2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
23 #include <netinet/in.h>
26 #include "classifier.h"
28 #include "discovery.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
33 #include "mac-learning.h"
37 #include "ofp-print.h"
38 #include "ofproto-sflow.h"
40 #include "openflow/nicira-ext.h"
41 #include "openflow/openflow.h"
42 #include "openvswitch/datapath-protocol.h"
46 #include "poll-loop.h"
47 #include "port-array.h"
52 #include "stream-ssl.h"
60 #define THIS_MODULE VLM_ofproto
63 #include "sflow_api.h"
67 TABLEID_CLASSIFIER = 1
71 struct netdev *netdev;
72 struct ofp_phy_port opp; /* In host byte order. */
75 static void ofport_free(struct ofport *);
76 static void hton_ofp_phy_port(struct ofp_phy_port *);
78 static int xlate_actions(const union ofp_action *in, size_t n_in,
79 const flow_t *flow, struct ofproto *ofproto,
80 const struct ofpbuf *packet,
81 struct odp_actions *out, tag_type *tags,
82 bool *may_set_up_flow, uint16_t *nf_output_iface);
87 uint64_t flow_cookie; /* Controller-issued identifier.
88 (Kept in network-byte order.) */
89 uint16_t idle_timeout; /* In seconds from time of last use. */
90 uint16_t hard_timeout; /* In seconds from time of creation. */
91 bool send_flow_removed; /* Send a flow removed message? */
92 long long int used; /* Last-used time (0 if never used). */
93 long long int created; /* Creation time. */
94 uint64_t packet_count; /* Number of packets received. */
95 uint64_t byte_count; /* Number of bytes received. */
96 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
97 tag_type tags; /* Tags (set only by hooks). */
98 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
100 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
101 * exact-match rule (having cr.wc.wildcards of 0) generated from the
102 * wildcard rule 'super'. In this case, 'list' is an element of the
105 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
106 * a list of subrules. A super-rule with no wildcards (where
107 * cr.wc.wildcards is 0) will never have any subrules. */
113 * 'n_actions' is the number of elements in the 'actions' array. A single
114 * action may take up more more than one element's worth of space.
116 * A subrule has no actions (it uses the super-rule's actions). */
118 union ofp_action *actions;
122 * A super-rule with wildcard fields never has ODP actions (since the
123 * datapath only supports exact-match flows). */
124 bool installed; /* Installed in datapath? */
125 bool may_install; /* True ordinarily; false if actions must
126 * be reassessed for every packet. */
128 union odp_action *odp_actions;
132 rule_is_hidden(const struct rule *rule)
134 /* Subrules are merely an implementation detail, so hide them from the
136 if (rule->super != NULL) {
140 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
141 * (e.g. by in-band control) and are intentionally hidden from the
143 if (rule->cr.priority > UINT16_MAX) {
150 static struct rule *rule_create(struct ofproto *, struct rule *super,
151 const union ofp_action *, size_t n_actions,
152 uint16_t idle_timeout, uint16_t hard_timeout,
153 uint64_t flow_cookie, bool send_flow_removed);
154 static void rule_free(struct rule *);
155 static void rule_destroy(struct ofproto *, struct rule *);
156 static struct rule *rule_from_cls_rule(const struct cls_rule *);
157 static void rule_insert(struct ofproto *, struct rule *,
158 struct ofpbuf *packet, uint16_t in_port);
159 static void rule_remove(struct ofproto *, struct rule *);
160 static bool rule_make_actions(struct ofproto *, struct rule *,
161 const struct ofpbuf *packet);
162 static void rule_install(struct ofproto *, struct rule *,
163 struct rule *displaced_rule);
164 static void rule_uninstall(struct ofproto *, struct rule *);
165 static void rule_post_uninstall(struct ofproto *, struct rule *);
166 static void send_flow_removed(struct ofproto *p, struct rule *rule,
167 long long int now, uint8_t reason);
169 /* ofproto supports two kinds of OpenFlow connections:
171 * - "Controller connections": Connections to ordinary OpenFlow controllers.
172 * ofproto maintains persistent connections to these controllers and by
173 * default sends them asynchronous messages such as packet-ins.
175 * - "Transient connections", e.g. from ovs-ofctl. When these connections
176 * drop, it is the other side's responsibility to reconnect them if
177 * necessary. ofproto does not send them asynchronous messages by default.
180 OFCONN_CONTROLLER, /* An OpenFlow controller. */
181 OFCONN_TRANSIENT /* A transient connection. */
184 /* An OpenFlow connection. */
186 struct ofproto *ofproto; /* The ofproto that owns this connection. */
187 struct list node; /* In struct ofproto's "all_conns" list. */
188 struct rconn *rconn; /* OpenFlow connection. */
189 enum ofconn_type type; /* Type. */
191 /* OFPT_PACKET_IN related data. */
192 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
193 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
194 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
195 int miss_send_len; /* Bytes to send of buffered packets. */
197 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
198 * requests, and the maximum number before we stop reading OpenFlow
200 #define OFCONN_REPLY_MAX 100
201 struct rconn_packet_counter *reply_counter;
203 /* type == OFCONN_CONTROLLER only. */
204 enum nx_role role; /* Role. */
205 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
206 struct discovery *discovery; /* Controller discovery object, if enabled. */
207 struct status_category *ss; /* Switch status category. */
208 enum ofproto_band band; /* In-band or out-of-band? */
211 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
212 * "schedulers" array. Their values are 0 and 1, and their meanings and values
213 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
214 * case anything ever changes, check their values here. */
215 #define N_SCHEDULERS 2
216 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
217 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
218 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
219 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
221 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
223 static void ofconn_destroy(struct ofconn *);
224 static void ofconn_run(struct ofconn *, struct ofproto *);
225 static void ofconn_wait(struct ofconn *);
226 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
227 struct rconn_packet_counter *counter);
229 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
230 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
234 uint64_t datapath_id; /* Datapath ID. */
235 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
236 char *mfr_desc; /* Manufacturer. */
237 char *hw_desc; /* Hardware. */
238 char *sw_desc; /* Software version. */
239 char *serial_desc; /* Serial number. */
240 char *dp_desc; /* Datapath description. */
244 struct netdev_monitor *netdev_monitor;
245 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
247 struct shash port_by_name;
251 struct switch_status *switch_status;
252 struct fail_open *fail_open;
253 struct netflow *netflow;
254 struct ofproto_sflow *sflow;
256 /* In-band control. */
257 struct in_band *in_band;
258 long long int next_in_band_update;
259 struct sockaddr_in *extra_in_band_remotes;
260 size_t n_extra_remotes;
263 struct classifier cls;
264 bool need_revalidate;
265 long long int next_expiration;
266 struct tag_set revalidate_set;
267 bool tun_id_from_cookie;
269 /* OpenFlow connections. */
270 struct hmap controllers; /* Controller "struct ofconn"s. */
271 struct list all_conns; /* Contains "struct ofconn"s. */
272 struct pvconn **listeners;
274 struct pvconn **snoops;
277 /* Hooks for ovs-vswitchd. */
278 const struct ofhooks *ofhooks;
281 /* Used by default ofhooks. */
282 struct mac_learning *ml;
285 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
287 static const struct ofhooks default_ofhooks;
289 static uint64_t pick_datapath_id(const struct ofproto *);
290 static uint64_t pick_fallback_dpid(void);
292 static void update_used(struct ofproto *);
293 static void update_stats(struct ofproto *, struct rule *,
294 const struct odp_flow_stats *);
295 static void expire_rule(struct cls_rule *, void *ofproto);
296 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
297 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
298 static void revalidate_cb(struct cls_rule *rule_, void *p_);
300 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
302 static void handle_openflow(struct ofconn *, struct ofproto *,
305 static void refresh_port_groups(struct ofproto *);
307 static void update_port(struct ofproto *, const char *devname);
308 static int init_ports(struct ofproto *);
309 static void reinit_ports(struct ofproto *);
312 ofproto_create(const char *datapath, const char *datapath_type,
313 const struct ofhooks *ofhooks, void *aux,
314 struct ofproto **ofprotop)
316 struct odp_stats stats;
323 /* Connect to datapath and start listening for messages. */
324 error = dpif_open(datapath, datapath_type, &dpif);
326 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
329 error = dpif_get_dp_stats(dpif, &stats);
331 VLOG_ERR("failed to obtain stats for datapath %s: %s",
332 datapath, strerror(error));
336 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
338 VLOG_ERR("failed to listen on datapath %s: %s",
339 datapath, strerror(error));
343 dpif_flow_flush(dpif);
344 dpif_recv_purge(dpif);
346 /* Initialize settings. */
347 p = xzalloc(sizeof *p);
348 p->fallback_dpid = pick_fallback_dpid();
349 p->datapath_id = p->fallback_dpid;
350 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
351 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
352 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
353 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
354 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
356 /* Initialize datapath. */
358 p->netdev_monitor = netdev_monitor_create();
359 port_array_init(&p->ports);
360 shash_init(&p->port_by_name);
361 p->max_ports = stats.max_ports;
363 /* Initialize submodules. */
364 p->switch_status = switch_status_create(p);
370 /* Initialize flow table. */
371 classifier_init(&p->cls);
372 p->need_revalidate = false;
373 p->next_expiration = time_msec() + 1000;
374 tag_set_init(&p->revalidate_set);
376 /* Initialize OpenFlow connections. */
377 list_init(&p->all_conns);
378 hmap_init(&p->controllers);
384 /* Initialize hooks. */
386 p->ofhooks = ofhooks;
390 p->ofhooks = &default_ofhooks;
392 p->ml = mac_learning_create();
395 /* Pick final datapath ID. */
396 p->datapath_id = pick_datapath_id(p);
397 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
404 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
406 uint64_t old_dpid = p->datapath_id;
407 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
408 if (p->datapath_id != old_dpid) {
409 struct ofconn *ofconn;
411 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
413 /* Force all active connections to reconnect, since there is no way to
414 * notify a controller that the datapath ID has changed. */
415 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
416 rconn_reconnect(ofconn->rconn);
422 is_discovery_controller(const struct ofproto_controller *c)
424 return !strcmp(c->target, "discover");
428 is_in_band_controller(const struct ofproto_controller *c)
430 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
433 /* Creates a new controller in 'ofproto'. Some of the settings are initially
434 * drawn from 'c', but update_controller() needs to be called later to finish
435 * the new ofconn's configuration. */
437 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
439 struct discovery *discovery;
440 struct ofconn *ofconn;
442 if (is_discovery_controller(c)) {
443 int error = discovery_create(c->accept_re, c->update_resolv_conf,
444 ofproto->dpif, ofproto->switch_status,
453 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
454 ofconn->pktbuf = pktbuf_create();
455 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
457 ofconn->discovery = discovery;
459 rconn_connect(ofconn->rconn, c->target);
461 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
462 hash_string(c->target, 0));
465 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
466 * target or turn discovery on or off (these are done by creating new ofconns
467 * and deleting old ones), but it can update the rest of an ofconn's
470 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
472 struct ofproto *ofproto = ofconn->ofproto;
476 ofconn->band = (is_in_band_controller(c)
477 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
479 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
481 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
482 rconn_set_probe_interval(ofconn->rconn, probe_interval);
484 if (ofconn->discovery) {
485 discovery_set_update_resolv_conf(ofconn->discovery,
486 c->update_resolv_conf);
487 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
490 for (i = 0; i < N_SCHEDULERS; i++) {
491 struct pinsched **s = &ofconn->schedulers[i];
493 if (c->rate_limit > 0) {
495 *s = pinsched_create(c->rate_limit, c->burst_limit,
496 ofproto->switch_status);
498 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
501 pinsched_destroy(*s);
508 ofconn_get_target(const struct ofconn *ofconn)
510 return ofconn->discovery ? "discover" : rconn_get_name(ofconn->rconn);
513 static struct ofconn *
514 find_controller_by_target(struct ofproto *ofproto, const char *target)
516 struct ofconn *ofconn;
518 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
519 hash_string(target, 0), &ofproto->controllers) {
520 if (!strcmp(ofconn_get_target(ofconn), target)) {
528 update_in_band_remotes(struct ofproto *ofproto)
530 const struct ofconn *ofconn;
531 struct sockaddr_in *addrs;
532 size_t max_addrs, n_addrs;
536 /* Allocate enough memory for as many remotes as we could possibly have. */
537 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
538 addrs = xmalloc(max_addrs * sizeof *addrs);
541 /* Add all the remotes. */
543 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
544 struct sockaddr_in *sin = &addrs[n_addrs];
546 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
547 if (sin->sin_addr.s_addr) {
548 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
551 if (ofconn->discovery) {
555 for (i = 0; i < ofproto->n_extra_remotes; i++) {
556 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
559 /* Create or update or destroy in-band.
561 * Ordinarily we only enable in-band if there's at least one remote
562 * address, but discovery needs the in-band rules for DHCP to be installed
563 * even before we know any remote addresses. */
564 if (n_addrs || discovery) {
565 if (!ofproto->in_band) {
566 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
569 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
570 ofproto->next_in_band_update = time_msec() + 1000;
572 in_band_destroy(ofproto->in_band);
573 ofproto->in_band = NULL;
581 ofproto_set_controllers(struct ofproto *p,
582 const struct ofproto_controller *controllers,
583 size_t n_controllers)
585 struct shash new_controllers;
586 enum ofproto_fail_mode fail_mode;
587 struct ofconn *ofconn, *next;
591 shash_init(&new_controllers);
592 for (i = 0; i < n_controllers; i++) {
593 const struct ofproto_controller *c = &controllers[i];
595 shash_add_once(&new_controllers, c->target, &controllers[i]);
596 if (!find_controller_by_target(p, c->target)) {
597 add_controller(p, c);
601 fail_mode = OFPROTO_FAIL_STANDALONE;
603 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
605 struct ofproto_controller *c;
607 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
609 ofconn_destroy(ofconn);
611 update_controller(ofconn, c);
615 if (c->fail == OFPROTO_FAIL_SECURE) {
616 fail_mode = OFPROTO_FAIL_SECURE;
620 shash_destroy(&new_controllers);
622 update_in_band_remotes(p);
624 if (!hmap_is_empty(&p->controllers)
625 && fail_mode == OFPROTO_FAIL_STANDALONE) {
626 struct rconn **rconns;
630 p->fail_open = fail_open_create(p, p->switch_status);
634 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
635 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
636 rconns[n++] = ofconn->rconn;
639 fail_open_set_controllers(p->fail_open, rconns, n);
640 /* p->fail_open takes ownership of 'rconns'. */
642 fail_open_destroy(p->fail_open);
646 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
647 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
648 struct ofconn, hmap_node);
649 ofconn->ss = switch_status_register(p->switch_status, "remote",
650 rconn_status_cb, ofconn->rconn);
655 any_extras_changed(const struct ofproto *ofproto,
656 const struct sockaddr_in *extras, size_t n)
660 if (n != ofproto->n_extra_remotes) {
664 for (i = 0; i < n; i++) {
665 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
666 const struct sockaddr_in *new = &extras[i];
668 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
669 old->sin_port != new->sin_port) {
677 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
678 * in-band control should guarantee access, in the same way that in-band
679 * control guarantees access to OpenFlow controllers. */
681 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
682 const struct sockaddr_in *extras, size_t n)
684 if (!any_extras_changed(ofproto, extras, n)) {
688 free(ofproto->extra_in_band_remotes);
689 ofproto->n_extra_remotes = n;
690 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
692 update_in_band_remotes(ofproto);
696 ofproto_set_desc(struct ofproto *p,
697 const char *mfr_desc, const char *hw_desc,
698 const char *sw_desc, const char *serial_desc,
701 struct ofp_desc_stats *ods;
704 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
705 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
706 sizeof ods->mfr_desc);
709 p->mfr_desc = xstrdup(mfr_desc);
712 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
713 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
714 sizeof ods->hw_desc);
717 p->hw_desc = xstrdup(hw_desc);
720 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
721 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
722 sizeof ods->sw_desc);
725 p->sw_desc = xstrdup(sw_desc);
728 if (strlen(serial_desc) >= sizeof ods->serial_num) {
729 VLOG_WARN("truncating serial_desc, must be less than %zu "
731 sizeof ods->serial_num);
733 free(p->serial_desc);
734 p->serial_desc = xstrdup(serial_desc);
737 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
738 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
739 sizeof ods->dp_desc);
742 p->dp_desc = xstrdup(dp_desc);
747 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
748 const struct svec *svec)
750 struct pvconn **pvconns = *pvconnsp;
751 size_t n_pvconns = *n_pvconnsp;
755 for (i = 0; i < n_pvconns; i++) {
756 pvconn_close(pvconns[i]);
760 pvconns = xmalloc(svec->n * sizeof *pvconns);
762 for (i = 0; i < svec->n; i++) {
763 const char *name = svec->names[i];
764 struct pvconn *pvconn;
767 error = pvconn_open(name, &pvconn);
769 pvconns[n_pvconns++] = pvconn;
771 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
779 *n_pvconnsp = n_pvconns;
785 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
787 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
791 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
793 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
797 ofproto_set_netflow(struct ofproto *ofproto,
798 const struct netflow_options *nf_options)
800 if (nf_options && nf_options->collectors.n) {
801 if (!ofproto->netflow) {
802 ofproto->netflow = netflow_create();
804 return netflow_set_options(ofproto->netflow, nf_options);
806 netflow_destroy(ofproto->netflow);
807 ofproto->netflow = NULL;
813 ofproto_set_sflow(struct ofproto *ofproto,
814 const struct ofproto_sflow_options *oso)
816 struct ofproto_sflow *os = ofproto->sflow;
819 struct ofport *ofport;
820 unsigned int odp_port;
822 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
823 refresh_port_groups(ofproto);
824 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
825 ofproto_sflow_add_port(os, odp_port,
826 netdev_get_name(ofport->netdev));
829 ofproto_sflow_set_options(os, oso);
831 ofproto_sflow_destroy(os);
832 ofproto->sflow = NULL;
837 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
841 VLOG_WARN("STP is not yet implemented");
849 ofproto_get_datapath_id(const struct ofproto *ofproto)
851 return ofproto->datapath_id;
855 ofproto_has_controller(const struct ofproto *ofproto)
857 return !hmap_is_empty(&ofproto->controllers);
861 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
865 for (i = 0; i < ofproto->n_listeners; i++) {
866 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
871 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
875 for (i = 0; i < ofproto->n_snoops; i++) {
876 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
881 ofproto_destroy(struct ofproto *p)
883 struct ofconn *ofconn, *next_ofconn;
884 struct ofport *ofport;
885 unsigned int port_no;
892 /* Destroy fail-open and in-band early, since they touch the classifier. */
893 fail_open_destroy(p->fail_open);
896 in_band_destroy(p->in_band);
898 free(p->extra_in_band_remotes);
900 ofproto_flush_flows(p);
901 classifier_destroy(&p->cls);
903 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
905 ofconn_destroy(ofconn);
907 hmap_destroy(&p->controllers);
910 netdev_monitor_destroy(p->netdev_monitor);
911 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
914 shash_destroy(&p->port_by_name);
916 switch_status_destroy(p->switch_status);
917 netflow_destroy(p->netflow);
918 ofproto_sflow_destroy(p->sflow);
920 for (i = 0; i < p->n_listeners; i++) {
921 pvconn_close(p->listeners[i]);
925 for (i = 0; i < p->n_snoops; i++) {
926 pvconn_close(p->snoops[i]);
930 mac_learning_destroy(p->ml);
935 free(p->serial_desc);
938 port_array_destroy(&p->ports);
944 ofproto_run(struct ofproto *p)
946 int error = ofproto_run1(p);
948 error = ofproto_run2(p, false);
954 process_port_change(struct ofproto *ofproto, int error, char *devname)
956 if (error == ENOBUFS) {
957 reinit_ports(ofproto);
959 update_port(ofproto, devname);
964 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
965 * Connects this vconn to a controller. */
967 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
969 struct ofconn *ofconn;
971 /* Arbitrarily pick the first controller in the list for monitoring. We
972 * could do something smarter or more flexible later, if it ever proves
974 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
975 if (ofconn->type == OFCONN_CONTROLLER) {
976 rconn_add_monitor(ofconn->rconn, vconn);
981 VLOG_INFO_RL(&rl, "no controller connection to monitor");
986 ofproto_run1(struct ofproto *p)
988 struct ofconn *ofconn, *next_ofconn;
993 if (shash_is_empty(&p->port_by_name)) {
997 for (i = 0; i < 50; i++) {
1001 error = dpif_recv(p->dpif, &buf);
1003 if (error == ENODEV) {
1004 /* Someone destroyed the datapath behind our back. The caller
1005 * better destroy us and give up, because we're just going to
1006 * spin from here on out. */
1007 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1008 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1009 dpif_name(p->dpif));
1015 handle_odp_msg(p, buf);
1018 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1019 process_port_change(p, error, devname);
1021 while ((error = netdev_monitor_poll(p->netdev_monitor,
1022 &devname)) != EAGAIN) {
1023 process_port_change(p, error, devname);
1027 if (time_msec() >= p->next_in_band_update) {
1028 update_in_band_remotes(p);
1030 in_band_run(p->in_band);
1033 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1035 ofconn_run(ofconn, p);
1038 /* Fail-open maintenance. Do this after processing the ofconns since
1039 * fail-open checks the status of the controller rconn. */
1041 fail_open_run(p->fail_open);
1044 for (i = 0; i < p->n_listeners; i++) {
1045 struct vconn *vconn;
1048 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
1050 ofconn_create(p, rconn_new_from_vconn("passive", vconn),
1052 } else if (retval != EAGAIN) {
1053 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1057 for (i = 0; i < p->n_snoops; i++) {
1058 struct vconn *vconn;
1061 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1063 add_snooper(p, vconn);
1064 } else if (retval != EAGAIN) {
1065 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1069 if (time_msec() >= p->next_expiration) {
1070 COVERAGE_INC(ofproto_expiration);
1071 p->next_expiration = time_msec() + 1000;
1074 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1076 /* Let the hook know that we're at a stable point: all outstanding data
1077 * in existing flows has been accounted to the account_cb. Thus, the
1078 * hook can now reasonably do operations that depend on having accurate
1079 * flow volume accounting (currently, that's just bond rebalancing). */
1080 if (p->ofhooks->account_checkpoint_cb) {
1081 p->ofhooks->account_checkpoint_cb(p->aux);
1086 netflow_run(p->netflow);
1089 ofproto_sflow_run(p->sflow);
1095 struct revalidate_cbdata {
1096 struct ofproto *ofproto;
1097 bool revalidate_all; /* Revalidate all exact-match rules? */
1098 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1099 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1103 ofproto_run2(struct ofproto *p, bool revalidate_all)
1105 if (p->need_revalidate || revalidate_all
1106 || !tag_set_is_empty(&p->revalidate_set)) {
1107 struct revalidate_cbdata cbdata;
1109 cbdata.revalidate_all = revalidate_all;
1110 cbdata.revalidate_subrules = p->need_revalidate;
1111 cbdata.revalidate_set = p->revalidate_set;
1112 tag_set_init(&p->revalidate_set);
1113 COVERAGE_INC(ofproto_revalidate);
1114 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1115 p->need_revalidate = false;
1122 ofproto_wait(struct ofproto *p)
1124 struct ofconn *ofconn;
1127 dpif_recv_wait(p->dpif);
1128 dpif_port_poll_wait(p->dpif);
1129 netdev_monitor_poll_wait(p->netdev_monitor);
1130 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1131 ofconn_wait(ofconn);
1134 poll_timer_wait(p->next_in_band_update - time_msec());
1135 in_band_wait(p->in_band);
1138 fail_open_wait(p->fail_open);
1141 ofproto_sflow_wait(p->sflow);
1143 if (!tag_set_is_empty(&p->revalidate_set)) {
1144 poll_immediate_wake();
1146 if (p->need_revalidate) {
1147 /* Shouldn't happen, but if it does just go around again. */
1148 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1149 poll_immediate_wake();
1150 } else if (p->next_expiration != LLONG_MAX) {
1151 poll_timer_wait(p->next_expiration - time_msec());
1153 for (i = 0; i < p->n_listeners; i++) {
1154 pvconn_wait(p->listeners[i]);
1156 for (i = 0; i < p->n_snoops; i++) {
1157 pvconn_wait(p->snoops[i]);
1162 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1164 tag_set_add(&ofproto->revalidate_set, tag);
1168 ofproto_get_revalidate_set(struct ofproto *ofproto)
1170 return &ofproto->revalidate_set;
1174 ofproto_is_alive(const struct ofproto *p)
1176 return !hmap_is_empty(&p->controllers);
1180 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1181 const union ofp_action *actions, size_t n_actions,
1182 const struct ofpbuf *packet)
1184 struct odp_actions odp_actions;
1187 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1193 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1195 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1196 odp_actions.n_actions, packet);
1201 ofproto_add_flow(struct ofproto *p,
1202 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1203 const union ofp_action *actions, size_t n_actions,
1207 rule = rule_create(p, NULL, actions, n_actions,
1208 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1210 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1211 rule_insert(p, rule, NULL, 0);
1215 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1216 uint32_t wildcards, unsigned int priority)
1220 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1224 rule_remove(ofproto, rule);
1229 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1231 struct rule *rule = rule_from_cls_rule(rule_);
1232 struct ofproto *ofproto = ofproto_;
1234 /* Mark the flow as not installed, even though it might really be
1235 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1236 * There is no point in uninstalling it individually since we are about to
1237 * blow away all the flows with dpif_flow_flush(). */
1238 rule->installed = false;
1240 rule_remove(ofproto, rule);
1244 ofproto_flush_flows(struct ofproto *ofproto)
1246 COVERAGE_INC(ofproto_flush);
1247 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1248 dpif_flow_flush(ofproto->dpif);
1249 if (ofproto->in_band) {
1250 in_band_flushed(ofproto->in_band);
1252 if (ofproto->fail_open) {
1253 fail_open_flushed(ofproto->fail_open);
1258 reinit_ports(struct ofproto *p)
1260 struct svec devnames;
1261 struct ofport *ofport;
1262 unsigned int port_no;
1263 struct odp_port *odp_ports;
1267 svec_init(&devnames);
1268 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1269 svec_add (&devnames, (char *) ofport->opp.name);
1271 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1272 for (i = 0; i < n_odp_ports; i++) {
1273 svec_add (&devnames, odp_ports[i].devname);
1277 svec_sort_unique(&devnames);
1278 for (i = 0; i < devnames.n; i++) {
1279 update_port(p, devnames.names[i]);
1281 svec_destroy(&devnames);
1285 refresh_port_group(struct ofproto *p, unsigned int group)
1289 struct ofport *port;
1290 unsigned int port_no;
1292 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1294 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1296 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1297 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1298 ports[n_ports++] = port_no;
1301 dpif_port_group_set(p->dpif, group, ports, n_ports);
1308 refresh_port_groups(struct ofproto *p)
1310 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1311 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1313 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1317 static struct ofport *
1318 make_ofport(const struct odp_port *odp_port)
1320 struct netdev_options netdev_options;
1321 enum netdev_flags flags;
1322 struct ofport *ofport;
1323 struct netdev *netdev;
1327 memset(&netdev_options, 0, sizeof netdev_options);
1328 netdev_options.name = odp_port->devname;
1329 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1330 netdev_options.may_open = true;
1332 error = netdev_open(&netdev_options, &netdev);
1334 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1335 "cannot be opened (%s)",
1336 odp_port->devname, odp_port->port,
1337 odp_port->devname, strerror(error));
1341 ofport = xmalloc(sizeof *ofport);
1342 ofport->netdev = netdev;
1343 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1344 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1345 memcpy(ofport->opp.name, odp_port->devname,
1346 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1347 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1349 netdev_get_flags(netdev, &flags);
1350 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1352 netdev_get_carrier(netdev, &carrier);
1353 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1355 netdev_get_features(netdev,
1356 &ofport->opp.curr, &ofport->opp.advertised,
1357 &ofport->opp.supported, &ofport->opp.peer);
1362 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1364 if (port_array_get(&p->ports, odp_port->port)) {
1365 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1368 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1369 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1378 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1380 const struct ofp_phy_port *a = &a_->opp;
1381 const struct ofp_phy_port *b = &b_->opp;
1383 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1384 return (a->port_no == b->port_no
1385 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1386 && !strcmp((char *) a->name, (char *) b->name)
1387 && a->state == b->state
1388 && a->config == b->config
1389 && a->curr == b->curr
1390 && a->advertised == b->advertised
1391 && a->supported == b->supported
1392 && a->peer == b->peer);
1396 send_port_status(struct ofproto *p, const struct ofport *ofport,
1399 /* XXX Should limit the number of queued port status change messages. */
1400 struct ofconn *ofconn;
1401 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1402 struct ofp_port_status *ops;
1405 if (ofconn->role == NX_ROLE_SLAVE) {
1409 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1410 ops->reason = reason;
1411 ops->desc = ofport->opp;
1412 hton_ofp_phy_port(&ops->desc);
1413 queue_tx(b, ofconn, NULL);
1415 if (p->ofhooks->port_changed_cb) {
1416 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1421 ofport_install(struct ofproto *p, struct ofport *ofport)
1423 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1424 const char *netdev_name = (const char *) ofport->opp.name;
1426 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1427 port_array_set(&p->ports, odp_port, ofport);
1428 shash_add(&p->port_by_name, netdev_name, ofport);
1430 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1435 ofport_remove(struct ofproto *p, struct ofport *ofport)
1437 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1439 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1440 port_array_set(&p->ports, odp_port, NULL);
1441 shash_delete(&p->port_by_name,
1442 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1444 ofproto_sflow_del_port(p->sflow, odp_port);
1449 ofport_free(struct ofport *ofport)
1452 netdev_close(ofport->netdev);
1458 update_port(struct ofproto *p, const char *devname)
1460 struct odp_port odp_port;
1461 struct ofport *old_ofport;
1462 struct ofport *new_ofport;
1465 COVERAGE_INC(ofproto_update_port);
1467 /* Query the datapath for port information. */
1468 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1470 /* Find the old ofport. */
1471 old_ofport = shash_find_data(&p->port_by_name, devname);
1474 /* There's no port named 'devname' but there might be a port with
1475 * the same port number. This could happen if a port is deleted
1476 * and then a new one added in its place very quickly, or if a port
1477 * is renamed. In the former case we want to send an OFPPR_DELETE
1478 * and an OFPPR_ADD, and in the latter case we want to send a
1479 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1480 * the old port's ifindex against the new port, or perhaps less
1481 * reliably but more portably by comparing the old port's MAC
1482 * against the new port's MAC. However, this code isn't that smart
1483 * and always sends an OFPPR_MODIFY (XXX). */
1484 old_ofport = port_array_get(&p->ports, odp_port.port);
1486 } else if (error != ENOENT && error != ENODEV) {
1487 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1488 "%s", strerror(error));
1492 /* Create a new ofport. */
1493 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1495 /* Eliminate a few pathological cases. */
1496 if (!old_ofport && !new_ofport) {
1498 } else if (old_ofport && new_ofport) {
1499 /* Most of the 'config' bits are OpenFlow soft state, but
1500 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1501 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1502 * leaves the other bits 0.) */
1503 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1505 if (ofport_equal(old_ofport, new_ofport)) {
1506 /* False alarm--no change. */
1507 ofport_free(new_ofport);
1512 /* Now deal with the normal cases. */
1514 ofport_remove(p, old_ofport);
1517 ofport_install(p, new_ofport);
1519 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1520 (!old_ofport ? OFPPR_ADD
1521 : !new_ofport ? OFPPR_DELETE
1523 ofport_free(old_ofport);
1525 /* Update port groups. */
1526 refresh_port_groups(p);
1530 init_ports(struct ofproto *p)
1532 struct odp_port *ports;
1537 error = dpif_port_list(p->dpif, &ports, &n_ports);
1542 for (i = 0; i < n_ports; i++) {
1543 const struct odp_port *odp_port = &ports[i];
1544 if (!ofport_conflicts(p, odp_port)) {
1545 struct ofport *ofport = make_ofport(odp_port);
1547 ofport_install(p, ofport);
1552 refresh_port_groups(p);
1556 static struct ofconn *
1557 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1559 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1560 ofconn->ofproto = p;
1561 list_push_back(&p->all_conns, &ofconn->node);
1562 ofconn->rconn = rconn;
1563 ofconn->type = type;
1564 ofconn->role = NX_ROLE_OTHER;
1565 ofconn->packet_in_counter = rconn_packet_counter_create ();
1566 ofconn->pktbuf = NULL;
1567 ofconn->miss_send_len = 0;
1568 ofconn->reply_counter = rconn_packet_counter_create ();
1573 ofconn_destroy(struct ofconn *ofconn)
1575 if (ofconn->type == OFCONN_CONTROLLER) {
1576 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1578 discovery_destroy(ofconn->discovery);
1580 list_remove(&ofconn->node);
1581 switch_status_unregister(ofconn->ss);
1582 rconn_destroy(ofconn->rconn);
1583 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1584 rconn_packet_counter_destroy(ofconn->reply_counter);
1585 pktbuf_destroy(ofconn->pktbuf);
1590 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1595 if (ofconn->discovery) {
1596 char *controller_name;
1597 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1598 discovery_question_connectivity(ofconn->discovery);
1600 if (discovery_run(ofconn->discovery, &controller_name)) {
1601 if (controller_name) {
1602 rconn_connect(ofconn->rconn, controller_name);
1604 rconn_disconnect(ofconn->rconn);
1609 for (i = 0; i < N_SCHEDULERS; i++) {
1610 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1613 rconn_run(ofconn->rconn);
1615 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1616 /* Limit the number of iterations to prevent other tasks from
1618 for (iteration = 0; iteration < 50; iteration++) {
1619 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1624 fail_open_maybe_recover(p->fail_open);
1626 handle_openflow(ofconn, p, of_msg);
1627 ofpbuf_delete(of_msg);
1631 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1632 ofconn_destroy(ofconn);
1637 ofconn_wait(struct ofconn *ofconn)
1641 if (ofconn->discovery) {
1642 discovery_wait(ofconn->discovery);
1644 for (i = 0; i < N_SCHEDULERS; i++) {
1645 pinsched_wait(ofconn->schedulers[i]);
1647 rconn_run_wait(ofconn->rconn);
1648 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1649 rconn_recv_wait(ofconn->rconn);
1651 COVERAGE_INC(ofproto_ofconn_stuck);
1655 /* Caller is responsible for initializing the 'cr' member of the returned
1657 static struct rule *
1658 rule_create(struct ofproto *ofproto, struct rule *super,
1659 const union ofp_action *actions, size_t n_actions,
1660 uint16_t idle_timeout, uint16_t hard_timeout,
1661 uint64_t flow_cookie, bool send_flow_removed)
1663 struct rule *rule = xzalloc(sizeof *rule);
1664 rule->idle_timeout = idle_timeout;
1665 rule->hard_timeout = hard_timeout;
1666 rule->flow_cookie = flow_cookie;
1667 rule->used = rule->created = time_msec();
1668 rule->send_flow_removed = send_flow_removed;
1669 rule->super = super;
1671 list_push_back(&super->list, &rule->list);
1673 list_init(&rule->list);
1675 rule->n_actions = n_actions;
1676 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1677 netflow_flow_clear(&rule->nf_flow);
1678 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1683 static struct rule *
1684 rule_from_cls_rule(const struct cls_rule *cls_rule)
1686 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1690 rule_free(struct rule *rule)
1692 free(rule->actions);
1693 free(rule->odp_actions);
1697 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1698 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1699 * through all of its subrules and revalidates them, destroying any that no
1700 * longer has a super-rule (which is probably all of them).
1702 * Before calling this function, the caller must make have removed 'rule' from
1703 * the classifier. If 'rule' is an exact-match rule, the caller is also
1704 * responsible for ensuring that it has been uninstalled from the datapath. */
1706 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1709 struct rule *subrule, *next;
1710 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1711 revalidate_rule(ofproto, subrule);
1714 list_remove(&rule->list);
1720 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1722 const union ofp_action *oa;
1723 struct actions_iterator i;
1725 if (out_port == htons(OFPP_NONE)) {
1728 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1729 oa = actions_next(&i)) {
1730 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1737 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1738 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1740 * The flow that 'packet' actually contains does not need to actually match
1741 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1742 * the packet and byte counters for 'rule' will be credited for the packet sent
1743 * out whether or not the packet actually matches 'rule'.
1745 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1746 * the caller must already have accurately composed ODP actions for it given
1747 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1748 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1749 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1750 * actions and apply them to 'packet'. */
1752 rule_execute(struct ofproto *ofproto, struct rule *rule,
1753 struct ofpbuf *packet, const flow_t *flow)
1755 const union odp_action *actions;
1757 struct odp_actions a;
1759 /* Grab or compose the ODP actions.
1761 * The special case for an exact-match 'rule' where 'flow' is not the
1762 * rule's flow is important to avoid, e.g., sending a packet out its input
1763 * port simply because the ODP actions were composed for the wrong
1765 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1766 struct rule *super = rule->super ? rule->super : rule;
1767 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1768 packet, &a, NULL, 0, NULL)) {
1771 actions = a.actions;
1772 n_actions = a.n_actions;
1774 actions = rule->odp_actions;
1775 n_actions = rule->n_odp_actions;
1778 /* Execute the ODP actions. */
1779 if (!dpif_execute(ofproto->dpif, flow->in_port,
1780 actions, n_actions, packet)) {
1781 struct odp_flow_stats stats;
1782 flow_extract_stats(flow, packet, &stats);
1783 update_stats(ofproto, rule, &stats);
1784 rule->used = time_msec();
1785 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1790 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1793 struct rule *displaced_rule;
1795 /* Insert the rule in the classifier. */
1796 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1797 if (!rule->cr.wc.wildcards) {
1798 rule_make_actions(p, rule, packet);
1801 /* Send the packet and credit it to the rule. */
1804 flow_extract(packet, 0, in_port, &flow);
1805 rule_execute(p, rule, packet, &flow);
1808 /* Install the rule in the datapath only after sending the packet, to
1809 * avoid packet reordering. */
1810 if (rule->cr.wc.wildcards) {
1811 COVERAGE_INC(ofproto_add_wc_flow);
1812 p->need_revalidate = true;
1814 rule_install(p, rule, displaced_rule);
1817 /* Free the rule that was displaced, if any. */
1818 if (displaced_rule) {
1819 rule_destroy(p, displaced_rule);
1823 static struct rule *
1824 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1827 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1828 rule->idle_timeout, rule->hard_timeout,
1830 COVERAGE_INC(ofproto_subrule_create);
1831 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1832 : rule->cr.priority), &subrule->cr);
1833 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1839 rule_remove(struct ofproto *ofproto, struct rule *rule)
1841 if (rule->cr.wc.wildcards) {
1842 COVERAGE_INC(ofproto_del_wc_flow);
1843 ofproto->need_revalidate = true;
1845 rule_uninstall(ofproto, rule);
1847 classifier_remove(&ofproto->cls, &rule->cr);
1848 rule_destroy(ofproto, rule);
1851 /* Returns true if the actions changed, false otherwise. */
1853 rule_make_actions(struct ofproto *p, struct rule *rule,
1854 const struct ofpbuf *packet)
1856 const struct rule *super;
1857 struct odp_actions a;
1860 assert(!rule->cr.wc.wildcards);
1862 super = rule->super ? rule->super : rule;
1864 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1865 packet, &a, &rule->tags, &rule->may_install,
1866 &rule->nf_flow.output_iface);
1868 actions_len = a.n_actions * sizeof *a.actions;
1869 if (rule->n_odp_actions != a.n_actions
1870 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1871 COVERAGE_INC(ofproto_odp_unchanged);
1872 free(rule->odp_actions);
1873 rule->n_odp_actions = a.n_actions;
1874 rule->odp_actions = xmemdup(a.actions, actions_len);
1882 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1883 struct odp_flow_put *put)
1885 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1886 put->flow.key = rule->cr.flow;
1887 put->flow.actions = rule->odp_actions;
1888 put->flow.n_actions = rule->n_odp_actions;
1889 put->flow.flags = 0;
1891 return dpif_flow_put(ofproto->dpif, put);
1895 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1897 assert(!rule->cr.wc.wildcards);
1899 if (rule->may_install) {
1900 struct odp_flow_put put;
1901 if (!do_put_flow(p, rule,
1902 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1904 rule->installed = true;
1905 if (displaced_rule) {
1906 update_stats(p, displaced_rule, &put.flow.stats);
1907 rule_post_uninstall(p, displaced_rule);
1910 } else if (displaced_rule) {
1911 rule_uninstall(p, displaced_rule);
1916 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1918 if (rule->installed) {
1919 struct odp_flow_put put;
1920 COVERAGE_INC(ofproto_dp_missed);
1921 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1923 rule_install(ofproto, rule, NULL);
1928 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1930 bool actions_changed;
1931 uint16_t new_out_iface, old_out_iface;
1933 old_out_iface = rule->nf_flow.output_iface;
1934 actions_changed = rule_make_actions(ofproto, rule, NULL);
1936 if (rule->may_install) {
1937 if (rule->installed) {
1938 if (actions_changed) {
1939 struct odp_flow_put put;
1940 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1941 | ODPPF_ZERO_STATS, &put);
1942 update_stats(ofproto, rule, &put.flow.stats);
1944 /* Temporarily set the old output iface so that NetFlow
1945 * messages have the correct output interface for the old
1947 new_out_iface = rule->nf_flow.output_iface;
1948 rule->nf_flow.output_iface = old_out_iface;
1949 rule_post_uninstall(ofproto, rule);
1950 rule->nf_flow.output_iface = new_out_iface;
1953 rule_install(ofproto, rule, NULL);
1956 rule_uninstall(ofproto, rule);
1961 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1963 uint64_t total_bytes = rule->byte_count + extra_bytes;
1965 if (ofproto->ofhooks->account_flow_cb
1966 && total_bytes > rule->accounted_bytes)
1968 ofproto->ofhooks->account_flow_cb(
1969 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1970 total_bytes - rule->accounted_bytes, ofproto->aux);
1971 rule->accounted_bytes = total_bytes;
1976 rule_uninstall(struct ofproto *p, struct rule *rule)
1978 assert(!rule->cr.wc.wildcards);
1979 if (rule->installed) {
1980 struct odp_flow odp_flow;
1982 odp_flow.key = rule->cr.flow;
1983 odp_flow.actions = NULL;
1984 odp_flow.n_actions = 0;
1986 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1987 update_stats(p, rule, &odp_flow.stats);
1989 rule->installed = false;
1991 rule_post_uninstall(p, rule);
1996 is_controller_rule(struct rule *rule)
1998 /* If the only action is send to the controller then don't report
1999 * NetFlow expiration messages since it is just part of the control
2000 * logic for the network and not real traffic. */
2002 if (rule && rule->super) {
2003 struct rule *super = rule->super;
2005 return super->n_actions == 1 &&
2006 super->actions[0].type == htons(OFPAT_OUTPUT) &&
2007 super->actions[0].output.port == htons(OFPP_CONTROLLER);
2014 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2016 struct rule *super = rule->super;
2018 rule_account(ofproto, rule, 0);
2020 if (ofproto->netflow && !is_controller_rule(rule)) {
2021 struct ofexpired expired;
2022 expired.flow = rule->cr.flow;
2023 expired.packet_count = rule->packet_count;
2024 expired.byte_count = rule->byte_count;
2025 expired.used = rule->used;
2026 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2029 super->packet_count += rule->packet_count;
2030 super->byte_count += rule->byte_count;
2032 /* Reset counters to prevent double counting if the rule ever gets
2034 rule->packet_count = 0;
2035 rule->byte_count = 0;
2036 rule->accounted_bytes = 0;
2038 netflow_flow_clear(&rule->nf_flow);
2043 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2044 struct rconn_packet_counter *counter)
2046 update_openflow_length(msg);
2047 if (rconn_send(ofconn->rconn, msg, counter)) {
2053 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2054 int error, const void *data, size_t len)
2057 struct ofp_error_msg *oem;
2059 if (!(error >> 16)) {
2060 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2065 COVERAGE_INC(ofproto_error);
2066 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2067 oh ? oh->xid : 0, &buf);
2068 oem->type = htons((unsigned int) error >> 16);
2069 oem->code = htons(error & 0xffff);
2070 memcpy(oem->data, data, len);
2071 queue_tx(buf, ofconn, ofconn->reply_counter);
2075 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2078 size_t oh_length = ntohs(oh->length);
2079 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2083 hton_ofp_phy_port(struct ofp_phy_port *opp)
2085 opp->port_no = htons(opp->port_no);
2086 opp->config = htonl(opp->config);
2087 opp->state = htonl(opp->state);
2088 opp->curr = htonl(opp->curr);
2089 opp->advertised = htonl(opp->advertised);
2090 opp->supported = htonl(opp->supported);
2091 opp->peer = htonl(opp->peer);
2095 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2097 struct ofp_header *rq = oh;
2098 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2103 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2104 struct ofp_header *oh)
2106 struct ofp_switch_features *osf;
2108 unsigned int port_no;
2109 struct ofport *port;
2111 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2112 osf->datapath_id = htonll(p->datapath_id);
2113 osf->n_buffers = htonl(pktbuf_capacity());
2115 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2116 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2117 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2118 (1u << OFPAT_SET_VLAN_VID) |
2119 (1u << OFPAT_SET_VLAN_PCP) |
2120 (1u << OFPAT_STRIP_VLAN) |
2121 (1u << OFPAT_SET_DL_SRC) |
2122 (1u << OFPAT_SET_DL_DST) |
2123 (1u << OFPAT_SET_NW_SRC) |
2124 (1u << OFPAT_SET_NW_DST) |
2125 (1u << OFPAT_SET_NW_TOS) |
2126 (1u << OFPAT_SET_TP_SRC) |
2127 (1u << OFPAT_SET_TP_DST));
2129 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2130 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2133 queue_tx(buf, ofconn, ofconn->reply_counter);
2138 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2139 struct ofp_header *oh)
2142 struct ofp_switch_config *osc;
2146 /* Figure out flags. */
2147 dpif_get_drop_frags(p->dpif, &drop_frags);
2148 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2151 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2152 osc->flags = htons(flags);
2153 osc->miss_send_len = htons(ofconn->miss_send_len);
2154 queue_tx(buf, ofconn, ofconn->reply_counter);
2160 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2161 struct ofp_switch_config *osc)
2166 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2170 flags = ntohs(osc->flags);
2172 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2173 switch (flags & OFPC_FRAG_MASK) {
2174 case OFPC_FRAG_NORMAL:
2175 dpif_set_drop_frags(p->dpif, false);
2177 case OFPC_FRAG_DROP:
2178 dpif_set_drop_frags(p->dpif, true);
2181 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2187 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2193 add_output_group_action(struct odp_actions *actions, uint16_t group,
2194 uint16_t *nf_output_iface)
2196 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2198 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2199 *nf_output_iface = NF_OUT_FLOOD;
2204 add_controller_action(struct odp_actions *actions,
2205 const struct ofp_action_output *oao)
2207 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2208 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
2211 struct action_xlate_ctx {
2213 flow_t flow; /* Flow to which these actions correspond. */
2214 int recurse; /* Recursion level, via xlate_table_action. */
2215 struct ofproto *ofproto;
2216 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2217 * null pointer if we are revalidating
2218 * without a packet to refer to. */
2221 struct odp_actions *out; /* Datapath actions. */
2222 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2223 bool may_set_up_flow; /* True ordinarily; false if the actions must
2224 * be reassessed for every packet. */
2225 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2228 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2229 struct action_xlate_ctx *ctx);
2232 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2234 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2237 if (ofport->opp.config & OFPPC_NO_FWD) {
2238 /* Forwarding disabled on port. */
2243 * We don't have an ofport record for this port, but it doesn't hurt to
2244 * allow forwarding to it anyhow. Maybe such a port will appear later
2245 * and we're pre-populating the flow table.
2249 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2250 ctx->nf_output_iface = port;
2253 static struct rule *
2254 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2257 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2259 /* The rule we found might not be valid, since we could be in need of
2260 * revalidation. If it is not valid, don't return it. */
2263 && ofproto->need_revalidate
2264 && !revalidate_rule(ofproto, rule)) {
2265 COVERAGE_INC(ofproto_invalidated);
2273 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2275 if (!ctx->recurse) {
2276 uint16_t old_in_port;
2279 /* Look up a flow with 'in_port' as the input port. Then restore the
2280 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2281 * have surprising behavior). */
2282 old_in_port = ctx->flow.in_port;
2283 ctx->flow.in_port = in_port;
2284 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2285 ctx->flow.in_port = old_in_port;
2293 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2300 xlate_output_action(struct action_xlate_ctx *ctx,
2301 const struct ofp_action_output *oao)
2304 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2306 ctx->nf_output_iface = NF_OUT_DROP;
2308 switch (ntohs(oao->port)) {
2310 add_output_action(ctx, ctx->flow.in_port);
2313 xlate_table_action(ctx, ctx->flow.in_port);
2316 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2317 ctx->out, ctx->tags,
2318 &ctx->nf_output_iface,
2319 ctx->ofproto->aux)) {
2320 COVERAGE_INC(ofproto_uninstallable);
2321 ctx->may_set_up_flow = false;
2325 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2326 &ctx->nf_output_iface);
2329 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2331 case OFPP_CONTROLLER:
2332 add_controller_action(ctx->out, oao);
2335 add_output_action(ctx, ODPP_LOCAL);
2338 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2339 if (odp_port != ctx->flow.in_port) {
2340 add_output_action(ctx, odp_port);
2345 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2346 ctx->nf_output_iface = NF_OUT_FLOOD;
2347 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2348 ctx->nf_output_iface = prev_nf_output_iface;
2349 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2350 ctx->nf_output_iface != NF_OUT_FLOOD) {
2351 ctx->nf_output_iface = NF_OUT_MULTI;
2356 xlate_nicira_action(struct action_xlate_ctx *ctx,
2357 const struct nx_action_header *nah)
2359 const struct nx_action_resubmit *nar;
2360 const struct nx_action_set_tunnel *nast;
2361 union odp_action *oa;
2362 int subtype = ntohs(nah->subtype);
2364 assert(nah->vendor == htonl(NX_VENDOR_ID));
2366 case NXAST_RESUBMIT:
2367 nar = (const struct nx_action_resubmit *) nah;
2368 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2371 case NXAST_SET_TUNNEL:
2372 nast = (const struct nx_action_set_tunnel *) nah;
2373 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2374 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2377 /* If you add a new action here that modifies flow data, don't forget to
2378 * update the flow key in ctx->flow in the same key. */
2381 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2387 do_xlate_actions(const union ofp_action *in, size_t n_in,
2388 struct action_xlate_ctx *ctx)
2390 struct actions_iterator iter;
2391 const union ofp_action *ia;
2392 const struct ofport *port;
2394 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2395 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2396 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2397 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2398 /* Drop this flow. */
2402 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2403 uint16_t type = ntohs(ia->type);
2404 union odp_action *oa;
2408 xlate_output_action(ctx, &ia->output);
2411 case OFPAT_SET_VLAN_VID:
2412 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2413 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2416 case OFPAT_SET_VLAN_PCP:
2417 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2418 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2421 case OFPAT_STRIP_VLAN:
2422 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2423 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2424 ctx->flow.dl_vlan_pcp = 0;
2427 case OFPAT_SET_DL_SRC:
2428 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2429 memcpy(oa->dl_addr.dl_addr,
2430 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2431 memcpy(ctx->flow.dl_src,
2432 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2435 case OFPAT_SET_DL_DST:
2436 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2437 memcpy(oa->dl_addr.dl_addr,
2438 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2439 memcpy(ctx->flow.dl_dst,
2440 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2443 case OFPAT_SET_NW_SRC:
2444 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2445 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2448 case OFPAT_SET_NW_DST:
2449 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2450 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2453 case OFPAT_SET_NW_TOS:
2454 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2455 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2458 case OFPAT_SET_TP_SRC:
2459 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2460 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2463 case OFPAT_SET_TP_DST:
2464 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2465 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2469 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2473 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2480 xlate_actions(const union ofp_action *in, size_t n_in,
2481 const flow_t *flow, struct ofproto *ofproto,
2482 const struct ofpbuf *packet,
2483 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2484 uint16_t *nf_output_iface)
2486 tag_type no_tags = 0;
2487 struct action_xlate_ctx ctx;
2488 COVERAGE_INC(ofproto_ofp2odp);
2489 odp_actions_init(out);
2492 ctx.ofproto = ofproto;
2493 ctx.packet = packet;
2495 ctx.tags = tags ? tags : &no_tags;
2496 ctx.may_set_up_flow = true;
2497 ctx.nf_output_iface = NF_OUT_DROP;
2498 do_xlate_actions(in, n_in, &ctx);
2500 /* Check with in-band control to see if we're allowed to set up this
2502 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2503 ctx.may_set_up_flow = false;
2506 if (may_set_up_flow) {
2507 *may_set_up_flow = ctx.may_set_up_flow;
2509 if (nf_output_iface) {
2510 *nf_output_iface = ctx.nf_output_iface;
2512 if (odp_actions_overflow(out)) {
2513 odp_actions_init(out);
2514 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2519 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2520 * error message code (composed with ofp_mkerr()) for the caller to propagate
2521 * upward. Otherwise, returns 0.
2523 * 'oh' is used to make log messages more informative. */
2525 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2527 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2528 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2531 type_name = ofp_message_type_to_string(oh->type);
2532 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2536 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2543 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2544 struct ofp_header *oh)
2546 struct ofp_packet_out *opo;
2547 struct ofpbuf payload, *buffer;
2548 struct odp_actions actions;
2554 error = reject_slave_controller(ofconn, oh);
2559 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2563 opo = (struct ofp_packet_out *) oh;
2565 COVERAGE_INC(ofproto_packet_out);
2566 if (opo->buffer_id != htonl(UINT32_MAX)) {
2567 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2569 if (error || !buffer) {
2577 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2578 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2579 &flow, p, &payload, &actions, NULL, NULL, NULL);
2584 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2586 ofpbuf_delete(buffer);
2592 update_port_config(struct ofproto *p, struct ofport *port,
2593 uint32_t config, uint32_t mask)
2595 mask &= config ^ port->opp.config;
2596 if (mask & OFPPC_PORT_DOWN) {
2597 if (config & OFPPC_PORT_DOWN) {
2598 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2600 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2603 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2604 if (mask & REVALIDATE_BITS) {
2605 COVERAGE_INC(ofproto_costly_flags);
2606 port->opp.config ^= mask & REVALIDATE_BITS;
2607 p->need_revalidate = true;
2609 #undef REVALIDATE_BITS
2610 if (mask & OFPPC_NO_FLOOD) {
2611 port->opp.config ^= OFPPC_NO_FLOOD;
2612 refresh_port_groups(p);
2614 if (mask & OFPPC_NO_PACKET_IN) {
2615 port->opp.config ^= OFPPC_NO_PACKET_IN;
2620 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2621 struct ofp_header *oh)
2623 const struct ofp_port_mod *opm;
2624 struct ofport *port;
2627 error = reject_slave_controller(ofconn, oh);
2631 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2635 opm = (struct ofp_port_mod *) oh;
2637 port = port_array_get(&p->ports,
2638 ofp_port_to_odp_port(ntohs(opm->port_no)));
2640 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2641 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2642 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2644 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2645 if (opm->advertise) {
2646 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2652 static struct ofpbuf *
2653 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2655 struct ofp_stats_reply *osr;
2658 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2659 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2661 osr->flags = htons(0);
2665 static struct ofpbuf *
2666 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2668 return make_stats_reply(request->header.xid, request->type, body_len);
2672 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2674 struct ofpbuf *msg = *msgp;
2675 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2676 if (nbytes + msg->size > UINT16_MAX) {
2677 struct ofp_stats_reply *reply = msg->data;
2678 reply->flags = htons(OFPSF_REPLY_MORE);
2679 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2680 queue_tx(msg, ofconn, ofconn->reply_counter);
2682 return ofpbuf_put_uninit(*msgp, nbytes);
2686 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2687 struct ofp_stats_request *request)
2689 struct ofp_desc_stats *ods;
2692 msg = start_stats_reply(request, sizeof *ods);
2693 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2694 memset(ods, 0, sizeof *ods);
2695 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2696 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2697 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2698 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2699 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2700 queue_tx(msg, ofconn, ofconn->reply_counter);
2706 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2708 struct rule *rule = rule_from_cls_rule(cls_rule);
2709 int *n_subrules = n_subrules_;
2717 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2718 struct ofp_stats_request *request)
2720 struct ofp_table_stats *ots;
2722 struct odp_stats dpstats;
2723 int n_exact, n_subrules, n_wild;
2725 msg = start_stats_reply(request, sizeof *ots * 2);
2727 /* Count rules of various kinds. */
2729 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2730 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2731 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2734 dpif_get_dp_stats(p->dpif, &dpstats);
2735 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2736 memset(ots, 0, sizeof *ots);
2737 ots->table_id = TABLEID_HASH;
2738 strcpy(ots->name, "hash");
2739 ots->wildcards = htonl(0);
2740 ots->max_entries = htonl(dpstats.max_capacity);
2741 ots->active_count = htonl(n_exact);
2742 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2744 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2746 /* Classifier table. */
2747 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2748 memset(ots, 0, sizeof *ots);
2749 ots->table_id = TABLEID_CLASSIFIER;
2750 strcpy(ots->name, "classifier");
2751 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2753 ots->max_entries = htonl(65536);
2754 ots->active_count = htonl(n_wild);
2755 ots->lookup_count = htonll(0); /* XXX */
2756 ots->matched_count = htonll(0); /* XXX */
2758 queue_tx(msg, ofconn, ofconn->reply_counter);
2763 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2764 struct ofpbuf **msgp)
2766 struct netdev_stats stats;
2767 struct ofp_port_stats *ops;
2769 /* Intentionally ignore return value, since errors will set
2770 * 'stats' to all-1s, which is correct for OpenFlow, and
2771 * netdev_get_stats() will log errors. */
2772 netdev_get_stats(port->netdev, &stats);
2774 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2775 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2776 memset(ops->pad, 0, sizeof ops->pad);
2777 ops->rx_packets = htonll(stats.rx_packets);
2778 ops->tx_packets = htonll(stats.tx_packets);
2779 ops->rx_bytes = htonll(stats.rx_bytes);
2780 ops->tx_bytes = htonll(stats.tx_bytes);
2781 ops->rx_dropped = htonll(stats.rx_dropped);
2782 ops->tx_dropped = htonll(stats.tx_dropped);
2783 ops->rx_errors = htonll(stats.rx_errors);
2784 ops->tx_errors = htonll(stats.tx_errors);
2785 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2786 ops->rx_over_err = htonll(stats.rx_over_errors);
2787 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2788 ops->collisions = htonll(stats.collisions);
2792 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2793 struct ofp_stats_request *osr,
2796 struct ofp_port_stats_request *psr;
2797 struct ofp_port_stats *ops;
2799 struct ofport *port;
2800 unsigned int port_no;
2802 if (arg_size != sizeof *psr) {
2803 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2805 psr = (struct ofp_port_stats_request *) osr->body;
2807 msg = start_stats_reply(osr, sizeof *ops * 16);
2808 if (psr->port_no != htons(OFPP_NONE)) {
2809 port = port_array_get(&p->ports,
2810 ofp_port_to_odp_port(ntohs(psr->port_no)));
2812 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2815 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2816 append_port_stat(port, port_no, ofconn, &msg);
2820 queue_tx(msg, ofconn, ofconn->reply_counter);
2824 struct flow_stats_cbdata {
2825 struct ofproto *ofproto;
2826 struct ofconn *ofconn;
2831 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2832 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2833 * returned statistic include statistics for all of 'rule''s subrules. */
2835 query_stats(struct ofproto *p, struct rule *rule,
2836 uint64_t *packet_countp, uint64_t *byte_countp)
2838 uint64_t packet_count, byte_count;
2839 struct rule *subrule;
2840 struct odp_flow *odp_flows;
2843 /* Start from historical data for 'rule' itself that are no longer tracked
2844 * by the datapath. This counts, for example, subrules that have
2846 packet_count = rule->packet_count;
2847 byte_count = rule->byte_count;
2849 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2850 * wildcarded then on all of its subrules.
2852 * Also, add any statistics that are not tracked by the datapath for each
2853 * subrule. This includes, for example, statistics for packets that were
2854 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2856 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2857 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2858 if (rule->cr.wc.wildcards) {
2860 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2861 odp_flows[i++].key = subrule->cr.flow;
2862 packet_count += subrule->packet_count;
2863 byte_count += subrule->byte_count;
2866 odp_flows[0].key = rule->cr.flow;
2869 /* Fetch up-to-date statistics from the datapath and add them in. */
2870 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2872 for (i = 0; i < n_odp_flows; i++) {
2873 struct odp_flow *odp_flow = &odp_flows[i];
2874 packet_count += odp_flow->stats.n_packets;
2875 byte_count += odp_flow->stats.n_bytes;
2880 /* Return the stats to the caller. */
2881 *packet_countp = packet_count;
2882 *byte_countp = byte_count;
2886 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2888 struct rule *rule = rule_from_cls_rule(rule_);
2889 struct flow_stats_cbdata *cbdata = cbdata_;
2890 struct ofp_flow_stats *ofs;
2891 uint64_t packet_count, byte_count;
2892 size_t act_len, len;
2893 long long int tdiff = time_msec() - rule->created;
2894 uint32_t sec = tdiff / 1000;
2895 uint32_t msec = tdiff - (sec * 1000);
2897 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2901 act_len = sizeof *rule->actions * rule->n_actions;
2902 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2904 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2906 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2907 ofs->length = htons(len);
2908 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2910 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2911 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2912 ofs->duration_sec = htonl(sec);
2913 ofs->duration_nsec = htonl(msec * 1000000);
2914 ofs->cookie = rule->flow_cookie;
2915 ofs->priority = htons(rule->cr.priority);
2916 ofs->idle_timeout = htons(rule->idle_timeout);
2917 ofs->hard_timeout = htons(rule->hard_timeout);
2918 memset(ofs->pad2, 0, sizeof ofs->pad2);
2919 ofs->packet_count = htonll(packet_count);
2920 ofs->byte_count = htonll(byte_count);
2921 memcpy(ofs->actions, rule->actions, act_len);
2925 table_id_to_include(uint8_t table_id)
2927 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2928 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2929 : table_id == 0xff ? CLS_INC_ALL
2934 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2935 const struct ofp_stats_request *osr,
2938 struct ofp_flow_stats_request *fsr;
2939 struct flow_stats_cbdata cbdata;
2940 struct cls_rule target;
2942 if (arg_size != sizeof *fsr) {
2943 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2945 fsr = (struct ofp_flow_stats_request *) osr->body;
2947 COVERAGE_INC(ofproto_flows_req);
2949 cbdata.ofconn = ofconn;
2950 cbdata.out_port = fsr->out_port;
2951 cbdata.msg = start_stats_reply(osr, 1024);
2952 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
2953 classifier_for_each_match(&p->cls, &target,
2954 table_id_to_include(fsr->table_id),
2955 flow_stats_cb, &cbdata);
2956 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2960 struct flow_stats_ds_cbdata {
2961 struct ofproto *ofproto;
2966 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2968 struct rule *rule = rule_from_cls_rule(rule_);
2969 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2970 struct ds *results = cbdata->results;
2971 struct ofp_match match;
2972 uint64_t packet_count, byte_count;
2973 size_t act_len = sizeof *rule->actions * rule->n_actions;
2975 /* Don't report on subrules. */
2976 if (rule->super != NULL) {
2980 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2981 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2982 cbdata->ofproto->tun_id_from_cookie, &match);
2984 ds_put_format(results, "duration=%llds, ",
2985 (time_msec() - rule->created) / 1000);
2986 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2987 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2988 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2989 ofp_print_match(results, &match, true);
2990 ofp_print_actions(results, &rule->actions->header, act_len);
2991 ds_put_cstr(results, "\n");
2994 /* Adds a pretty-printed description of all flows to 'results', including
2995 * those marked hidden by secchan (e.g., by in-band control). */
2997 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2999 struct ofp_match match;
3000 struct cls_rule target;
3001 struct flow_stats_ds_cbdata cbdata;
3003 memset(&match, 0, sizeof match);
3004 match.wildcards = htonl(OVSFW_ALL);
3007 cbdata.results = results;
3009 cls_rule_from_match(&match, 0, false, 0, &target);
3010 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3011 flow_stats_ds_cb, &cbdata);
3014 struct aggregate_stats_cbdata {
3015 struct ofproto *ofproto;
3017 uint64_t packet_count;
3018 uint64_t byte_count;
3023 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3025 struct rule *rule = rule_from_cls_rule(rule_);
3026 struct aggregate_stats_cbdata *cbdata = cbdata_;
3027 uint64_t packet_count, byte_count;
3029 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3033 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3035 cbdata->packet_count += packet_count;
3036 cbdata->byte_count += byte_count;
3041 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3042 const struct ofp_stats_request *osr,
3045 struct ofp_aggregate_stats_request *asr;
3046 struct ofp_aggregate_stats_reply *reply;
3047 struct aggregate_stats_cbdata cbdata;
3048 struct cls_rule target;
3051 if (arg_size != sizeof *asr) {
3052 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3054 asr = (struct ofp_aggregate_stats_request *) osr->body;
3056 COVERAGE_INC(ofproto_agg_request);
3058 cbdata.out_port = asr->out_port;
3059 cbdata.packet_count = 0;
3060 cbdata.byte_count = 0;
3062 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3063 classifier_for_each_match(&p->cls, &target,
3064 table_id_to_include(asr->table_id),
3065 aggregate_stats_cb, &cbdata);
3067 msg = start_stats_reply(osr, sizeof *reply);
3068 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3069 reply->flow_count = htonl(cbdata.n_flows);
3070 reply->packet_count = htonll(cbdata.packet_count);
3071 reply->byte_count = htonll(cbdata.byte_count);
3072 queue_tx(msg, ofconn, ofconn->reply_counter);
3077 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3078 struct ofp_header *oh)
3080 struct ofp_stats_request *osr;
3084 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3089 osr = (struct ofp_stats_request *) oh;
3091 switch (ntohs(osr->type)) {
3093 return handle_desc_stats_request(p, ofconn, osr);
3096 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3098 case OFPST_AGGREGATE:
3099 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3102 return handle_table_stats_request(p, ofconn, osr);
3105 return handle_port_stats_request(p, ofconn, osr, arg_size);
3108 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3111 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3115 static long long int
3116 msec_from_nsec(uint64_t sec, uint32_t nsec)
3118 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3122 update_time(struct ofproto *ofproto, struct rule *rule,
3123 const struct odp_flow_stats *stats)
3125 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3126 if (used > rule->used) {
3128 if (rule->super && used > rule->super->used) {
3129 rule->super->used = used;
3131 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3136 update_stats(struct ofproto *ofproto, struct rule *rule,
3137 const struct odp_flow_stats *stats)
3139 if (stats->n_packets) {
3140 update_time(ofproto, rule, stats);
3141 rule->packet_count += stats->n_packets;
3142 rule->byte_count += stats->n_bytes;
3143 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3148 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3149 * in which no matching flow already exists in the flow table.
3151 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3152 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3153 * code as encoded by ofp_mkerr() on failure.
3155 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3158 add_flow(struct ofproto *p, struct ofconn *ofconn,
3159 const struct ofp_flow_mod *ofm, size_t n_actions)
3161 struct ofpbuf *packet;
3166 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3170 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3172 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3173 ntohs(ofm->priority))) {
3174 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3178 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3179 n_actions, ntohs(ofm->idle_timeout),
3180 ntohs(ofm->hard_timeout), ofm->cookie,
3181 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3182 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3183 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3186 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3187 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3191 in_port = UINT16_MAX;
3194 rule_insert(p, rule, packet, in_port);
3195 ofpbuf_delete(packet);
3199 static struct rule *
3200 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3205 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3207 return rule_from_cls_rule(classifier_find_rule_exactly(
3208 &p->cls, &flow, wildcards,
3209 ntohs(ofm->priority)));
3213 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3214 struct rule *rule, const struct ofp_flow_mod *ofm)
3216 struct ofpbuf *packet;
3221 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3225 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3231 flow_extract(packet, 0, in_port, &flow);
3232 rule_execute(ofproto, rule, packet, &flow);
3233 ofpbuf_delete(packet);
3238 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3240 struct modify_flows_cbdata {
3241 struct ofproto *ofproto;
3242 const struct ofp_flow_mod *ofm;
3247 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3248 size_t n_actions, struct rule *);
3249 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3251 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3252 * encoded by ofp_mkerr() on failure.
3254 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3257 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3258 const struct ofp_flow_mod *ofm, size_t n_actions)
3260 struct modify_flows_cbdata cbdata;
3261 struct cls_rule target;
3265 cbdata.n_actions = n_actions;
3266 cbdata.match = NULL;
3268 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3271 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3272 modify_flows_cb, &cbdata);
3274 /* This credits the packet to whichever flow happened to happened to
3275 * match last. That's weird. Maybe we should do a lookup for the
3276 * flow that actually matches the packet? Who knows. */
3277 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3280 return add_flow(p, ofconn, ofm, n_actions);
3284 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3285 * code as encoded by ofp_mkerr() on failure.
3287 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3290 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3291 struct ofp_flow_mod *ofm, size_t n_actions)
3293 struct rule *rule = find_flow_strict(p, ofm);
3294 if (rule && !rule_is_hidden(rule)) {
3295 modify_flow(p, ofm, n_actions, rule);
3296 return send_buffered_packet(p, ofconn, rule, ofm);
3298 return add_flow(p, ofconn, ofm, n_actions);
3302 /* Callback for modify_flows_loose(). */
3304 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3306 struct rule *rule = rule_from_cls_rule(rule_);
3307 struct modify_flows_cbdata *cbdata = cbdata_;
3309 if (!rule_is_hidden(rule)) {
3310 cbdata->match = rule;
3311 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3315 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3316 * been identified as a flow in 'p''s flow table to be modified, by changing
3317 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3318 * ofp_action[] structures). */
3320 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3321 size_t n_actions, struct rule *rule)
3323 size_t actions_len = n_actions * sizeof *rule->actions;
3325 rule->flow_cookie = ofm->cookie;
3327 /* If the actions are the same, do nothing. */
3328 if (n_actions == rule->n_actions
3329 && !memcmp(ofm->actions, rule->actions, actions_len))
3334 /* Replace actions. */
3335 free(rule->actions);
3336 rule->actions = xmemdup(ofm->actions, actions_len);
3337 rule->n_actions = n_actions;
3339 /* Make sure that the datapath gets updated properly. */
3340 if (rule->cr.wc.wildcards) {
3341 COVERAGE_INC(ofproto_mod_wc_flow);
3342 p->need_revalidate = true;
3344 rule_update_actions(p, rule);
3350 /* OFPFC_DELETE implementation. */
3352 struct delete_flows_cbdata {
3353 struct ofproto *ofproto;
3357 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3358 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3360 /* Implements OFPFC_DELETE. */
3362 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3364 struct delete_flows_cbdata cbdata;
3365 struct cls_rule target;
3368 cbdata.out_port = ofm->out_port;
3370 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3373 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3374 delete_flows_cb, &cbdata);
3377 /* Implements OFPFC_DELETE_STRICT. */
3379 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3381 struct rule *rule = find_flow_strict(p, ofm);
3383 delete_flow(p, rule, ofm->out_port);
3387 /* Callback for delete_flows_loose(). */
3389 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3391 struct rule *rule = rule_from_cls_rule(rule_);
3392 struct delete_flows_cbdata *cbdata = cbdata_;
3394 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3397 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3398 * been identified as a flow to delete from 'p''s flow table, by deleting the
3399 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3402 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3403 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3404 * specified 'out_port'. */
3406 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3408 if (rule_is_hidden(rule)) {
3412 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3416 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3417 rule_remove(p, rule);
3421 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3422 struct ofp_flow_mod *ofm)
3427 error = reject_slave_controller(ofconn, &ofm->header);
3431 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3432 sizeof *ofm->actions, &n_actions);
3437 /* We do not support the emergency flow cache. It will hopefully
3438 * get dropped from OpenFlow in the near future. */
3439 if (ofm->flags & htons(OFPFF_EMERG)) {
3440 /* There isn't a good fit for an error code, so just state that the
3441 * flow table is full. */
3442 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3445 normalize_match(&ofm->match);
3446 if (!ofm->match.wildcards) {
3447 ofm->priority = htons(UINT16_MAX);
3450 error = validate_actions((const union ofp_action *) ofm->actions,
3451 n_actions, p->max_ports);
3456 switch (ntohs(ofm->command)) {
3458 return add_flow(p, ofconn, ofm, n_actions);
3461 return modify_flows_loose(p, ofconn, ofm, n_actions);
3463 case OFPFC_MODIFY_STRICT:
3464 return modify_flow_strict(p, ofconn, ofm, n_actions);
3467 delete_flows_loose(p, ofm);
3470 case OFPFC_DELETE_STRICT:
3471 delete_flow_strict(p, ofm);
3475 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3480 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3484 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3489 p->tun_id_from_cookie = !!msg->set;
3494 handle_role_request(struct ofproto *ofproto,
3495 struct ofconn *ofconn, struct nicira_header *msg)
3497 struct nx_role_request *nrr;
3498 struct nx_role_request *reply;
3502 if (ntohs(msg->header.length) != sizeof *nrr) {
3503 VLOG_WARN_RL(&rl, "received role request of length %zu (expected %zu)",
3504 ntohs(msg->header.length), sizeof *nrr);
3505 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3507 nrr = (struct nx_role_request *) msg;
3509 if (ofconn->type != OFCONN_CONTROLLER) {
3510 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3512 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3515 role = ntohl(nrr->role);
3516 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3517 && role != NX_ROLE_SLAVE) {
3518 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3520 /* There's no good error code for this. */
3521 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3524 if (role == NX_ROLE_MASTER) {
3525 struct ofconn *other;
3527 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3528 &ofproto->controllers) {
3529 if (other->role == NX_ROLE_MASTER) {
3530 other->role = NX_ROLE_SLAVE;
3534 ofconn->role = role;
3536 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3538 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3539 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3540 reply->role = htonl(role);
3541 queue_tx(buf, ofconn, ofconn->reply_counter);
3547 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3549 struct ofp_vendor_header *ovh = msg;
3550 struct nicira_header *nh;
3552 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3553 VLOG_WARN_RL(&rl, "received vendor message of length %zu "
3554 "(expected at least %zu)",
3555 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3556 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3558 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3559 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3561 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3562 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %zu "
3563 "(expected at least %zu)",
3564 ntohs(ovh->header.length), sizeof(struct nicira_header));
3565 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3569 switch (ntohl(nh->subtype)) {
3570 case NXT_STATUS_REQUEST:
3571 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3574 case NXT_TUN_ID_FROM_COOKIE:
3575 return handle_tun_id_from_cookie(p, msg);
3577 case NXT_ROLE_REQUEST:
3578 return handle_role_request(p, ofconn, msg);
3581 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3585 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3587 struct ofp_header *ob;
3590 /* Currently, everything executes synchronously, so we can just
3591 * immediately send the barrier reply. */
3592 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3593 queue_tx(buf, ofconn, ofconn->reply_counter);
3598 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3599 struct ofpbuf *ofp_msg)
3601 struct ofp_header *oh = ofp_msg->data;
3604 COVERAGE_INC(ofproto_recv_openflow);
3606 case OFPT_ECHO_REQUEST:
3607 error = handle_echo_request(ofconn, oh);
3610 case OFPT_ECHO_REPLY:
3614 case OFPT_FEATURES_REQUEST:
3615 error = handle_features_request(p, ofconn, oh);
3618 case OFPT_GET_CONFIG_REQUEST:
3619 error = handle_get_config_request(p, ofconn, oh);
3622 case OFPT_SET_CONFIG:
3623 error = handle_set_config(p, ofconn, ofp_msg->data);
3626 case OFPT_PACKET_OUT:
3627 error = handle_packet_out(p, ofconn, ofp_msg->data);
3631 error = handle_port_mod(p, ofconn, oh);
3635 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3638 case OFPT_STATS_REQUEST:
3639 error = handle_stats_request(p, ofconn, oh);
3643 error = handle_vendor(p, ofconn, ofp_msg->data);
3646 case OFPT_BARRIER_REQUEST:
3647 error = handle_barrier_request(ofconn, oh);
3651 if (VLOG_IS_WARN_ENABLED()) {
3652 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3653 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3656 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3661 send_error_oh(ofconn, ofp_msg->data, error);
3666 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3668 struct odp_msg *msg = packet->data;
3670 struct ofpbuf payload;
3673 payload.data = msg + 1;
3674 payload.size = msg->length - sizeof *msg;
3675 flow_extract(&payload, msg->arg, msg->port, &flow);
3677 /* Check with in-band control to see if this packet should be sent
3678 * to the local port regardless of the flow table. */
3679 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3680 union odp_action action;
3682 memset(&action, 0, sizeof(action));
3683 action.output.type = ODPAT_OUTPUT;
3684 action.output.port = ODPP_LOCAL;
3685 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3688 rule = lookup_valid_rule(p, &flow);
3690 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3691 struct ofport *port = port_array_get(&p->ports, msg->port);
3693 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3694 COVERAGE_INC(ofproto_no_packet_in);
3695 /* XXX install 'drop' flow entry */
3696 ofpbuf_delete(packet);
3700 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3703 COVERAGE_INC(ofproto_packet_in);
3704 send_packet_in(p, packet);
3708 if (rule->cr.wc.wildcards) {
3709 rule = rule_create_subrule(p, rule, &flow);
3710 rule_make_actions(p, rule, packet);
3712 if (!rule->may_install) {
3713 /* The rule is not installable, that is, we need to process every
3714 * packet, so process the current packet and set its actions into
3716 rule_make_actions(p, rule, packet);
3718 /* XXX revalidate rule if it needs it */
3722 rule_execute(p, rule, &payload, &flow);
3723 rule_reinstall(p, rule);
3725 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3727 * Extra-special case for fail-open mode.
3729 * We are in fail-open mode and the packet matched the fail-open rule,
3730 * but we are connected to a controller too. We should send the packet
3731 * up to the controller in the hope that it will try to set up a flow
3732 * and thereby allow us to exit fail-open.
3734 * See the top-level comment in fail-open.c for more information.
3736 send_packet_in(p, packet);
3738 ofpbuf_delete(packet);
3743 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3745 struct odp_msg *msg = packet->data;
3747 switch (msg->type) {
3748 case _ODPL_ACTION_NR:
3749 COVERAGE_INC(ofproto_ctlr_action);
3750 send_packet_in(p, packet);
3753 case _ODPL_SFLOW_NR:
3755 ofproto_sflow_received(p->sflow, msg);
3757 ofpbuf_delete(packet);
3761 handle_odp_miss_msg(p, packet);
3765 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3772 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3774 struct rule *sub = rule_from_cls_rule(sub_);
3775 struct revalidate_cbdata *cbdata = cbdata_;
3777 if (cbdata->revalidate_all
3778 || (cbdata->revalidate_subrules && sub->super)
3779 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3780 revalidate_rule(cbdata->ofproto, sub);
3785 revalidate_rule(struct ofproto *p, struct rule *rule)
3787 const flow_t *flow = &rule->cr.flow;
3789 COVERAGE_INC(ofproto_revalidate_rule);
3792 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3794 rule_remove(p, rule);
3796 } else if (super != rule->super) {
3797 COVERAGE_INC(ofproto_revalidate_moved);
3798 list_remove(&rule->list);
3799 list_push_back(&super->list, &rule->list);
3800 rule->super = super;
3801 rule->hard_timeout = super->hard_timeout;
3802 rule->idle_timeout = super->idle_timeout;
3803 rule->created = super->created;
3808 rule_update_actions(p, rule);
3812 static struct ofpbuf *
3813 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3814 long long int now, uint8_t reason)
3816 struct ofp_flow_removed *ofr;
3818 long long int tdiff = now - rule->created;
3819 uint32_t sec = tdiff / 1000;
3820 uint32_t msec = tdiff - (sec * 1000);
3822 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3823 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3825 ofr->cookie = rule->flow_cookie;
3826 ofr->priority = htons(rule->cr.priority);
3827 ofr->reason = reason;
3828 ofr->duration_sec = htonl(sec);
3829 ofr->duration_nsec = htonl(msec * 1000000);
3830 ofr->idle_timeout = htons(rule->idle_timeout);
3831 ofr->packet_count = htonll(rule->packet_count);
3832 ofr->byte_count = htonll(rule->byte_count);
3838 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3840 assert(rule->installed);
3841 assert(!rule->cr.wc.wildcards);
3844 rule_remove(ofproto, rule);
3846 rule_uninstall(ofproto, rule);
3851 send_flow_removed(struct ofproto *p, struct rule *rule,
3852 long long int now, uint8_t reason)
3854 struct ofconn *ofconn;
3855 struct ofconn *prev;
3856 struct ofpbuf *buf = NULL;
3858 /* We limit the maximum number of queued flow expirations it by accounting
3859 * them under the counter for replies. That works because preventing
3860 * OpenFlow requests from being processed also prevents new flows from
3861 * being added (and expiring). (It also prevents processing OpenFlow
3862 * requests that would not add new flows, so it is imperfect.) */
3865 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3866 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
3867 && ofconn->role != NX_ROLE_SLAVE) {
3869 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3871 buf = compose_flow_removed(p, rule, now, reason);
3877 queue_tx(buf, prev, prev->reply_counter);
3883 expire_rule(struct cls_rule *cls_rule, void *p_)
3885 struct ofproto *p = p_;
3886 struct rule *rule = rule_from_cls_rule(cls_rule);
3887 long long int hard_expire, idle_expire, expire, now;
3889 hard_expire = (rule->hard_timeout
3890 ? rule->created + rule->hard_timeout * 1000
3892 idle_expire = (rule->idle_timeout
3893 && (rule->super || list_is_empty(&rule->list))
3894 ? rule->used + rule->idle_timeout * 1000
3896 expire = MIN(hard_expire, idle_expire);
3900 if (rule->installed && now >= rule->used + 5000) {
3901 uninstall_idle_flow(p, rule);
3902 } else if (!rule->cr.wc.wildcards) {
3903 active_timeout(p, rule);
3909 COVERAGE_INC(ofproto_expired);
3911 /* Update stats. This code will be a no-op if the rule expired
3912 * due to an idle timeout. */
3913 if (rule->cr.wc.wildcards) {
3914 struct rule *subrule, *next;
3915 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3916 rule_remove(p, subrule);
3919 rule_uninstall(p, rule);
3922 if (!rule_is_hidden(rule)) {
3923 send_flow_removed(p, rule, now,
3925 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3927 rule_remove(p, rule);
3931 active_timeout(struct ofproto *ofproto, struct rule *rule)
3933 if (ofproto->netflow && !is_controller_rule(rule) &&
3934 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3935 struct ofexpired expired;
3936 struct odp_flow odp_flow;
3938 /* Get updated flow stats. */
3939 memset(&odp_flow, 0, sizeof odp_flow);
3940 if (rule->installed) {
3941 odp_flow.key = rule->cr.flow;
3942 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3943 dpif_flow_get(ofproto->dpif, &odp_flow);
3945 if (odp_flow.stats.n_packets) {
3946 update_time(ofproto, rule, &odp_flow.stats);
3947 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3948 odp_flow.stats.tcp_flags);
3952 expired.flow = rule->cr.flow;
3953 expired.packet_count = rule->packet_count +
3954 odp_flow.stats.n_packets;
3955 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3956 expired.used = rule->used;
3958 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3960 /* Schedule us to send the accumulated records once we have
3961 * collected all of them. */
3962 poll_immediate_wake();
3967 update_used(struct ofproto *p)
3969 struct odp_flow *flows;
3974 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3979 for (i = 0; i < n_flows; i++) {
3980 struct odp_flow *f = &flows[i];
3983 rule = rule_from_cls_rule(
3984 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3985 if (!rule || !rule->installed) {
3986 COVERAGE_INC(ofproto_unexpected_rule);
3987 dpif_flow_del(p->dpif, f);
3991 update_time(p, rule, &f->stats);
3992 rule_account(p, rule, f->stats.n_bytes);
3997 /* pinsched callback for sending 'packet' on 'ofconn'. */
3999 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4001 struct ofconn *ofconn = ofconn_;
4003 rconn_send_with_limit(ofconn->rconn, packet,
4004 ofconn->packet_in_counter, 100);
4007 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4008 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4009 * packet scheduler for sending.
4011 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4012 * ownership is transferred to this function. */
4014 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, bool clone)
4016 struct ofproto *ofproto = ofconn->ofproto;
4017 struct ofp_packet_in *opi = packet->data;
4018 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4019 int send_len, trim_size;
4023 if (opi->reason == OFPR_ACTION) {
4024 buffer_id = UINT32_MAX;
4025 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4026 buffer_id = pktbuf_get_null();
4028 struct ofpbuf payload;
4029 payload.data = opi->data;
4030 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4031 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4034 /* Figure out how much of the packet to send. */
4035 send_len = ntohs(opi->total_len);
4036 if (buffer_id != UINT32_MAX) {
4037 send_len = MIN(send_len, ofconn->miss_send_len);
4040 /* Adjust packet length and clone if necessary. */
4041 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4043 packet = ofpbuf_clone_data(packet->data, trim_size);
4046 packet->size = trim_size;
4049 /* Update packet headers. */
4050 opi->buffer_id = htonl(buffer_id);
4051 update_openflow_length(packet);
4053 /* Hand over to packet scheduler. It might immediately call into
4054 * do_send_packet_in() or it might buffer it for a while (until a later
4055 * call to pinsched_run()). */
4056 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4057 packet, do_send_packet_in, ofconn);
4060 /* Replace struct odp_msg header in 'packet' by equivalent struct
4061 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4062 * returned by dpif_recv()).
4064 * The conversion is not complete: the caller still needs to trim any unneeded
4065 * payload off the end of the buffer, set the length in the OpenFlow header,
4066 * and set buffer_id. Those require us to know the controller settings and so
4067 * must be done on a per-controller basis. */
4069 do_convert_to_packet_in(struct ofpbuf *packet)
4071 struct odp_msg *msg = packet->data;
4072 struct ofp_packet_in *opi;
4077 /* Extract relevant header fields */
4078 reason = (msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH);
4079 total_len = msg->length - sizeof *msg;
4080 in_port = odp_port_to_ofp_port(msg->port);
4082 /* Repurpose packet buffer by overwriting header. */
4083 ofpbuf_pull(packet, sizeof(struct odp_msg));
4084 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4085 opi->header.version = OFP_VERSION;
4086 opi->header.type = OFPT_PACKET_IN;
4087 opi->total_len = htons(total_len);
4088 opi->in_port = htons(in_port);
4089 opi->reason = reason;
4092 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4093 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4094 * as necessary according to their individual configurations.
4096 * 'packet' must have sufficient headroom to convert it into a struct
4097 * ofp_packet_in (e.g. as returned by dpif_recv()).
4099 * Takes ownership of 'packet'. */
4101 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4103 struct ofconn *ofconn, *prev;
4105 do_convert_to_packet_in(packet);
4108 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4109 if (ofconn->role != NX_ROLE_SLAVE) {
4111 schedule_packet_in(prev, packet, true);
4117 schedule_packet_in(prev, packet, false);
4119 ofpbuf_delete(packet);
4124 pick_datapath_id(const struct ofproto *ofproto)
4126 const struct ofport *port;
4128 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4130 uint8_t ea[ETH_ADDR_LEN];
4133 error = netdev_get_etheraddr(port->netdev, ea);
4135 return eth_addr_to_uint64(ea);
4137 VLOG_WARN("could not get MAC address for %s (%s)",
4138 netdev_get_name(port->netdev), strerror(error));
4140 return ofproto->fallback_dpid;
4144 pick_fallback_dpid(void)
4146 uint8_t ea[ETH_ADDR_LEN];
4147 eth_addr_nicira_random(ea);
4148 return eth_addr_to_uint64(ea);
4152 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4153 struct odp_actions *actions, tag_type *tags,
4154 uint16_t *nf_output_iface, void *ofproto_)
4156 struct ofproto *ofproto = ofproto_;
4159 /* Drop frames for reserved multicast addresses. */
4160 if (eth_addr_is_reserved(flow->dl_dst)) {
4164 /* Learn source MAC (but don't try to learn from revalidation). */
4165 if (packet != NULL) {
4166 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4169 /* The log messages here could actually be useful in debugging,
4170 * so keep the rate limit relatively high. */
4171 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4172 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4173 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4174 ofproto_revalidate(ofproto, rev_tag);
4178 /* Determine output port. */
4179 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
4181 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4182 } else if (out_port != flow->in_port) {
4183 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4184 *nf_output_iface = out_port;
4192 static const struct ofhooks default_ofhooks = {
4194 default_normal_ofhook_cb,