2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
30 #include "fail-open.h"
32 #include "mac-learning.h"
36 #include "ofp-print.h"
37 #include "ofproto-sflow.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openvswitch/datapath-protocol.h"
45 #include "poll-loop.h"
46 #include "port-array.h"
51 #include "stream-ssl.h"
59 #define THIS_MODULE VLM_ofproto
62 #include "sflow_api.h"
66 TABLEID_CLASSIFIER = 1
70 struct netdev *netdev;
71 struct ofp_phy_port opp; /* In host byte order. */
74 static void ofport_free(struct ofport *);
75 static void hton_ofp_phy_port(struct ofp_phy_port *);
77 static int xlate_actions(const union ofp_action *in, size_t n_in,
78 const flow_t *flow, struct ofproto *ofproto,
79 const struct ofpbuf *packet,
80 struct odp_actions *out, tag_type *tags,
81 bool *may_set_up_flow, uint16_t *nf_output_iface);
86 uint64_t flow_cookie; /* Controller-issued identifier.
87 (Kept in network-byte order.) */
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 bool send_flow_removed; /* Send a flow removed message? */
91 long long int used; /* Last-used time (0 if never used). */
92 long long int created; /* Creation time. */
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
95 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
96 tag_type tags; /* Tags (set only by hooks). */
97 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
99 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
100 * exact-match rule (having cr.wc.wildcards of 0) generated from the
101 * wildcard rule 'super'. In this case, 'list' is an element of the
104 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
105 * a list of subrules. A super-rule with no wildcards (where
106 * cr.wc.wildcards is 0) will never have any subrules. */
112 * 'n_actions' is the number of elements in the 'actions' array. A single
113 * action may take up more more than one element's worth of space.
115 * A subrule has no actions (it uses the super-rule's actions). */
117 union ofp_action *actions;
121 * A super-rule with wildcard fields never has ODP actions (since the
122 * datapath only supports exact-match flows). */
123 bool installed; /* Installed in datapath? */
124 bool may_install; /* True ordinarily; false if actions must
125 * be reassessed for every packet. */
127 union odp_action *odp_actions;
131 rule_is_hidden(const struct rule *rule)
133 /* Subrules are merely an implementation detail, so hide them from the
135 if (rule->super != NULL) {
139 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
140 * (e.g. by in-band control) and are intentionally hidden from the
142 if (rule->cr.priority > UINT16_MAX) {
149 static struct rule *rule_create(struct ofproto *, struct rule *super,
150 const union ofp_action *, size_t n_actions,
151 uint16_t idle_timeout, uint16_t hard_timeout,
152 uint64_t flow_cookie, bool send_flow_removed);
153 static void rule_free(struct rule *);
154 static void rule_destroy(struct ofproto *, struct rule *);
155 static struct rule *rule_from_cls_rule(const struct cls_rule *);
156 static void rule_insert(struct ofproto *, struct rule *,
157 struct ofpbuf *packet, uint16_t in_port);
158 static void rule_remove(struct ofproto *, struct rule *);
159 static bool rule_make_actions(struct ofproto *, struct rule *,
160 const struct ofpbuf *packet);
161 static void rule_install(struct ofproto *, struct rule *,
162 struct rule *displaced_rule);
163 static void rule_uninstall(struct ofproto *, struct rule *);
164 static void rule_post_uninstall(struct ofproto *, struct rule *);
165 static void send_flow_removed(struct ofproto *p, struct rule *rule,
166 long long int now, uint8_t reason);
168 /* ofproto supports two kinds of OpenFlow connections:
170 * - "Controller connections": Connections to ordinary OpenFlow controllers.
171 * ofproto maintains persistent connections to these controllers and by
172 * default sends them asynchronous messages such as packet-ins.
174 * - "Transient connections", e.g. from ovs-ofctl. When these connections
175 * drop, it is the other side's responsibility to reconnect them if
176 * necessary. ofproto does not send them asynchronous messages by default.
179 OFCONN_CONTROLLER, /* An OpenFlow controller. */
180 OFCONN_TRANSIENT /* A transient connection. */
183 /* An OpenFlow connection. */
185 struct ofproto *ofproto; /* The ofproto that owns this connection. */
186 struct list node; /* In struct ofproto's "all_conns" list. */
187 struct rconn *rconn; /* OpenFlow connection. */
188 enum ofconn_type type; /* Type. */
190 /* OFPT_PACKET_IN related data. */
191 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
192 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
193 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
194 int miss_send_len; /* Bytes to send of buffered packets. */
196 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
197 * requests, and the maximum number before we stop reading OpenFlow
199 #define OFCONN_REPLY_MAX 100
200 struct rconn_packet_counter *reply_counter;
202 /* type == OFCONN_CONTROLLER only. */
203 enum nx_role role; /* Role. */
204 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
205 struct discovery *discovery; /* Controller discovery object, if enabled. */
206 struct status_category *ss; /* Switch status category. */
209 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
210 * "schedulers" array. Their values are 0 and 1, and their meanings and values
211 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
212 * case anything ever changes, check their values here. */
213 #define N_SCHEDULERS 2
214 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
215 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
216 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
217 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
219 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
221 static void ofconn_destroy(struct ofconn *);
222 static void ofconn_run(struct ofconn *, struct ofproto *);
223 static void ofconn_wait(struct ofconn *);
224 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
225 struct rconn_packet_counter *counter);
227 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
228 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
232 uint64_t datapath_id; /* Datapath ID. */
233 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
234 char *mfr_desc; /* Manufacturer. */
235 char *hw_desc; /* Hardware. */
236 char *sw_desc; /* Software version. */
237 char *serial_desc; /* Serial number. */
238 char *dp_desc; /* Datapath description. */
242 struct netdev_monitor *netdev_monitor;
243 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
245 struct shash port_by_name;
249 struct switch_status *switch_status;
250 struct in_band *in_band;
251 struct fail_open *fail_open;
252 struct netflow *netflow;
253 struct ofproto_sflow *sflow;
256 struct classifier cls;
257 bool need_revalidate;
258 long long int next_expiration;
259 struct tag_set revalidate_set;
260 bool tun_id_from_cookie;
262 /* OpenFlow connections. */
263 struct hmap controllers; /* Controller "struct ofconn"s. */
264 struct list all_conns; /* Contains "struct ofconn"s. */
265 struct pvconn **listeners;
267 struct pvconn **snoops;
270 /* Hooks for ovs-vswitchd. */
271 const struct ofhooks *ofhooks;
274 /* Used by default ofhooks. */
275 struct mac_learning *ml;
278 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
280 static const struct ofhooks default_ofhooks;
282 static uint64_t pick_datapath_id(const struct ofproto *);
283 static uint64_t pick_fallback_dpid(void);
285 static void update_used(struct ofproto *);
286 static void update_stats(struct ofproto *, struct rule *,
287 const struct odp_flow_stats *);
288 static void expire_rule(struct cls_rule *, void *ofproto);
289 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
290 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
291 static void revalidate_cb(struct cls_rule *rule_, void *p_);
293 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
295 static void handle_openflow(struct ofconn *, struct ofproto *,
298 static void refresh_port_groups(struct ofproto *);
300 static void update_port(struct ofproto *, const char *devname);
301 static int init_ports(struct ofproto *);
302 static void reinit_ports(struct ofproto *);
305 ofproto_create(const char *datapath, const char *datapath_type,
306 const struct ofhooks *ofhooks, void *aux,
307 struct ofproto **ofprotop)
309 struct odp_stats stats;
316 /* Connect to datapath and start listening for messages. */
317 error = dpif_open(datapath, datapath_type, &dpif);
319 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
322 error = dpif_get_dp_stats(dpif, &stats);
324 VLOG_ERR("failed to obtain stats for datapath %s: %s",
325 datapath, strerror(error));
329 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
331 VLOG_ERR("failed to listen on datapath %s: %s",
332 datapath, strerror(error));
336 dpif_flow_flush(dpif);
337 dpif_recv_purge(dpif);
339 /* Initialize settings. */
340 p = xzalloc(sizeof *p);
341 p->fallback_dpid = pick_fallback_dpid();
342 p->datapath_id = p->fallback_dpid;
343 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
344 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
345 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
346 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
347 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
349 /* Initialize datapath. */
351 p->netdev_monitor = netdev_monitor_create();
352 port_array_init(&p->ports);
353 shash_init(&p->port_by_name);
354 p->max_ports = stats.max_ports;
356 /* Initialize submodules. */
357 p->switch_status = switch_status_create(p);
363 /* Initialize flow table. */
364 classifier_init(&p->cls);
365 p->need_revalidate = false;
366 p->next_expiration = time_msec() + 1000;
367 tag_set_init(&p->revalidate_set);
369 /* Initialize OpenFlow connections. */
370 list_init(&p->all_conns);
371 hmap_init(&p->controllers);
377 /* Initialize hooks. */
379 p->ofhooks = ofhooks;
383 p->ofhooks = &default_ofhooks;
385 p->ml = mac_learning_create();
388 /* Pick final datapath ID. */
389 p->datapath_id = pick_datapath_id(p);
390 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
397 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
399 uint64_t old_dpid = p->datapath_id;
400 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
401 if (p->datapath_id != old_dpid) {
402 struct ofconn *ofconn;
404 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
406 /* Force all active connections to reconnect, since there is no way to
407 * notify a controller that the datapath ID has changed. */
408 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
409 rconn_reconnect(ofconn->rconn);
415 is_discovery_controller(const struct ofproto_controller *c)
417 return !strcmp(c->target, "discover");
421 is_in_band_controller(const struct ofproto_controller *c)
423 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
426 /* Creates a new controller in 'ofproto'. Some of the settings are initially
427 * drawn from 'c', but update_controller() needs to be called later to finish
428 * the new ofconn's configuration. */
430 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
432 struct discovery *discovery;
433 struct ofconn *ofconn;
435 if (is_discovery_controller(c)) {
436 int error = discovery_create(c->accept_re, c->update_resolv_conf,
437 ofproto->dpif, ofproto->switch_status,
446 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
447 ofconn->pktbuf = pktbuf_create();
448 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
450 ofconn->discovery = discovery;
452 rconn_connect(ofconn->rconn, c->target);
454 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
455 hash_string(c->target, 0));
458 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
459 * target or turn discovery on or off (these are done by creating new ofconns
460 * and deleting old ones), but it can update the rest of an ofconn's
463 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
465 struct ofproto *ofproto = ofconn->ofproto;
469 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
471 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
472 rconn_set_probe_interval(ofconn->rconn, probe_interval);
474 if (ofconn->discovery) {
475 discovery_set_update_resolv_conf(ofconn->discovery,
476 c->update_resolv_conf);
477 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
480 for (i = 0; i < N_SCHEDULERS; i++) {
481 struct pinsched **s = &ofconn->schedulers[i];
483 if (c->rate_limit > 0) {
485 *s = pinsched_create(c->rate_limit, c->burst_limit,
486 ofproto->switch_status);
488 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
491 pinsched_destroy(*s);
498 ofconn_get_target(const struct ofconn *ofconn)
500 return ofconn->discovery ? "discover" : rconn_get_name(ofconn->rconn);
503 static struct ofconn *
504 find_controller_by_target(struct ofproto *ofproto, const char *target)
506 struct ofconn *ofconn;
508 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
509 hash_string(target, 0), &ofproto->controllers) {
510 if (!strcmp(ofconn_get_target(ofconn), target)) {
518 ofproto_set_controllers(struct ofproto *p,
519 const struct ofproto_controller *controllers,
520 size_t n_controllers)
522 struct shash new_controllers;
523 struct rconn **in_band_rconns;
524 enum ofproto_fail_mode fail_mode;
525 struct ofconn *ofconn, *next;
530 shash_init(&new_controllers);
531 for (i = 0; i < n_controllers; i++) {
532 const struct ofproto_controller *c = &controllers[i];
534 shash_add_once(&new_controllers, c->target, &controllers[i]);
535 if (!find_controller_by_target(p, c->target)) {
536 add_controller(p, c);
540 in_band_rconns = xmalloc(n_controllers * sizeof *in_band_rconns);
542 fail_mode = OFPROTO_FAIL_STANDALONE;
544 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
546 struct ofproto_controller *c;
548 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
550 ofconn_destroy(ofconn);
552 update_controller(ofconn, c);
557 if (is_in_band_controller(c)) {
558 in_band_rconns[n_in_band++] = ofconn->rconn;
561 if (c->fail == OFPROTO_FAIL_SECURE) {
562 fail_mode = OFPROTO_FAIL_SECURE;
566 shash_destroy(&new_controllers);
570 in_band_create(p, p->dpif, p->switch_status, &p->in_band);
573 in_band_set_remotes(p->in_band, in_band_rconns, n_in_band);
576 in_band_destroy(p->in_band);
579 free(in_band_rconns);
581 if (!hmap_is_empty(&p->controllers)
582 && fail_mode == OFPROTO_FAIL_STANDALONE) {
583 struct rconn **rconns;
587 p->fail_open = fail_open_create(p, p->switch_status);
591 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
592 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
593 rconns[n++] = ofconn->rconn;
596 fail_open_set_controllers(p->fail_open, rconns, n);
597 /* p->fail_open takes ownership of 'rconns'. */
599 fail_open_destroy(p->fail_open);
603 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
604 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
605 struct ofconn, hmap_node);
606 ofconn->ss = switch_status_register(p->switch_status, "remote",
607 rconn_status_cb, ofconn->rconn);
612 ofproto_set_desc(struct ofproto *p,
613 const char *mfr_desc, const char *hw_desc,
614 const char *sw_desc, const char *serial_desc,
617 struct ofp_desc_stats *ods;
620 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
621 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
622 sizeof ods->mfr_desc);
625 p->mfr_desc = xstrdup(mfr_desc);
628 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
629 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
630 sizeof ods->hw_desc);
633 p->hw_desc = xstrdup(hw_desc);
636 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
637 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
638 sizeof ods->sw_desc);
641 p->sw_desc = xstrdup(sw_desc);
644 if (strlen(serial_desc) >= sizeof ods->serial_num) {
645 VLOG_WARN("truncating serial_desc, must be less than %zu "
647 sizeof ods->serial_num);
649 free(p->serial_desc);
650 p->serial_desc = xstrdup(serial_desc);
653 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
654 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
655 sizeof ods->dp_desc);
658 p->dp_desc = xstrdup(dp_desc);
663 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
664 const struct svec *svec)
666 struct pvconn **pvconns = *pvconnsp;
667 size_t n_pvconns = *n_pvconnsp;
671 for (i = 0; i < n_pvconns; i++) {
672 pvconn_close(pvconns[i]);
676 pvconns = xmalloc(svec->n * sizeof *pvconns);
678 for (i = 0; i < svec->n; i++) {
679 const char *name = svec->names[i];
680 struct pvconn *pvconn;
683 error = pvconn_open(name, &pvconn);
685 pvconns[n_pvconns++] = pvconn;
687 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
695 *n_pvconnsp = n_pvconns;
701 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
703 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
707 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
709 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
713 ofproto_set_netflow(struct ofproto *ofproto,
714 const struct netflow_options *nf_options)
716 if (nf_options && nf_options->collectors.n) {
717 if (!ofproto->netflow) {
718 ofproto->netflow = netflow_create();
720 return netflow_set_options(ofproto->netflow, nf_options);
722 netflow_destroy(ofproto->netflow);
723 ofproto->netflow = NULL;
729 ofproto_set_sflow(struct ofproto *ofproto,
730 const struct ofproto_sflow_options *oso)
732 struct ofproto_sflow *os = ofproto->sflow;
735 struct ofport *ofport;
736 unsigned int odp_port;
738 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
739 refresh_port_groups(ofproto);
740 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
741 ofproto_sflow_add_port(os, odp_port,
742 netdev_get_name(ofport->netdev));
745 ofproto_sflow_set_options(os, oso);
747 ofproto_sflow_destroy(os);
748 ofproto->sflow = NULL;
753 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
757 VLOG_WARN("STP is not yet implemented");
765 ofproto_get_datapath_id(const struct ofproto *ofproto)
767 return ofproto->datapath_id;
771 ofproto_has_controller(const struct ofproto *ofproto)
773 return !hmap_is_empty(&ofproto->controllers);
777 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
781 for (i = 0; i < ofproto->n_listeners; i++) {
782 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
787 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
791 for (i = 0; i < ofproto->n_snoops; i++) {
792 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
797 ofproto_destroy(struct ofproto *p)
799 struct ofconn *ofconn, *next_ofconn;
800 struct ofport *ofport;
801 unsigned int port_no;
808 /* Destroy fail-open and in-band early, since they touch the classifier. */
809 fail_open_destroy(p->fail_open);
812 in_band_destroy(p->in_band);
815 ofproto_flush_flows(p);
816 classifier_destroy(&p->cls);
818 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
820 ofconn_destroy(ofconn);
822 hmap_destroy(&p->controllers);
825 netdev_monitor_destroy(p->netdev_monitor);
826 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
829 shash_destroy(&p->port_by_name);
831 switch_status_destroy(p->switch_status);
832 netflow_destroy(p->netflow);
833 ofproto_sflow_destroy(p->sflow);
835 for (i = 0; i < p->n_listeners; i++) {
836 pvconn_close(p->listeners[i]);
840 for (i = 0; i < p->n_snoops; i++) {
841 pvconn_close(p->snoops[i]);
845 mac_learning_destroy(p->ml);
850 free(p->serial_desc);
853 port_array_destroy(&p->ports);
859 ofproto_run(struct ofproto *p)
861 int error = ofproto_run1(p);
863 error = ofproto_run2(p, false);
869 process_port_change(struct ofproto *ofproto, int error, char *devname)
871 if (error == ENOBUFS) {
872 reinit_ports(ofproto);
874 update_port(ofproto, devname);
879 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
880 * Connects this vconn to a controller. */
882 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
884 struct ofconn *ofconn;
886 /* Arbitrarily pick the first controller in the list for monitoring. We
887 * could do something smarter or more flexible later, if it ever proves
889 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
890 if (ofconn->type == OFCONN_CONTROLLER) {
891 rconn_add_monitor(ofconn->rconn, vconn);
896 VLOG_INFO_RL(&rl, "no controller connection to monitor");
901 ofproto_run1(struct ofproto *p)
903 struct ofconn *ofconn, *next_ofconn;
908 if (shash_is_empty(&p->port_by_name)) {
912 for (i = 0; i < 50; i++) {
916 error = dpif_recv(p->dpif, &buf);
918 if (error == ENODEV) {
919 /* Someone destroyed the datapath behind our back. The caller
920 * better destroy us and give up, because we're just going to
921 * spin from here on out. */
922 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
923 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
930 handle_odp_msg(p, buf);
933 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
934 process_port_change(p, error, devname);
936 while ((error = netdev_monitor_poll(p->netdev_monitor,
937 &devname)) != EAGAIN) {
938 process_port_change(p, error, devname);
942 in_band_run(p->in_band);
945 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
947 ofconn_run(ofconn, p);
950 /* Fail-open maintenance. Do this after processing the ofconns since
951 * fail-open checks the status of the controller rconn. */
953 fail_open_run(p->fail_open);
956 for (i = 0; i < p->n_listeners; i++) {
960 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
962 ofconn_create(p, rconn_new_from_vconn("passive", vconn),
964 } else if (retval != EAGAIN) {
965 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
969 for (i = 0; i < p->n_snoops; i++) {
973 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
975 add_snooper(p, vconn);
976 } else if (retval != EAGAIN) {
977 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
981 if (time_msec() >= p->next_expiration) {
982 COVERAGE_INC(ofproto_expiration);
983 p->next_expiration = time_msec() + 1000;
986 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
988 /* Let the hook know that we're at a stable point: all outstanding data
989 * in existing flows has been accounted to the account_cb. Thus, the
990 * hook can now reasonably do operations that depend on having accurate
991 * flow volume accounting (currently, that's just bond rebalancing). */
992 if (p->ofhooks->account_checkpoint_cb) {
993 p->ofhooks->account_checkpoint_cb(p->aux);
998 netflow_run(p->netflow);
1001 ofproto_sflow_run(p->sflow);
1007 struct revalidate_cbdata {
1008 struct ofproto *ofproto;
1009 bool revalidate_all; /* Revalidate all exact-match rules? */
1010 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1011 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1015 ofproto_run2(struct ofproto *p, bool revalidate_all)
1017 if (p->need_revalidate || revalidate_all
1018 || !tag_set_is_empty(&p->revalidate_set)) {
1019 struct revalidate_cbdata cbdata;
1021 cbdata.revalidate_all = revalidate_all;
1022 cbdata.revalidate_subrules = p->need_revalidate;
1023 cbdata.revalidate_set = p->revalidate_set;
1024 tag_set_init(&p->revalidate_set);
1025 COVERAGE_INC(ofproto_revalidate);
1026 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1027 p->need_revalidate = false;
1034 ofproto_wait(struct ofproto *p)
1036 struct ofconn *ofconn;
1039 dpif_recv_wait(p->dpif);
1040 dpif_port_poll_wait(p->dpif);
1041 netdev_monitor_poll_wait(p->netdev_monitor);
1042 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1043 ofconn_wait(ofconn);
1046 in_band_wait(p->in_band);
1049 fail_open_wait(p->fail_open);
1052 ofproto_sflow_wait(p->sflow);
1054 if (!tag_set_is_empty(&p->revalidate_set)) {
1055 poll_immediate_wake();
1057 if (p->need_revalidate) {
1058 /* Shouldn't happen, but if it does just go around again. */
1059 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1060 poll_immediate_wake();
1061 } else if (p->next_expiration != LLONG_MAX) {
1062 poll_timer_wait(p->next_expiration - time_msec());
1064 for (i = 0; i < p->n_listeners; i++) {
1065 pvconn_wait(p->listeners[i]);
1067 for (i = 0; i < p->n_snoops; i++) {
1068 pvconn_wait(p->snoops[i]);
1073 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1075 tag_set_add(&ofproto->revalidate_set, tag);
1079 ofproto_get_revalidate_set(struct ofproto *ofproto)
1081 return &ofproto->revalidate_set;
1085 ofproto_is_alive(const struct ofproto *p)
1087 return !hmap_is_empty(&p->controllers);
1091 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1092 const union ofp_action *actions, size_t n_actions,
1093 const struct ofpbuf *packet)
1095 struct odp_actions odp_actions;
1098 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1104 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1106 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1107 odp_actions.n_actions, packet);
1112 ofproto_add_flow(struct ofproto *p,
1113 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1114 const union ofp_action *actions, size_t n_actions,
1118 rule = rule_create(p, NULL, actions, n_actions,
1119 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1121 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1122 rule_insert(p, rule, NULL, 0);
1126 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1127 uint32_t wildcards, unsigned int priority)
1131 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1135 rule_remove(ofproto, rule);
1140 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1142 struct rule *rule = rule_from_cls_rule(rule_);
1143 struct ofproto *ofproto = ofproto_;
1145 /* Mark the flow as not installed, even though it might really be
1146 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1147 * There is no point in uninstalling it individually since we are about to
1148 * blow away all the flows with dpif_flow_flush(). */
1149 rule->installed = false;
1151 rule_remove(ofproto, rule);
1155 ofproto_flush_flows(struct ofproto *ofproto)
1157 COVERAGE_INC(ofproto_flush);
1158 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1159 dpif_flow_flush(ofproto->dpif);
1160 if (ofproto->in_band) {
1161 in_band_flushed(ofproto->in_band);
1163 if (ofproto->fail_open) {
1164 fail_open_flushed(ofproto->fail_open);
1169 reinit_ports(struct ofproto *p)
1171 struct svec devnames;
1172 struct ofport *ofport;
1173 unsigned int port_no;
1174 struct odp_port *odp_ports;
1178 svec_init(&devnames);
1179 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1180 svec_add (&devnames, (char *) ofport->opp.name);
1182 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1183 for (i = 0; i < n_odp_ports; i++) {
1184 svec_add (&devnames, odp_ports[i].devname);
1188 svec_sort_unique(&devnames);
1189 for (i = 0; i < devnames.n; i++) {
1190 update_port(p, devnames.names[i]);
1192 svec_destroy(&devnames);
1196 refresh_port_group(struct ofproto *p, unsigned int group)
1200 struct ofport *port;
1201 unsigned int port_no;
1203 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1205 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1207 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1208 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1209 ports[n_ports++] = port_no;
1212 dpif_port_group_set(p->dpif, group, ports, n_ports);
1219 refresh_port_groups(struct ofproto *p)
1221 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1222 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1224 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1228 static struct ofport *
1229 make_ofport(const struct odp_port *odp_port)
1231 struct netdev_options netdev_options;
1232 enum netdev_flags flags;
1233 struct ofport *ofport;
1234 struct netdev *netdev;
1238 memset(&netdev_options, 0, sizeof netdev_options);
1239 netdev_options.name = odp_port->devname;
1240 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1241 netdev_options.may_open = true;
1243 error = netdev_open(&netdev_options, &netdev);
1245 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1246 "cannot be opened (%s)",
1247 odp_port->devname, odp_port->port,
1248 odp_port->devname, strerror(error));
1252 ofport = xmalloc(sizeof *ofport);
1253 ofport->netdev = netdev;
1254 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1255 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1256 memcpy(ofport->opp.name, odp_port->devname,
1257 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1258 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1260 netdev_get_flags(netdev, &flags);
1261 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1263 netdev_get_carrier(netdev, &carrier);
1264 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1266 netdev_get_features(netdev,
1267 &ofport->opp.curr, &ofport->opp.advertised,
1268 &ofport->opp.supported, &ofport->opp.peer);
1273 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1275 if (port_array_get(&p->ports, odp_port->port)) {
1276 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1279 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1280 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1289 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1291 const struct ofp_phy_port *a = &a_->opp;
1292 const struct ofp_phy_port *b = &b_->opp;
1294 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1295 return (a->port_no == b->port_no
1296 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1297 && !strcmp((char *) a->name, (char *) b->name)
1298 && a->state == b->state
1299 && a->config == b->config
1300 && a->curr == b->curr
1301 && a->advertised == b->advertised
1302 && a->supported == b->supported
1303 && a->peer == b->peer);
1307 send_port_status(struct ofproto *p, const struct ofport *ofport,
1310 /* XXX Should limit the number of queued port status change messages. */
1311 struct ofconn *ofconn;
1312 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1313 struct ofp_port_status *ops;
1316 if (ofconn->role == NX_ROLE_SLAVE) {
1320 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1321 ops->reason = reason;
1322 ops->desc = ofport->opp;
1323 hton_ofp_phy_port(&ops->desc);
1324 queue_tx(b, ofconn, NULL);
1326 if (p->ofhooks->port_changed_cb) {
1327 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1332 ofport_install(struct ofproto *p, struct ofport *ofport)
1334 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1335 const char *netdev_name = (const char *) ofport->opp.name;
1337 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1338 port_array_set(&p->ports, odp_port, ofport);
1339 shash_add(&p->port_by_name, netdev_name, ofport);
1341 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1346 ofport_remove(struct ofproto *p, struct ofport *ofport)
1348 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1350 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1351 port_array_set(&p->ports, odp_port, NULL);
1352 shash_delete(&p->port_by_name,
1353 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1355 ofproto_sflow_del_port(p->sflow, odp_port);
1360 ofport_free(struct ofport *ofport)
1363 netdev_close(ofport->netdev);
1369 update_port(struct ofproto *p, const char *devname)
1371 struct odp_port odp_port;
1372 struct ofport *old_ofport;
1373 struct ofport *new_ofport;
1376 COVERAGE_INC(ofproto_update_port);
1378 /* Query the datapath for port information. */
1379 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1381 /* Find the old ofport. */
1382 old_ofport = shash_find_data(&p->port_by_name, devname);
1385 /* There's no port named 'devname' but there might be a port with
1386 * the same port number. This could happen if a port is deleted
1387 * and then a new one added in its place very quickly, or if a port
1388 * is renamed. In the former case we want to send an OFPPR_DELETE
1389 * and an OFPPR_ADD, and in the latter case we want to send a
1390 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1391 * the old port's ifindex against the new port, or perhaps less
1392 * reliably but more portably by comparing the old port's MAC
1393 * against the new port's MAC. However, this code isn't that smart
1394 * and always sends an OFPPR_MODIFY (XXX). */
1395 old_ofport = port_array_get(&p->ports, odp_port.port);
1397 } else if (error != ENOENT && error != ENODEV) {
1398 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1399 "%s", strerror(error));
1403 /* Create a new ofport. */
1404 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1406 /* Eliminate a few pathological cases. */
1407 if (!old_ofport && !new_ofport) {
1409 } else if (old_ofport && new_ofport) {
1410 /* Most of the 'config' bits are OpenFlow soft state, but
1411 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1412 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1413 * leaves the other bits 0.) */
1414 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1416 if (ofport_equal(old_ofport, new_ofport)) {
1417 /* False alarm--no change. */
1418 ofport_free(new_ofport);
1423 /* Now deal with the normal cases. */
1425 ofport_remove(p, old_ofport);
1428 ofport_install(p, new_ofport);
1430 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1431 (!old_ofport ? OFPPR_ADD
1432 : !new_ofport ? OFPPR_DELETE
1434 ofport_free(old_ofport);
1436 /* Update port groups. */
1437 refresh_port_groups(p);
1441 init_ports(struct ofproto *p)
1443 struct odp_port *ports;
1448 error = dpif_port_list(p->dpif, &ports, &n_ports);
1453 for (i = 0; i < n_ports; i++) {
1454 const struct odp_port *odp_port = &ports[i];
1455 if (!ofport_conflicts(p, odp_port)) {
1456 struct ofport *ofport = make_ofport(odp_port);
1458 ofport_install(p, ofport);
1463 refresh_port_groups(p);
1467 static struct ofconn *
1468 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1470 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1471 ofconn->ofproto = p;
1472 list_push_back(&p->all_conns, &ofconn->node);
1473 ofconn->rconn = rconn;
1474 ofconn->type = type;
1475 ofconn->role = NX_ROLE_OTHER;
1476 ofconn->packet_in_counter = rconn_packet_counter_create ();
1477 ofconn->pktbuf = NULL;
1478 ofconn->miss_send_len = 0;
1479 ofconn->reply_counter = rconn_packet_counter_create ();
1484 ofconn_destroy(struct ofconn *ofconn)
1486 if (ofconn->type == OFCONN_CONTROLLER) {
1487 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1489 discovery_destroy(ofconn->discovery);
1491 list_remove(&ofconn->node);
1492 switch_status_unregister(ofconn->ss);
1493 rconn_destroy(ofconn->rconn);
1494 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1495 rconn_packet_counter_destroy(ofconn->reply_counter);
1496 pktbuf_destroy(ofconn->pktbuf);
1501 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1506 if (ofconn->discovery) {
1507 char *controller_name;
1508 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1509 discovery_question_connectivity(ofconn->discovery);
1511 if (discovery_run(ofconn->discovery, &controller_name)) {
1512 if (controller_name) {
1513 rconn_connect(ofconn->rconn, controller_name);
1515 rconn_disconnect(ofconn->rconn);
1520 for (i = 0; i < N_SCHEDULERS; i++) {
1521 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1524 rconn_run(ofconn->rconn);
1526 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1527 /* Limit the number of iterations to prevent other tasks from
1529 for (iteration = 0; iteration < 50; iteration++) {
1530 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1535 fail_open_maybe_recover(p->fail_open);
1537 handle_openflow(ofconn, p, of_msg);
1538 ofpbuf_delete(of_msg);
1542 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1543 ofconn_destroy(ofconn);
1548 ofconn_wait(struct ofconn *ofconn)
1552 if (ofconn->discovery) {
1553 discovery_wait(ofconn->discovery);
1555 for (i = 0; i < N_SCHEDULERS; i++) {
1556 pinsched_wait(ofconn->schedulers[i]);
1558 rconn_run_wait(ofconn->rconn);
1559 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1560 rconn_recv_wait(ofconn->rconn);
1562 COVERAGE_INC(ofproto_ofconn_stuck);
1566 /* Caller is responsible for initializing the 'cr' member of the returned
1568 static struct rule *
1569 rule_create(struct ofproto *ofproto, struct rule *super,
1570 const union ofp_action *actions, size_t n_actions,
1571 uint16_t idle_timeout, uint16_t hard_timeout,
1572 uint64_t flow_cookie, bool send_flow_removed)
1574 struct rule *rule = xzalloc(sizeof *rule);
1575 rule->idle_timeout = idle_timeout;
1576 rule->hard_timeout = hard_timeout;
1577 rule->flow_cookie = flow_cookie;
1578 rule->used = rule->created = time_msec();
1579 rule->send_flow_removed = send_flow_removed;
1580 rule->super = super;
1582 list_push_back(&super->list, &rule->list);
1584 list_init(&rule->list);
1586 rule->n_actions = n_actions;
1587 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1588 netflow_flow_clear(&rule->nf_flow);
1589 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1594 static struct rule *
1595 rule_from_cls_rule(const struct cls_rule *cls_rule)
1597 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1601 rule_free(struct rule *rule)
1603 free(rule->actions);
1604 free(rule->odp_actions);
1608 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1609 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1610 * through all of its subrules and revalidates them, destroying any that no
1611 * longer has a super-rule (which is probably all of them).
1613 * Before calling this function, the caller must make have removed 'rule' from
1614 * the classifier. If 'rule' is an exact-match rule, the caller is also
1615 * responsible for ensuring that it has been uninstalled from the datapath. */
1617 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1620 struct rule *subrule, *next;
1621 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1622 revalidate_rule(ofproto, subrule);
1625 list_remove(&rule->list);
1631 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1633 const union ofp_action *oa;
1634 struct actions_iterator i;
1636 if (out_port == htons(OFPP_NONE)) {
1639 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1640 oa = actions_next(&i)) {
1641 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1648 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1649 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1651 * The flow that 'packet' actually contains does not need to actually match
1652 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1653 * the packet and byte counters for 'rule' will be credited for the packet sent
1654 * out whether or not the packet actually matches 'rule'.
1656 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1657 * the caller must already have accurately composed ODP actions for it given
1658 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1659 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1660 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1661 * actions and apply them to 'packet'. */
1663 rule_execute(struct ofproto *ofproto, struct rule *rule,
1664 struct ofpbuf *packet, const flow_t *flow)
1666 const union odp_action *actions;
1668 struct odp_actions a;
1670 /* Grab or compose the ODP actions.
1672 * The special case for an exact-match 'rule' where 'flow' is not the
1673 * rule's flow is important to avoid, e.g., sending a packet out its input
1674 * port simply because the ODP actions were composed for the wrong
1676 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1677 struct rule *super = rule->super ? rule->super : rule;
1678 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1679 packet, &a, NULL, 0, NULL)) {
1682 actions = a.actions;
1683 n_actions = a.n_actions;
1685 actions = rule->odp_actions;
1686 n_actions = rule->n_odp_actions;
1689 /* Execute the ODP actions. */
1690 if (!dpif_execute(ofproto->dpif, flow->in_port,
1691 actions, n_actions, packet)) {
1692 struct odp_flow_stats stats;
1693 flow_extract_stats(flow, packet, &stats);
1694 update_stats(ofproto, rule, &stats);
1695 rule->used = time_msec();
1696 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1701 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1704 struct rule *displaced_rule;
1706 /* Insert the rule in the classifier. */
1707 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1708 if (!rule->cr.wc.wildcards) {
1709 rule_make_actions(p, rule, packet);
1712 /* Send the packet and credit it to the rule. */
1715 flow_extract(packet, 0, in_port, &flow);
1716 rule_execute(p, rule, packet, &flow);
1719 /* Install the rule in the datapath only after sending the packet, to
1720 * avoid packet reordering. */
1721 if (rule->cr.wc.wildcards) {
1722 COVERAGE_INC(ofproto_add_wc_flow);
1723 p->need_revalidate = true;
1725 rule_install(p, rule, displaced_rule);
1728 /* Free the rule that was displaced, if any. */
1729 if (displaced_rule) {
1730 rule_destroy(p, displaced_rule);
1734 static struct rule *
1735 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1738 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1739 rule->idle_timeout, rule->hard_timeout,
1741 COVERAGE_INC(ofproto_subrule_create);
1742 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1743 : rule->cr.priority), &subrule->cr);
1744 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1750 rule_remove(struct ofproto *ofproto, struct rule *rule)
1752 if (rule->cr.wc.wildcards) {
1753 COVERAGE_INC(ofproto_del_wc_flow);
1754 ofproto->need_revalidate = true;
1756 rule_uninstall(ofproto, rule);
1758 classifier_remove(&ofproto->cls, &rule->cr);
1759 rule_destroy(ofproto, rule);
1762 /* Returns true if the actions changed, false otherwise. */
1764 rule_make_actions(struct ofproto *p, struct rule *rule,
1765 const struct ofpbuf *packet)
1767 const struct rule *super;
1768 struct odp_actions a;
1771 assert(!rule->cr.wc.wildcards);
1773 super = rule->super ? rule->super : rule;
1775 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1776 packet, &a, &rule->tags, &rule->may_install,
1777 &rule->nf_flow.output_iface);
1779 actions_len = a.n_actions * sizeof *a.actions;
1780 if (rule->n_odp_actions != a.n_actions
1781 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1782 COVERAGE_INC(ofproto_odp_unchanged);
1783 free(rule->odp_actions);
1784 rule->n_odp_actions = a.n_actions;
1785 rule->odp_actions = xmemdup(a.actions, actions_len);
1793 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1794 struct odp_flow_put *put)
1796 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1797 put->flow.key = rule->cr.flow;
1798 put->flow.actions = rule->odp_actions;
1799 put->flow.n_actions = rule->n_odp_actions;
1800 put->flow.flags = 0;
1802 return dpif_flow_put(ofproto->dpif, put);
1806 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1808 assert(!rule->cr.wc.wildcards);
1810 if (rule->may_install) {
1811 struct odp_flow_put put;
1812 if (!do_put_flow(p, rule,
1813 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1815 rule->installed = true;
1816 if (displaced_rule) {
1817 update_stats(p, displaced_rule, &put.flow.stats);
1818 rule_post_uninstall(p, displaced_rule);
1821 } else if (displaced_rule) {
1822 rule_uninstall(p, displaced_rule);
1827 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1829 if (rule->installed) {
1830 struct odp_flow_put put;
1831 COVERAGE_INC(ofproto_dp_missed);
1832 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1834 rule_install(ofproto, rule, NULL);
1839 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1841 bool actions_changed;
1842 uint16_t new_out_iface, old_out_iface;
1844 old_out_iface = rule->nf_flow.output_iface;
1845 actions_changed = rule_make_actions(ofproto, rule, NULL);
1847 if (rule->may_install) {
1848 if (rule->installed) {
1849 if (actions_changed) {
1850 struct odp_flow_put put;
1851 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1852 | ODPPF_ZERO_STATS, &put);
1853 update_stats(ofproto, rule, &put.flow.stats);
1855 /* Temporarily set the old output iface so that NetFlow
1856 * messages have the correct output interface for the old
1858 new_out_iface = rule->nf_flow.output_iface;
1859 rule->nf_flow.output_iface = old_out_iface;
1860 rule_post_uninstall(ofproto, rule);
1861 rule->nf_flow.output_iface = new_out_iface;
1864 rule_install(ofproto, rule, NULL);
1867 rule_uninstall(ofproto, rule);
1872 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1874 uint64_t total_bytes = rule->byte_count + extra_bytes;
1876 if (ofproto->ofhooks->account_flow_cb
1877 && total_bytes > rule->accounted_bytes)
1879 ofproto->ofhooks->account_flow_cb(
1880 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1881 total_bytes - rule->accounted_bytes, ofproto->aux);
1882 rule->accounted_bytes = total_bytes;
1887 rule_uninstall(struct ofproto *p, struct rule *rule)
1889 assert(!rule->cr.wc.wildcards);
1890 if (rule->installed) {
1891 struct odp_flow odp_flow;
1893 odp_flow.key = rule->cr.flow;
1894 odp_flow.actions = NULL;
1895 odp_flow.n_actions = 0;
1897 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1898 update_stats(p, rule, &odp_flow.stats);
1900 rule->installed = false;
1902 rule_post_uninstall(p, rule);
1907 is_controller_rule(struct rule *rule)
1909 /* If the only action is send to the controller then don't report
1910 * NetFlow expiration messages since it is just part of the control
1911 * logic for the network and not real traffic. */
1913 if (rule && rule->super) {
1914 struct rule *super = rule->super;
1916 return super->n_actions == 1 &&
1917 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1918 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1925 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1927 struct rule *super = rule->super;
1929 rule_account(ofproto, rule, 0);
1931 if (ofproto->netflow && !is_controller_rule(rule)) {
1932 struct ofexpired expired;
1933 expired.flow = rule->cr.flow;
1934 expired.packet_count = rule->packet_count;
1935 expired.byte_count = rule->byte_count;
1936 expired.used = rule->used;
1937 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1940 super->packet_count += rule->packet_count;
1941 super->byte_count += rule->byte_count;
1943 /* Reset counters to prevent double counting if the rule ever gets
1945 rule->packet_count = 0;
1946 rule->byte_count = 0;
1947 rule->accounted_bytes = 0;
1949 netflow_flow_clear(&rule->nf_flow);
1954 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1955 struct rconn_packet_counter *counter)
1957 update_openflow_length(msg);
1958 if (rconn_send(ofconn->rconn, msg, counter)) {
1964 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1965 int error, const void *data, size_t len)
1968 struct ofp_error_msg *oem;
1970 if (!(error >> 16)) {
1971 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1976 COVERAGE_INC(ofproto_error);
1977 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1978 oh ? oh->xid : 0, &buf);
1979 oem->type = htons((unsigned int) error >> 16);
1980 oem->code = htons(error & 0xffff);
1981 memcpy(oem->data, data, len);
1982 queue_tx(buf, ofconn, ofconn->reply_counter);
1986 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1989 size_t oh_length = ntohs(oh->length);
1990 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1994 hton_ofp_phy_port(struct ofp_phy_port *opp)
1996 opp->port_no = htons(opp->port_no);
1997 opp->config = htonl(opp->config);
1998 opp->state = htonl(opp->state);
1999 opp->curr = htonl(opp->curr);
2000 opp->advertised = htonl(opp->advertised);
2001 opp->supported = htonl(opp->supported);
2002 opp->peer = htonl(opp->peer);
2006 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2008 struct ofp_header *rq = oh;
2009 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2014 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2015 struct ofp_header *oh)
2017 struct ofp_switch_features *osf;
2019 unsigned int port_no;
2020 struct ofport *port;
2022 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2023 osf->datapath_id = htonll(p->datapath_id);
2024 osf->n_buffers = htonl(pktbuf_capacity());
2026 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2027 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2028 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2029 (1u << OFPAT_SET_VLAN_VID) |
2030 (1u << OFPAT_SET_VLAN_PCP) |
2031 (1u << OFPAT_STRIP_VLAN) |
2032 (1u << OFPAT_SET_DL_SRC) |
2033 (1u << OFPAT_SET_DL_DST) |
2034 (1u << OFPAT_SET_NW_SRC) |
2035 (1u << OFPAT_SET_NW_DST) |
2036 (1u << OFPAT_SET_NW_TOS) |
2037 (1u << OFPAT_SET_TP_SRC) |
2038 (1u << OFPAT_SET_TP_DST));
2040 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2041 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2044 queue_tx(buf, ofconn, ofconn->reply_counter);
2049 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2050 struct ofp_header *oh)
2053 struct ofp_switch_config *osc;
2057 /* Figure out flags. */
2058 dpif_get_drop_frags(p->dpif, &drop_frags);
2059 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2062 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2063 osc->flags = htons(flags);
2064 osc->miss_send_len = htons(ofconn->miss_send_len);
2065 queue_tx(buf, ofconn, ofconn->reply_counter);
2071 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2072 struct ofp_switch_config *osc)
2077 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2081 flags = ntohs(osc->flags);
2083 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2084 switch (flags & OFPC_FRAG_MASK) {
2085 case OFPC_FRAG_NORMAL:
2086 dpif_set_drop_frags(p->dpif, false);
2088 case OFPC_FRAG_DROP:
2089 dpif_set_drop_frags(p->dpif, true);
2092 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2098 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2104 add_output_group_action(struct odp_actions *actions, uint16_t group,
2105 uint16_t *nf_output_iface)
2107 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2109 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2110 *nf_output_iface = NF_OUT_FLOOD;
2115 add_controller_action(struct odp_actions *actions,
2116 const struct ofp_action_output *oao)
2118 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2119 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
2122 struct action_xlate_ctx {
2124 flow_t flow; /* Flow to which these actions correspond. */
2125 int recurse; /* Recursion level, via xlate_table_action. */
2126 struct ofproto *ofproto;
2127 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2128 * null pointer if we are revalidating
2129 * without a packet to refer to. */
2132 struct odp_actions *out; /* Datapath actions. */
2133 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2134 bool may_set_up_flow; /* True ordinarily; false if the actions must
2135 * be reassessed for every packet. */
2136 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2139 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2140 struct action_xlate_ctx *ctx);
2143 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2145 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2148 if (ofport->opp.config & OFPPC_NO_FWD) {
2149 /* Forwarding disabled on port. */
2154 * We don't have an ofport record for this port, but it doesn't hurt to
2155 * allow forwarding to it anyhow. Maybe such a port will appear later
2156 * and we're pre-populating the flow table.
2160 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2161 ctx->nf_output_iface = port;
2164 static struct rule *
2165 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2168 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2170 /* The rule we found might not be valid, since we could be in need of
2171 * revalidation. If it is not valid, don't return it. */
2174 && ofproto->need_revalidate
2175 && !revalidate_rule(ofproto, rule)) {
2176 COVERAGE_INC(ofproto_invalidated);
2184 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2186 if (!ctx->recurse) {
2187 uint16_t old_in_port;
2190 /* Look up a flow with 'in_port' as the input port. Then restore the
2191 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2192 * have surprising behavior). */
2193 old_in_port = ctx->flow.in_port;
2194 ctx->flow.in_port = in_port;
2195 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2196 ctx->flow.in_port = old_in_port;
2204 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2211 xlate_output_action(struct action_xlate_ctx *ctx,
2212 const struct ofp_action_output *oao)
2215 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2217 ctx->nf_output_iface = NF_OUT_DROP;
2219 switch (ntohs(oao->port)) {
2221 add_output_action(ctx, ctx->flow.in_port);
2224 xlate_table_action(ctx, ctx->flow.in_port);
2227 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2228 ctx->out, ctx->tags,
2229 &ctx->nf_output_iface,
2230 ctx->ofproto->aux)) {
2231 COVERAGE_INC(ofproto_uninstallable);
2232 ctx->may_set_up_flow = false;
2236 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2237 &ctx->nf_output_iface);
2240 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2242 case OFPP_CONTROLLER:
2243 add_controller_action(ctx->out, oao);
2246 add_output_action(ctx, ODPP_LOCAL);
2249 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2250 if (odp_port != ctx->flow.in_port) {
2251 add_output_action(ctx, odp_port);
2256 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2257 ctx->nf_output_iface = NF_OUT_FLOOD;
2258 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2259 ctx->nf_output_iface = prev_nf_output_iface;
2260 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2261 ctx->nf_output_iface != NF_OUT_FLOOD) {
2262 ctx->nf_output_iface = NF_OUT_MULTI;
2267 xlate_nicira_action(struct action_xlate_ctx *ctx,
2268 const struct nx_action_header *nah)
2270 const struct nx_action_resubmit *nar;
2271 const struct nx_action_set_tunnel *nast;
2272 union odp_action *oa;
2273 int subtype = ntohs(nah->subtype);
2275 assert(nah->vendor == htonl(NX_VENDOR_ID));
2277 case NXAST_RESUBMIT:
2278 nar = (const struct nx_action_resubmit *) nah;
2279 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2282 case NXAST_SET_TUNNEL:
2283 nast = (const struct nx_action_set_tunnel *) nah;
2284 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2285 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2288 /* If you add a new action here that modifies flow data, don't forget to
2289 * update the flow key in ctx->flow in the same key. */
2292 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2298 do_xlate_actions(const union ofp_action *in, size_t n_in,
2299 struct action_xlate_ctx *ctx)
2301 struct actions_iterator iter;
2302 const union ofp_action *ia;
2303 const struct ofport *port;
2305 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2306 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2307 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2308 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2309 /* Drop this flow. */
2313 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2314 uint16_t type = ntohs(ia->type);
2315 union odp_action *oa;
2319 xlate_output_action(ctx, &ia->output);
2322 case OFPAT_SET_VLAN_VID:
2323 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2324 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2327 case OFPAT_SET_VLAN_PCP:
2328 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2329 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2332 case OFPAT_STRIP_VLAN:
2333 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2334 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2335 ctx->flow.dl_vlan_pcp = 0;
2338 case OFPAT_SET_DL_SRC:
2339 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2340 memcpy(oa->dl_addr.dl_addr,
2341 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2342 memcpy(ctx->flow.dl_src,
2343 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2346 case OFPAT_SET_DL_DST:
2347 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2348 memcpy(oa->dl_addr.dl_addr,
2349 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2350 memcpy(ctx->flow.dl_dst,
2351 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2354 case OFPAT_SET_NW_SRC:
2355 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2356 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2359 case OFPAT_SET_NW_DST:
2360 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2361 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2364 case OFPAT_SET_NW_TOS:
2365 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2366 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2369 case OFPAT_SET_TP_SRC:
2370 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2371 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2374 case OFPAT_SET_TP_DST:
2375 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2376 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2380 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2384 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2391 xlate_actions(const union ofp_action *in, size_t n_in,
2392 const flow_t *flow, struct ofproto *ofproto,
2393 const struct ofpbuf *packet,
2394 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2395 uint16_t *nf_output_iface)
2397 tag_type no_tags = 0;
2398 struct action_xlate_ctx ctx;
2399 COVERAGE_INC(ofproto_ofp2odp);
2400 odp_actions_init(out);
2403 ctx.ofproto = ofproto;
2404 ctx.packet = packet;
2406 ctx.tags = tags ? tags : &no_tags;
2407 ctx.may_set_up_flow = true;
2408 ctx.nf_output_iface = NF_OUT_DROP;
2409 do_xlate_actions(in, n_in, &ctx);
2411 /* Check with in-band control to see if we're allowed to set up this
2413 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2414 ctx.may_set_up_flow = false;
2417 if (may_set_up_flow) {
2418 *may_set_up_flow = ctx.may_set_up_flow;
2420 if (nf_output_iface) {
2421 *nf_output_iface = ctx.nf_output_iface;
2423 if (odp_actions_overflow(out)) {
2424 odp_actions_init(out);
2425 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2430 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2431 * error message code (composed with ofp_mkerr()) for the caller to propagate
2432 * upward. Otherwise, returns 0.
2434 * 'oh' is used to make log messages more informative. */
2436 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2438 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2439 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2442 type_name = ofp_message_type_to_string(oh->type);
2443 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2447 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2454 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2455 struct ofp_header *oh)
2457 struct ofp_packet_out *opo;
2458 struct ofpbuf payload, *buffer;
2459 struct odp_actions actions;
2465 error = reject_slave_controller(ofconn, oh);
2470 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2474 opo = (struct ofp_packet_out *) oh;
2476 COVERAGE_INC(ofproto_packet_out);
2477 if (opo->buffer_id != htonl(UINT32_MAX)) {
2478 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2480 if (error || !buffer) {
2488 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2489 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2490 &flow, p, &payload, &actions, NULL, NULL, NULL);
2495 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2497 ofpbuf_delete(buffer);
2503 update_port_config(struct ofproto *p, struct ofport *port,
2504 uint32_t config, uint32_t mask)
2506 mask &= config ^ port->opp.config;
2507 if (mask & OFPPC_PORT_DOWN) {
2508 if (config & OFPPC_PORT_DOWN) {
2509 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2511 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2514 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2515 if (mask & REVALIDATE_BITS) {
2516 COVERAGE_INC(ofproto_costly_flags);
2517 port->opp.config ^= mask & REVALIDATE_BITS;
2518 p->need_revalidate = true;
2520 #undef REVALIDATE_BITS
2521 if (mask & OFPPC_NO_FLOOD) {
2522 port->opp.config ^= OFPPC_NO_FLOOD;
2523 refresh_port_groups(p);
2525 if (mask & OFPPC_NO_PACKET_IN) {
2526 port->opp.config ^= OFPPC_NO_PACKET_IN;
2531 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2532 struct ofp_header *oh)
2534 const struct ofp_port_mod *opm;
2535 struct ofport *port;
2538 error = reject_slave_controller(ofconn, oh);
2542 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2546 opm = (struct ofp_port_mod *) oh;
2548 port = port_array_get(&p->ports,
2549 ofp_port_to_odp_port(ntohs(opm->port_no)));
2551 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2552 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2553 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2555 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2556 if (opm->advertise) {
2557 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2563 static struct ofpbuf *
2564 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2566 struct ofp_stats_reply *osr;
2569 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2570 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2572 osr->flags = htons(0);
2576 static struct ofpbuf *
2577 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2579 return make_stats_reply(request->header.xid, request->type, body_len);
2583 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2585 struct ofpbuf *msg = *msgp;
2586 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2587 if (nbytes + msg->size > UINT16_MAX) {
2588 struct ofp_stats_reply *reply = msg->data;
2589 reply->flags = htons(OFPSF_REPLY_MORE);
2590 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2591 queue_tx(msg, ofconn, ofconn->reply_counter);
2593 return ofpbuf_put_uninit(*msgp, nbytes);
2597 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2598 struct ofp_stats_request *request)
2600 struct ofp_desc_stats *ods;
2603 msg = start_stats_reply(request, sizeof *ods);
2604 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2605 memset(ods, 0, sizeof *ods);
2606 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2607 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2608 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2609 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2610 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2611 queue_tx(msg, ofconn, ofconn->reply_counter);
2617 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2619 struct rule *rule = rule_from_cls_rule(cls_rule);
2620 int *n_subrules = n_subrules_;
2628 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2629 struct ofp_stats_request *request)
2631 struct ofp_table_stats *ots;
2633 struct odp_stats dpstats;
2634 int n_exact, n_subrules, n_wild;
2636 msg = start_stats_reply(request, sizeof *ots * 2);
2638 /* Count rules of various kinds. */
2640 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2641 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2642 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2645 dpif_get_dp_stats(p->dpif, &dpstats);
2646 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2647 memset(ots, 0, sizeof *ots);
2648 ots->table_id = TABLEID_HASH;
2649 strcpy(ots->name, "hash");
2650 ots->wildcards = htonl(0);
2651 ots->max_entries = htonl(dpstats.max_capacity);
2652 ots->active_count = htonl(n_exact);
2653 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2655 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2657 /* Classifier table. */
2658 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2659 memset(ots, 0, sizeof *ots);
2660 ots->table_id = TABLEID_CLASSIFIER;
2661 strcpy(ots->name, "classifier");
2662 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2664 ots->max_entries = htonl(65536);
2665 ots->active_count = htonl(n_wild);
2666 ots->lookup_count = htonll(0); /* XXX */
2667 ots->matched_count = htonll(0); /* XXX */
2669 queue_tx(msg, ofconn, ofconn->reply_counter);
2674 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2677 struct netdev_stats stats;
2678 struct ofp_port_stats *ops;
2680 /* Intentionally ignore return value, since errors will set
2681 * 'stats' to all-1s, which is correct for OpenFlow, and
2682 * netdev_get_stats() will log errors. */
2683 netdev_get_stats(port->netdev, &stats);
2685 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2686 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2687 memset(ops->pad, 0, sizeof ops->pad);
2688 ops->rx_packets = htonll(stats.rx_packets);
2689 ops->tx_packets = htonll(stats.tx_packets);
2690 ops->rx_bytes = htonll(stats.rx_bytes);
2691 ops->tx_bytes = htonll(stats.tx_bytes);
2692 ops->rx_dropped = htonll(stats.rx_dropped);
2693 ops->tx_dropped = htonll(stats.tx_dropped);
2694 ops->rx_errors = htonll(stats.rx_errors);
2695 ops->tx_errors = htonll(stats.tx_errors);
2696 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2697 ops->rx_over_err = htonll(stats.rx_over_errors);
2698 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2699 ops->collisions = htonll(stats.collisions);
2703 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2704 struct ofp_stats_request *osr,
2707 struct ofp_port_stats_request *psr;
2708 struct ofp_port_stats *ops;
2710 struct ofport *port;
2711 unsigned int port_no;
2713 if (arg_size != sizeof *psr) {
2714 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2716 psr = (struct ofp_port_stats_request *) osr->body;
2718 msg = start_stats_reply(osr, sizeof *ops * 16);
2719 if (psr->port_no != htons(OFPP_NONE)) {
2720 port = port_array_get(&p->ports,
2721 ofp_port_to_odp_port(ntohs(psr->port_no)));
2723 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2726 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2727 append_port_stat(port, port_no, ofconn, msg);
2731 queue_tx(msg, ofconn, ofconn->reply_counter);
2735 struct flow_stats_cbdata {
2736 struct ofproto *ofproto;
2737 struct ofconn *ofconn;
2742 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2743 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2744 * returned statistic include statistics for all of 'rule''s subrules. */
2746 query_stats(struct ofproto *p, struct rule *rule,
2747 uint64_t *packet_countp, uint64_t *byte_countp)
2749 uint64_t packet_count, byte_count;
2750 struct rule *subrule;
2751 struct odp_flow *odp_flows;
2754 /* Start from historical data for 'rule' itself that are no longer tracked
2755 * by the datapath. This counts, for example, subrules that have
2757 packet_count = rule->packet_count;
2758 byte_count = rule->byte_count;
2760 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2761 * wildcarded then on all of its subrules.
2763 * Also, add any statistics that are not tracked by the datapath for each
2764 * subrule. This includes, for example, statistics for packets that were
2765 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2767 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2768 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2769 if (rule->cr.wc.wildcards) {
2771 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2772 odp_flows[i++].key = subrule->cr.flow;
2773 packet_count += subrule->packet_count;
2774 byte_count += subrule->byte_count;
2777 odp_flows[0].key = rule->cr.flow;
2780 /* Fetch up-to-date statistics from the datapath and add them in. */
2781 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2783 for (i = 0; i < n_odp_flows; i++) {
2784 struct odp_flow *odp_flow = &odp_flows[i];
2785 packet_count += odp_flow->stats.n_packets;
2786 byte_count += odp_flow->stats.n_bytes;
2791 /* Return the stats to the caller. */
2792 *packet_countp = packet_count;
2793 *byte_countp = byte_count;
2797 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2799 struct rule *rule = rule_from_cls_rule(rule_);
2800 struct flow_stats_cbdata *cbdata = cbdata_;
2801 struct ofp_flow_stats *ofs;
2802 uint64_t packet_count, byte_count;
2803 size_t act_len, len;
2804 long long int tdiff = time_msec() - rule->created;
2805 uint32_t sec = tdiff / 1000;
2806 uint32_t msec = tdiff - (sec * 1000);
2808 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2812 act_len = sizeof *rule->actions * rule->n_actions;
2813 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2815 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2817 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2818 ofs->length = htons(len);
2819 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2821 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2822 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2823 ofs->duration_sec = htonl(sec);
2824 ofs->duration_nsec = htonl(msec * 1000000);
2825 ofs->cookie = rule->flow_cookie;
2826 ofs->priority = htons(rule->cr.priority);
2827 ofs->idle_timeout = htons(rule->idle_timeout);
2828 ofs->hard_timeout = htons(rule->hard_timeout);
2829 memset(ofs->pad2, 0, sizeof ofs->pad2);
2830 ofs->packet_count = htonll(packet_count);
2831 ofs->byte_count = htonll(byte_count);
2832 memcpy(ofs->actions, rule->actions, act_len);
2836 table_id_to_include(uint8_t table_id)
2838 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2839 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2840 : table_id == 0xff ? CLS_INC_ALL
2845 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2846 const struct ofp_stats_request *osr,
2849 struct ofp_flow_stats_request *fsr;
2850 struct flow_stats_cbdata cbdata;
2851 struct cls_rule target;
2853 if (arg_size != sizeof *fsr) {
2854 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2856 fsr = (struct ofp_flow_stats_request *) osr->body;
2858 COVERAGE_INC(ofproto_flows_req);
2860 cbdata.ofconn = ofconn;
2861 cbdata.out_port = fsr->out_port;
2862 cbdata.msg = start_stats_reply(osr, 1024);
2863 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
2864 classifier_for_each_match(&p->cls, &target,
2865 table_id_to_include(fsr->table_id),
2866 flow_stats_cb, &cbdata);
2867 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2871 struct flow_stats_ds_cbdata {
2872 struct ofproto *ofproto;
2877 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2879 struct rule *rule = rule_from_cls_rule(rule_);
2880 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2881 struct ds *results = cbdata->results;
2882 struct ofp_match match;
2883 uint64_t packet_count, byte_count;
2884 size_t act_len = sizeof *rule->actions * rule->n_actions;
2886 /* Don't report on subrules. */
2887 if (rule->super != NULL) {
2891 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2892 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2893 cbdata->ofproto->tun_id_from_cookie, &match);
2895 ds_put_format(results, "duration=%llds, ",
2896 (time_msec() - rule->created) / 1000);
2897 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2898 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2899 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2900 ofp_print_match(results, &match, true);
2901 ofp_print_actions(results, &rule->actions->header, act_len);
2902 ds_put_cstr(results, "\n");
2905 /* Adds a pretty-printed description of all flows to 'results', including
2906 * those marked hidden by secchan (e.g., by in-band control). */
2908 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2910 struct ofp_match match;
2911 struct cls_rule target;
2912 struct flow_stats_ds_cbdata cbdata;
2914 memset(&match, 0, sizeof match);
2915 match.wildcards = htonl(OVSFW_ALL);
2918 cbdata.results = results;
2920 cls_rule_from_match(&match, 0, false, 0, &target);
2921 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2922 flow_stats_ds_cb, &cbdata);
2925 struct aggregate_stats_cbdata {
2926 struct ofproto *ofproto;
2928 uint64_t packet_count;
2929 uint64_t byte_count;
2934 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2936 struct rule *rule = rule_from_cls_rule(rule_);
2937 struct aggregate_stats_cbdata *cbdata = cbdata_;
2938 uint64_t packet_count, byte_count;
2940 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2944 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2946 cbdata->packet_count += packet_count;
2947 cbdata->byte_count += byte_count;
2952 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2953 const struct ofp_stats_request *osr,
2956 struct ofp_aggregate_stats_request *asr;
2957 struct ofp_aggregate_stats_reply *reply;
2958 struct aggregate_stats_cbdata cbdata;
2959 struct cls_rule target;
2962 if (arg_size != sizeof *asr) {
2963 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2965 asr = (struct ofp_aggregate_stats_request *) osr->body;
2967 COVERAGE_INC(ofproto_agg_request);
2969 cbdata.out_port = asr->out_port;
2970 cbdata.packet_count = 0;
2971 cbdata.byte_count = 0;
2973 cls_rule_from_match(&asr->match, 0, false, 0, &target);
2974 classifier_for_each_match(&p->cls, &target,
2975 table_id_to_include(asr->table_id),
2976 aggregate_stats_cb, &cbdata);
2978 msg = start_stats_reply(osr, sizeof *reply);
2979 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2980 reply->flow_count = htonl(cbdata.n_flows);
2981 reply->packet_count = htonll(cbdata.packet_count);
2982 reply->byte_count = htonll(cbdata.byte_count);
2983 queue_tx(msg, ofconn, ofconn->reply_counter);
2988 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2989 struct ofp_header *oh)
2991 struct ofp_stats_request *osr;
2995 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3000 osr = (struct ofp_stats_request *) oh;
3002 switch (ntohs(osr->type)) {
3004 return handle_desc_stats_request(p, ofconn, osr);
3007 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3009 case OFPST_AGGREGATE:
3010 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3013 return handle_table_stats_request(p, ofconn, osr);
3016 return handle_port_stats_request(p, ofconn, osr, arg_size);
3019 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3022 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3026 static long long int
3027 msec_from_nsec(uint64_t sec, uint32_t nsec)
3029 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3033 update_time(struct ofproto *ofproto, struct rule *rule,
3034 const struct odp_flow_stats *stats)
3036 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3037 if (used > rule->used) {
3039 if (rule->super && used > rule->super->used) {
3040 rule->super->used = used;
3042 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3047 update_stats(struct ofproto *ofproto, struct rule *rule,
3048 const struct odp_flow_stats *stats)
3050 if (stats->n_packets) {
3051 update_time(ofproto, rule, stats);
3052 rule->packet_count += stats->n_packets;
3053 rule->byte_count += stats->n_bytes;
3054 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3059 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3060 * in which no matching flow already exists in the flow table.
3062 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3063 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3064 * code as encoded by ofp_mkerr() on failure.
3066 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3069 add_flow(struct ofproto *p, struct ofconn *ofconn,
3070 const struct ofp_flow_mod *ofm, size_t n_actions)
3072 struct ofpbuf *packet;
3077 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3081 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3083 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3084 ntohs(ofm->priority))) {
3085 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3089 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3090 n_actions, ntohs(ofm->idle_timeout),
3091 ntohs(ofm->hard_timeout), ofm->cookie,
3092 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3093 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3094 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3097 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3098 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3102 in_port = UINT16_MAX;
3105 rule_insert(p, rule, packet, in_port);
3106 ofpbuf_delete(packet);
3110 static struct rule *
3111 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3116 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3118 return rule_from_cls_rule(classifier_find_rule_exactly(
3119 &p->cls, &flow, wildcards,
3120 ntohs(ofm->priority)));
3124 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3125 struct rule *rule, const struct ofp_flow_mod *ofm)
3127 struct ofpbuf *packet;
3132 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3136 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3142 flow_extract(packet, 0, in_port, &flow);
3143 rule_execute(ofproto, rule, packet, &flow);
3144 ofpbuf_delete(packet);
3149 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3151 struct modify_flows_cbdata {
3152 struct ofproto *ofproto;
3153 const struct ofp_flow_mod *ofm;
3158 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3159 size_t n_actions, struct rule *);
3160 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3162 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3163 * encoded by ofp_mkerr() on failure.
3165 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3168 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3169 const struct ofp_flow_mod *ofm, size_t n_actions)
3171 struct modify_flows_cbdata cbdata;
3172 struct cls_rule target;
3176 cbdata.n_actions = n_actions;
3177 cbdata.match = NULL;
3179 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3182 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3183 modify_flows_cb, &cbdata);
3185 /* This credits the packet to whichever flow happened to happened to
3186 * match last. That's weird. Maybe we should do a lookup for the
3187 * flow that actually matches the packet? Who knows. */
3188 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3191 return add_flow(p, ofconn, ofm, n_actions);
3195 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3196 * code as encoded by ofp_mkerr() on failure.
3198 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3201 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3202 struct ofp_flow_mod *ofm, size_t n_actions)
3204 struct rule *rule = find_flow_strict(p, ofm);
3205 if (rule && !rule_is_hidden(rule)) {
3206 modify_flow(p, ofm, n_actions, rule);
3207 return send_buffered_packet(p, ofconn, rule, ofm);
3209 return add_flow(p, ofconn, ofm, n_actions);
3213 /* Callback for modify_flows_loose(). */
3215 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3217 struct rule *rule = rule_from_cls_rule(rule_);
3218 struct modify_flows_cbdata *cbdata = cbdata_;
3220 if (!rule_is_hidden(rule)) {
3221 cbdata->match = rule;
3222 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3226 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3227 * been identified as a flow in 'p''s flow table to be modified, by changing
3228 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3229 * ofp_action[] structures). */
3231 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3232 size_t n_actions, struct rule *rule)
3234 size_t actions_len = n_actions * sizeof *rule->actions;
3236 rule->flow_cookie = ofm->cookie;
3238 /* If the actions are the same, do nothing. */
3239 if (n_actions == rule->n_actions
3240 && !memcmp(ofm->actions, rule->actions, actions_len))
3245 /* Replace actions. */
3246 free(rule->actions);
3247 rule->actions = xmemdup(ofm->actions, actions_len);
3248 rule->n_actions = n_actions;
3250 /* Make sure that the datapath gets updated properly. */
3251 if (rule->cr.wc.wildcards) {
3252 COVERAGE_INC(ofproto_mod_wc_flow);
3253 p->need_revalidate = true;
3255 rule_update_actions(p, rule);
3261 /* OFPFC_DELETE implementation. */
3263 struct delete_flows_cbdata {
3264 struct ofproto *ofproto;
3268 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3269 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3271 /* Implements OFPFC_DELETE. */
3273 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3275 struct delete_flows_cbdata cbdata;
3276 struct cls_rule target;
3279 cbdata.out_port = ofm->out_port;
3281 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3284 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3285 delete_flows_cb, &cbdata);
3288 /* Implements OFPFC_DELETE_STRICT. */
3290 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3292 struct rule *rule = find_flow_strict(p, ofm);
3294 delete_flow(p, rule, ofm->out_port);
3298 /* Callback for delete_flows_loose(). */
3300 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3302 struct rule *rule = rule_from_cls_rule(rule_);
3303 struct delete_flows_cbdata *cbdata = cbdata_;
3305 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3308 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3309 * been identified as a flow to delete from 'p''s flow table, by deleting the
3310 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3313 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3314 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3315 * specified 'out_port'. */
3317 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3319 if (rule_is_hidden(rule)) {
3323 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3327 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3328 rule_remove(p, rule);
3332 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3333 struct ofp_flow_mod *ofm)
3338 error = reject_slave_controller(ofconn, &ofm->header);
3342 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3343 sizeof *ofm->actions, &n_actions);
3348 /* We do not support the emergency flow cache. It will hopefully
3349 * get dropped from OpenFlow in the near future. */
3350 if (ofm->flags & htons(OFPFF_EMERG)) {
3351 /* There isn't a good fit for an error code, so just state that the
3352 * flow table is full. */
3353 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3356 normalize_match(&ofm->match);
3357 if (!ofm->match.wildcards) {
3358 ofm->priority = htons(UINT16_MAX);
3361 error = validate_actions((const union ofp_action *) ofm->actions,
3362 n_actions, p->max_ports);
3367 switch (ntohs(ofm->command)) {
3369 return add_flow(p, ofconn, ofm, n_actions);
3372 return modify_flows_loose(p, ofconn, ofm, n_actions);
3374 case OFPFC_MODIFY_STRICT:
3375 return modify_flow_strict(p, ofconn, ofm, n_actions);
3378 delete_flows_loose(p, ofm);
3381 case OFPFC_DELETE_STRICT:
3382 delete_flow_strict(p, ofm);
3386 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3391 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3395 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3400 p->tun_id_from_cookie = !!msg->set;
3405 handle_role_request(struct ofproto *ofproto,
3406 struct ofconn *ofconn, struct nicira_header *msg)
3408 struct nx_role_request *nrr;
3409 struct nx_role_request *reply;
3413 if (ntohs(msg->header.length) != sizeof *nrr) {
3414 VLOG_WARN_RL(&rl, "received role request of length %zu (expected %zu)",
3415 ntohs(msg->header.length), sizeof *nrr);
3416 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3418 nrr = (struct nx_role_request *) msg;
3420 if (ofconn->type != OFCONN_CONTROLLER) {
3421 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3423 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3426 role = ntohl(nrr->role);
3427 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3428 && role != NX_ROLE_SLAVE) {
3429 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3431 /* There's no good error code for this. */
3432 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3435 if (role == NX_ROLE_MASTER) {
3436 struct ofconn *other;
3438 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3439 &ofproto->controllers) {
3440 if (other->role == NX_ROLE_MASTER) {
3441 other->role = NX_ROLE_SLAVE;
3445 ofconn->role = role;
3447 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3449 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3450 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3451 reply->role = htonl(role);
3452 queue_tx(buf, ofconn, ofconn->reply_counter);
3458 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3460 struct ofp_vendor_header *ovh = msg;
3461 struct nicira_header *nh;
3463 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3464 VLOG_WARN_RL(&rl, "received vendor message of length %zu "
3465 "(expected at least %zu)",
3466 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3467 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3469 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3470 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3472 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3473 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %zu "
3474 "(expected at least %zu)",
3475 ntohs(ovh->header.length), sizeof(struct nicira_header));
3476 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3480 switch (ntohl(nh->subtype)) {
3481 case NXT_STATUS_REQUEST:
3482 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3485 case NXT_TUN_ID_FROM_COOKIE:
3486 return handle_tun_id_from_cookie(p, msg);
3488 case NXT_ROLE_REQUEST:
3489 return handle_role_request(p, ofconn, msg);
3492 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3496 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3498 struct ofp_header *ob;
3501 /* Currently, everything executes synchronously, so we can just
3502 * immediately send the barrier reply. */
3503 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3504 queue_tx(buf, ofconn, ofconn->reply_counter);
3509 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3510 struct ofpbuf *ofp_msg)
3512 struct ofp_header *oh = ofp_msg->data;
3515 COVERAGE_INC(ofproto_recv_openflow);
3517 case OFPT_ECHO_REQUEST:
3518 error = handle_echo_request(ofconn, oh);
3521 case OFPT_ECHO_REPLY:
3525 case OFPT_FEATURES_REQUEST:
3526 error = handle_features_request(p, ofconn, oh);
3529 case OFPT_GET_CONFIG_REQUEST:
3530 error = handle_get_config_request(p, ofconn, oh);
3533 case OFPT_SET_CONFIG:
3534 error = handle_set_config(p, ofconn, ofp_msg->data);
3537 case OFPT_PACKET_OUT:
3538 error = handle_packet_out(p, ofconn, ofp_msg->data);
3542 error = handle_port_mod(p, ofconn, oh);
3546 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3549 case OFPT_STATS_REQUEST:
3550 error = handle_stats_request(p, ofconn, oh);
3554 error = handle_vendor(p, ofconn, ofp_msg->data);
3557 case OFPT_BARRIER_REQUEST:
3558 error = handle_barrier_request(ofconn, oh);
3562 if (VLOG_IS_WARN_ENABLED()) {
3563 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3564 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3567 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3572 send_error_oh(ofconn, ofp_msg->data, error);
3577 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3579 struct odp_msg *msg = packet->data;
3581 struct ofpbuf payload;
3584 payload.data = msg + 1;
3585 payload.size = msg->length - sizeof *msg;
3586 flow_extract(&payload, msg->arg, msg->port, &flow);
3588 /* Check with in-band control to see if this packet should be sent
3589 * to the local port regardless of the flow table. */
3590 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3591 union odp_action action;
3593 memset(&action, 0, sizeof(action));
3594 action.output.type = ODPAT_OUTPUT;
3595 action.output.port = ODPP_LOCAL;
3596 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3599 rule = lookup_valid_rule(p, &flow);
3601 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3602 struct ofport *port = port_array_get(&p->ports, msg->port);
3604 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3605 COVERAGE_INC(ofproto_no_packet_in);
3606 /* XXX install 'drop' flow entry */
3607 ofpbuf_delete(packet);
3611 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3614 COVERAGE_INC(ofproto_packet_in);
3615 send_packet_in(p, packet);
3619 if (rule->cr.wc.wildcards) {
3620 rule = rule_create_subrule(p, rule, &flow);
3621 rule_make_actions(p, rule, packet);
3623 if (!rule->may_install) {
3624 /* The rule is not installable, that is, we need to process every
3625 * packet, so process the current packet and set its actions into
3627 rule_make_actions(p, rule, packet);
3629 /* XXX revalidate rule if it needs it */
3633 rule_execute(p, rule, &payload, &flow);
3634 rule_reinstall(p, rule);
3636 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3638 * Extra-special case for fail-open mode.
3640 * We are in fail-open mode and the packet matched the fail-open rule,
3641 * but we are connected to a controller too. We should send the packet
3642 * up to the controller in the hope that it will try to set up a flow
3643 * and thereby allow us to exit fail-open.
3645 * See the top-level comment in fail-open.c for more information.
3647 send_packet_in(p, packet);
3649 ofpbuf_delete(packet);
3654 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3656 struct odp_msg *msg = packet->data;
3658 switch (msg->type) {
3659 case _ODPL_ACTION_NR:
3660 COVERAGE_INC(ofproto_ctlr_action);
3661 send_packet_in(p, packet);
3664 case _ODPL_SFLOW_NR:
3666 ofproto_sflow_received(p->sflow, msg);
3668 ofpbuf_delete(packet);
3672 handle_odp_miss_msg(p, packet);
3676 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3683 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3685 struct rule *sub = rule_from_cls_rule(sub_);
3686 struct revalidate_cbdata *cbdata = cbdata_;
3688 if (cbdata->revalidate_all
3689 || (cbdata->revalidate_subrules && sub->super)
3690 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3691 revalidate_rule(cbdata->ofproto, sub);
3696 revalidate_rule(struct ofproto *p, struct rule *rule)
3698 const flow_t *flow = &rule->cr.flow;
3700 COVERAGE_INC(ofproto_revalidate_rule);
3703 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3705 rule_remove(p, rule);
3707 } else if (super != rule->super) {
3708 COVERAGE_INC(ofproto_revalidate_moved);
3709 list_remove(&rule->list);
3710 list_push_back(&super->list, &rule->list);
3711 rule->super = super;
3712 rule->hard_timeout = super->hard_timeout;
3713 rule->idle_timeout = super->idle_timeout;
3714 rule->created = super->created;
3719 rule_update_actions(p, rule);
3723 static struct ofpbuf *
3724 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3725 long long int now, uint8_t reason)
3727 struct ofp_flow_removed *ofr;
3729 long long int tdiff = now - rule->created;
3730 uint32_t sec = tdiff / 1000;
3731 uint32_t msec = tdiff - (sec * 1000);
3733 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3734 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3736 ofr->cookie = rule->flow_cookie;
3737 ofr->priority = htons(rule->cr.priority);
3738 ofr->reason = reason;
3739 ofr->duration_sec = htonl(sec);
3740 ofr->duration_nsec = htonl(msec * 1000000);
3741 ofr->idle_timeout = htons(rule->idle_timeout);
3742 ofr->packet_count = htonll(rule->packet_count);
3743 ofr->byte_count = htonll(rule->byte_count);
3749 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3751 assert(rule->installed);
3752 assert(!rule->cr.wc.wildcards);
3755 rule_remove(ofproto, rule);
3757 rule_uninstall(ofproto, rule);
3762 send_flow_removed(struct ofproto *p, struct rule *rule,
3763 long long int now, uint8_t reason)
3765 struct ofconn *ofconn;
3766 struct ofconn *prev;
3767 struct ofpbuf *buf = NULL;
3769 /* We limit the maximum number of queued flow expirations it by accounting
3770 * them under the counter for replies. That works because preventing
3771 * OpenFlow requests from being processed also prevents new flows from
3772 * being added (and expiring). (It also prevents processing OpenFlow
3773 * requests that would not add new flows, so it is imperfect.) */
3776 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3777 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
3778 && ofconn->role != NX_ROLE_SLAVE) {
3780 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3782 buf = compose_flow_removed(p, rule, now, reason);
3788 queue_tx(buf, prev, prev->reply_counter);
3794 expire_rule(struct cls_rule *cls_rule, void *p_)
3796 struct ofproto *p = p_;
3797 struct rule *rule = rule_from_cls_rule(cls_rule);
3798 long long int hard_expire, idle_expire, expire, now;
3800 hard_expire = (rule->hard_timeout
3801 ? rule->created + rule->hard_timeout * 1000
3803 idle_expire = (rule->idle_timeout
3804 && (rule->super || list_is_empty(&rule->list))
3805 ? rule->used + rule->idle_timeout * 1000
3807 expire = MIN(hard_expire, idle_expire);
3811 if (rule->installed && now >= rule->used + 5000) {
3812 uninstall_idle_flow(p, rule);
3813 } else if (!rule->cr.wc.wildcards) {
3814 active_timeout(p, rule);
3820 COVERAGE_INC(ofproto_expired);
3822 /* Update stats. This code will be a no-op if the rule expired
3823 * due to an idle timeout. */
3824 if (rule->cr.wc.wildcards) {
3825 struct rule *subrule, *next;
3826 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3827 rule_remove(p, subrule);
3830 rule_uninstall(p, rule);
3833 if (!rule_is_hidden(rule)) {
3834 send_flow_removed(p, rule, now,
3836 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3838 rule_remove(p, rule);
3842 active_timeout(struct ofproto *ofproto, struct rule *rule)
3844 if (ofproto->netflow && !is_controller_rule(rule) &&
3845 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3846 struct ofexpired expired;
3847 struct odp_flow odp_flow;
3849 /* Get updated flow stats. */
3850 memset(&odp_flow, 0, sizeof odp_flow);
3851 if (rule->installed) {
3852 odp_flow.key = rule->cr.flow;
3853 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3854 dpif_flow_get(ofproto->dpif, &odp_flow);
3856 if (odp_flow.stats.n_packets) {
3857 update_time(ofproto, rule, &odp_flow.stats);
3858 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3859 odp_flow.stats.tcp_flags);
3863 expired.flow = rule->cr.flow;
3864 expired.packet_count = rule->packet_count +
3865 odp_flow.stats.n_packets;
3866 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3867 expired.used = rule->used;
3869 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3871 /* Schedule us to send the accumulated records once we have
3872 * collected all of them. */
3873 poll_immediate_wake();
3878 update_used(struct ofproto *p)
3880 struct odp_flow *flows;
3885 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3890 for (i = 0; i < n_flows; i++) {
3891 struct odp_flow *f = &flows[i];
3894 rule = rule_from_cls_rule(
3895 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3896 if (!rule || !rule->installed) {
3897 COVERAGE_INC(ofproto_unexpected_rule);
3898 dpif_flow_del(p->dpif, f);
3902 update_time(p, rule, &f->stats);
3903 rule_account(p, rule, f->stats.n_bytes);
3909 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
3911 struct ofconn *ofconn = ofconn_;
3912 struct ofproto *ofproto = ofconn->ofproto;
3913 struct odp_msg *msg = packet->data;
3914 struct ofpbuf payload;
3919 /* Extract packet payload from 'msg'. */
3920 payload.data = msg + 1;
3921 payload.size = msg->length - sizeof *msg;
3923 /* Construct packet-in message. */
3925 if (msg->type == _ODPL_ACTION_NR) {
3926 buffer_id = UINT32_MAX;
3928 if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
3929 buffer_id = pktbuf_get_null();
3931 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, msg->port);
3933 if (buffer_id != UINT32_MAX) {
3934 send_len = ofconn->miss_send_len;
3937 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port),
3938 msg->type, &payload, send_len);
3941 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3943 ofpbuf_delete(packet);
3947 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
3949 struct odp_msg *msg = packet->data;
3950 struct ofconn *ofconn, *prev;
3952 assert(msg->type == _ODPL_MISS_NR || msg->type == _ODPL_ACTION_NR);
3955 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
3956 if (ofconn->role != NX_ROLE_SLAVE) {
3958 pinsched_send(prev->schedulers[msg->type], msg->port,
3959 ofpbuf_clone(packet), do_send_packet_in, prev);
3965 pinsched_send(prev->schedulers[msg->type], msg->port,
3966 packet, do_send_packet_in, prev);
3968 ofpbuf_delete(packet);
3973 pick_datapath_id(const struct ofproto *ofproto)
3975 const struct ofport *port;
3977 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3979 uint8_t ea[ETH_ADDR_LEN];
3982 error = netdev_get_etheraddr(port->netdev, ea);
3984 return eth_addr_to_uint64(ea);
3986 VLOG_WARN("could not get MAC address for %s (%s)",
3987 netdev_get_name(port->netdev), strerror(error));
3989 return ofproto->fallback_dpid;
3993 pick_fallback_dpid(void)
3995 uint8_t ea[ETH_ADDR_LEN];
3996 eth_addr_nicira_random(ea);
3997 return eth_addr_to_uint64(ea);
4001 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4002 struct odp_actions *actions, tag_type *tags,
4003 uint16_t *nf_output_iface, void *ofproto_)
4005 struct ofproto *ofproto = ofproto_;
4008 /* Drop frames for reserved multicast addresses. */
4009 if (eth_addr_is_reserved(flow->dl_dst)) {
4013 /* Learn source MAC (but don't try to learn from revalidation). */
4014 if (packet != NULL) {
4015 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4018 /* The log messages here could actually be useful in debugging,
4019 * so keep the rate limit relatively high. */
4020 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4021 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4022 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4023 ofproto_revalidate(ofproto, rev_tag);
4027 /* Determine output port. */
4028 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
4030 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4031 } else if (out_port != flow->in_port) {
4032 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4033 *nf_output_iface = out_port;
4041 static const struct ofhooks default_ofhooks = {
4043 default_normal_ofhook_cb,