p->services) {
1128 struct vconn *vconn;
1131 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1133 struct rconn *rconn;
1136 rconn = rconn_create(ofservice->probe_interval, 0);
1137 name = ofconn_make_name(p, vconn_get_name(vconn));
1138 rconn_connect_unreliably(rconn, vconn, name);
1141 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1142 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1143 ofservice->burst_limit);
1144 } else if (retval != EAGAIN) {
1145 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1149 for (i = 0; i < p->n_snoops; i++) {
1150 struct vconn *vconn;
1153 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1155 add_snooper(p, vconn);
1156 } else if (retval != EAGAIN) {
1157 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1161 if (time_msec() >= p->next_expiration) {
1162 int delay = ofproto_expire(p);
1163 p->next_expiration = time_msec() + delay;
1164 COVERAGE_INC(ofproto_expiration);
1168 netflow_run(p->netflow);
1171 ofproto_sflow_run(p->sflow);
1178 ofproto_run2(struct ofproto *p, bool revalidate_all)
1180 /* Figure out what we need to revalidate now, if anything. */
1181 struct tag_set revalidate_set = p->revalidate_set;
1182 if (p->need_revalidate) {
1183 revalidate_all = true;
1186 /* Clear the revalidation flags. */
1187 tag_set_init(&p->revalidate_set);
1188 p->need_revalidate = false;
1190 /* Now revalidate if there's anything to do. */
1191 if (revalidate_all || !tag_set_is_empty(&revalidate_set)) {
1192 struct facet *facet, *next;
1194 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) {
1196 || tag_set_intersects(&revalidate_set, facet->tags)) {
1197 facet_revalidate(p, facet);
1206 ofproto_wait(struct ofproto *p)
1208 struct ofservice *ofservice;
1209 struct ofconn *ofconn;
1212 dpif_recv_wait(p->dpif);
1213 dpif_port_poll_wait(p->dpif);
1214 netdev_monitor_poll_wait(p->netdev_monitor);
1215 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1216 ofconn_wait(ofconn);
1219 poll_timer_wait_until(p->next_in_band_update);
1220 in_band_wait(p->in_band);
1223 fail_open_wait(p->fail_open);
1226 ofproto_sflow_wait(p->sflow);
1228 if (!tag_set_is_empty(&p->revalidate_set)) {
1229 poll_immediate_wake();
1231 if (p->need_revalidate) {
1232 /* Shouldn't happen, but if it does just go around again. */
1233 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1234 poll_immediate_wake();
1235 } else if (p->next_expiration != LLONG_MAX) {
1236 poll_timer_wait_until(p->next_expiration);
1238 HMAP_FOR_EACH (ofservice, node, &p->services) {
1239 pvconn_wait(ofservice->pvconn);
1241 for (i = 0; i < p->n_snoops; i++) {
1242 pvconn_wait(p->snoops[i]);
1247 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1249 tag_set_add(&ofproto->revalidate_set, tag);
1253 ofproto_get_revalidate_set(struct ofproto *ofproto)
1255 return &ofproto->revalidate_set;
1259 ofproto_is_alive(const struct ofproto *p)
1261 return !hmap_is_empty(&p->controllers);
1264 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1266 * This is almost the same as calling dpif_port_del() directly on the
1267 * datapath, but it also makes 'ofproto' close its open netdev for the port
1268 * (if any). This makes it possible to create a new netdev of a different
1269 * type under the same name, which otherwise the netdev library would refuse
1270 * to do because of the conflict. (The netdev would eventually get closed on
1271 * the next trip through ofproto_run(), but this interface is more direct.)
1273 * Returns 0 if successful, otherwise a positive errno. */
1275 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1277 struct ofport *ofport = get_port(ofproto, odp_port);
1278 const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
1281 error = dpif_port_del(ofproto->dpif, odp_port);
1283 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1284 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1285 } else if (ofport) {
1286 /* 'name' is ofport->opp.name and update_port() is going to destroy
1287 * 'ofport'. Just in case update_port() refers to 'name' after it
1288 * destroys 'ofport', make a copy of it around the update_port()
1290 char *devname = xstrdup(name);
1291 update_port(ofproto, devname);
1297 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1298 * true if 'odp_port' exists and should be included, false otherwise. */
1300 ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port)
1302 struct ofport *ofport = get_port(ofproto, odp_port);
1303 return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD);
1307 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1308 const union ofp_action *actions, size_t n_actions,
1309 const struct ofpbuf *packet)
1311 struct odp_actions odp_actions;
1314 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1320 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1322 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
1326 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1327 * performs the 'n_actions' actions in 'actions'. The new flow will not
1330 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1331 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1332 * controllers; otherwise, it will be hidden.
1334 * The caller retains ownership of 'cls_rule' and 'actions'. */
1336 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
1337 const union ofp_action *actions, size_t n_actions)
1340 rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false);
1341 rule_insert(p, rule, NULL, 0);
1345 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1349 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1352 rule_remove(ofproto, rule);
1357 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1359 struct rule *rule = rule_from_cls_rule(rule_);
1360 struct ofproto *ofproto = ofproto_;
1362 rule_remove(ofproto, rule);
1366 ofproto_flush_flows(struct ofproto *ofproto)
1368 struct facet *facet, *next_facet;
1370 COVERAGE_INC(ofproto_flush);
1372 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
1373 /* Mark the facet as not installed so that facet_remove() doesn't
1374 * bother trying to uninstall it. There is no point in uninstalling it
1375 * individually since we are about to blow away all the facets with
1376 * dpif_flow_flush(). */
1377 facet->installed = false;
1378 facet_remove(ofproto, facet);
1380 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1381 dpif_flow_flush(ofproto->dpif);
1382 if (ofproto->in_band) {
1383 in_band_flushed(ofproto->in_band);
1385 if (ofproto->fail_open) {
1386 fail_open_flushed(ofproto->fail_open);
1391 reinit_ports(struct ofproto *p)
1393 struct svec devnames;
1394 struct ofport *ofport;
1395 struct odp_port *odp_ports;
1399 COVERAGE_INC(ofproto_reinit_ports);
1401 svec_init(&devnames);
1402 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1403 svec_add (&devnames, (char *) ofport->opp.name);
1405 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1406 for (i = 0; i < n_odp_ports; i++) {
1407 svec_add (&devnames, odp_ports[i].devname);
1411 svec_sort_unique(&devnames);
1412 for (i = 0; i < devnames.n; i++) {
1413 update_port(p, devnames.names[i]);
1415 svec_destroy(&devnames);
1418 static struct ofport *
1419 make_ofport(const struct odp_port *odp_port)
1421 struct netdev_options netdev_options;
1422 enum netdev_flags flags;
1423 struct ofport *ofport;
1424 struct netdev *netdev;
1427 memset(&netdev_options, 0, sizeof netdev_options);
1428 netdev_options.name = odp_port->devname;
1429 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1431 error = netdev_open(&netdev_options, &netdev);
1433 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1434 "cannot be opened (%s)",
1435 odp_port->devname, odp_port->port,
1436 odp_port->devname, strerror(error));
1440 ofport = xmalloc(sizeof *ofport);
1441 ofport->netdev = netdev;
1442 ofport->odp_port = odp_port->port;
1443 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1444 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1445 memcpy(ofport->opp.name, odp_port->devname,
1446 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1447 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1449 netdev_get_flags(netdev, &flags);
1450 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1452 ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN;
1454 netdev_get_features(netdev,
1455 &ofport->opp.curr, &ofport->opp.advertised,
1456 &ofport->opp.supported, &ofport->opp.peer);
1461 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1463 if (get_port(p, odp_port->port)) {
1464 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1467 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1468 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1477 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1479 const struct ofp_phy_port *a = &a_->opp;
1480 const struct ofp_phy_port *b = &b_->opp;
1482 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1483 return (a->port_no == b->port_no
1484 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1485 && !strcmp((char *) a->name, (char *) b->name)
1486 && a->state == b->state
1487 && a->config == b->config
1488 && a->curr == b->curr
1489 && a->advertised == b->advertised
1490 && a->supported == b->supported
1491 && a->peer == b->peer);
1495 send_port_status(struct ofproto *p, const struct ofport *ofport,
1498 /* XXX Should limit the number of queued port status change messages. */
1499 struct ofconn *ofconn;
1500 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1501 struct ofp_port_status *ops;
1504 if (!ofconn_receives_async_msgs(ofconn)) {
1508 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1509 ops->reason = reason;
1510 ops->desc = ofport->opp;
1511 hton_ofp_phy_port(&ops->desc);
1512 queue_tx(b, ofconn, NULL);
1517 ofport_install(struct ofproto *p, struct ofport *ofport)
1519 const char *netdev_name = (const char *) ofport->opp.name;
1521 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1522 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1523 shash_add(&p->port_by_name, netdev_name, ofport);
1525 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1530 ofport_remove(struct ofproto *p, struct ofport *ofport)
1532 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1533 hmap_remove(&p->ports, &ofport->hmap_node);
1534 shash_delete(&p->port_by_name,
1535 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1537 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1542 ofport_free(struct ofport *ofport)
1545 netdev_close(ofport->netdev);
1550 static struct ofport *
1551 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1553 struct ofport *port;
1555 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1556 hash_int(odp_port, 0), &ofproto->ports) {
1557 if (port->odp_port == odp_port) {
1565 update_port(struct ofproto *p, const char *devname)
1567 struct odp_port odp_port;
1568 struct ofport *old_ofport;
1569 struct ofport *new_ofport;
1572 COVERAGE_INC(ofproto_update_port);
1574 /* Query the datapath for port information. */
1575 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1577 /* Find the old ofport. */
1578 old_ofport = shash_find_data(&p->port_by_name, devname);
1581 /* There's no port named 'devname' but there might be a port with
1582 * the same port number. This could happen if a port is deleted
1583 * and then a new one added in its place very quickly, or if a port
1584 * is renamed. In the former case we want to send an OFPPR_DELETE
1585 * and an OFPPR_ADD, and in the latter case we want to send a
1586 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1587 * the old port's ifindex against the new port, or perhaps less
1588 * reliably but more portably by comparing the old port's MAC
1589 * against the new port's MAC. However, this code isn't that smart
1590 * and always sends an OFPPR_MODIFY (XXX). */
1591 old_ofport = get_port(p, odp_port.port);
1593 } else if (error != ENOENT && error != ENODEV) {
1594 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1595 "%s", strerror(error));
1599 /* Create a new ofport. */
1600 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1602 /* Eliminate a few pathological cases. */
1603 if (!old_ofport && !new_ofport) {
1605 } else if (old_ofport && new_ofport) {
1606 /* Most of the 'config' bits are OpenFlow soft state, but
1607 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1608 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1609 * leaves the other bits 0.) */
1610 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1612 if (ofport_equal(old_ofport, new_ofport)) {
1613 /* False alarm--no change. */
1614 ofport_free(new_ofport);
1619 /* Now deal with the normal cases. */
1621 ofport_remove(p, old_ofport);
1624 ofport_install(p, new_ofport);
1626 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1627 (!old_ofport ? OFPPR_ADD
1628 : !new_ofport ? OFPPR_DELETE
1630 ofport_free(old_ofport);
1634 init_ports(struct ofproto *p)
1636 struct odp_port *ports;
1641 error = dpif_port_list(p->dpif, &ports, &n_ports);
1646 for (i = 0; i < n_ports; i++) {
1647 const struct odp_port *odp_port = &ports[i];
1648 if (!ofport_conflicts(p, odp_port)) {
1649 struct ofport *ofport = make_ofport(odp_port);
1651 ofport_install(p, ofport);
1659 static struct ofconn *
1660 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1662 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1663 ofconn->ofproto = p;
1664 list_push_back(&p->all_conns, &ofconn->node);
1665 ofconn->rconn = rconn;
1666 ofconn->type = type;
1667 ofconn->flow_format = NXFF_OPENFLOW10;
1668 ofconn->role = NX_ROLE_OTHER;
1669 ofconn->packet_in_counter = rconn_packet_counter_create ();
1670 ofconn->pktbuf = NULL;
1671 ofconn->miss_send_len = 0;
1672 ofconn->reply_counter = rconn_packet_counter_create ();
1677 ofconn_destroy(struct ofconn *ofconn)
1679 if (ofconn->type == OFCONN_PRIMARY) {
1680 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1682 discovery_destroy(ofconn->discovery);
1684 list_remove(&ofconn->node);
1685 switch_status_unregister(ofconn->ss);
1686 rconn_destroy(ofconn->rconn);
1687 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1688 rconn_packet_counter_destroy(ofconn->reply_counter);
1689 pktbuf_destroy(ofconn->pktbuf);
1694 ofconn_run(struct ofconn *ofconn)
1696 struct ofproto *p = ofconn->ofproto;
1700 if (ofconn->discovery) {
1701 char *controller_name;
1702 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1703 discovery_question_connectivity(ofconn->discovery);
1705 if (discovery_run(ofconn->discovery, &controller_name)) {
1706 if (controller_name) {
1707 char *ofconn_name = ofconn_make_name(p, controller_name);
1708 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1711 rconn_disconnect(ofconn->rconn);
1716 for (i = 0; i < N_SCHEDULERS; i++) {
1717 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1720 rconn_run(ofconn->rconn);
1722 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1723 /* Limit the number of iterations to prevent other tasks from
1725 for (iteration = 0; iteration < 50; iteration++) {
1726 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1731 fail_open_maybe_recover(p->fail_open);
1733 handle_openflow(ofconn, of_msg);
1734 ofpbuf_delete(of_msg);
1738 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1739 ofconn_destroy(ofconn);
1744 ofconn_wait(struct ofconn *ofconn)
1748 if (ofconn->discovery) {
1749 discovery_wait(ofconn->discovery);
1751 for (i = 0; i < N_SCHEDULERS; i++) {
1752 pinsched_wait(ofconn->schedulers[i]);
1754 rconn_run_wait(ofconn->rconn);
1755 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1756 rconn_recv_wait(ofconn->rconn);
1758 COVERAGE_INC(ofproto_ofconn_stuck);
1762 /* Returns true if 'ofconn' should receive asynchronous messages. */
1764 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1766 if (ofconn->type == OFCONN_PRIMARY) {
1767 /* Primary controllers always get asynchronous messages unless they
1768 * have configured themselves as "slaves". */
1769 return ofconn->role != NX_ROLE_SLAVE;
1771 /* Service connections don't get asynchronous messages unless they have
1772 * explicitly asked for them by setting a nonzero miss send length. */
1773 return ofconn->miss_send_len > 0;
1777 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1778 * and 'target', suitable for use in log messages for identifying the
1781 * The name is dynamically allocated. The caller should free it (with free())
1782 * when it is no longer needed. */
1784 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1786 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1790 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1794 for (i = 0; i < N_SCHEDULERS; i++) {
1795 struct pinsched **s = &ofconn->schedulers[i];
1799 *s = pinsched_create(rate, burst,
1800 ofconn->ofproto->switch_status);
1802 pinsched_set_limits(*s, rate, burst);
1805 pinsched_destroy(*s);
1812 ofservice_reconfigure(struct ofservice *ofservice,
1813 const struct ofproto_controller *c)
1815 ofservice->probe_interval = c->probe_interval;
1816 ofservice->rate_limit = c->rate_limit;
1817 ofservice->burst_limit = c->burst_limit;
1820 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1821 * positive errno value. */
1823 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1825 struct ofservice *ofservice;
1826 struct pvconn *pvconn;
1829 error = pvconn_open(c->target, &pvconn);
1834 ofservice = xzalloc(sizeof *ofservice);
1835 hmap_insert(&ofproto->services, &ofservice->node,
1836 hash_string(c->target, 0));
1837 ofservice->pvconn = pvconn;
1839 ofservice_reconfigure(ofservice, c);
1845 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1847 hmap_remove(&ofproto->services, &ofservice->node);
1848 pvconn_close(ofservice->pvconn);
1852 /* Finds and returns the ofservice within 'ofproto' that has the given
1853 * 'target', or a null pointer if none exists. */
1854 static struct ofservice *
1855 ofservice_lookup(struct ofproto *ofproto, const char *target)
1857 struct ofservice *ofservice;
1859 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1860 &ofproto->services) {
1861 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1868 /* Returns true if 'rule' should be hidden from the controller.
1870 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1871 * (e.g. by in-band control) and are intentionally hidden from the
1874 rule_is_hidden(const struct rule *rule)
1876 return rule->cr.priority > UINT16_MAX;
1879 /* Creates and returns a new rule initialized as specified.
1881 * The caller is responsible for inserting the rule into the classifier (with
1882 * rule_insert()). */
1883 static struct rule *
1884 rule_create(const struct cls_rule *cls_rule,
1885 const union ofp_action *actions, size_t n_actions,
1886 uint16_t idle_timeout, uint16_t hard_timeout,
1887 ovs_be64 flow_cookie, bool send_flow_removed)
1889 struct rule *rule = xzalloc(sizeof *rule);
1890 rule->cr = *cls_rule;
1891 rule->idle_timeout = idle_timeout;
1892 rule->hard_timeout = hard_timeout;
1893 rule->flow_cookie = flow_cookie;
1894 rule->used = rule->created = time_msec();
1895 rule->send_flow_removed = send_flow_removed;
1896 list_init(&rule->facets);
1897 if (n_actions > 0) {
1898 rule->n_actions = n_actions;
1899 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1905 static struct rule *
1906 rule_from_cls_rule(const struct cls_rule *cls_rule)
1908 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1912 rule_free(struct rule *rule)
1914 free(rule->actions);
1918 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1919 * destroying any that no longer has a rule (which is probably all of them).
1921 * The caller must have already removed 'rule' from the classifier. */
1923 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1925 struct facet *facet, *next_facet;
1926 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
1927 facet_revalidate(ofproto, facet);
1932 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1933 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1936 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1938 const union ofp_action *oa;
1939 struct actions_iterator i;
1941 if (out_port == htons(OFPP_NONE)) {
1944 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1945 oa = actions_next(&i)) {
1946 if (action_outputs_to_port(oa, out_port)) {
1953 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1954 * 'packet', which arrived on 'in_port'.
1956 * Takes ownership of 'packet'. */
1958 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1959 const union odp_action *actions, size_t n_actions,
1960 struct ofpbuf *packet)
1962 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1963 /* As an optimization, avoid a round-trip from userspace to kernel to
1964 * userspace. This also avoids possibly filling up kernel packet
1965 * buffers along the way. */
1966 struct odp_msg *msg;
1968 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1969 msg->type = _ODPL_ACTION_NR;
1970 msg->length = sizeof(struct odp_msg) + packet->size;
1971 msg->port = in_port;
1973 msg->arg = actions[0].controller.arg;
1975 send_packet_in(ofproto, packet);
1981 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
1982 ofpbuf_delete(packet);
1987 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1988 * statistics appropriately. 'packet' must have at least sizeof(struct
1989 * ofp_packet_in) bytes of headroom.
1991 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1992 * applying flow_extract() to 'packet' would yield the same flow as
1995 * 'facet' must have accurately composed ODP actions; that is, it must not be
1996 * in need of revalidation.
1998 * Takes ownership of 'packet'. */
2000 facet_execute(struct ofproto *ofproto, struct facet *facet,
2001 struct ofpbuf *packet)
2003 struct odp_flow_stats stats;
2005 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2007 flow_extract_stats(&facet->flow, packet, &stats);
2008 if (execute_odp_actions(ofproto, facet->flow.in_port,
2009 facet->actions, facet->n_actions, packet)) {
2010 facet_update_stats(ofproto, facet, &stats);
2011 facet->used = time_msec();
2012 netflow_flow_update_time(ofproto->netflow,
2013 &facet->nf_flow, facet->used);
2017 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2018 * statistics (or the statistics for one of its facets) appropriately.
2019 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2021 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2022 * with statistics for 'packet' either way.
2024 * Takes ownership of 'packet'. */
2026 rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port,
2027 struct ofpbuf *packet)
2029 struct facet *facet;
2030 struct odp_actions a;
2034 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2036 flow_extract(packet, 0, in_port, &flow);
2038 /* First look for a related facet. If we find one, account it to that. */
2039 facet = facet_lookup_valid(ofproto, &flow);
2040 if (facet && facet->rule == rule) {
2041 facet_execute(ofproto, facet, packet);
2045 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2046 * create a new facet for it and use that. */
2047 if (rule_lookup(ofproto, &flow) == rule) {
2048 facet = facet_create(ofproto, rule, &flow, packet);
2049 facet_execute(ofproto, facet, packet);
2050 facet_install(ofproto, facet, true);
2054 /* We can't account anything to a facet. If we were to try, then that
2055 * facet would have a non-matching rule, busting our invariants. */
2056 if (xlate_actions(rule->actions, rule->n_actions, &flow, ofproto,
2057 packet, &a, NULL, 0, NULL)) {
2058 ofpbuf_delete(packet);
2061 size = packet->size;
2062 if (execute_odp_actions(ofproto, in_port,
2063 a.actions, a.n_actions, packet)) {
2064 rule->used = time_msec();
2065 rule->packet_count++;
2066 rule->byte_count += size;
2070 /* Inserts 'rule' into 'p''s flow table.
2072 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2073 * actions on it and credits the statistics for sending the packet to 'rule'.
2074 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2077 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2080 struct rule *displaced_rule;
2082 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2083 if (displaced_rule) {
2084 rule_destroy(p, displaced_rule);
2086 p->need_revalidate = true;
2089 rule_execute(p, rule, in_port, packet);
2093 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2094 * 'flow' and an example 'packet' within that flow.
2096 * The caller must already have determined that no facet with an identical
2097 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2098 * 'ofproto''s classifier table. */
2099 static struct facet *
2100 facet_create(struct ofproto *ofproto, struct rule *rule,
2101 const struct flow *flow, const struct ofpbuf *packet)
2103 struct facet *facet;
2105 facet = xzalloc(sizeof *facet);
2106 facet->used = time_msec();
2107 hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
2108 list_push_back(&rule->facets, &facet->list_node);
2110 facet->flow = *flow;
2111 netflow_flow_init(&facet->nf_flow);
2112 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
2114 facet_make_actions(ofproto, facet, packet);
2120 facet_free(struct facet *facet)
2122 free(facet->actions);
2126 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2128 * - Removes 'rule' from the classifier.
2130 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2131 * destroys them), via rule_destroy().
2134 rule_remove(struct ofproto *ofproto, struct rule *rule)
2136 COVERAGE_INC(ofproto_del_rule);
2137 ofproto->need_revalidate = true;
2138 classifier_remove(&ofproto->cls, &rule->cr);
2139 rule_destroy(ofproto, rule);
2142 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2144 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2145 * rule's statistics, via facet_uninstall().
2147 * - Removes 'facet' from its rule and from ofproto->facets.
2150 facet_remove(struct ofproto *ofproto, struct facet *facet)
2152 facet_uninstall(ofproto, facet);
2153 facet_flush_stats(ofproto, facet);
2154 hmap_remove(&ofproto->facets, &facet->hmap_node);
2155 list_remove(&facet->list_node);
2159 /* Composes the ODP actions for 'facet' based on its rule's actions.
2160 * Returns true if the actions changed, false otherwise. */
2162 facet_make_actions(struct ofproto *p, struct facet *facet,
2163 const struct ofpbuf *packet)
2165 const struct rule *rule = facet->rule;
2166 struct odp_actions a;
2169 xlate_actions(rule->actions, rule->n_actions, &facet->flow, p,
2170 packet, &a, &facet->tags, &facet->may_install,
2171 &facet->nf_flow.output_iface);
2173 actions_len = a.n_actions * sizeof *a.actions;
2174 if (facet->n_actions == a.n_actions
2175 && !memcmp(facet->actions, a.actions, actions_len)) {
2179 free(facet->actions);
2180 facet->n_actions = a.n_actions;
2181 facet->actions = xmemdup(a.actions, actions_len);
2186 facet_put__(struct ofproto *ofproto, struct facet *facet, int flags,
2187 struct odp_flow_put *put)
2189 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2190 odp_flow_key_from_flow(&put->flow.key, &facet->flow);
2191 put->flow.actions = facet->actions;
2192 put->flow.n_actions = facet->n_actions;
2193 put->flow.flags = 0;
2195 return dpif_flow_put(ofproto->dpif, put);
2198 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2199 * 'zero_stats' is true, clears any existing statistics from the datapath for
2202 facet_install(struct ofproto *p, struct facet *facet, bool zero_stats)
2204 if (facet->may_install) {
2205 struct odp_flow_put put;
2208 flags = ODPPF_CREATE | ODPPF_MODIFY;
2210 flags |= ODPPF_ZERO_STATS;
2212 if (!facet_put__(p, facet, flags, &put)) {
2213 facet->installed = true;
2218 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2219 * to the accounting hook function in the ofhooks structure. */
2221 facet_account(struct ofproto *ofproto,
2222 struct facet *facet, uint64_t extra_bytes)
2224 uint64_t total_bytes = facet->byte_count + extra_bytes;
2226 if (ofproto->ofhooks->account_flow_cb
2227 && total_bytes > facet->accounted_bytes)
2229 ofproto->ofhooks->account_flow_cb(
2230 &facet->flow, facet->tags, facet->actions, facet->n_actions,
2231 total_bytes - facet->accounted_bytes, ofproto->aux);
2232 facet->accounted_bytes = total_bytes;
2236 /* If 'rule' is installed in the datapath, uninstalls it. */
2238 facet_uninstall(struct ofproto *p, struct facet *facet)
2240 if (facet->installed) {
2241 struct odp_flow odp_flow;
2243 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
2244 odp_flow.actions = NULL;
2245 odp_flow.n_actions = 0;
2247 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2248 facet_update_stats(p, facet, &odp_flow.stats);
2250 facet->installed = false;
2254 /* Returns true if the only action for 'facet' is to send to the controller.
2255 * (We don't report NetFlow expiration messages for such facets because they
2256 * are just part of the control logic for the network, not real traffic). */
2258 facet_is_controller_flow(struct facet *facet)
2261 && facet->rule->n_actions == 1
2262 && action_outputs_to_port(&facet->rule->actions[0],
2263 htons(OFPP_CONTROLLER)));
2266 /* Folds all of 'facet''s statistics into its rule. Also updates the
2267 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2269 facet_flush_stats(struct ofproto *ofproto, struct facet *facet)
2271 facet_account(ofproto, facet, 0);
2273 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
2274 struct ofexpired expired;
2275 expired.flow = facet->flow;
2276 expired.packet_count = facet->packet_count;
2277 expired.byte_count = facet->byte_count;
2278 expired.used = facet->used;
2279 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
2282 facet->rule->packet_count += facet->packet_count;
2283 facet->rule->byte_count += facet->byte_count;
2285 /* Reset counters to prevent double counting if 'facet' ever gets
2287 facet->packet_count = 0;
2288 facet->byte_count = 0;
2289 facet->accounted_bytes = 0;
2291 netflow_flow_clear(&facet->nf_flow);
2294 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2295 * Returns it if found, otherwise a null pointer.
2297 * The returned facet might need revalidation; use facet_lookup_valid()
2298 * instead if that is important. */
2299 static struct facet *
2300 facet_find(struct ofproto *ofproto, const struct flow *flow)
2302 struct facet *facet;
2304 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
2306 if (flow_equal(flow, &facet->flow)) {
2314 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2315 * Returns it if found, otherwise a null pointer.
2317 * The returned facet is guaranteed to be valid. */
2318 static struct facet *
2319 facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow)
2321 struct facet *facet = facet_find(ofproto, flow);
2323 /* The facet we found might not be valid, since we could be in need of
2324 * revalidation. If it is not valid, don't return it. */
2326 && ofproto->need_revalidate
2327 && !facet_revalidate(ofproto, facet)) {
2328 COVERAGE_INC(ofproto_invalidated);
2335 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2337 * - If the rule found is different from 'facet''s current rule, moves
2338 * 'facet' to the new rule and recompiles its actions.
2340 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2341 * where it is and recompiles its actions anyway.
2343 * - If there is none, destroys 'facet'.
2345 * Returns true if 'facet' still exists, false if it has been destroyed. */
2347 facet_revalidate(struct ofproto *ofproto, struct facet *facet)
2349 struct rule *new_rule;
2350 struct odp_actions a;
2352 uint16_t new_nf_output_iface;
2353 bool actions_changed;
2355 COVERAGE_INC(facet_revalidate);
2357 /* Determine the new rule. */
2358 new_rule = rule_lookup(ofproto, &facet->flow);
2360 /* No new rule, so delete the facet. */
2361 facet_remove(ofproto, facet);
2365 /* Calculate new ODP actions.
2367 * We are very cautious about actually modifying 'facet' state at this
2368 * point, because we might need to, e.g., emit a NetFlow expiration and, if
2369 * so, we need to have the old state around to properly compose it. */
2370 xlate_actions(new_rule->actions, new_rule->n_actions, &facet->flow,
2371 ofproto, NULL, &a, &facet->tags, &facet->may_install,
2372 &new_nf_output_iface);
2373 actions_len = a.n_actions * sizeof *a.actions;
2374 actions_changed = (facet->n_actions != a.n_actions
2375 || memcmp(facet->actions, a.actions, actions_len));
2377 /* If the ODP actions changed or the installability changed, then we need
2378 * to talk to the datapath. */
2379 if (actions_changed || facet->may_install != facet->installed) {
2380 if (facet->may_install) {
2381 struct odp_flow_put put;
2383 memset(&put.flow.stats, 0, sizeof put.flow.stats);
2384 odp_flow_key_from_flow(&put.flow.key, &facet->flow);
2385 put.flow.actions = a.actions;
2386 put.flow.n_actions = a.n_actions;
2388 put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS;
2389 dpif_flow_put(ofproto->dpif, &put);
2391 facet_update_stats(ofproto, facet, &put.flow.stats);
2393 facet_uninstall(ofproto, facet);
2396 /* The datapath flow is gone or has zeroed stats, so push stats out of
2397 * 'facet' into 'rule'. */
2398 facet_flush_stats(ofproto, facet);
2401 /* Update 'facet' now that we've taken care of all the old state. */
2402 facet->nf_flow.output_iface = new_nf_output_iface;
2403 if (actions_changed) {
2404 free(facet->actions);
2405 facet->n_actions = a.n_actions;
2406 facet->actions = xmemdup(a.actions, actions_len);
2408 if (facet->rule != new_rule) {
2409 COVERAGE_INC(facet_changed_rule);
2410 list_remove(&facet->list_node);
2411 list_push_back(&new_rule->facets, &facet->list_node);
2412 facet->rule = new_rule;
2413 facet->used = new_rule->created;
2420 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2421 struct rconn_packet_counter *counter)
2423 update_openflow_length(msg);
2424 if (rconn_send(ofconn->rconn, msg, counter)) {
2430 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2433 struct ofpbuf *buf = make_ofp_error_msg(error, oh);
2435 COVERAGE_INC(ofproto_error);
2436 queue_tx(buf, ofconn, ofconn->reply_counter);
2441 hton_ofp_phy_port(struct ofp_phy_port *opp)
2443 opp->port_no = htons(opp->port_no);
2444 opp->config = htonl(opp->config);
2445 opp->state = htonl(opp->state);
2446 opp->curr = htonl(opp->curr);
2447 opp->advertised = htonl(opp->advertised);
2448 opp->supported = htonl(opp->supported);
2449 opp->peer = htonl(opp->peer);
2453 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2455 struct ofp_header *rq = oh;
2456 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2461 handle_features_request(struct ofconn *ofconn, struct ofp_header *oh)
2463 struct ofp_switch_features *osf;
2465 struct ofport *port;
2467 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2468 osf->datapath_id = htonll(ofconn->ofproto->datapath_id);
2469 osf->n_buffers = htonl(pktbuf_capacity());
2471 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2472 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2473 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2474 (1u << OFPAT_SET_VLAN_VID) |
2475 (1u << OFPAT_SET_VLAN_PCP) |
2476 (1u << OFPAT_STRIP_VLAN) |
2477 (1u << OFPAT_SET_DL_SRC) |
2478 (1u << OFPAT_SET_DL_DST) |
2479 (1u << OFPAT_SET_NW_SRC) |
2480 (1u << OFPAT_SET_NW_DST) |
2481 (1u << OFPAT_SET_NW_TOS) |
2482 (1u << OFPAT_SET_TP_SRC) |
2483 (1u << OFPAT_SET_TP_DST) |
2484 (1u << OFPAT_ENQUEUE));
2486 HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) {
2487 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2490 queue_tx(buf, ofconn, ofconn->reply_counter);
2495 handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh)
2498 struct ofp_switch_config *osc;
2502 /* Figure out flags. */
2503 dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags);
2504 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2507 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2508 osc->flags = htons(flags);
2509 osc->miss_send_len = htons(ofconn->miss_send_len);
2510 queue_tx(buf, ofconn, ofconn->reply_counter);
2516 handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc)
2521 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2525 flags = ntohs(osc->flags);
2527 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2528 switch (flags & OFPC_FRAG_MASK) {
2529 case OFPC_FRAG_NORMAL:
2530 dpif_set_drop_frags(ofconn->ofproto->dpif, false);
2532 case OFPC_FRAG_DROP:
2533 dpif_set_drop_frags(ofconn->ofproto->dpif, true);
2536 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2542 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2548 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2550 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2551 a->controller.arg = max_len;
2554 struct action_xlate_ctx {
2556 struct flow flow; /* Flow to which these actions correspond. */
2557 int recurse; /* Recursion level, via xlate_table_action. */
2558 struct ofproto *ofproto;
2559 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2560 * null pointer if we are revalidating
2561 * without a packet to refer to. */
2564 struct odp_actions *out; /* Datapath actions. */
2565 tag_type tags; /* Tags associated with OFPP_NORMAL actions. */
2566 bool may_set_up_flow; /* True ordinarily; false if the actions must
2567 * be reassessed for every packet. */
2568 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2571 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2572 * flow translation. */
2573 #define MAX_RESUBMIT_RECURSION 8
2575 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2576 struct action_xlate_ctx *ctx);
2579 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2581 const struct ofport *ofport = get_port(ctx->ofproto, port);
2584 if (ofport->opp.config & OFPPC_NO_FWD) {
2585 /* Forwarding disabled on port. */
2590 * We don't have an ofport record for this port, but it doesn't hurt to
2591 * allow forwarding to it anyhow. Maybe such a port will appear later
2592 * and we're pre-populating the flow table.
2596 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2597 ctx->nf_output_iface = port;
2600 static struct rule *
2601 rule_lookup(struct ofproto *ofproto, const struct flow *flow)
2603 return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow,
2608 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2610 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2611 uint16_t old_in_port;
2614 /* Look up a flow with 'in_port' as the input port. Then restore the
2615 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2616 * have surprising behavior). */
2617 old_in_port = ctx->flow.in_port;
2618 ctx->flow.in_port = in_port;
2619 rule = rule_lookup(ctx->ofproto, &ctx->flow);
2620 ctx->flow.in_port = old_in_port;
2624 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2628 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2630 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2631 MAX_RESUBMIT_RECURSION);
2636 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2637 uint16_t *nf_output_iface, struct odp_actions *actions)
2639 struct ofport *ofport;
2641 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2642 uint16_t odp_port = ofport->odp_port;
2643 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2644 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2647 *nf_output_iface = NF_OUT_FLOOD;
2651 xlate_output_action__(struct action_xlate_ctx *ctx,
2652 uint16_t port, uint16_t max_len)
2655 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2657 ctx->nf_output_iface = NF_OUT_DROP;
2661 add_output_action(ctx, ctx->flow.in_port);
2664 xlate_table_action(ctx, ctx->flow.in_port);
2667 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2668 ctx->out, &ctx->tags,
2669 &ctx->nf_output_iface,
2670 ctx->ofproto->aux)) {
2671 COVERAGE_INC(ofproto_uninstallable);
2672 ctx->may_set_up_flow = false;
2676 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2677 &ctx->nf_output_iface, ctx->out);
2680 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2681 &ctx->nf_output_iface, ctx->out);
2683 case OFPP_CONTROLLER:
2684 add_controller_action(ctx->out, max_len);
2687 add_output_action(ctx, ODPP_LOCAL);
2690 odp_port = ofp_port_to_odp_port(port);
2691 if (odp_port != ctx->flow.in_port) {
2692 add_output_action(ctx, odp_port);
2697 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2698 ctx->nf_output_iface = NF_OUT_FLOOD;
2699 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2700 ctx->nf_output_iface = prev_nf_output_iface;
2701 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2702 ctx->nf_output_iface != NF_OUT_FLOOD) {
2703 ctx->nf_output_iface = NF_OUT_MULTI;
2708 xlate_output_action(struct action_xlate_ctx *ctx,
2709 const struct ofp_action_output *oao)
2711 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2714 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2715 * optimization, because we're going to add another action that sets the
2716 * priority immediately after, or because there are no actions following the
2719 remove_pop_action(struct action_xlate_ctx *ctx)
2721 size_t n = ctx->out->n_actions;
2722 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2723 ctx->out->n_actions--;
2728 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2729 const struct ofp_action_enqueue *oae)
2731 uint16_t ofp_port, odp_port;
2735 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2738 /* Fall back to ordinary output action. */
2739 xlate_output_action__(ctx, ntohs(oae->port), 0);
2743 /* Figure out ODP output port. */
2744 ofp_port = ntohs(oae->port);
2745 if (ofp_port != OFPP_IN_PORT) {
2746 odp_port = ofp_port_to_odp_port(ofp_port);
2748 odp_port = ctx->flow.in_port;
2751 /* Add ODP actions. */
2752 remove_pop_action(ctx);
2753 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2755 add_output_action(ctx, odp_port);
2756 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2758 /* Update NetFlow output port. */
2759 if (ctx->nf_output_iface == NF_OUT_DROP) {
2760 ctx->nf_output_iface = odp_port;
2761 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2762 ctx->nf_output_iface = NF_OUT_MULTI;
2767 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2768 const struct nx_action_set_queue *nasq)
2773 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2776 /* Couldn't translate queue to a priority, so ignore. A warning
2777 * has already been logged. */
2781 remove_pop_action(ctx);
2782 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2787 xlate_set_dl_tci(struct action_xlate_ctx *ctx)
2789 ovs_be16 dl_vlan = ctx->flow.dl_vlan;
2790 uint8_t dl_vlan_pcp = ctx->flow.dl_vlan_pcp;
2792 if (dl_vlan == htons(OFP_VLAN_NONE)) {
2793 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2795 union odp_action *oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2796 oa->dl_tci.tci = htons(ntohs(dl_vlan & htons(VLAN_VID_MASK))
2797 | (dl_vlan_pcp << VLAN_PCP_SHIFT)
2803 xlate_reg_move_action(struct action_xlate_ctx *ctx,
2804 const struct nx_action_reg_move *narm)
2806 ovs_be16 old_vlan = ctx->flow.dl_vlan;
2807 uint8_t old_pcp = ctx->flow.dl_vlan_pcp;
2809 nxm_execute_reg_move(narm, &ctx->flow);
2811 if (ctx->flow.dl_vlan != old_vlan || ctx->flow.dl_vlan_pcp != old_pcp) {
2812 xlate_set_dl_tci(ctx);
2817 xlate_nicira_action(struct action_xlate_ctx *ctx,
2818 const struct nx_action_header *nah)
2820 const struct nx_action_resubmit *nar;
2821 const struct nx_action_set_tunnel *nast;
2822 const struct nx_action_set_queue *nasq;
2823 union odp_action *oa;
2824 int subtype = ntohs(nah->subtype);
2826 assert(nah->vendor == htonl(NX_VENDOR_ID));
2828 case NXAST_RESUBMIT:
2829 nar = (const struct nx_action_resubmit *) nah;
2830 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2833 case NXAST_SET_TUNNEL:
2834 nast = (const struct nx_action_set_tunnel *) nah;
2835 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2836 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2839 case NXAST_DROP_SPOOFED_ARP:
2840 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2841 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2845 case NXAST_SET_QUEUE:
2846 nasq = (const struct nx_action_set_queue *) nah;
2847 xlate_set_queue_action(ctx, nasq);
2850 case NXAST_POP_QUEUE:
2851 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2854 case NXAST_REG_MOVE:
2855 xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah);
2858 case NXAST_REG_LOAD:
2859 nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
2863 /* If you add a new action here that modifies flow data, don't forget to
2864 * update the flow key in ctx->flow at the same time. */
2867 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2873 do_xlate_actions(const union ofp_action *in, size_t n_in,
2874 struct action_xlate_ctx *ctx)
2876 struct actions_iterator iter;
2877 const union ofp_action *ia;
2878 const struct ofport *port;
2880 port = get_port(ctx->ofproto, ctx->flow.in_port);
2881 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2882 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2883 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2884 /* Drop this flow. */
2888 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2889 uint16_t type = ntohs(ia->type);
2890 union odp_action *oa;
2894 xlate_output_action(ctx, &ia->output);
2897 case OFPAT_SET_VLAN_VID:
2898 ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid;
2899 xlate_set_dl_tci(ctx);
2902 case OFPAT_SET_VLAN_PCP:
2903 ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp;
2904 xlate_set_dl_tci(ctx);
2907 case OFPAT_STRIP_VLAN:
2908 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2909 ctx->flow.dl_vlan_pcp = 0;
2910 xlate_set_dl_tci(ctx);
2913 case OFPAT_SET_DL_SRC:
2914 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2915 memcpy(oa->dl_addr.dl_addr,
2916 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2917 memcpy(ctx->flow.dl_src,
2918 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2921 case OFPAT_SET_DL_DST:
2922 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2923 memcpy(oa->dl_addr.dl_addr,
2924 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2925 memcpy(ctx->flow.dl_dst,
2926 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2929 case OFPAT_SET_NW_SRC:
2930 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2931 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2934 case OFPAT_SET_NW_DST:
2935 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2936 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2939 case OFPAT_SET_NW_TOS:
2940 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2941 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2944 case OFPAT_SET_TP_SRC:
2945 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2946 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2949 case OFPAT_SET_TP_DST:
2950 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2951 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2955 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2959 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2963 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2970 xlate_actions(const union ofp_action *in, size_t n_in,
2971 const struct flow *flow, struct ofproto *ofproto,
2972 const struct ofpbuf *packet,
2973 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2974 uint16_t *nf_output_iface)
2976 struct action_xlate_ctx ctx;
2978 COVERAGE_INC(ofproto_ofp2odp);
2979 odp_actions_init(out);
2982 ctx.ofproto = ofproto;
2983 ctx.packet = packet;
2986 ctx.may_set_up_flow = true;
2987 ctx.nf_output_iface = NF_OUT_DROP;
2988 do_xlate_actions(in, n_in, &ctx);
2989 remove_pop_action(&ctx);
2991 /* Check with in-band control to see if we're allowed to set up this
2993 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2994 ctx.may_set_up_flow = false;
3000 if (may_set_up_flow) {
3001 *may_set_up_flow = ctx.may_set_up_flow;
3003 if (nf_output_iface) {
3004 *nf_output_iface = ctx.nf_output_iface;
3006 if (odp_actions_overflow(out)) {
3007 COVERAGE_INC(odp_overflow);
3008 odp_actions_init(out);
3009 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
3014 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3015 * error message code (composed with ofp_mkerr()) for the caller to propagate
3016 * upward. Otherwise, returns 0.
3018 * The log message mentions 'msg_type'. */
3020 reject_slave_controller(struct ofconn *ofconn, const const char *msg_type)
3022 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
3023 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
3024 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
3027 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3034 handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh)
3036 struct ofproto *p = ofconn->ofproto;
3037 struct ofp_packet_out *opo;
3038 struct ofpbuf payload, *buffer;
3039 union ofp_action *ofp_actions;
3040 struct odp_actions odp_actions;
3041 struct ofpbuf request;
3043 size_t n_ofp_actions;
3047 COVERAGE_INC(ofproto_packet_out);
3049 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
3054 /* Get ofp_packet_out. */
3056 request.size = ntohs(oh->length);
3057 opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions));
3059 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3063 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
3064 &ofp_actions, &n_ofp_actions);
3070 if (opo->buffer_id != htonl(UINT32_MAX)) {
3071 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
3073 if (error || !buffer) {
3082 /* Extract flow, check actions. */
3083 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)),
3085 error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports);
3091 error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload,
3092 &odp_actions, NULL, NULL, NULL);
3094 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions,
3099 ofpbuf_delete(buffer);
3104 update_port_config(struct ofproto *p, struct ofport *port,
3105 uint32_t config, uint32_t mask)
3107 mask &= config ^ port->opp.config;
3108 if (mask & OFPPC_PORT_DOWN) {
3109 if (config & OFPPC_PORT_DOWN) {
3110 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
3112 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3115 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3116 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3117 if (mask & REVALIDATE_BITS) {
3118 COVERAGE_INC(ofproto_costly_flags);
3119 port->opp.config ^= mask & REVALIDATE_BITS;
3120 p->need_revalidate = true;
3122 #undef REVALIDATE_BITS
3123 if (mask & OFPPC_NO_PACKET_IN) {
3124 port->opp.config ^= OFPPC_NO_PACKET_IN;
3129 handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh)
3131 struct ofproto *p = ofconn->ofproto;
3132 const struct ofp_port_mod *opm;
3133 struct ofport *port;
3136 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
3140 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
3144 opm = (struct ofp_port_mod *) oh;
3146 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3148 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3149 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3150 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3152 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3153 if (opm->advertise) {
3154 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3160 static struct ofpbuf *
3161 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
3163 struct ofp_stats_reply *osr;
3166 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3167 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3169 osr->flags = htons(0);
3173 static struct ofpbuf *
3174 start_ofp_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3176 return make_ofp_stats_reply(request->header.xid, request->type, body_len);
3180 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
3181 struct ofpbuf **msgp)
3183 struct ofpbuf *msg = *msgp;
3184 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3185 if (nbytes + msg->size > UINT16_MAX) {
3186 struct ofp_stats_reply *reply = msg->data;
3187 reply->flags = htons(OFPSF_REPLY_MORE);
3188 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
3189 queue_tx(msg, ofconn, ofconn->reply_counter);
3191 return ofpbuf_put_uninit(*msgp, nbytes);
3194 static struct ofpbuf *
3195 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
3197 struct nicira_stats_msg *nsm;
3200 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
3201 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
3202 nsm->type = htons(OFPST_VENDOR);
3203 nsm->flags = htons(0);
3204 nsm->vendor = htonl(NX_VENDOR_ID);
3205 nsm->subtype = htonl(subtype);
3209 static struct ofpbuf *
3210 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
3212 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
3216 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
3217 struct ofpbuf **msgp)
3219 struct ofpbuf *msg = *msgp;
3220 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
3221 if (nbytes + msg->size > UINT16_MAX) {
3222 struct nicira_stats_msg *reply = msg->data;
3223 reply->flags = htons(OFPSF_REPLY_MORE);
3224 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
3225 queue_tx(msg, ofconn, ofconn->reply_counter);
3227 ofpbuf_prealloc_tailroom(*msgp, nbytes);
3231 handle_desc_stats_request(struct ofconn *ofconn,
3232 struct ofp_stats_request *request)
3234 struct ofproto *p = ofconn->ofproto;
3235 struct ofp_desc_stats *ods;
3238 msg = start_ofp_stats_reply(request, sizeof *ods);
3239 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
3240 memset(ods, 0, sizeof *ods);
3241 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3242 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3243 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3244 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3245 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3246 queue_tx(msg, ofconn, ofconn->reply_counter);
3252 handle_table_stats_request(struct ofconn *ofconn,
3253 struct ofp_stats_request *request)
3255 struct ofproto *p = ofconn->ofproto;
3256 struct ofp_table_stats *ots;
3259 msg = start_ofp_stats_reply(request, sizeof *ots * 2);
3261 /* Classifier table. */
3262 ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg);
3263 memset(ots, 0, sizeof *ots);
3264 strcpy(ots->name, "classifier");
3265 ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10
3266 ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL));
3267 ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */
3268 ots->active_count = htonl(classifier_count(&p->cls));
3269 ots->lookup_count = htonll(0); /* XXX */
3270 ots->matched_count = htonll(0); /* XXX */
3272 queue_tx(msg, ofconn, ofconn->reply_counter);
3277 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3278 struct ofpbuf **msgp)
3280 struct netdev_stats stats;
3281 struct ofp_port_stats *ops;
3283 /* Intentionally ignore return value, since errors will set
3284 * 'stats' to all-1s, which is correct for OpenFlow, and
3285 * netdev_get_stats() will log errors. */
3286 netdev_get_stats(port->netdev, &stats);
3288 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
3289 ops->port_no = htons(port->opp.port_no);
3290 memset(ops->pad, 0, sizeof ops->pad);
3291 ops->rx_packets = htonll(stats.rx_packets);
3292 ops->tx_packets = htonll(stats.tx_packets);
3293 ops->rx_bytes = htonll(stats.rx_bytes);
3294 ops->tx_bytes = htonll(stats.tx_bytes);
3295 ops->rx_dropped = htonll(stats.rx_dropped);
3296 ops->tx_dropped = htonll(stats.tx_dropped);
3297 ops->rx_errors = htonll(stats.rx_errors);
3298 ops->tx_errors = htonll(stats.tx_errors);
3299 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3300 ops->rx_over_err = htonll(stats.rx_over_errors);
3301 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3302 ops->collisions = htonll(stats.collisions);
3306 handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr,
3309 struct ofproto *p = ofconn->ofproto;
3310 struct ofp_port_stats_request *psr;
3311 struct ofp_port_stats *ops;
3313 struct ofport *port;
3315 if (arg_size != sizeof *psr) {
3316 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3318 psr = (struct ofp_port_stats_request *) osr->body;
3320 msg = start_ofp_stats_reply(osr, sizeof *ops * 16);
3321 if (psr->port_no != htons(OFPP_NONE)) {
3322 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3324 append_port_stat(port, ofconn, &msg);
3327 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3328 append_port_stat(port, ofconn, &msg);
3332 queue_tx(msg, ofconn, ofconn->reply_counter);
3336 struct flow_stats_cbdata {
3337 struct ofconn *ofconn;
3342 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3343 * '*packet_countp' and '*byte_countp'. The returned statistics include
3344 * statistics for all of 'rule''s facets. */
3346 query_stats(struct ofproto *p, struct rule *rule,
3347 uint64_t *packet_countp, uint64_t *byte_countp)
3349 uint64_t packet_count, byte_count;
3350 struct facet *facet;
3351 struct odp_flow *odp_flows;
3354 /* Start from historical data for 'rule' itself that are no longer tracked
3355 * by the datapath. This counts, for example, facets that have expired. */
3356 packet_count = rule->packet_count;
3357 byte_count = rule->byte_count;
3359 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3361 * Also, add any statistics that are not tracked by the datapath for each
3362 * facet. This includes, for example, statistics for packets that were
3363 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3365 odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows);
3367 LIST_FOR_EACH (facet, list_node, &rule->facets) {
3368 struct odp_flow *odp_flow = &odp_flows[n_odp_flows++];
3369 odp_flow_key_from_flow(&odp_flow->key, &facet->flow);
3370 packet_count += facet->packet_count;
3371 byte_count += facet->byte_count;
3374 /* Fetch up-to-date statistics from the datapath and add them in. */
3375 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3378 for (i = 0; i < n_odp_flows; i++) {
3379 struct odp_flow *odp_flow = &odp_flows[i];
3380 packet_count += odp_flow->stats.n_packets;
3381 byte_count += odp_flow->stats.n_bytes;
3386 /* Return the stats to the caller. */
3387 *packet_countp = packet_count;
3388 *byte_countp = byte_count;
3392 calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec)
3394 long long int msecs = time_msec() - start;
3395 *sec = htonl(msecs / 1000);
3396 *nsec = htonl((msecs % 1000) * (1000 * 1000));
3400 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3402 struct rule *rule = rule_from_cls_rule(rule_);
3403 struct flow_stats_cbdata *cbdata = cbdata_;
3404 struct ofp_flow_stats *ofs;
3405 uint64_t packet_count, byte_count;
3406 size_t act_len, len;
3408 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3412 act_len = sizeof *rule->actions * rule->n_actions;
3413 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3415 query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count);
3417 ofs = append_ofp_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3418 ofs->length = htons(len);
3421 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3422 cbdata->ofconn->flow_format, &ofs->match);
3423 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
3424 ofs->cookie = rule->flow_cookie;
3425 ofs->priority = htons(rule->cr.priority);
3426 ofs->idle_timeout = htons(rule->idle_timeout);
3427 ofs->hard_timeout = htons(rule->hard_timeout);
3428 memset(ofs->pad2, 0, sizeof ofs->pad2);
3429 ofs->packet_count = htonll(packet_count);
3430 ofs->byte_count = htonll(byte_count);
3431 if (rule->n_actions > 0) {
3432 memcpy(ofs->actions, rule->actions, act_len);
3437 table_id_to_include(uint8_t table_id)
3439 return table_id == 0 || table_id == 0xff ? CLS_INC_ALL : 0;
3443 handle_flow_stats_request(struct ofconn *ofconn,
3444 const struct ofp_stats_request *osr, size_t arg_size)
3446 struct ofp_flow_stats_request *fsr;
3447 struct flow_stats_cbdata cbdata;
3448 struct cls_rule target;
3450 if (arg_size != sizeof *fsr) {
3451 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3453 fsr = (struct ofp_flow_stats_request *) osr->body;
3455 COVERAGE_INC(ofproto_flows_req);
3456 cbdata.ofconn = ofconn;
3457 cbdata.out_port = fsr->out_port;
3458 cbdata.msg = start_ofp_stats_reply(osr, 1024);
3459 cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, &target);
3460 classifier_for_each_match(&ofconn->ofproto->cls, &target,
3461 table_id_to_include(fsr->table_id),
3462 flow_stats_cb, &cbdata);
3463 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3468 nx_flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3470 struct rule *rule = rule_from_cls_rule(rule_);
3471 struct flow_stats_cbdata *cbdata = cbdata_;
3472 struct nx_flow_stats *nfs;
3473 uint64_t packet_count, byte_count;
3474 size_t act_len, start_len;
3476 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3480 query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count);
3482 act_len = sizeof *rule->actions * rule->n_actions;
3484 start_len = cbdata->msg->size;
3485 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len,
3486 cbdata->ofconn, &cbdata->msg);
3487 nfs = ofpbuf_put_uninit(cbdata->msg, sizeof *nfs);
3490 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
3491 nfs->cookie = rule->flow_cookie;
3492 nfs->priority = htons(rule->cr.priority);
3493 nfs->idle_timeout = htons(rule->idle_timeout);
3494 nfs->hard_timeout = htons(rule->hard_timeout);
3495 nfs->match_len = htons(nx_put_match(cbdata->msg, &rule->cr));
3496 memset(nfs->pad2, 0, sizeof nfs->pad2);
3497 nfs->packet_count = htonll(packet_count);
3498 nfs->byte_count = htonll(byte_count);
3499 if (rule->n_actions > 0) {
3500 ofpbuf_put(cbdata->msg, rule->actions, act_len);
3502 nfs->length = htons(cbdata->msg->size - start_len);
3506 handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b)
3508 struct nx_flow_stats_request *nfsr;
3509 struct flow_stats_cbdata cbdata;
3510 struct cls_rule target;
3513 /* Dissect the message. */
3514 nfsr = ofpbuf_try_pull(b, sizeof *nfsr);
3516 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3518 error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target);
3523 COVERAGE_INC(ofproto_flows_req);
3524 cbdata.ofconn = ofconn;
3525 cbdata.out_port = nfsr->out_port;
3526 cbdata.msg = start_nxstats_reply(&nfsr->nsm, 1024);
3527 classifier_for_each_match(&ofconn->ofproto->cls, &target,
3528 table_id_to_include(nfsr->table_id),
3529 nx_flow_stats_cb, &cbdata);
3530 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3534 struct flow_stats_ds_cbdata {
3535 struct ofproto *ofproto;
3540 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3542 struct rule *rule = rule_from_cls_rule(rule_);
3543 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3544 struct ds *results = cbdata->results;
3545 struct ofp_match match;
3546 uint64_t packet_count, byte_count;
3547 size_t act_len = sizeof *rule->actions * rule->n_actions;
3549 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3550 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3551 NXFF_OPENFLOW10, &match);
3553 ds_put_format(results, "duration=%llds, ",
3554 (time_msec() - rule->created) / 1000);
3555 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3556 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3557 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3558 ofp_print_match(results, &match, true);
3560 ofp_print_actions(results, &rule->actions->header, act_len);
3562 ds_put_cstr(results, "drop");
3564 ds_put_cstr(results, "\n");
3567 /* Adds a pretty-printed description of all flows to 'results', including
3568 * those marked hidden by secchan (e.g., by in-band control). */
3570 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3572 struct ofp_match match;
3573 struct cls_rule target;
3574 struct flow_stats_ds_cbdata cbdata;
3576 memset(&match, 0, sizeof match);
3577 match.wildcards = htonl(OVSFW_ALL);
3580 cbdata.results = results;
3582 cls_rule_from_match(&match, 0, NXFF_OPENFLOW10, 0, &target);
3583 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3584 flow_stats_ds_cb, &cbdata);
3587 struct aggregate_stats_cbdata {
3588 struct ofproto *ofproto;
3590 uint64_t packet_count;
3591 uint64_t byte_count;
3596 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3598 struct rule *rule = rule_from_cls_rule(rule_);
3599 struct aggregate_stats_cbdata *cbdata = cbdata_;
3600 uint64_t packet_count, byte_count;
3602 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3606 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3608 cbdata->packet_count += packet_count;
3609 cbdata->byte_count += byte_count;
3614 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
3615 ovs_be16 out_port, uint8_t table_id,
3616 struct ofp_aggregate_stats_reply *oasr)
3618 struct aggregate_stats_cbdata cbdata;
3620 COVERAGE_INC(ofproto_agg_request);
3621 cbdata.ofproto = ofproto;
3622 cbdata.out_port = out_port;
3623 cbdata.packet_count = 0;
3624 cbdata.byte_count = 0;
3626 classifier_for_each_match(&ofproto->cls, target,
3627 table_id_to_include(table_id),
3628 aggregate_stats_cb, &cbdata);
3630 oasr->flow_count = htonl(cbdata.n_flows);
3631 oasr->packet_count = htonll(cbdata.packet_count);
3632 oasr->byte_count = htonll(cbdata.byte_count);
3633 memset(oasr->pad, 0, sizeof oasr->pad);
3637 handle_aggregate_stats_request(struct ofconn *ofconn,
3638 const struct ofp_stats_request *osr,
3641 struct ofp_aggregate_stats_request *request;
3642 struct ofp_aggregate_stats_reply *reply;
3643 struct cls_rule target;
3646 if (arg_size != sizeof *request) {
3647 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3649 request = (struct ofp_aggregate_stats_request *) osr->body;
3651 cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, &target);
3653 msg = start_ofp_stats_reply(osr, sizeof *reply);
3654 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
3655 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3656 request->table_id, reply);
3657 queue_tx(msg, ofconn, ofconn->reply_counter);
3662 handle_nxst_aggregate(struct ofconn *ofconn, struct ofpbuf *b)
3664 struct nx_aggregate_stats_request *request;
3665 struct ofp_aggregate_stats_reply *reply;
3666 struct cls_rule target;
3670 /* Dissect the message. */
3671 request = ofpbuf_try_pull(b, sizeof *request);
3673 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3675 error = nx_pull_match(b, ntohs(request->match_len), 0, &target);
3681 COVERAGE_INC(ofproto_flows_req);
3682 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
3683 reply = ofpbuf_put_uninit(buf, sizeof *reply);
3684 query_aggregate_stats(ofconn->ofproto, &target, request->out_port,
3685 request->table_id, reply);
3686 queue_tx(buf, ofconn, ofconn->reply_counter);
3691 struct queue_stats_cbdata {
3692 struct ofconn *ofconn;
3693 struct ofport *ofport;
3698 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3699 const struct netdev_queue_stats *stats)
3701 struct ofp_queue_stats *reply;
3703 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3704 reply->port_no = htons(cbdata->ofport->opp.port_no);
3705 memset(reply->pad, 0, sizeof reply->pad);
3706 reply->queue_id = htonl(queue_id);
3707 reply->tx_bytes = htonll(stats->tx_bytes);
3708 reply->tx_packets = htonll(stats->tx_packets);
3709 reply->tx_errors = htonll(stats->tx_errors);
3713 handle_queue_stats_dump_cb(uint32_t queue_id,
3714 struct netdev_queue_stats *stats,
3717 struct queue_stats_cbdata *cbdata = cbdata_;
3719 put_queue_stats(cbdata, queue_id, stats);
3723 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3724 struct queue_stats_cbdata *cbdata)
3726 cbdata->ofport = port;
3727 if (queue_id == OFPQ_ALL) {
3728 netdev_dump_queue_stats(port->netdev,
3729 handle_queue_stats_dump_cb, cbdata);
3731 struct netdev_queue_stats stats;
3733 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3734 put_queue_stats(cbdata, queue_id, &stats);
3740 handle_queue_stats_request(struct ofconn *ofconn,
3741 const struct ofp_stats_request *osr,
3744 struct ofproto *ofproto = ofconn->ofproto;
3745 struct ofp_queue_stats_request *qsr;
3746 struct queue_stats_cbdata cbdata;
3747 struct ofport *port;
3748 unsigned int port_no;
3751 if (arg_size != sizeof *qsr) {
3752 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3754 qsr = (struct ofp_queue_stats_request *) osr->body;
3756 COVERAGE_INC(ofproto_queue_req);
3758 cbdata.ofconn = ofconn;
3759 cbdata.msg = start_ofp_stats_reply(osr, 128);
3761 port_no = ntohs(qsr->port_no);
3762 queue_id = ntohl(qsr->queue_id);
3763 if (port_no == OFPP_ALL) {
3764 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3765 handle_queue_stats_for_port(port, queue_id, &cbdata);
3767 } else if (port_no < ofproto->max_ports) {
3768 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3770 handle_queue_stats_for_port(port, queue_id, &cbdata);
3773 ofpbuf_delete(cbdata.msg);
3774 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3776 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3782 handle_vendor_stats_request(struct ofconn *ofconn,
3783 struct ofp_stats_request *osr, size_t arg_size)
3785 struct nicira_stats_msg *nsm;
3790 VLOG_WARN_RL(&rl, "truncated vendor stats request body");
3791 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3794 memcpy(&vendor, osr->body, sizeof vendor);
3795 if (vendor != htonl(NX_VENDOR_ID)) {
3796 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3799 if (ntohs(osr->header.length) < sizeof(struct nicira_stats_msg)) {
3800 VLOG_WARN_RL(&rl, "truncated Nicira stats request");
3801 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3804 nsm = (struct nicira_stats_msg *) osr;
3806 b.size = ntohs(nsm->header.length);
3807 switch (ntohl(nsm->subtype)) {
3809 return handle_nxst_flow(ofconn, &b);
3811 case NXST_AGGREGATE:
3812 return handle_nxst_aggregate(ofconn, &b);
3815 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3820 handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh)
3822 struct ofp_stats_request *osr;
3826 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3831 osr = (struct ofp_stats_request *) oh;
3833 switch (ntohs(osr->type)) {
3835 return handle_desc_stats_request(ofconn, osr);
3838 return handle_flow_stats_request(ofconn, osr, arg_size);
3840 case OFPST_AGGREGATE:
3841 return handle_aggregate_stats_request(ofconn, osr, arg_size);
3844 return handle_table_stats_request(ofconn, osr);
3847 return handle_port_stats_request(ofconn, osr, arg_size);
3850 return handle_queue_stats_request(ofconn, osr, arg_size);
3853 return handle_vendor_stats_request(ofconn, osr, arg_size);
3856 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3860 static long long int
3861 msec_from_nsec(uint64_t sec, uint32_t nsec)
3863 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3867 facet_update_time(struct ofproto *ofproto, struct facet *facet,
3868 const struct odp_flow_stats *stats)
3870 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3871 if (used > facet->used) {
3873 if (used > facet->rule->used) {
3874 facet->rule->used = used;
3876 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
3880 /* Folds the statistics from 'stats' into the counters in 'facet'.
3882 * Because of the meaning of a facet's counters, it only makes sense to do this
3883 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3884 * packet that was sent by hand or if it represents statistics that have been
3885 * cleared out of the datapath. */
3887 facet_update_stats(struct ofproto *ofproto, struct facet *facet,
3888 const struct odp_flow_stats *stats)
3890 if (stats->n_packets) {
3891 facet_update_time(ofproto, facet, stats);
3892 facet->packet_count += stats->n_packets;
3893 facet->byte_count += stats->n_bytes;
3894 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
3902 uint16_t idle_timeout;
3903 uint16_t hard_timeout;
3907 union ofp_action *actions;
3911 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3912 * in which no matching flow already exists in the flow table.
3914 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3915 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3916 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3918 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3921 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
3923 struct ofproto *p = ofconn->ofproto;
3924 struct ofpbuf *packet;
3929 if (fm->flags & OFPFF_CHECK_OVERLAP
3930 && classifier_rule_overlaps(&p->cls, &fm->cr)) {
3931 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3935 if (fm->buffer_id != UINT32_MAX) {
3936 error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id,
3940 in_port = UINT16_MAX;
3943 rule = rule_create(&fm->cr, fm->actions, fm->n_actions,
3944 fm->idle_timeout, fm->hard_timeout, fm->cookie,
3945 fm->flags & OFPFF_SEND_FLOW_REM);
3946 rule_insert(p, rule, packet, in_port);
3950 static struct rule *
3951 find_flow_strict(struct ofproto *p, const struct flow_mod *fm)
3953 return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr));
3957 send_buffered_packet(struct ofconn *ofconn,
3958 struct rule *rule, uint32_t buffer_id)
3960 struct ofpbuf *packet;
3964 if (buffer_id == UINT32_MAX) {
3968 error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port);
3973 rule_execute(ofconn->ofproto, rule, in_port, packet);
3978 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3980 struct modify_flows_cbdata {
3981 struct ofproto *ofproto;
3982 const struct flow_mod *fm;
3986 static int modify_flow(struct ofproto *, const struct flow_mod *,
3988 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3990 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3991 * encoded by ofp_mkerr() on failure.
3993 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3996 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
3998 struct modify_flows_cbdata cbdata;
4000 cbdata.ofproto = ofconn->ofproto;
4002 cbdata.match = NULL;
4004 classifier_for_each_match(&ofconn->ofproto->cls, &fm->cr, CLS_INC_ALL,
4005 modify_flows_cb, &cbdata);
4007 /* This credits the packet to whichever flow happened to happened to
4008 * match last. That's weird. Maybe we should do a lookup for the
4009 * flow that actually matches the packet? Who knows. */
4010 send_buffered_packet(ofconn, cbdata.match, fm->buffer_id);
4013 return add_flow(ofconn, fm);
4017 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4018 * code as encoded by ofp_mkerr() on failure.
4020 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4023 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
4025 struct ofproto *p = ofconn->ofproto;
4026 struct rule *rule = find_flow_strict(p, fm);
4027 if (rule && !rule_is_hidden(rule)) {
4028 modify_flow(p, fm, rule);
4029 return send_buffered_packet(ofconn, rule, fm->buffer_id);
4031 return add_flow(ofconn, fm);
4035 /* Callback for modify_flows_loose(). */
4037 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
4039 struct rule *rule = rule_from_cls_rule(rule_);
4040 struct modify_flows_cbdata *cbdata = cbdata_;
4042 if (!rule_is_hidden(rule)) {
4043 cbdata->match = rule;
4044 modify_flow(cbdata->ofproto, cbdata->fm, rule);
4048 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4049 * been identified as a flow in 'p''s flow table to be modified, by changing
4050 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4051 * ofp_action[] structures). */
4053 modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule)
4055 size_t actions_len = fm->n_actions * sizeof *rule->actions;
4057 rule->flow_cookie = fm->cookie;
4059 /* If the actions are the same, do nothing. */
4060 if (fm->n_actions == rule->n_actions
4062 || !memcmp(fm->actions, rule->actions, actions_len))) {
4066 /* Replace actions. */
4067 free(rule->actions);
4068 rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL;
4069 rule->n_actions = fm->n_actions;
4071 p->need_revalidate = true;
4076 /* OFPFC_DELETE implementation. */
4078 struct delete_flows_cbdata {
4079 struct ofproto *ofproto;
4083 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
4084 static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port);
4086 /* Implements OFPFC_DELETE. */
4088 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
4090 struct delete_flows_cbdata cbdata;
4093 cbdata.out_port = htons(fm->out_port);
4095 classifier_for_each_match(&p->cls, &fm->cr, CLS_INC_ALL,
4096 delete_flows_cb, &cbdata);
4099 /* Implements OFPFC_DELETE_STRICT. */
4101 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
4103 struct rule *rule = find_flow_strict(p, fm);
4105 delete_flow(p, rule, htons(fm->out_port));
4109 /* Callback for delete_flows_loose(). */
4111 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
4113 struct rule *rule = rule_from_cls_rule(rule_);
4114 struct delete_flows_cbdata *cbdata = cbdata_;
4116 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
4119 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4120 * been identified as a flow to delete from 'p''s flow table, by deleting the
4121 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4124 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4125 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4126 * specified 'out_port'. */
4128 delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port)
4130 if (rule_is_hidden(rule)) {
4134 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
4138 rule_send_removed(p, rule, OFPRR_DELETE);
4139 rule_remove(p, rule);
4143 flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm)
4145 struct ofproto *p = ofconn->ofproto;
4148 error = reject_slave_controller(ofconn, "flow_mod");
4153 error = validate_actions(fm->actions, fm->n_actions,
4154 &fm->cr.flow, p->max_ports);
4159 /* We do not support the emergency flow cache. It will hopefully
4160 * get dropped from OpenFlow in the near future. */
4161 if (fm->flags & OFPFF_EMERG) {
4162 /* There isn't a good fit for an error code, so just state that the
4163 * flow table is full. */
4164 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
4167 switch (fm->command) {
4169 return add_flow(ofconn, fm);
4172 return modify_flows_loose(ofconn, fm);
4174 case OFPFC_MODIFY_STRICT:
4175 return modify_flow_strict(ofconn, fm);
4178 delete_flows_loose(p, fm);
4181 case OFPFC_DELETE_STRICT:
4182 delete_flow_strict(p, fm);
4186 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
4191 handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
4193 struct ofp_match orig_match;
4194 struct ofp_flow_mod *ofm;
4200 b.size = ntohs(oh->length);
4202 /* Dissect the message. */
4203 ofm = ofpbuf_try_pull(&b, sizeof *ofm);
4205 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4207 error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
4212 /* Normalize ofm->match. If normalization actually changes anything, then
4213 * log the differences. */
4214 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
4215 orig_match = ofm->match;
4216 normalize_match(&ofm->match);
4217 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
4218 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
4219 if (!VLOG_DROP_INFO(&normal_rl)) {
4220 char *old = ofp_match_to_literal_string(&orig_match);
4221 char *new = ofp_match_to_literal_string(&ofm->match);
4222 VLOG_INFO("%s: normalization changed ofp_match, details:",
4223 rconn_get_name(ofconn->rconn));
4224 VLOG_INFO(" pre: %s", old);
4225 VLOG_INFO("post: %s", new);
4231 /* Translate the message. */
4232 cls_rule_from_match(&ofm->match, ntohs(ofm->priority), ofconn->flow_format,
4233 ofm->cookie, &fm.cr);
4234 fm.cookie = ofm->cookie;
4235 fm.command = ntohs(ofm->command);
4236 fm.idle_timeout = ntohs(ofm->idle_timeout);
4237 fm.hard_timeout = ntohs(ofm->hard_timeout);
4238 fm.buffer_id = ntohl(ofm->buffer_id);
4239 fm.out_port = ntohs(ofm->out_port);
4240 fm.flags = ntohs(ofm->flags);
4242 /* Execute the command. */
4243 return flow_mod_core(ofconn, &fm);
4247 handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh)
4249 struct nx_flow_mod *nfm;
4255 b.size = ntohs(oh->length);
4257 /* Dissect the message. */
4258 nfm = ofpbuf_try_pull(&b, sizeof *nfm);
4260 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4262 error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority),
4267 error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions);
4272 /* Translate the message. */
4273 fm.cookie = nfm->cookie;
4274 fm.command = ntohs(nfm->command);
4275 fm.idle_timeout = ntohs(nfm->idle_timeout);
4276 fm.hard_timeout = ntohs(nfm->hard_timeout);
4277 fm.buffer_id = ntohl(nfm->buffer_id);
4278 fm.out_port = ntohs(nfm->out_port);
4279 fm.flags = ntohs(nfm->flags);
4281 /* Execute the command. */
4282 return flow_mod_core(ofconn, &fm);
4286 handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg)
4290 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
4295 ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10;
4300 handle_role_request(struct ofconn *ofconn, struct nicira_header *msg)
4302 struct nx_role_request *nrr;
4303 struct nx_role_request *reply;
4307 if (ntohs(msg->header.length) != sizeof *nrr) {
4308 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
4309 ntohs(msg->header.length), sizeof *nrr);
4310 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4312 nrr = (struct nx_role_request *) msg;
4314 if (ofconn->type != OFCONN_PRIMARY) {
4315 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4317 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4320 role = ntohl(nrr->role);
4321 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4322 && role != NX_ROLE_SLAVE) {
4323 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4325 /* There's no good error code for this. */
4326 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4329 if (role == NX_ROLE_MASTER) {
4330 struct ofconn *other;
4332 HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) {
4333 if (other->role == NX_ROLE_MASTER) {
4334 other->role = NX_ROLE_SLAVE;
4338 ofconn->role = role;
4340 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid,
4342 reply->role = htonl(role);
4343 queue_tx(buf, ofconn, ofconn->reply_counter);
4349 handle_nxt_set_flow_format(struct ofconn *ofconn,
4350 struct nxt_set_flow_format *msg)
4355 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
4360 format = ntohl(msg->format);
4361 if (format == NXFF_OPENFLOW10
4362 || format == NXFF_TUN_ID_FROM_COOKIE
4363 || format == NXFF_NXM) {
4364 ofconn->flow_format = format;
4367 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4372 handle_vendor(struct ofconn *ofconn, void *msg)
4374 struct ofproto *p = ofconn->ofproto;
4375 struct ofp_vendor_header *ovh = msg;
4376 struct nicira_header *nh;
4378 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
4379 VLOG_WARN_RL(&rl, "received vendor message of length %u "
4380 "(expected at least %zu)",
4381 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
4382 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4384 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
4385 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
4387 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
4388 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
4389 "(expected at least %zu)",
4390 ntohs(ovh->header.length), sizeof(struct nicira_header));
4391 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4395 switch (ntohl(nh->subtype)) {
4396 case NXT_STATUS_REQUEST:
4397 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4400 case NXT_TUN_ID_FROM_COOKIE:
4401 return handle_tun_id_from_cookie(ofconn, msg);
4403 case NXT_ROLE_REQUEST:
4404 return handle_role_request(ofconn, msg);
4406 case NXT_SET_FLOW_FORMAT:
4407 return handle_nxt_set_flow_format(ofconn, msg);
4410 return handle_nxt_flow_mod(ofconn, &ovh->header);
4413 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4417 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4419 struct ofp_header *ob;
4422 /* Currently, everything executes synchronously, so we can just
4423 * immediately send the barrier reply. */
4424 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4425 queue_tx(buf, ofconn, ofconn->reply_counter);
4430 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4432 struct ofp_header *oh = ofp_msg->data;
4435 COVERAGE_INC(ofproto_recv_openflow);
4437 case OFPT_ECHO_REQUEST:
4438 error = handle_echo_request(ofconn, oh);
4441 case OFPT_ECHO_REPLY:
4445 case OFPT_FEATURES_REQUEST:
4446 error = handle_features_request(ofconn, oh);
4449 case OFPT_GET_CONFIG_REQUEST:
4450 error = handle_get_config_request(ofconn, oh);
4453 case OFPT_SET_CONFIG:
4454 error = handle_set_config(ofconn, ofp_msg->data);
4457 case OFPT_PACKET_OUT:
4458 error = handle_packet_out(ofconn, ofp_msg->data);
4462 error = handle_port_mod(ofconn, oh);
4466 error = handle_ofpt_flow_mod(ofconn, ofp_msg->data);
4469 case OFPT_STATS_REQUEST:
4470 error = handle_stats_request(ofconn, oh);
4474 error = handle_vendor(ofconn, ofp_msg->data);
4477 case OFPT_BARRIER_REQUEST:
4478 error = handle_barrier_request(ofconn, oh);
4482 if (VLOG_IS_WARN_ENABLED()) {
4483 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4484 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4487 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4492 send_error_oh(ofconn, ofp_msg->data, error);
4497 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4499 struct odp_msg *msg = packet->data;
4500 struct ofpbuf payload;
4501 struct facet *facet;
4504 payload.data = msg + 1;
4505 payload.size = msg->length - sizeof *msg;
4506 flow_extract(&payload, msg->arg, msg->port, &flow);
4508 /* Check with in-band control to see if this packet should be sent
4509 * to the local port regardless of the flow table. */
4510 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4511 union odp_action action;
4513 memset(&action, 0, sizeof(action));
4514 action.output.type = ODPAT_OUTPUT;
4515 action.output.port = ODPP_LOCAL;
4516 dpif_execute(p->dpif, &action, 1, &payload);
4519 facet = facet_lookup_valid(p, &flow);
4521 struct rule *rule = rule_lookup(p, &flow);
4523 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4524 struct ofport *port = get_port(p, msg->port);
4526 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4527 COVERAGE_INC(ofproto_no_packet_in);
4528 /* XXX install 'drop' flow entry */
4529 ofpbuf_delete(packet);
4533 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
4537 COVERAGE_INC(ofproto_packet_in);
4538 send_packet_in(p, packet);
4542 facet = facet_create(p, rule, &flow, packet);
4543 } else if (!facet->may_install) {
4544 /* The facet is not installable, that is, we need to process every
4545 * packet, so process the current packet's actions into 'facet'. */
4546 facet_make_actions(p, facet, packet);
4549 if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) {
4551 * Extra-special case for fail-open mode.
4553 * We are in fail-open mode and the packet matched the fail-open rule,
4554 * but we are connected to a controller too. We should send the packet
4555 * up to the controller in the hope that it will try to set up a flow
4556 * and thereby allow us to exit fail-open.
4558 * See the top-level comment in fail-open.c for more information.
4560 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4561 DPIF_RECV_MSG_PADDING));
4564 ofpbuf_pull(packet, sizeof *msg);
4565 facet_execute(p, facet, packet);
4566 facet_install(p, facet, false);
4570 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4572 struct odp_msg *msg = packet->data;
4574 switch (msg->type) {
4575 case _ODPL_ACTION_NR:
4576 COVERAGE_INC(ofproto_ctlr_action);
4577 send_packet_in(p, packet);
4580 case _ODPL_SFLOW_NR:
4582 ofproto_sflow_received(p->sflow, msg);
4584 ofpbuf_delete(packet);
4588 handle_odp_miss_msg(p, packet);
4592 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4598 /* Flow expiration. */
4600 struct expire_cbdata {
4601 struct ofproto *ofproto;
4605 static int ofproto_dp_max_idle(const struct ofproto *);
4606 static void ofproto_update_used(struct ofproto *);
4607 static void rule_expire(struct cls_rule *, void *cbdata);
4608 static void ofproto_expire_facets(struct ofproto *, int dp_max_idle);
4610 /* This function is called periodically by ofproto_run(). Its job is to
4611 * collect updates for the flows that have been installed into the datapath,
4612 * most importantly when they last were used, and then use that information to
4613 * expire flows that have not been used recently.
4615 * Returns the number of milliseconds after which it should be called again. */
4617 ofproto_expire(struct ofproto *ofproto)
4619 struct expire_cbdata cbdata;
4621 /* Update 'used' for each flow in the datapath. */
4622 ofproto_update_used(ofproto);
4624 /* Expire facets that have been idle too long. */
4625 cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
4626 ofproto_expire_facets(ofproto, cbdata.dp_max_idle);
4628 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4629 cbdata.ofproto = ofproto;
4630 classifier_for_each(&ofproto->cls, CLS_INC_ALL, rule_expire, &cbdata);
4632 /* Let the hook know that we're at a stable point: all outstanding data
4633 * in existing flows has been accounted to the account_cb. Thus, the
4634 * hook can now reasonably do operations that depend on having accurate
4635 * flow volume accounting (currently, that's just bond rebalancing). */
4636 if (ofproto->ofhooks->account_checkpoint_cb) {
4637 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4640 return MIN(cbdata.dp_max_idle, 1000);
4643 /* Update 'used' member of installed facets. */
4645 ofproto_update_used(struct ofproto *p)
4647 struct odp_flow *flows;
4652 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4657 for (i = 0; i < n_flows; i++) {
4658 struct odp_flow *f = &flows[i];
4659 struct facet *facet;
4662 odp_flow_key_to_flow(&f->key, &flow);
4663 facet = facet_find(p, &flow);
4665 if (facet && facet->installed) {
4666 facet_update_time(p, facet, &f->stats);
4667 facet_account(p, facet, f->stats.n_bytes);
4669 /* There's a flow in the datapath that we know nothing about.
4671 COVERAGE_INC(ofproto_unexpected_rule);
4672 dpif_flow_del(p->dpif, f);
4679 /* Calculates and returns the number of milliseconds of idle time after which
4680 * facets should expire from the datapath and we should fold their statistics
4681 * into their parent rules in userspace. */
4683 ofproto_dp_max_idle(const struct ofproto *ofproto)
4686 * Idle time histogram.
4688 * Most of the time a switch has a relatively small number of facets. When
4689 * this is the case we might as well keep statistics for all of them in
4690 * userspace and to cache them in the kernel datapath for performance as
4693 * As the number of facets increases, the memory required to maintain
4694 * statistics about them in userspace and in the kernel becomes
4695 * significant. However, with a large number of facets it is likely that
4696 * only a few of them are "heavy hitters" that consume a large amount of
4697 * bandwidth. At this point, only heavy hitters are worth caching in the
4698 * kernel and maintaining in userspaces; other facets we can discard.
4700 * The technique used to compute the idle time is to build a histogram with
4701 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4702 * that is installed in the kernel gets dropped in the appropriate bucket.
4703 * After the histogram has been built, we compute the cutoff so that only
4704 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4705 * cached. At least the most-recently-used bucket of facets is kept, so
4706 * actually an arbitrary number of facets can be kept in any given
4707 * expiration run (though the next run will delete most of those unless
4708 * they receive additional data).
4710 * This requires a second pass through the facets, in addition to the pass
4711 * made by ofproto_update_used(), because the former function never looks
4712 * at uninstallable facets.
4714 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4715 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4716 int buckets[N_BUCKETS] = { 0 };
4717 struct facet *facet;
4722 total = hmap_count(&ofproto->facets);
4723 if (total <= 1000) {
4724 return N_BUCKETS * BUCKET_WIDTH;
4727 /* Build histogram. */
4729 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
4730 long long int idle = now - facet->used;
4731 int bucket = (idle <= 0 ? 0
4732 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4733 : (unsigned int) idle / BUCKET_WIDTH);
4737 /* Find the first bucket whose flows should be expired. */
4738 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4739 if (buckets[bucket]) {
4742 subtotal += buckets[bucket++];
4743 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4748 if (VLOG_IS_DBG_ENABLED()) {
4752 ds_put_cstr(&s, "keep");
4753 for (i = 0; i < N_BUCKETS; i++) {
4755 ds_put_cstr(&s, ", drop");
4758 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4761 VLOG_INFO("%s: %s (msec:count)",
4762 dpif_name(ofproto->dpif), ds_cstr(&s));
4766 return bucket * BUCKET_WIDTH;
4770 facet_active_timeout(struct ofproto *ofproto, struct facet *facet)
4772 if (ofproto->netflow && !facet_is_controller_flow(facet) &&
4773 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4774 struct ofexpired expired;
4775 struct odp_flow odp_flow;
4777 /* Get updated flow stats.
4779 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4780 * updated TCP flags and (2) the dpif_flow_list_all() in
4781 * ofproto_update_used() zeroed TCP flags. */
4782 memset(&odp_flow, 0, sizeof odp_flow);
4783 if (facet->installed) {
4784 odp_flow_key_from_flow(&odp_flow.key, &facet->flow);
4785 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4786 dpif_flow_get(ofproto->dpif, &odp_flow);
4788 if (odp_flow.stats.n_packets) {
4789 facet_update_time(ofproto, facet, &odp_flow.stats);
4790 netflow_flow_update_flags(&facet->nf_flow,
4791 odp_flow.stats.tcp_flags);
4795 expired.flow = facet->flow;
4796 expired.packet_count = facet->packet_count +
4797 odp_flow.stats.n_packets;
4798 expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes;
4799 expired.used = facet->used;
4801 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4806 ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle)
4808 long long int cutoff = time_msec() - dp_max_idle;
4809 struct facet *facet, *next_facet;
4811 HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
4812 facet_active_timeout(ofproto, facet);
4813 if (facet->used < cutoff) {
4814 facet_remove(ofproto, facet);
4819 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4820 * rules, then delete it entirely.
4822 * (This is a callback function for classifier_for_each().) */
4824 rule_expire(struct cls_rule *cls_rule, void *cbdata_)
4826 struct expire_cbdata *cbdata = cbdata_;
4827 struct rule *rule = rule_from_cls_rule(cls_rule);
4828 struct facet *facet, *next_facet;
4832 /* Has 'rule' expired? */
4834 if (rule->hard_timeout
4835 && now > rule->created + rule->hard_timeout * 1000) {
4836 reason = OFPRR_HARD_TIMEOUT;
4837 } else if (rule->idle_timeout && list_is_empty(&rule->facets)
4838 && now >rule->used + rule->idle_timeout * 1000) {
4839 reason = OFPRR_IDLE_TIMEOUT;
4844 COVERAGE_INC(ofproto_expired);
4846 /* Update stats. (This is a no-op if the rule expired due to an idle
4847 * timeout, because that only happens when the rule has no facets left.) */
4848 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4849 facet_remove(cbdata->ofproto, facet);
4852 /* Get rid of the rule. */
4853 if (!rule_is_hidden(rule)) {
4854 rule_send_removed(cbdata->ofproto, rule, reason);
4856 rule_remove(cbdata->ofproto, rule);
4859 static struct ofpbuf *
4860 compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule,
4863 struct ofp_flow_removed *ofr;
4866 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4867 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, ofconn->flow_format,
4869 ofr->cookie = rule->flow_cookie;
4870 ofr->priority = htons(rule->cr.priority);
4871 ofr->reason = reason;
4872 calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec);
4873 ofr->idle_timeout = htons(rule->idle_timeout);
4874 ofr->packet_count = htonll(rule->packet_count);
4875 ofr->byte_count = htonll(rule->byte_count);
4880 static struct ofpbuf *
4881 compose_nx_flow_removed(const struct rule *rule, uint8_t reason)
4883 struct nx_flow_removed *nfr;
4887 nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf);
4889 match_len = nx_put_match(buf, &rule->cr);
4891 nfr->cookie = rule->flow_cookie;
4892 nfr->priority = htons(rule->cr.priority);
4893 nfr->reason = reason;
4894 calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec);
4895 nfr->idle_timeout = htons(rule->idle_timeout);
4896 nfr->match_len = htons(match_len);
4897 nfr->packet_count = htonll(rule->packet_count);
4898 nfr->byte_count = htonll(rule->byte_count);
4904 rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason)
4906 struct ofconn *ofconn;
4908 if (!rule->send_flow_removed) {
4912 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4915 if (!rconn_is_connected(ofconn->rconn)
4916 || !ofconn_receives_async_msgs(ofconn)) {
4920 msg = (ofconn->flow_format == NXFF_NXM
4921 ? compose_nx_flow_removed(rule, reason)
4922 : compose_ofp_flow_removed(ofconn, rule, reason));
4924 /* Account flow expirations under ofconn->reply_counter, the counter
4925 * for replies to OpenFlow requests. That works because preventing
4926 * OpenFlow requests from being processed also prevents new flows from
4927 * being added (and expiring). (It also prevents processing OpenFlow
4928 * requests that would not add new flows, so it is imperfect.) */
4929 queue_tx(msg, ofconn, ofconn->reply_counter);
4933 /* pinsched callback for sending 'packet' on 'ofconn'. */
4935 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4937 struct ofconn *ofconn = ofconn_;
4939 rconn_send_with_limit(ofconn->rconn, packet,
4940 ofconn->packet_in_counter, 100);
4943 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4944 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4945 * packet scheduler for sending.
4947 * 'max_len' specifies the maximum number of bytes of the packet to send on
4948 * 'ofconn' (INT_MAX specifies no limit).
4950 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4951 * ownership is transferred to this function. */
4953 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4956 struct ofproto *ofproto = ofconn->ofproto;
4957 struct ofp_packet_in *opi = packet->data;
4958 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4959 int send_len, trim_size;
4963 if (opi->reason == OFPR_ACTION) {
4964 buffer_id = UINT32_MAX;
4965 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4966 buffer_id = pktbuf_get_null();
4967 } else if (!ofconn->pktbuf) {
4968 buffer_id = UINT32_MAX;
4970 struct ofpbuf payload;
4971 payload.data = opi->data;
4972 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4973 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4976 /* Figure out how much of the packet to send. */
4977 send_len = ntohs(opi->total_len);
4978 if (buffer_id != UINT32_MAX) {
4979 send_len = MIN(send_len, ofconn->miss_send_len);
4981 send_len = MIN(send_len, max_len);
4983 /* Adjust packet length and clone if necessary. */
4984 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4986 packet = ofpbuf_clone_data(packet->data, trim_size);
4989 packet->size = trim_size;
4992 /* Update packet headers. */
4993 opi->buffer_id = htonl(buffer_id);
4994 update_openflow_length(packet);
4996 /* Hand over to packet scheduler. It might immediately call into
4997 * do_send_packet_in() or it might buffer it for a while (until a later
4998 * call to pinsched_run()). */
4999 pinsched_send(ofconn->schedulers[opi->reason], in_port,
5000 packet, do_send_packet_in, ofconn);
5003 /* Replace struct odp_msg header in 'packet' by equivalent struct
5004 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
5005 * returned by dpif_recv()).
5007 * The conversion is not complete: the caller still needs to trim any unneeded
5008 * payload off the end of the buffer, set the length in the OpenFlow header,
5009 * and set buffer_id. Those require us to know the controller settings and so
5010 * must be done on a per-controller basis.
5012 * Returns the maximum number of bytes of the packet that should be sent to
5013 * the controller (INT_MAX if no limit). */
5015 do_convert_to_packet_in(struct ofpbuf *packet)
5017 struct odp_msg *msg = packet->data;
5018 struct ofp_packet_in *opi;
5024 /* Extract relevant header fields */
5025 if (msg->type == _ODPL_ACTION_NR) {
5026 reason = OFPR_ACTION;
5029 reason = OFPR_NO_MATCH;
5032 total_len = msg->length - sizeof *msg;
5033 in_port = odp_port_to_ofp_port(msg->port);
5035 /* Repurpose packet buffer by overwriting header. */
5036 ofpbuf_pull(packet, sizeof(struct odp_msg));
5037 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
5038 opi->header.version = OFP_VERSION;
5039 opi->header.type = OFPT_PACKET_IN;
5040 opi->total_len = htons(total_len);
5041 opi->in_port = htons(in_port);
5042 opi->reason = reason;
5047 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
5048 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
5049 * as necessary according to their individual configurations.
5051 * 'packet' must have sufficient headroom to convert it into a struct
5052 * ofp_packet_in (e.g. as returned by dpif_recv()).
5054 * Takes ownership of 'packet'. */
5056 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
5058 struct ofconn *ofconn, *prev;
5061 max_len = do_convert_to_packet_in(packet);
5064 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
5065 if (ofconn_receives_async_msgs(ofconn)) {
5067 schedule_packet_in(prev, packet, max_len, true);
5073 schedule_packet_in(prev, packet, max_len, false);
5075 ofpbuf_delete(packet);
5080 pick_datapath_id(const struct ofproto *ofproto)
5082 const struct ofport *port;
5084 port = get_port(ofproto, ODPP_LOCAL);
5086 uint8_t ea[ETH_ADDR_LEN];
5089 error = netdev_get_etheraddr(port->netdev, ea);
5091 return eth_addr_to_uint64(ea);
5093 VLOG_WARN("could not get MAC address for %s (%s)",
5094 netdev_get_name(port->netdev), strerror(error));
5096 return ofproto->fallback_dpid;
5100 pick_fallback_dpid(void)
5102 uint8_t ea[ETH_ADDR_LEN];
5103 eth_addr_nicira_random(ea);
5104 return eth_addr_to_uint64(ea);
5108 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
5109 struct odp_actions *actions, tag_type *tags,
5110 uint16_t *nf_output_iface, void *ofproto_)
5112 struct ofproto *ofproto = ofproto_;
5115 /* Drop frames for reserved multicast addresses. */
5116 if (eth_addr_is_reserved(flow->dl_dst)) {
5120 /* Learn source MAC (but don't try to learn from revalidation). */
5121 if (packet != NULL) {
5122 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
5124 GRAT_ARP_LOCK_NONE);
5126 /* The log messages here could actually be useful in debugging,
5127 * so keep the rate limit relatively high. */
5128 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
5129 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
5130 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
5131 ofproto_revalidate(ofproto, rev_tag);
5135 /* Determine output port. */
5136 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
5139 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
5140 nf_output_iface, actions);
5141 } else if (out_port != flow->in_port) {
5142 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
5143 *nf_output_iface = out_port;
5151 static const struct ofhooks default_ofhooks = {
5152 default_normal_ofhook_cb,