2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "classifier.h"
29 #include "discovery.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
34 #include "mac-learning.h"
38 #include "ofp-print.h"
39 #include "ofproto-sflow.h"
41 #include "openflow/nicira-ext.h"
42 #include "openflow/openflow.h"
43 #include "openvswitch/datapath-protocol.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
53 #include "stream-ssl.h"
61 #define THIS_MODULE VLM_ofproto
64 #include "sflow_api.h"
68 TABLEID_CLASSIFIER = 1
72 struct netdev *netdev;
73 struct ofp_phy_port opp; /* In host byte order. */
76 static void ofport_free(struct ofport *);
77 static void hton_ofp_phy_port(struct ofp_phy_port *);
79 static int xlate_actions(const union ofp_action *in, size_t n_in,
80 const flow_t *flow, struct ofproto *ofproto,
81 const struct ofpbuf *packet,
82 struct odp_actions *out, tag_type *tags,
83 bool *may_set_up_flow, uint16_t *nf_output_iface);
88 uint64_t flow_cookie; /* Controller-issued identifier.
89 (Kept in network-byte order.) */
90 uint16_t idle_timeout; /* In seconds from time of last use. */
91 uint16_t hard_timeout; /* In seconds from time of creation. */
92 bool send_flow_removed; /* Send a flow removed message? */
93 long long int used; /* Last-used time (0 if never used). */
94 long long int created; /* Creation time. */
95 uint64_t packet_count; /* Number of packets received. */
96 uint64_t byte_count; /* Number of bytes received. */
97 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
98 tag_type tags; /* Tags (set only by hooks). */
99 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
101 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
102 * exact-match rule (having cr.wc.wildcards of 0) generated from the
103 * wildcard rule 'super'. In this case, 'list' is an element of the
106 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
107 * a list of subrules. A super-rule with no wildcards (where
108 * cr.wc.wildcards is 0) will never have any subrules. */
114 * 'n_actions' is the number of elements in the 'actions' array. A single
115 * action may take up more more than one element's worth of space.
117 * A subrule has no actions (it uses the super-rule's actions). */
119 union ofp_action *actions;
123 * A super-rule with wildcard fields never has ODP actions (since the
124 * datapath only supports exact-match flows). */
125 bool installed; /* Installed in datapath? */
126 bool may_install; /* True ordinarily; false if actions must
127 * be reassessed for every packet. */
129 union odp_action *odp_actions;
133 rule_is_hidden(const struct rule *rule)
135 /* Subrules are merely an implementation detail, so hide them from the
137 if (rule->super != NULL) {
141 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
142 * (e.g. by in-band control) and are intentionally hidden from the
144 if (rule->cr.priority > UINT16_MAX) {
151 static struct rule *rule_create(struct ofproto *, struct rule *super,
152 const union ofp_action *, size_t n_actions,
153 uint16_t idle_timeout, uint16_t hard_timeout,
154 uint64_t flow_cookie, bool send_flow_removed);
155 static void rule_free(struct rule *);
156 static void rule_destroy(struct ofproto *, struct rule *);
157 static struct rule *rule_from_cls_rule(const struct cls_rule *);
158 static void rule_insert(struct ofproto *, struct rule *,
159 struct ofpbuf *packet, uint16_t in_port);
160 static void rule_remove(struct ofproto *, struct rule *);
161 static bool rule_make_actions(struct ofproto *, struct rule *,
162 const struct ofpbuf *packet);
163 static void rule_install(struct ofproto *, struct rule *,
164 struct rule *displaced_rule);
165 static void rule_uninstall(struct ofproto *, struct rule *);
166 static void rule_post_uninstall(struct ofproto *, struct rule *);
167 static void send_flow_removed(struct ofproto *p, struct rule *rule,
168 long long int now, uint8_t reason);
170 /* ofproto supports two kinds of OpenFlow connections:
172 * - "Controller connections": Connections to ordinary OpenFlow controllers.
173 * ofproto maintains persistent connections to these controllers and by
174 * default sends them asynchronous messages such as packet-ins.
176 * - "Transient connections", e.g. from ovs-ofctl. When these connections
177 * drop, it is the other side's responsibility to reconnect them if
178 * necessary. ofproto does not send them asynchronous messages by default.
181 OFCONN_CONTROLLER, /* An OpenFlow controller. */
182 OFCONN_TRANSIENT /* A transient connection. */
185 /* An OpenFlow connection. */
187 struct ofproto *ofproto; /* The ofproto that owns this connection. */
188 struct list node; /* In struct ofproto's "all_conns" list. */
189 struct rconn *rconn; /* OpenFlow connection. */
190 enum ofconn_type type; /* Type. */
192 /* OFPT_PACKET_IN related data. */
193 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
194 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
195 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
196 int miss_send_len; /* Bytes to send of buffered packets. */
198 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
199 * requests, and the maximum number before we stop reading OpenFlow
201 #define OFCONN_REPLY_MAX 100
202 struct rconn_packet_counter *reply_counter;
204 /* type == OFCONN_CONTROLLER only. */
205 enum nx_role role; /* Role. */
206 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
207 struct discovery *discovery; /* Controller discovery object, if enabled. */
208 struct status_category *ss; /* Switch status category. */
209 enum ofproto_band band; /* In-band or out-of-band? */
212 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
213 * "schedulers" array. Their values are 0 and 1, and their meanings and values
214 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
215 * case anything ever changes, check their values here. */
216 #define N_SCHEDULERS 2
217 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
218 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
219 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
220 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
222 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
224 static void ofconn_destroy(struct ofconn *);
225 static void ofconn_run(struct ofconn *, struct ofproto *);
226 static void ofconn_wait(struct ofconn *);
227 static bool ofconn_receives_async_msgs(const struct ofconn *);
229 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
230 struct rconn_packet_counter *counter);
232 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
233 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
237 uint64_t datapath_id; /* Datapath ID. */
238 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
239 char *mfr_desc; /* Manufacturer. */
240 char *hw_desc; /* Hardware. */
241 char *sw_desc; /* Software version. */
242 char *serial_desc; /* Serial number. */
243 char *dp_desc; /* Datapath description. */
247 struct netdev_monitor *netdev_monitor;
248 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
250 struct shash port_by_name;
254 struct switch_status *switch_status;
255 struct fail_open *fail_open;
256 struct netflow *netflow;
257 struct ofproto_sflow *sflow;
259 /* In-band control. */
260 struct in_band *in_band;
261 long long int next_in_band_update;
262 struct sockaddr_in *extra_in_band_remotes;
263 size_t n_extra_remotes;
266 struct classifier cls;
267 bool need_revalidate;
268 long long int next_expiration;
269 struct tag_set revalidate_set;
270 bool tun_id_from_cookie;
272 /* OpenFlow connections. */
273 struct hmap controllers; /* Controller "struct ofconn"s. */
274 struct list all_conns; /* Contains "struct ofconn"s. */
275 struct pvconn **listeners;
277 struct pvconn **snoops;
280 /* Hooks for ovs-vswitchd. */
281 const struct ofhooks *ofhooks;
284 /* Used by default ofhooks. */
285 struct mac_learning *ml;
288 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
290 static const struct ofhooks default_ofhooks;
292 static uint64_t pick_datapath_id(const struct ofproto *);
293 static uint64_t pick_fallback_dpid(void);
295 static void update_used(struct ofproto *);
296 static void update_stats(struct ofproto *, struct rule *,
297 const struct odp_flow_stats *);
298 static void expire_rule(struct cls_rule *, void *ofproto);
299 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
300 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
301 static void revalidate_cb(struct cls_rule *rule_, void *p_);
303 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
305 static void handle_openflow(struct ofconn *, struct ofproto *,
308 static void refresh_port_groups(struct ofproto *);
310 static void update_port(struct ofproto *, const char *devname);
311 static int init_ports(struct ofproto *);
312 static void reinit_ports(struct ofproto *);
315 ofproto_create(const char *datapath, const char *datapath_type,
316 const struct ofhooks *ofhooks, void *aux,
317 struct ofproto **ofprotop)
319 struct odp_stats stats;
326 /* Connect to datapath and start listening for messages. */
327 error = dpif_open(datapath, datapath_type, &dpif);
329 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
332 error = dpif_get_dp_stats(dpif, &stats);
334 VLOG_ERR("failed to obtain stats for datapath %s: %s",
335 datapath, strerror(error));
339 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
341 VLOG_ERR("failed to listen on datapath %s: %s",
342 datapath, strerror(error));
346 dpif_flow_flush(dpif);
347 dpif_recv_purge(dpif);
349 /* Initialize settings. */
350 p = xzalloc(sizeof *p);
351 p->fallback_dpid = pick_fallback_dpid();
352 p->datapath_id = p->fallback_dpid;
353 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
354 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
355 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
356 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
357 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
359 /* Initialize datapath. */
361 p->netdev_monitor = netdev_monitor_create();
362 port_array_init(&p->ports);
363 shash_init(&p->port_by_name);
364 p->max_ports = stats.max_ports;
366 /* Initialize submodules. */
367 p->switch_status = switch_status_create(p);
373 /* Initialize flow table. */
374 classifier_init(&p->cls);
375 p->need_revalidate = false;
376 p->next_expiration = time_msec() + 1000;
377 tag_set_init(&p->revalidate_set);
379 /* Initialize OpenFlow connections. */
380 list_init(&p->all_conns);
381 hmap_init(&p->controllers);
387 /* Initialize hooks. */
389 p->ofhooks = ofhooks;
393 p->ofhooks = &default_ofhooks;
395 p->ml = mac_learning_create();
398 /* Pick final datapath ID. */
399 p->datapath_id = pick_datapath_id(p);
400 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
407 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
409 uint64_t old_dpid = p->datapath_id;
410 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
411 if (p->datapath_id != old_dpid) {
412 struct ofconn *ofconn;
414 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
416 /* Force all active connections to reconnect, since there is no way to
417 * notify a controller that the datapath ID has changed. */
418 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
419 rconn_reconnect(ofconn->rconn);
425 is_discovery_controller(const struct ofproto_controller *c)
427 return !strcmp(c->target, "discover");
431 is_in_band_controller(const struct ofproto_controller *c)
433 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
436 /* Creates a new controller in 'ofproto'. Some of the settings are initially
437 * drawn from 'c', but update_controller() needs to be called later to finish
438 * the new ofconn's configuration. */
440 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
442 struct discovery *discovery;
443 struct ofconn *ofconn;
445 if (is_discovery_controller(c)) {
446 int error = discovery_create(c->accept_re, c->update_resolv_conf,
447 ofproto->dpif, ofproto->switch_status,
456 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
457 ofconn->pktbuf = pktbuf_create();
458 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
460 ofconn->discovery = discovery;
462 rconn_connect(ofconn->rconn, c->target);
464 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
465 hash_string(c->target, 0));
468 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
469 * target or turn discovery on or off (these are done by creating new ofconns
470 * and deleting old ones), but it can update the rest of an ofconn's
473 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
475 struct ofproto *ofproto = ofconn->ofproto;
479 ofconn->band = (is_in_band_controller(c)
480 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
482 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
484 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
485 rconn_set_probe_interval(ofconn->rconn, probe_interval);
487 if (ofconn->discovery) {
488 discovery_set_update_resolv_conf(ofconn->discovery,
489 c->update_resolv_conf);
490 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
493 for (i = 0; i < N_SCHEDULERS; i++) {
494 struct pinsched **s = &ofconn->schedulers[i];
496 if (c->rate_limit > 0) {
498 *s = pinsched_create(c->rate_limit, c->burst_limit,
499 ofproto->switch_status);
501 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
504 pinsched_destroy(*s);
511 ofconn_get_target(const struct ofconn *ofconn)
513 return ofconn->discovery ? "discover" : rconn_get_name(ofconn->rconn);
516 static struct ofconn *
517 find_controller_by_target(struct ofproto *ofproto, const char *target)
519 struct ofconn *ofconn;
521 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
522 hash_string(target, 0), &ofproto->controllers) {
523 if (!strcmp(ofconn_get_target(ofconn), target)) {
531 update_in_band_remotes(struct ofproto *ofproto)
533 const struct ofconn *ofconn;
534 struct sockaddr_in *addrs;
535 size_t max_addrs, n_addrs;
539 /* Allocate enough memory for as many remotes as we could possibly have. */
540 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
541 addrs = xmalloc(max_addrs * sizeof *addrs);
544 /* Add all the remotes. */
546 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
547 struct sockaddr_in *sin = &addrs[n_addrs];
549 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
553 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
554 if (sin->sin_addr.s_addr) {
555 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
558 if (ofconn->discovery) {
562 for (i = 0; i < ofproto->n_extra_remotes; i++) {
563 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
566 /* Create or update or destroy in-band.
568 * Ordinarily we only enable in-band if there's at least one remote
569 * address, but discovery needs the in-band rules for DHCP to be installed
570 * even before we know any remote addresses. */
571 if (n_addrs || discovery) {
572 if (!ofproto->in_band) {
573 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
576 if (ofproto->in_band) {
577 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
579 ofproto->next_in_band_update = time_msec() + 1000;
581 in_band_destroy(ofproto->in_band);
582 ofproto->in_band = NULL;
590 ofproto_set_controllers(struct ofproto *p,
591 const struct ofproto_controller *controllers,
592 size_t n_controllers)
594 struct shash new_controllers;
595 enum ofproto_fail_mode fail_mode;
596 struct ofconn *ofconn, *next;
600 shash_init(&new_controllers);
601 for (i = 0; i < n_controllers; i++) {
602 const struct ofproto_controller *c = &controllers[i];
604 shash_add_once(&new_controllers, c->target, &controllers[i]);
605 if (!find_controller_by_target(p, c->target)) {
606 add_controller(p, c);
610 fail_mode = OFPROTO_FAIL_STANDALONE;
612 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
614 struct ofproto_controller *c;
616 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
618 ofconn_destroy(ofconn);
620 update_controller(ofconn, c);
624 if (c->fail == OFPROTO_FAIL_SECURE) {
625 fail_mode = OFPROTO_FAIL_SECURE;
629 shash_destroy(&new_controllers);
631 update_in_band_remotes(p);
633 if (!hmap_is_empty(&p->controllers)
634 && fail_mode == OFPROTO_FAIL_STANDALONE) {
635 struct rconn **rconns;
639 p->fail_open = fail_open_create(p, p->switch_status);
643 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
644 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
645 rconns[n++] = ofconn->rconn;
648 fail_open_set_controllers(p->fail_open, rconns, n);
649 /* p->fail_open takes ownership of 'rconns'. */
651 fail_open_destroy(p->fail_open);
655 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
656 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
657 struct ofconn, hmap_node);
658 ofconn->ss = switch_status_register(p->switch_status, "remote",
659 rconn_status_cb, ofconn->rconn);
664 any_extras_changed(const struct ofproto *ofproto,
665 const struct sockaddr_in *extras, size_t n)
669 if (n != ofproto->n_extra_remotes) {
673 for (i = 0; i < n; i++) {
674 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
675 const struct sockaddr_in *new = &extras[i];
677 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
678 old->sin_port != new->sin_port) {
686 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
687 * in-band control should guarantee access, in the same way that in-band
688 * control guarantees access to OpenFlow controllers. */
690 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
691 const struct sockaddr_in *extras, size_t n)
693 if (!any_extras_changed(ofproto, extras, n)) {
697 free(ofproto->extra_in_band_remotes);
698 ofproto->n_extra_remotes = n;
699 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
701 update_in_band_remotes(ofproto);
705 ofproto_set_desc(struct ofproto *p,
706 const char *mfr_desc, const char *hw_desc,
707 const char *sw_desc, const char *serial_desc,
710 struct ofp_desc_stats *ods;
713 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
714 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
715 sizeof ods->mfr_desc);
718 p->mfr_desc = xstrdup(mfr_desc);
721 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
722 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
723 sizeof ods->hw_desc);
726 p->hw_desc = xstrdup(hw_desc);
729 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
730 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
731 sizeof ods->sw_desc);
734 p->sw_desc = xstrdup(sw_desc);
737 if (strlen(serial_desc) >= sizeof ods->serial_num) {
738 VLOG_WARN("truncating serial_desc, must be less than %zu "
740 sizeof ods->serial_num);
742 free(p->serial_desc);
743 p->serial_desc = xstrdup(serial_desc);
746 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
747 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
748 sizeof ods->dp_desc);
751 p->dp_desc = xstrdup(dp_desc);
756 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
757 const struct svec *svec)
759 struct pvconn **pvconns = *pvconnsp;
760 size_t n_pvconns = *n_pvconnsp;
764 for (i = 0; i < n_pvconns; i++) {
765 pvconn_close(pvconns[i]);
769 pvconns = xmalloc(svec->n * sizeof *pvconns);
771 for (i = 0; i < svec->n; i++) {
772 const char *name = svec->names[i];
773 struct pvconn *pvconn;
776 error = pvconn_open(name, &pvconn);
778 pvconns[n_pvconns++] = pvconn;
780 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
788 *n_pvconnsp = n_pvconns;
794 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
796 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
800 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
802 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
806 ofproto_set_netflow(struct ofproto *ofproto,
807 const struct netflow_options *nf_options)
809 if (nf_options && nf_options->collectors.n) {
810 if (!ofproto->netflow) {
811 ofproto->netflow = netflow_create();
813 return netflow_set_options(ofproto->netflow, nf_options);
815 netflow_destroy(ofproto->netflow);
816 ofproto->netflow = NULL;
822 ofproto_set_sflow(struct ofproto *ofproto,
823 const struct ofproto_sflow_options *oso)
825 struct ofproto_sflow *os = ofproto->sflow;
828 struct ofport *ofport;
829 unsigned int odp_port;
831 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
832 refresh_port_groups(ofproto);
833 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
834 ofproto_sflow_add_port(os, odp_port,
835 netdev_get_name(ofport->netdev));
838 ofproto_sflow_set_options(os, oso);
840 ofproto_sflow_destroy(os);
841 ofproto->sflow = NULL;
846 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
850 VLOG_WARN("STP is not yet implemented");
858 ofproto_get_datapath_id(const struct ofproto *ofproto)
860 return ofproto->datapath_id;
864 ofproto_has_controller(const struct ofproto *ofproto)
866 return !hmap_is_empty(&ofproto->controllers);
870 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
874 for (i = 0; i < ofproto->n_listeners; i++) {
875 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
880 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
884 for (i = 0; i < ofproto->n_snoops; i++) {
885 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
890 ofproto_destroy(struct ofproto *p)
892 struct ofconn *ofconn, *next_ofconn;
893 struct ofport *ofport;
894 unsigned int port_no;
901 /* Destroy fail-open and in-band early, since they touch the classifier. */
902 fail_open_destroy(p->fail_open);
905 in_band_destroy(p->in_band);
907 free(p->extra_in_band_remotes);
909 ofproto_flush_flows(p);
910 classifier_destroy(&p->cls);
912 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
914 ofconn_destroy(ofconn);
916 hmap_destroy(&p->controllers);
919 netdev_monitor_destroy(p->netdev_monitor);
920 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
923 shash_destroy(&p->port_by_name);
925 switch_status_destroy(p->switch_status);
926 netflow_destroy(p->netflow);
927 ofproto_sflow_destroy(p->sflow);
929 for (i = 0; i < p->n_listeners; i++) {
930 pvconn_close(p->listeners[i]);
934 for (i = 0; i < p->n_snoops; i++) {
935 pvconn_close(p->snoops[i]);
939 mac_learning_destroy(p->ml);
944 free(p->serial_desc);
947 port_array_destroy(&p->ports);
953 ofproto_run(struct ofproto *p)
955 int error = ofproto_run1(p);
957 error = ofproto_run2(p, false);
963 process_port_change(struct ofproto *ofproto, int error, char *devname)
965 if (error == ENOBUFS) {
966 reinit_ports(ofproto);
968 update_port(ofproto, devname);
973 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
974 * means that 'ofconn' is more interesting for monitoring than a lower return
977 snoop_preference(const struct ofconn *ofconn)
979 switch (ofconn->role) {
987 /* Shouldn't happen. */
992 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
993 * Connects this vconn to a controller. */
995 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
997 struct ofconn *ofconn, *best;
999 /* Pick a controller for monitoring. */
1001 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1002 if (ofconn->type == OFCONN_CONTROLLER
1003 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1009 rconn_add_monitor(best->rconn, vconn);
1011 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1017 ofproto_run1(struct ofproto *p)
1019 struct ofconn *ofconn, *next_ofconn;
1024 if (shash_is_empty(&p->port_by_name)) {
1028 for (i = 0; i < 50; i++) {
1032 error = dpif_recv(p->dpif, &buf);
1034 if (error == ENODEV) {
1035 /* Someone destroyed the datapath behind our back. The caller
1036 * better destroy us and give up, because we're just going to
1037 * spin from here on out. */
1038 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1039 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1040 dpif_name(p->dpif));
1046 handle_odp_msg(p, buf);
1049 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1050 process_port_change(p, error, devname);
1052 while ((error = netdev_monitor_poll(p->netdev_monitor,
1053 &devname)) != EAGAIN) {
1054 process_port_change(p, error, devname);
1058 if (time_msec() >= p->next_in_band_update) {
1059 update_in_band_remotes(p);
1061 in_band_run(p->in_band);
1064 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1066 ofconn_run(ofconn, p);
1069 /* Fail-open maintenance. Do this after processing the ofconns since
1070 * fail-open checks the status of the controller rconn. */
1072 fail_open_run(p->fail_open);
1075 for (i = 0; i < p->n_listeners; i++) {
1076 struct vconn *vconn;
1079 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
1081 ofconn_create(p, rconn_new_from_vconn("passive", vconn),
1083 } else if (retval != EAGAIN) {
1084 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1088 for (i = 0; i < p->n_snoops; i++) {
1089 struct vconn *vconn;
1092 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1094 add_snooper(p, vconn);
1095 } else if (retval != EAGAIN) {
1096 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1100 if (time_msec() >= p->next_expiration) {
1101 COVERAGE_INC(ofproto_expiration);
1102 p->next_expiration = time_msec() + 1000;
1105 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1107 /* Let the hook know that we're at a stable point: all outstanding data
1108 * in existing flows has been accounted to the account_cb. Thus, the
1109 * hook can now reasonably do operations that depend on having accurate
1110 * flow volume accounting (currently, that's just bond rebalancing). */
1111 if (p->ofhooks->account_checkpoint_cb) {
1112 p->ofhooks->account_checkpoint_cb(p->aux);
1117 netflow_run(p->netflow);
1120 ofproto_sflow_run(p->sflow);
1126 struct revalidate_cbdata {
1127 struct ofproto *ofproto;
1128 bool revalidate_all; /* Revalidate all exact-match rules? */
1129 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1130 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1134 ofproto_run2(struct ofproto *p, bool revalidate_all)
1136 if (p->need_revalidate || revalidate_all
1137 || !tag_set_is_empty(&p->revalidate_set)) {
1138 struct revalidate_cbdata cbdata;
1140 cbdata.revalidate_all = revalidate_all;
1141 cbdata.revalidate_subrules = p->need_revalidate;
1142 cbdata.revalidate_set = p->revalidate_set;
1143 tag_set_init(&p->revalidate_set);
1144 COVERAGE_INC(ofproto_revalidate);
1145 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1146 p->need_revalidate = false;
1153 ofproto_wait(struct ofproto *p)
1155 struct ofconn *ofconn;
1158 dpif_recv_wait(p->dpif);
1159 dpif_port_poll_wait(p->dpif);
1160 netdev_monitor_poll_wait(p->netdev_monitor);
1161 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1162 ofconn_wait(ofconn);
1165 poll_timer_wait_until(p->next_in_band_update);
1166 in_band_wait(p->in_band);
1169 fail_open_wait(p->fail_open);
1172 ofproto_sflow_wait(p->sflow);
1174 if (!tag_set_is_empty(&p->revalidate_set)) {
1175 poll_immediate_wake();
1177 if (p->need_revalidate) {
1178 /* Shouldn't happen, but if it does just go around again. */
1179 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1180 poll_immediate_wake();
1181 } else if (p->next_expiration != LLONG_MAX) {
1182 poll_timer_wait_until(p->next_expiration);
1184 for (i = 0; i < p->n_listeners; i++) {
1185 pvconn_wait(p->listeners[i]);
1187 for (i = 0; i < p->n_snoops; i++) {
1188 pvconn_wait(p->snoops[i]);
1193 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1195 tag_set_add(&ofproto->revalidate_set, tag);
1199 ofproto_get_revalidate_set(struct ofproto *ofproto)
1201 return &ofproto->revalidate_set;
1205 ofproto_is_alive(const struct ofproto *p)
1207 return !hmap_is_empty(&p->controllers);
1211 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1212 const union ofp_action *actions, size_t n_actions,
1213 const struct ofpbuf *packet)
1215 struct odp_actions odp_actions;
1218 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1224 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1226 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1227 odp_actions.n_actions, packet);
1232 ofproto_add_flow(struct ofproto *p,
1233 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1234 const union ofp_action *actions, size_t n_actions,
1238 rule = rule_create(p, NULL, actions, n_actions,
1239 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1241 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1242 rule_insert(p, rule, NULL, 0);
1246 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1247 uint32_t wildcards, unsigned int priority)
1251 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1255 rule_remove(ofproto, rule);
1260 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1262 struct rule *rule = rule_from_cls_rule(rule_);
1263 struct ofproto *ofproto = ofproto_;
1265 /* Mark the flow as not installed, even though it might really be
1266 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1267 * There is no point in uninstalling it individually since we are about to
1268 * blow away all the flows with dpif_flow_flush(). */
1269 rule->installed = false;
1271 rule_remove(ofproto, rule);
1275 ofproto_flush_flows(struct ofproto *ofproto)
1277 COVERAGE_INC(ofproto_flush);
1278 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1279 dpif_flow_flush(ofproto->dpif);
1280 if (ofproto->in_band) {
1281 in_band_flushed(ofproto->in_band);
1283 if (ofproto->fail_open) {
1284 fail_open_flushed(ofproto->fail_open);
1289 reinit_ports(struct ofproto *p)
1291 struct svec devnames;
1292 struct ofport *ofport;
1293 unsigned int port_no;
1294 struct odp_port *odp_ports;
1298 svec_init(&devnames);
1299 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1300 svec_add (&devnames, (char *) ofport->opp.name);
1302 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1303 for (i = 0; i < n_odp_ports; i++) {
1304 svec_add (&devnames, odp_ports[i].devname);
1308 svec_sort_unique(&devnames);
1309 for (i = 0; i < devnames.n; i++) {
1310 update_port(p, devnames.names[i]);
1312 svec_destroy(&devnames);
1316 refresh_port_group(struct ofproto *p, unsigned int group)
1320 struct ofport *port;
1321 unsigned int port_no;
1323 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1325 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1327 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1328 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1329 ports[n_ports++] = port_no;
1332 dpif_port_group_set(p->dpif, group, ports, n_ports);
1339 refresh_port_groups(struct ofproto *p)
1341 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1342 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1344 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1348 static struct ofport *
1349 make_ofport(const struct odp_port *odp_port)
1351 struct netdev_options netdev_options;
1352 enum netdev_flags flags;
1353 struct ofport *ofport;
1354 struct netdev *netdev;
1358 memset(&netdev_options, 0, sizeof netdev_options);
1359 netdev_options.name = odp_port->devname;
1360 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1361 netdev_options.may_open = true;
1363 error = netdev_open(&netdev_options, &netdev);
1365 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1366 "cannot be opened (%s)",
1367 odp_port->devname, odp_port->port,
1368 odp_port->devname, strerror(error));
1372 ofport = xmalloc(sizeof *ofport);
1373 ofport->netdev = netdev;
1374 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1375 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1376 memcpy(ofport->opp.name, odp_port->devname,
1377 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1378 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1380 netdev_get_flags(netdev, &flags);
1381 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1383 netdev_get_carrier(netdev, &carrier);
1384 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1386 netdev_get_features(netdev,
1387 &ofport->opp.curr, &ofport->opp.advertised,
1388 &ofport->opp.supported, &ofport->opp.peer);
1393 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1395 if (port_array_get(&p->ports, odp_port->port)) {
1396 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1399 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1400 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1409 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1411 const struct ofp_phy_port *a = &a_->opp;
1412 const struct ofp_phy_port *b = &b_->opp;
1414 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1415 return (a->port_no == b->port_no
1416 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1417 && !strcmp((char *) a->name, (char *) b->name)
1418 && a->state == b->state
1419 && a->config == b->config
1420 && a->curr == b->curr
1421 && a->advertised == b->advertised
1422 && a->supported == b->supported
1423 && a->peer == b->peer);
1427 send_port_status(struct ofproto *p, const struct ofport *ofport,
1430 /* XXX Should limit the number of queued port status change messages. */
1431 struct ofconn *ofconn;
1432 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1433 struct ofp_port_status *ops;
1436 if (!ofconn_receives_async_msgs(ofconn)) {
1440 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1441 ops->reason = reason;
1442 ops->desc = ofport->opp;
1443 hton_ofp_phy_port(&ops->desc);
1444 queue_tx(b, ofconn, NULL);
1446 if (p->ofhooks->port_changed_cb) {
1447 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1452 ofport_install(struct ofproto *p, struct ofport *ofport)
1454 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1455 const char *netdev_name = (const char *) ofport->opp.name;
1457 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1458 port_array_set(&p->ports, odp_port, ofport);
1459 shash_add(&p->port_by_name, netdev_name, ofport);
1461 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1466 ofport_remove(struct ofproto *p, struct ofport *ofport)
1468 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1470 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1471 port_array_set(&p->ports, odp_port, NULL);
1472 shash_delete(&p->port_by_name,
1473 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1475 ofproto_sflow_del_port(p->sflow, odp_port);
1480 ofport_free(struct ofport *ofport)
1483 netdev_close(ofport->netdev);
1489 update_port(struct ofproto *p, const char *devname)
1491 struct odp_port odp_port;
1492 struct ofport *old_ofport;
1493 struct ofport *new_ofport;
1496 COVERAGE_INC(ofproto_update_port);
1498 /* Query the datapath for port information. */
1499 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1501 /* Find the old ofport. */
1502 old_ofport = shash_find_data(&p->port_by_name, devname);
1505 /* There's no port named 'devname' but there might be a port with
1506 * the same port number. This could happen if a port is deleted
1507 * and then a new one added in its place very quickly, or if a port
1508 * is renamed. In the former case we want to send an OFPPR_DELETE
1509 * and an OFPPR_ADD, and in the latter case we want to send a
1510 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1511 * the old port's ifindex against the new port, or perhaps less
1512 * reliably but more portably by comparing the old port's MAC
1513 * against the new port's MAC. However, this code isn't that smart
1514 * and always sends an OFPPR_MODIFY (XXX). */
1515 old_ofport = port_array_get(&p->ports, odp_port.port);
1517 } else if (error != ENOENT && error != ENODEV) {
1518 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1519 "%s", strerror(error));
1523 /* Create a new ofport. */
1524 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1526 /* Eliminate a few pathological cases. */
1527 if (!old_ofport && !new_ofport) {
1529 } else if (old_ofport && new_ofport) {
1530 /* Most of the 'config' bits are OpenFlow soft state, but
1531 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1532 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1533 * leaves the other bits 0.) */
1534 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1536 if (ofport_equal(old_ofport, new_ofport)) {
1537 /* False alarm--no change. */
1538 ofport_free(new_ofport);
1543 /* Now deal with the normal cases. */
1545 ofport_remove(p, old_ofport);
1548 ofport_install(p, new_ofport);
1550 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1551 (!old_ofport ? OFPPR_ADD
1552 : !new_ofport ? OFPPR_DELETE
1554 ofport_free(old_ofport);
1556 /* Update port groups. */
1557 refresh_port_groups(p);
1561 init_ports(struct ofproto *p)
1563 struct odp_port *ports;
1568 error = dpif_port_list(p->dpif, &ports, &n_ports);
1573 for (i = 0; i < n_ports; i++) {
1574 const struct odp_port *odp_port = &ports[i];
1575 if (!ofport_conflicts(p, odp_port)) {
1576 struct ofport *ofport = make_ofport(odp_port);
1578 ofport_install(p, ofport);
1583 refresh_port_groups(p);
1587 static struct ofconn *
1588 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1590 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1591 ofconn->ofproto = p;
1592 list_push_back(&p->all_conns, &ofconn->node);
1593 ofconn->rconn = rconn;
1594 ofconn->type = type;
1595 ofconn->role = NX_ROLE_OTHER;
1596 ofconn->packet_in_counter = rconn_packet_counter_create ();
1597 ofconn->pktbuf = NULL;
1598 ofconn->miss_send_len = 0;
1599 ofconn->reply_counter = rconn_packet_counter_create ();
1604 ofconn_destroy(struct ofconn *ofconn)
1606 if (ofconn->type == OFCONN_CONTROLLER) {
1607 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1609 discovery_destroy(ofconn->discovery);
1611 list_remove(&ofconn->node);
1612 switch_status_unregister(ofconn->ss);
1613 rconn_destroy(ofconn->rconn);
1614 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1615 rconn_packet_counter_destroy(ofconn->reply_counter);
1616 pktbuf_destroy(ofconn->pktbuf);
1621 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1626 if (ofconn->discovery) {
1627 char *controller_name;
1628 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1629 discovery_question_connectivity(ofconn->discovery);
1631 if (discovery_run(ofconn->discovery, &controller_name)) {
1632 if (controller_name) {
1633 rconn_connect(ofconn->rconn, controller_name);
1635 rconn_disconnect(ofconn->rconn);
1640 for (i = 0; i < N_SCHEDULERS; i++) {
1641 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1644 rconn_run(ofconn->rconn);
1646 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1647 /* Limit the number of iterations to prevent other tasks from
1649 for (iteration = 0; iteration < 50; iteration++) {
1650 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1655 fail_open_maybe_recover(p->fail_open);
1657 handle_openflow(ofconn, p, of_msg);
1658 ofpbuf_delete(of_msg);
1662 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1663 ofconn_destroy(ofconn);
1668 ofconn_wait(struct ofconn *ofconn)
1672 if (ofconn->discovery) {
1673 discovery_wait(ofconn->discovery);
1675 for (i = 0; i < N_SCHEDULERS; i++) {
1676 pinsched_wait(ofconn->schedulers[i]);
1678 rconn_run_wait(ofconn->rconn);
1679 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1680 rconn_recv_wait(ofconn->rconn);
1682 COVERAGE_INC(ofproto_ofconn_stuck);
1686 /* Returns true if 'ofconn' should receive asynchronous messages. */
1688 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1690 if (ofconn->type == OFCONN_CONTROLLER) {
1691 /* Ordinary controllers always get asynchronous messages unless they
1692 * have configured themselves as "slaves". */
1693 return ofconn->role != NX_ROLE_SLAVE;
1695 /* Transient connections don't get asynchronous messages unless they
1696 * have explicitly asked for them by setting a nonzero miss send
1698 return ofconn->miss_send_len > 0;
1702 /* Caller is responsible for initializing the 'cr' member of the returned
1704 static struct rule *
1705 rule_create(struct ofproto *ofproto, struct rule *super,
1706 const union ofp_action *actions, size_t n_actions,
1707 uint16_t idle_timeout, uint16_t hard_timeout,
1708 uint64_t flow_cookie, bool send_flow_removed)
1710 struct rule *rule = xzalloc(sizeof *rule);
1711 rule->idle_timeout = idle_timeout;
1712 rule->hard_timeout = hard_timeout;
1713 rule->flow_cookie = flow_cookie;
1714 rule->used = rule->created = time_msec();
1715 rule->send_flow_removed = send_flow_removed;
1716 rule->super = super;
1718 list_push_back(&super->list, &rule->list);
1720 list_init(&rule->list);
1722 rule->n_actions = n_actions;
1723 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1724 netflow_flow_clear(&rule->nf_flow);
1725 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1730 static struct rule *
1731 rule_from_cls_rule(const struct cls_rule *cls_rule)
1733 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1737 rule_free(struct rule *rule)
1739 free(rule->actions);
1740 free(rule->odp_actions);
1744 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1745 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1746 * through all of its subrules and revalidates them, destroying any that no
1747 * longer has a super-rule (which is probably all of them).
1749 * Before calling this function, the caller must make have removed 'rule' from
1750 * the classifier. If 'rule' is an exact-match rule, the caller is also
1751 * responsible for ensuring that it has been uninstalled from the datapath. */
1753 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1756 struct rule *subrule, *next;
1757 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1758 revalidate_rule(ofproto, subrule);
1761 list_remove(&rule->list);
1767 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1769 const union ofp_action *oa;
1770 struct actions_iterator i;
1772 if (out_port == htons(OFPP_NONE)) {
1775 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1776 oa = actions_next(&i)) {
1777 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1784 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1785 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1787 * The flow that 'packet' actually contains does not need to actually match
1788 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1789 * the packet and byte counters for 'rule' will be credited for the packet sent
1790 * out whether or not the packet actually matches 'rule'.
1792 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1793 * the caller must already have accurately composed ODP actions for it given
1794 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1795 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1796 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1797 * actions and apply them to 'packet'. */
1799 rule_execute(struct ofproto *ofproto, struct rule *rule,
1800 struct ofpbuf *packet, const flow_t *flow)
1802 const union odp_action *actions;
1804 struct odp_actions a;
1806 /* Grab or compose the ODP actions.
1808 * The special case for an exact-match 'rule' where 'flow' is not the
1809 * rule's flow is important to avoid, e.g., sending a packet out its input
1810 * port simply because the ODP actions were composed for the wrong
1812 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1813 struct rule *super = rule->super ? rule->super : rule;
1814 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1815 packet, &a, NULL, 0, NULL)) {
1818 actions = a.actions;
1819 n_actions = a.n_actions;
1821 actions = rule->odp_actions;
1822 n_actions = rule->n_odp_actions;
1825 /* Execute the ODP actions. */
1826 if (!dpif_execute(ofproto->dpif, flow->in_port,
1827 actions, n_actions, packet)) {
1828 struct odp_flow_stats stats;
1829 flow_extract_stats(flow, packet, &stats);
1830 update_stats(ofproto, rule, &stats);
1831 rule->used = time_msec();
1832 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1837 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1840 struct rule *displaced_rule;
1842 /* Insert the rule in the classifier. */
1843 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1844 if (!rule->cr.wc.wildcards) {
1845 rule_make_actions(p, rule, packet);
1848 /* Send the packet and credit it to the rule. */
1851 flow_extract(packet, 0, in_port, &flow);
1852 rule_execute(p, rule, packet, &flow);
1855 /* Install the rule in the datapath only after sending the packet, to
1856 * avoid packet reordering. */
1857 if (rule->cr.wc.wildcards) {
1858 COVERAGE_INC(ofproto_add_wc_flow);
1859 p->need_revalidate = true;
1861 rule_install(p, rule, displaced_rule);
1864 /* Free the rule that was displaced, if any. */
1865 if (displaced_rule) {
1866 rule_destroy(p, displaced_rule);
1870 static struct rule *
1871 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1874 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1875 rule->idle_timeout, rule->hard_timeout,
1877 COVERAGE_INC(ofproto_subrule_create);
1878 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1879 : rule->cr.priority), &subrule->cr);
1880 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1886 rule_remove(struct ofproto *ofproto, struct rule *rule)
1888 if (rule->cr.wc.wildcards) {
1889 COVERAGE_INC(ofproto_del_wc_flow);
1890 ofproto->need_revalidate = true;
1892 rule_uninstall(ofproto, rule);
1894 classifier_remove(&ofproto->cls, &rule->cr);
1895 rule_destroy(ofproto, rule);
1898 /* Returns true if the actions changed, false otherwise. */
1900 rule_make_actions(struct ofproto *p, struct rule *rule,
1901 const struct ofpbuf *packet)
1903 const struct rule *super;
1904 struct odp_actions a;
1907 assert(!rule->cr.wc.wildcards);
1909 super = rule->super ? rule->super : rule;
1911 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1912 packet, &a, &rule->tags, &rule->may_install,
1913 &rule->nf_flow.output_iface);
1915 actions_len = a.n_actions * sizeof *a.actions;
1916 if (rule->n_odp_actions != a.n_actions
1917 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1918 COVERAGE_INC(ofproto_odp_unchanged);
1919 free(rule->odp_actions);
1920 rule->n_odp_actions = a.n_actions;
1921 rule->odp_actions = xmemdup(a.actions, actions_len);
1929 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1930 struct odp_flow_put *put)
1932 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1933 put->flow.key = rule->cr.flow;
1934 put->flow.actions = rule->odp_actions;
1935 put->flow.n_actions = rule->n_odp_actions;
1936 put->flow.flags = 0;
1938 return dpif_flow_put(ofproto->dpif, put);
1942 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1944 assert(!rule->cr.wc.wildcards);
1946 if (rule->may_install) {
1947 struct odp_flow_put put;
1948 if (!do_put_flow(p, rule,
1949 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1951 rule->installed = true;
1952 if (displaced_rule) {
1953 update_stats(p, displaced_rule, &put.flow.stats);
1954 rule_post_uninstall(p, displaced_rule);
1957 } else if (displaced_rule) {
1958 rule_uninstall(p, displaced_rule);
1963 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1965 if (rule->installed) {
1966 struct odp_flow_put put;
1967 COVERAGE_INC(ofproto_dp_missed);
1968 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1970 rule_install(ofproto, rule, NULL);
1975 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1977 bool actions_changed;
1978 uint16_t new_out_iface, old_out_iface;
1980 old_out_iface = rule->nf_flow.output_iface;
1981 actions_changed = rule_make_actions(ofproto, rule, NULL);
1983 if (rule->may_install) {
1984 if (rule->installed) {
1985 if (actions_changed) {
1986 struct odp_flow_put put;
1987 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1988 | ODPPF_ZERO_STATS, &put);
1989 update_stats(ofproto, rule, &put.flow.stats);
1991 /* Temporarily set the old output iface so that NetFlow
1992 * messages have the correct output interface for the old
1994 new_out_iface = rule->nf_flow.output_iface;
1995 rule->nf_flow.output_iface = old_out_iface;
1996 rule_post_uninstall(ofproto, rule);
1997 rule->nf_flow.output_iface = new_out_iface;
2000 rule_install(ofproto, rule, NULL);
2003 rule_uninstall(ofproto, rule);
2008 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2010 uint64_t total_bytes = rule->byte_count + extra_bytes;
2012 if (ofproto->ofhooks->account_flow_cb
2013 && total_bytes > rule->accounted_bytes)
2015 ofproto->ofhooks->account_flow_cb(
2016 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
2017 total_bytes - rule->accounted_bytes, ofproto->aux);
2018 rule->accounted_bytes = total_bytes;
2023 rule_uninstall(struct ofproto *p, struct rule *rule)
2025 assert(!rule->cr.wc.wildcards);
2026 if (rule->installed) {
2027 struct odp_flow odp_flow;
2029 odp_flow.key = rule->cr.flow;
2030 odp_flow.actions = NULL;
2031 odp_flow.n_actions = 0;
2033 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2034 update_stats(p, rule, &odp_flow.stats);
2036 rule->installed = false;
2038 rule_post_uninstall(p, rule);
2043 is_controller_rule(struct rule *rule)
2045 /* If the only action is send to the controller then don't report
2046 * NetFlow expiration messages since it is just part of the control
2047 * logic for the network and not real traffic. */
2049 if (rule && rule->super) {
2050 struct rule *super = rule->super;
2052 return super->n_actions == 1 &&
2053 super->actions[0].type == htons(OFPAT_OUTPUT) &&
2054 super->actions[0].output.port == htons(OFPP_CONTROLLER);
2061 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2063 struct rule *super = rule->super;
2065 rule_account(ofproto, rule, 0);
2067 if (ofproto->netflow && !is_controller_rule(rule)) {
2068 struct ofexpired expired;
2069 expired.flow = rule->cr.flow;
2070 expired.packet_count = rule->packet_count;
2071 expired.byte_count = rule->byte_count;
2072 expired.used = rule->used;
2073 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2076 super->packet_count += rule->packet_count;
2077 super->byte_count += rule->byte_count;
2079 /* Reset counters to prevent double counting if the rule ever gets
2081 rule->packet_count = 0;
2082 rule->byte_count = 0;
2083 rule->accounted_bytes = 0;
2085 netflow_flow_clear(&rule->nf_flow);
2090 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2091 struct rconn_packet_counter *counter)
2093 update_openflow_length(msg);
2094 if (rconn_send(ofconn->rconn, msg, counter)) {
2100 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2101 int error, const void *data, size_t len)
2104 struct ofp_error_msg *oem;
2106 if (!(error >> 16)) {
2107 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2112 COVERAGE_INC(ofproto_error);
2113 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2114 oh ? oh->xid : 0, &buf);
2115 oem->type = htons((unsigned int) error >> 16);
2116 oem->code = htons(error & 0xffff);
2117 memcpy(oem->data, data, len);
2118 queue_tx(buf, ofconn, ofconn->reply_counter);
2122 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2125 size_t oh_length = ntohs(oh->length);
2126 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2130 hton_ofp_phy_port(struct ofp_phy_port *opp)
2132 opp->port_no = htons(opp->port_no);
2133 opp->config = htonl(opp->config);
2134 opp->state = htonl(opp->state);
2135 opp->curr = htonl(opp->curr);
2136 opp->advertised = htonl(opp->advertised);
2137 opp->supported = htonl(opp->supported);
2138 opp->peer = htonl(opp->peer);
2142 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2144 struct ofp_header *rq = oh;
2145 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2150 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2151 struct ofp_header *oh)
2153 struct ofp_switch_features *osf;
2155 unsigned int port_no;
2156 struct ofport *port;
2158 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2159 osf->datapath_id = htonll(p->datapath_id);
2160 osf->n_buffers = htonl(pktbuf_capacity());
2162 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2163 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2164 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2165 (1u << OFPAT_SET_VLAN_VID) |
2166 (1u << OFPAT_SET_VLAN_PCP) |
2167 (1u << OFPAT_STRIP_VLAN) |
2168 (1u << OFPAT_SET_DL_SRC) |
2169 (1u << OFPAT_SET_DL_DST) |
2170 (1u << OFPAT_SET_NW_SRC) |
2171 (1u << OFPAT_SET_NW_DST) |
2172 (1u << OFPAT_SET_NW_TOS) |
2173 (1u << OFPAT_SET_TP_SRC) |
2174 (1u << OFPAT_SET_TP_DST));
2176 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2177 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2180 queue_tx(buf, ofconn, ofconn->reply_counter);
2185 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2186 struct ofp_header *oh)
2189 struct ofp_switch_config *osc;
2193 /* Figure out flags. */
2194 dpif_get_drop_frags(p->dpif, &drop_frags);
2195 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2198 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2199 osc->flags = htons(flags);
2200 osc->miss_send_len = htons(ofconn->miss_send_len);
2201 queue_tx(buf, ofconn, ofconn->reply_counter);
2207 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2208 struct ofp_switch_config *osc)
2213 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2217 flags = ntohs(osc->flags);
2219 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2220 switch (flags & OFPC_FRAG_MASK) {
2221 case OFPC_FRAG_NORMAL:
2222 dpif_set_drop_frags(p->dpif, false);
2224 case OFPC_FRAG_DROP:
2225 dpif_set_drop_frags(p->dpif, true);
2228 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2234 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2240 add_output_group_action(struct odp_actions *actions, uint16_t group,
2241 uint16_t *nf_output_iface)
2243 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2245 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2246 *nf_output_iface = NF_OUT_FLOOD;
2251 add_controller_action(struct odp_actions *actions,
2252 const struct ofp_action_output *oao)
2254 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2255 a->controller.arg = ntohs(oao->max_len);
2258 struct action_xlate_ctx {
2260 flow_t flow; /* Flow to which these actions correspond. */
2261 int recurse; /* Recursion level, via xlate_table_action. */
2262 struct ofproto *ofproto;
2263 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2264 * null pointer if we are revalidating
2265 * without a packet to refer to. */
2268 struct odp_actions *out; /* Datapath actions. */
2269 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2270 bool may_set_up_flow; /* True ordinarily; false if the actions must
2271 * be reassessed for every packet. */
2272 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2275 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2276 struct action_xlate_ctx *ctx);
2279 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2281 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2284 if (ofport->opp.config & OFPPC_NO_FWD) {
2285 /* Forwarding disabled on port. */
2290 * We don't have an ofport record for this port, but it doesn't hurt to
2291 * allow forwarding to it anyhow. Maybe such a port will appear later
2292 * and we're pre-populating the flow table.
2296 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2297 ctx->nf_output_iface = port;
2300 static struct rule *
2301 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2304 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2306 /* The rule we found might not be valid, since we could be in need of
2307 * revalidation. If it is not valid, don't return it. */
2310 && ofproto->need_revalidate
2311 && !revalidate_rule(ofproto, rule)) {
2312 COVERAGE_INC(ofproto_invalidated);
2320 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2322 if (!ctx->recurse) {
2323 uint16_t old_in_port;
2326 /* Look up a flow with 'in_port' as the input port. Then restore the
2327 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2328 * have surprising behavior). */
2329 old_in_port = ctx->flow.in_port;
2330 ctx->flow.in_port = in_port;
2331 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2332 ctx->flow.in_port = old_in_port;
2340 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2347 xlate_output_action(struct action_xlate_ctx *ctx,
2348 const struct ofp_action_output *oao)
2351 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2353 ctx->nf_output_iface = NF_OUT_DROP;
2355 switch (ntohs(oao->port)) {
2357 add_output_action(ctx, ctx->flow.in_port);
2360 xlate_table_action(ctx, ctx->flow.in_port);
2363 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2364 ctx->out, ctx->tags,
2365 &ctx->nf_output_iface,
2366 ctx->ofproto->aux)) {
2367 COVERAGE_INC(ofproto_uninstallable);
2368 ctx->may_set_up_flow = false;
2372 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2373 &ctx->nf_output_iface);
2376 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2378 case OFPP_CONTROLLER:
2379 add_controller_action(ctx->out, oao);
2382 add_output_action(ctx, ODPP_LOCAL);
2385 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2386 if (odp_port != ctx->flow.in_port) {
2387 add_output_action(ctx, odp_port);
2392 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2393 ctx->nf_output_iface = NF_OUT_FLOOD;
2394 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2395 ctx->nf_output_iface = prev_nf_output_iface;
2396 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2397 ctx->nf_output_iface != NF_OUT_FLOOD) {
2398 ctx->nf_output_iface = NF_OUT_MULTI;
2403 xlate_nicira_action(struct action_xlate_ctx *ctx,
2404 const struct nx_action_header *nah)
2406 const struct nx_action_resubmit *nar;
2407 const struct nx_action_set_tunnel *nast;
2408 union odp_action *oa;
2409 int subtype = ntohs(nah->subtype);
2411 assert(nah->vendor == htonl(NX_VENDOR_ID));
2413 case NXAST_RESUBMIT:
2414 nar = (const struct nx_action_resubmit *) nah;
2415 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2418 case NXAST_SET_TUNNEL:
2419 nast = (const struct nx_action_set_tunnel *) nah;
2420 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2421 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2424 /* If you add a new action here that modifies flow data, don't forget to
2425 * update the flow key in ctx->flow in the same key. */
2428 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2434 do_xlate_actions(const union ofp_action *in, size_t n_in,
2435 struct action_xlate_ctx *ctx)
2437 struct actions_iterator iter;
2438 const union ofp_action *ia;
2439 const struct ofport *port;
2441 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2442 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2443 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2444 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2445 /* Drop this flow. */
2449 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2450 uint16_t type = ntohs(ia->type);
2451 union odp_action *oa;
2455 xlate_output_action(ctx, &ia->output);
2458 case OFPAT_SET_VLAN_VID:
2459 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2460 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2463 case OFPAT_SET_VLAN_PCP:
2464 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2465 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2468 case OFPAT_STRIP_VLAN:
2469 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2470 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2471 ctx->flow.dl_vlan_pcp = 0;
2474 case OFPAT_SET_DL_SRC:
2475 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2476 memcpy(oa->dl_addr.dl_addr,
2477 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2478 memcpy(ctx->flow.dl_src,
2479 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2482 case OFPAT_SET_DL_DST:
2483 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2484 memcpy(oa->dl_addr.dl_addr,
2485 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2486 memcpy(ctx->flow.dl_dst,
2487 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2490 case OFPAT_SET_NW_SRC:
2491 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2492 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2495 case OFPAT_SET_NW_DST:
2496 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2497 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2500 case OFPAT_SET_NW_TOS:
2501 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2502 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2505 case OFPAT_SET_TP_SRC:
2506 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2507 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2510 case OFPAT_SET_TP_DST:
2511 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2512 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2516 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2520 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2527 xlate_actions(const union ofp_action *in, size_t n_in,
2528 const flow_t *flow, struct ofproto *ofproto,
2529 const struct ofpbuf *packet,
2530 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2531 uint16_t *nf_output_iface)
2533 tag_type no_tags = 0;
2534 struct action_xlate_ctx ctx;
2535 COVERAGE_INC(ofproto_ofp2odp);
2536 odp_actions_init(out);
2539 ctx.ofproto = ofproto;
2540 ctx.packet = packet;
2542 ctx.tags = tags ? tags : &no_tags;
2543 ctx.may_set_up_flow = true;
2544 ctx.nf_output_iface = NF_OUT_DROP;
2545 do_xlate_actions(in, n_in, &ctx);
2547 /* Check with in-band control to see if we're allowed to set up this
2549 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2550 ctx.may_set_up_flow = false;
2553 if (may_set_up_flow) {
2554 *may_set_up_flow = ctx.may_set_up_flow;
2556 if (nf_output_iface) {
2557 *nf_output_iface = ctx.nf_output_iface;
2559 if (odp_actions_overflow(out)) {
2560 odp_actions_init(out);
2561 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2566 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2567 * error message code (composed with ofp_mkerr()) for the caller to propagate
2568 * upward. Otherwise, returns 0.
2570 * 'oh' is used to make log messages more informative. */
2572 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2574 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2575 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2578 type_name = ofp_message_type_to_string(oh->type);
2579 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2583 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2590 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2591 struct ofp_header *oh)
2593 struct ofp_packet_out *opo;
2594 struct ofpbuf payload, *buffer;
2595 struct odp_actions actions;
2601 error = reject_slave_controller(ofconn, oh);
2606 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2610 opo = (struct ofp_packet_out *) oh;
2612 COVERAGE_INC(ofproto_packet_out);
2613 if (opo->buffer_id != htonl(UINT32_MAX)) {
2614 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2616 if (error || !buffer) {
2624 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2625 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2626 &flow, p, &payload, &actions, NULL, NULL, NULL);
2631 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2633 ofpbuf_delete(buffer);
2639 update_port_config(struct ofproto *p, struct ofport *port,
2640 uint32_t config, uint32_t mask)
2642 mask &= config ^ port->opp.config;
2643 if (mask & OFPPC_PORT_DOWN) {
2644 if (config & OFPPC_PORT_DOWN) {
2645 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2647 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2650 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2651 if (mask & REVALIDATE_BITS) {
2652 COVERAGE_INC(ofproto_costly_flags);
2653 port->opp.config ^= mask & REVALIDATE_BITS;
2654 p->need_revalidate = true;
2656 #undef REVALIDATE_BITS
2657 if (mask & OFPPC_NO_FLOOD) {
2658 port->opp.config ^= OFPPC_NO_FLOOD;
2659 refresh_port_groups(p);
2661 if (mask & OFPPC_NO_PACKET_IN) {
2662 port->opp.config ^= OFPPC_NO_PACKET_IN;
2667 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2668 struct ofp_header *oh)
2670 const struct ofp_port_mod *opm;
2671 struct ofport *port;
2674 error = reject_slave_controller(ofconn, oh);
2678 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2682 opm = (struct ofp_port_mod *) oh;
2684 port = port_array_get(&p->ports,
2685 ofp_port_to_odp_port(ntohs(opm->port_no)));
2687 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2688 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2689 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2691 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2692 if (opm->advertise) {
2693 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2699 static struct ofpbuf *
2700 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2702 struct ofp_stats_reply *osr;
2705 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2706 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2708 osr->flags = htons(0);
2712 static struct ofpbuf *
2713 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2715 return make_stats_reply(request->header.xid, request->type, body_len);
2719 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2721 struct ofpbuf *msg = *msgp;
2722 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2723 if (nbytes + msg->size > UINT16_MAX) {
2724 struct ofp_stats_reply *reply = msg->data;
2725 reply->flags = htons(OFPSF_REPLY_MORE);
2726 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2727 queue_tx(msg, ofconn, ofconn->reply_counter);
2729 return ofpbuf_put_uninit(*msgp, nbytes);
2733 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2734 struct ofp_stats_request *request)
2736 struct ofp_desc_stats *ods;
2739 msg = start_stats_reply(request, sizeof *ods);
2740 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2741 memset(ods, 0, sizeof *ods);
2742 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2743 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2744 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2745 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2746 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2747 queue_tx(msg, ofconn, ofconn->reply_counter);
2753 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2755 struct rule *rule = rule_from_cls_rule(cls_rule);
2756 int *n_subrules = n_subrules_;
2764 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2765 struct ofp_stats_request *request)
2767 struct ofp_table_stats *ots;
2769 struct odp_stats dpstats;
2770 int n_exact, n_subrules, n_wild;
2772 msg = start_stats_reply(request, sizeof *ots * 2);
2774 /* Count rules of various kinds. */
2776 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2777 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2778 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2781 dpif_get_dp_stats(p->dpif, &dpstats);
2782 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2783 memset(ots, 0, sizeof *ots);
2784 ots->table_id = TABLEID_HASH;
2785 strcpy(ots->name, "hash");
2786 ots->wildcards = htonl(0);
2787 ots->max_entries = htonl(dpstats.max_capacity);
2788 ots->active_count = htonl(n_exact);
2789 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2791 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2793 /* Classifier table. */
2794 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2795 memset(ots, 0, sizeof *ots);
2796 ots->table_id = TABLEID_CLASSIFIER;
2797 strcpy(ots->name, "classifier");
2798 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2800 ots->max_entries = htonl(65536);
2801 ots->active_count = htonl(n_wild);
2802 ots->lookup_count = htonll(0); /* XXX */
2803 ots->matched_count = htonll(0); /* XXX */
2805 queue_tx(msg, ofconn, ofconn->reply_counter);
2810 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2811 struct ofpbuf **msgp)
2813 struct netdev_stats stats;
2814 struct ofp_port_stats *ops;
2816 /* Intentionally ignore return value, since errors will set
2817 * 'stats' to all-1s, which is correct for OpenFlow, and
2818 * netdev_get_stats() will log errors. */
2819 netdev_get_stats(port->netdev, &stats);
2821 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2822 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2823 memset(ops->pad, 0, sizeof ops->pad);
2824 ops->rx_packets = htonll(stats.rx_packets);
2825 ops->tx_packets = htonll(stats.tx_packets);
2826 ops->rx_bytes = htonll(stats.rx_bytes);
2827 ops->tx_bytes = htonll(stats.tx_bytes);
2828 ops->rx_dropped = htonll(stats.rx_dropped);
2829 ops->tx_dropped = htonll(stats.tx_dropped);
2830 ops->rx_errors = htonll(stats.rx_errors);
2831 ops->tx_errors = htonll(stats.tx_errors);
2832 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2833 ops->rx_over_err = htonll(stats.rx_over_errors);
2834 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2835 ops->collisions = htonll(stats.collisions);
2839 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2840 struct ofp_stats_request *osr,
2843 struct ofp_port_stats_request *psr;
2844 struct ofp_port_stats *ops;
2846 struct ofport *port;
2847 unsigned int port_no;
2849 if (arg_size != sizeof *psr) {
2850 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2852 psr = (struct ofp_port_stats_request *) osr->body;
2854 msg = start_stats_reply(osr, sizeof *ops * 16);
2855 if (psr->port_no != htons(OFPP_NONE)) {
2856 port = port_array_get(&p->ports,
2857 ofp_port_to_odp_port(ntohs(psr->port_no)));
2859 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2862 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2863 append_port_stat(port, port_no, ofconn, &msg);
2867 queue_tx(msg, ofconn, ofconn->reply_counter);
2871 struct flow_stats_cbdata {
2872 struct ofproto *ofproto;
2873 struct ofconn *ofconn;
2878 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2879 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2880 * returned statistic include statistics for all of 'rule''s subrules. */
2882 query_stats(struct ofproto *p, struct rule *rule,
2883 uint64_t *packet_countp, uint64_t *byte_countp)
2885 uint64_t packet_count, byte_count;
2886 struct rule *subrule;
2887 struct odp_flow *odp_flows;
2890 /* Start from historical data for 'rule' itself that are no longer tracked
2891 * by the datapath. This counts, for example, subrules that have
2893 packet_count = rule->packet_count;
2894 byte_count = rule->byte_count;
2896 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2897 * wildcarded then on all of its subrules.
2899 * Also, add any statistics that are not tracked by the datapath for each
2900 * subrule. This includes, for example, statistics for packets that were
2901 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2903 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2904 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2905 if (rule->cr.wc.wildcards) {
2907 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2908 odp_flows[i++].key = subrule->cr.flow;
2909 packet_count += subrule->packet_count;
2910 byte_count += subrule->byte_count;
2913 odp_flows[0].key = rule->cr.flow;
2916 /* Fetch up-to-date statistics from the datapath and add them in. */
2917 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2919 for (i = 0; i < n_odp_flows; i++) {
2920 struct odp_flow *odp_flow = &odp_flows[i];
2921 packet_count += odp_flow->stats.n_packets;
2922 byte_count += odp_flow->stats.n_bytes;
2927 /* Return the stats to the caller. */
2928 *packet_countp = packet_count;
2929 *byte_countp = byte_count;
2933 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2935 struct rule *rule = rule_from_cls_rule(rule_);
2936 struct flow_stats_cbdata *cbdata = cbdata_;
2937 struct ofp_flow_stats *ofs;
2938 uint64_t packet_count, byte_count;
2939 size_t act_len, len;
2940 long long int tdiff = time_msec() - rule->created;
2941 uint32_t sec = tdiff / 1000;
2942 uint32_t msec = tdiff - (sec * 1000);
2944 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2948 act_len = sizeof *rule->actions * rule->n_actions;
2949 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2951 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2953 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2954 ofs->length = htons(len);
2955 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2957 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2958 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2959 ofs->duration_sec = htonl(sec);
2960 ofs->duration_nsec = htonl(msec * 1000000);
2961 ofs->cookie = rule->flow_cookie;
2962 ofs->priority = htons(rule->cr.priority);
2963 ofs->idle_timeout = htons(rule->idle_timeout);
2964 ofs->hard_timeout = htons(rule->hard_timeout);
2965 memset(ofs->pad2, 0, sizeof ofs->pad2);
2966 ofs->packet_count = htonll(packet_count);
2967 ofs->byte_count = htonll(byte_count);
2968 memcpy(ofs->actions, rule->actions, act_len);
2972 table_id_to_include(uint8_t table_id)
2974 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2975 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2976 : table_id == 0xff ? CLS_INC_ALL
2981 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2982 const struct ofp_stats_request *osr,
2985 struct ofp_flow_stats_request *fsr;
2986 struct flow_stats_cbdata cbdata;
2987 struct cls_rule target;
2989 if (arg_size != sizeof *fsr) {
2990 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2992 fsr = (struct ofp_flow_stats_request *) osr->body;
2994 COVERAGE_INC(ofproto_flows_req);
2996 cbdata.ofconn = ofconn;
2997 cbdata.out_port = fsr->out_port;
2998 cbdata.msg = start_stats_reply(osr, 1024);
2999 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3000 classifier_for_each_match(&p->cls, &target,
3001 table_id_to_include(fsr->table_id),
3002 flow_stats_cb, &cbdata);
3003 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3007 struct flow_stats_ds_cbdata {
3008 struct ofproto *ofproto;
3013 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3015 struct rule *rule = rule_from_cls_rule(rule_);
3016 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3017 struct ds *results = cbdata->results;
3018 struct ofp_match match;
3019 uint64_t packet_count, byte_count;
3020 size_t act_len = sizeof *rule->actions * rule->n_actions;
3022 /* Don't report on subrules. */
3023 if (rule->super != NULL) {
3027 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3028 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3029 cbdata->ofproto->tun_id_from_cookie, &match);
3031 ds_put_format(results, "duration=%llds, ",
3032 (time_msec() - rule->created) / 1000);
3033 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3034 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3035 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3036 ofp_print_match(results, &match, true);
3037 ofp_print_actions(results, &rule->actions->header, act_len);
3038 ds_put_cstr(results, "\n");
3041 /* Adds a pretty-printed description of all flows to 'results', including
3042 * those marked hidden by secchan (e.g., by in-band control). */
3044 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3046 struct ofp_match match;
3047 struct cls_rule target;
3048 struct flow_stats_ds_cbdata cbdata;
3050 memset(&match, 0, sizeof match);
3051 match.wildcards = htonl(OVSFW_ALL);
3054 cbdata.results = results;
3056 cls_rule_from_match(&match, 0, false, 0, &target);
3057 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3058 flow_stats_ds_cb, &cbdata);
3061 struct aggregate_stats_cbdata {
3062 struct ofproto *ofproto;
3064 uint64_t packet_count;
3065 uint64_t byte_count;
3070 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3072 struct rule *rule = rule_from_cls_rule(rule_);
3073 struct aggregate_stats_cbdata *cbdata = cbdata_;
3074 uint64_t packet_count, byte_count;
3076 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3080 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3082 cbdata->packet_count += packet_count;
3083 cbdata->byte_count += byte_count;
3088 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3089 const struct ofp_stats_request *osr,
3092 struct ofp_aggregate_stats_request *asr;
3093 struct ofp_aggregate_stats_reply *reply;
3094 struct aggregate_stats_cbdata cbdata;
3095 struct cls_rule target;
3098 if (arg_size != sizeof *asr) {
3099 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3101 asr = (struct ofp_aggregate_stats_request *) osr->body;
3103 COVERAGE_INC(ofproto_agg_request);
3105 cbdata.out_port = asr->out_port;
3106 cbdata.packet_count = 0;
3107 cbdata.byte_count = 0;
3109 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3110 classifier_for_each_match(&p->cls, &target,
3111 table_id_to_include(asr->table_id),
3112 aggregate_stats_cb, &cbdata);
3114 msg = start_stats_reply(osr, sizeof *reply);
3115 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3116 reply->flow_count = htonl(cbdata.n_flows);
3117 reply->packet_count = htonll(cbdata.packet_count);
3118 reply->byte_count = htonll(cbdata.byte_count);
3119 queue_tx(msg, ofconn, ofconn->reply_counter);
3124 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3125 struct ofp_header *oh)
3127 struct ofp_stats_request *osr;
3131 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3136 osr = (struct ofp_stats_request *) oh;
3138 switch (ntohs(osr->type)) {
3140 return handle_desc_stats_request(p, ofconn, osr);
3143 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3145 case OFPST_AGGREGATE:
3146 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3149 return handle_table_stats_request(p, ofconn, osr);
3152 return handle_port_stats_request(p, ofconn, osr, arg_size);
3155 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3158 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3162 static long long int
3163 msec_from_nsec(uint64_t sec, uint32_t nsec)
3165 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3169 update_time(struct ofproto *ofproto, struct rule *rule,
3170 const struct odp_flow_stats *stats)
3172 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3173 if (used > rule->used) {
3175 if (rule->super && used > rule->super->used) {
3176 rule->super->used = used;
3178 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3183 update_stats(struct ofproto *ofproto, struct rule *rule,
3184 const struct odp_flow_stats *stats)
3186 if (stats->n_packets) {
3187 update_time(ofproto, rule, stats);
3188 rule->packet_count += stats->n_packets;
3189 rule->byte_count += stats->n_bytes;
3190 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3195 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3196 * in which no matching flow already exists in the flow table.
3198 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3199 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3200 * code as encoded by ofp_mkerr() on failure.
3202 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3205 add_flow(struct ofproto *p, struct ofconn *ofconn,
3206 const struct ofp_flow_mod *ofm, size_t n_actions)
3208 struct ofpbuf *packet;
3213 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3217 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3219 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3220 ntohs(ofm->priority))) {
3221 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3225 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3226 n_actions, ntohs(ofm->idle_timeout),
3227 ntohs(ofm->hard_timeout), ofm->cookie,
3228 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3229 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3230 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3233 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3234 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3238 in_port = UINT16_MAX;
3241 rule_insert(p, rule, packet, in_port);
3242 ofpbuf_delete(packet);
3246 static struct rule *
3247 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3252 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3254 return rule_from_cls_rule(classifier_find_rule_exactly(
3255 &p->cls, &flow, wildcards,
3256 ntohs(ofm->priority)));
3260 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3261 struct rule *rule, const struct ofp_flow_mod *ofm)
3263 struct ofpbuf *packet;
3268 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3272 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3278 flow_extract(packet, 0, in_port, &flow);
3279 rule_execute(ofproto, rule, packet, &flow);
3280 ofpbuf_delete(packet);
3285 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3287 struct modify_flows_cbdata {
3288 struct ofproto *ofproto;
3289 const struct ofp_flow_mod *ofm;
3294 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3295 size_t n_actions, struct rule *);
3296 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3298 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3299 * encoded by ofp_mkerr() on failure.
3301 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3304 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3305 const struct ofp_flow_mod *ofm, size_t n_actions)
3307 struct modify_flows_cbdata cbdata;
3308 struct cls_rule target;
3312 cbdata.n_actions = n_actions;
3313 cbdata.match = NULL;
3315 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3318 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3319 modify_flows_cb, &cbdata);
3321 /* This credits the packet to whichever flow happened to happened to
3322 * match last. That's weird. Maybe we should do a lookup for the
3323 * flow that actually matches the packet? Who knows. */
3324 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3327 return add_flow(p, ofconn, ofm, n_actions);
3331 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3332 * code as encoded by ofp_mkerr() on failure.
3334 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3337 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3338 struct ofp_flow_mod *ofm, size_t n_actions)
3340 struct rule *rule = find_flow_strict(p, ofm);
3341 if (rule && !rule_is_hidden(rule)) {
3342 modify_flow(p, ofm, n_actions, rule);
3343 return send_buffered_packet(p, ofconn, rule, ofm);
3345 return add_flow(p, ofconn, ofm, n_actions);
3349 /* Callback for modify_flows_loose(). */
3351 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3353 struct rule *rule = rule_from_cls_rule(rule_);
3354 struct modify_flows_cbdata *cbdata = cbdata_;
3356 if (!rule_is_hidden(rule)) {
3357 cbdata->match = rule;
3358 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3362 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3363 * been identified as a flow in 'p''s flow table to be modified, by changing
3364 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3365 * ofp_action[] structures). */
3367 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3368 size_t n_actions, struct rule *rule)
3370 size_t actions_len = n_actions * sizeof *rule->actions;
3372 rule->flow_cookie = ofm->cookie;
3374 /* If the actions are the same, do nothing. */
3375 if (n_actions == rule->n_actions
3376 && !memcmp(ofm->actions, rule->actions, actions_len))
3381 /* Replace actions. */
3382 free(rule->actions);
3383 rule->actions = xmemdup(ofm->actions, actions_len);
3384 rule->n_actions = n_actions;
3386 /* Make sure that the datapath gets updated properly. */
3387 if (rule->cr.wc.wildcards) {
3388 COVERAGE_INC(ofproto_mod_wc_flow);
3389 p->need_revalidate = true;
3391 rule_update_actions(p, rule);
3397 /* OFPFC_DELETE implementation. */
3399 struct delete_flows_cbdata {
3400 struct ofproto *ofproto;
3404 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3405 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3407 /* Implements OFPFC_DELETE. */
3409 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3411 struct delete_flows_cbdata cbdata;
3412 struct cls_rule target;
3415 cbdata.out_port = ofm->out_port;
3417 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3420 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3421 delete_flows_cb, &cbdata);
3424 /* Implements OFPFC_DELETE_STRICT. */
3426 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3428 struct rule *rule = find_flow_strict(p, ofm);
3430 delete_flow(p, rule, ofm->out_port);
3434 /* Callback for delete_flows_loose(). */
3436 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3438 struct rule *rule = rule_from_cls_rule(rule_);
3439 struct delete_flows_cbdata *cbdata = cbdata_;
3441 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3444 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3445 * been identified as a flow to delete from 'p''s flow table, by deleting the
3446 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3449 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3450 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3451 * specified 'out_port'. */
3453 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3455 if (rule_is_hidden(rule)) {
3459 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3463 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3464 rule_remove(p, rule);
3468 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3469 struct ofp_flow_mod *ofm)
3474 error = reject_slave_controller(ofconn, &ofm->header);
3478 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3479 sizeof *ofm->actions, &n_actions);
3484 /* We do not support the emergency flow cache. It will hopefully
3485 * get dropped from OpenFlow in the near future. */
3486 if (ofm->flags & htons(OFPFF_EMERG)) {
3487 /* There isn't a good fit for an error code, so just state that the
3488 * flow table is full. */
3489 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3492 normalize_match(&ofm->match);
3493 if (!ofm->match.wildcards) {
3494 ofm->priority = htons(UINT16_MAX);
3497 error = validate_actions((const union ofp_action *) ofm->actions,
3498 n_actions, p->max_ports);
3503 switch (ntohs(ofm->command)) {
3505 return add_flow(p, ofconn, ofm, n_actions);
3508 return modify_flows_loose(p, ofconn, ofm, n_actions);
3510 case OFPFC_MODIFY_STRICT:
3511 return modify_flow_strict(p, ofconn, ofm, n_actions);
3514 delete_flows_loose(p, ofm);
3517 case OFPFC_DELETE_STRICT:
3518 delete_flow_strict(p, ofm);
3522 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3527 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3531 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3536 p->tun_id_from_cookie = !!msg->set;
3541 handle_role_request(struct ofproto *ofproto,
3542 struct ofconn *ofconn, struct nicira_header *msg)
3544 struct nx_role_request *nrr;
3545 struct nx_role_request *reply;
3549 if (ntohs(msg->header.length) != sizeof *nrr) {
3550 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3551 ntohs(msg->header.length), sizeof *nrr);
3552 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3554 nrr = (struct nx_role_request *) msg;
3556 if (ofconn->type != OFCONN_CONTROLLER) {
3557 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3559 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3562 role = ntohl(nrr->role);
3563 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3564 && role != NX_ROLE_SLAVE) {
3565 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3567 /* There's no good error code for this. */
3568 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3571 if (role == NX_ROLE_MASTER) {
3572 struct ofconn *other;
3574 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3575 &ofproto->controllers) {
3576 if (other->role == NX_ROLE_MASTER) {
3577 other->role = NX_ROLE_SLAVE;
3581 ofconn->role = role;
3583 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3585 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3586 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3587 reply->role = htonl(role);
3588 queue_tx(buf, ofconn, ofconn->reply_counter);
3594 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3596 struct ofp_vendor_header *ovh = msg;
3597 struct nicira_header *nh;
3599 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3600 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3601 "(expected at least %zu)",
3602 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3603 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3605 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3606 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3608 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3609 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3610 "(expected at least %zu)",
3611 ntohs(ovh->header.length), sizeof(struct nicira_header));
3612 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3616 switch (ntohl(nh->subtype)) {
3617 case NXT_STATUS_REQUEST:
3618 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3621 case NXT_TUN_ID_FROM_COOKIE:
3622 return handle_tun_id_from_cookie(p, msg);
3624 case NXT_ROLE_REQUEST:
3625 return handle_role_request(p, ofconn, msg);
3628 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3632 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3634 struct ofp_header *ob;
3637 /* Currently, everything executes synchronously, so we can just
3638 * immediately send the barrier reply. */
3639 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3640 queue_tx(buf, ofconn, ofconn->reply_counter);
3645 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3646 struct ofpbuf *ofp_msg)
3648 struct ofp_header *oh = ofp_msg->data;
3651 COVERAGE_INC(ofproto_recv_openflow);
3653 case OFPT_ECHO_REQUEST:
3654 error = handle_echo_request(ofconn, oh);
3657 case OFPT_ECHO_REPLY:
3661 case OFPT_FEATURES_REQUEST:
3662 error = handle_features_request(p, ofconn, oh);
3665 case OFPT_GET_CONFIG_REQUEST:
3666 error = handle_get_config_request(p, ofconn, oh);
3669 case OFPT_SET_CONFIG:
3670 error = handle_set_config(p, ofconn, ofp_msg->data);
3673 case OFPT_PACKET_OUT:
3674 error = handle_packet_out(p, ofconn, ofp_msg->data);
3678 error = handle_port_mod(p, ofconn, oh);
3682 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3685 case OFPT_STATS_REQUEST:
3686 error = handle_stats_request(p, ofconn, oh);
3690 error = handle_vendor(p, ofconn, ofp_msg->data);
3693 case OFPT_BARRIER_REQUEST:
3694 error = handle_barrier_request(ofconn, oh);
3698 if (VLOG_IS_WARN_ENABLED()) {
3699 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3700 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3703 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3708 send_error_oh(ofconn, ofp_msg->data, error);
3713 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3715 struct odp_msg *msg = packet->data;
3717 struct ofpbuf payload;
3720 payload.data = msg + 1;
3721 payload.size = msg->length - sizeof *msg;
3722 flow_extract(&payload, msg->arg, msg->port, &flow);
3724 /* Check with in-band control to see if this packet should be sent
3725 * to the local port regardless of the flow table. */
3726 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3727 union odp_action action;
3729 memset(&action, 0, sizeof(action));
3730 action.output.type = ODPAT_OUTPUT;
3731 action.output.port = ODPP_LOCAL;
3732 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3735 rule = lookup_valid_rule(p, &flow);
3737 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3738 struct ofport *port = port_array_get(&p->ports, msg->port);
3740 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3741 COVERAGE_INC(ofproto_no_packet_in);
3742 /* XXX install 'drop' flow entry */
3743 ofpbuf_delete(packet);
3747 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3750 COVERAGE_INC(ofproto_packet_in);
3751 send_packet_in(p, packet);
3755 if (rule->cr.wc.wildcards) {
3756 rule = rule_create_subrule(p, rule, &flow);
3757 rule_make_actions(p, rule, packet);
3759 if (!rule->may_install) {
3760 /* The rule is not installable, that is, we need to process every
3761 * packet, so process the current packet and set its actions into
3763 rule_make_actions(p, rule, packet);
3765 /* XXX revalidate rule if it needs it */
3769 rule_execute(p, rule, &payload, &flow);
3770 rule_reinstall(p, rule);
3772 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3774 * Extra-special case for fail-open mode.
3776 * We are in fail-open mode and the packet matched the fail-open rule,
3777 * but we are connected to a controller too. We should send the packet
3778 * up to the controller in the hope that it will try to set up a flow
3779 * and thereby allow us to exit fail-open.
3781 * See the top-level comment in fail-open.c for more information.
3783 send_packet_in(p, packet);
3785 ofpbuf_delete(packet);
3790 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3792 struct odp_msg *msg = packet->data;
3794 switch (msg->type) {
3795 case _ODPL_ACTION_NR:
3796 COVERAGE_INC(ofproto_ctlr_action);
3797 send_packet_in(p, packet);
3800 case _ODPL_SFLOW_NR:
3802 ofproto_sflow_received(p->sflow, msg);
3804 ofpbuf_delete(packet);
3808 handle_odp_miss_msg(p, packet);
3812 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3819 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3821 struct rule *sub = rule_from_cls_rule(sub_);
3822 struct revalidate_cbdata *cbdata = cbdata_;
3824 if (cbdata->revalidate_all
3825 || (cbdata->revalidate_subrules && sub->super)
3826 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3827 revalidate_rule(cbdata->ofproto, sub);
3832 revalidate_rule(struct ofproto *p, struct rule *rule)
3834 const flow_t *flow = &rule->cr.flow;
3836 COVERAGE_INC(ofproto_revalidate_rule);
3839 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3841 rule_remove(p, rule);
3843 } else if (super != rule->super) {
3844 COVERAGE_INC(ofproto_revalidate_moved);
3845 list_remove(&rule->list);
3846 list_push_back(&super->list, &rule->list);
3847 rule->super = super;
3848 rule->hard_timeout = super->hard_timeout;
3849 rule->idle_timeout = super->idle_timeout;
3850 rule->created = super->created;
3855 rule_update_actions(p, rule);
3859 static struct ofpbuf *
3860 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3861 long long int now, uint8_t reason)
3863 struct ofp_flow_removed *ofr;
3865 long long int tdiff = now - rule->created;
3866 uint32_t sec = tdiff / 1000;
3867 uint32_t msec = tdiff - (sec * 1000);
3869 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3870 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3872 ofr->cookie = rule->flow_cookie;
3873 ofr->priority = htons(rule->cr.priority);
3874 ofr->reason = reason;
3875 ofr->duration_sec = htonl(sec);
3876 ofr->duration_nsec = htonl(msec * 1000000);
3877 ofr->idle_timeout = htons(rule->idle_timeout);
3878 ofr->packet_count = htonll(rule->packet_count);
3879 ofr->byte_count = htonll(rule->byte_count);
3885 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3887 assert(rule->installed);
3888 assert(!rule->cr.wc.wildcards);
3891 rule_remove(ofproto, rule);
3893 rule_uninstall(ofproto, rule);
3898 send_flow_removed(struct ofproto *p, struct rule *rule,
3899 long long int now, uint8_t reason)
3901 struct ofconn *ofconn;
3902 struct ofconn *prev;
3903 struct ofpbuf *buf = NULL;
3905 /* We limit the maximum number of queued flow expirations it by accounting
3906 * them under the counter for replies. That works because preventing
3907 * OpenFlow requests from being processed also prevents new flows from
3908 * being added (and expiring). (It also prevents processing OpenFlow
3909 * requests that would not add new flows, so it is imperfect.) */
3912 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3913 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
3914 && ofconn_receives_async_msgs(ofconn)) {
3916 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3918 buf = compose_flow_removed(p, rule, now, reason);
3924 queue_tx(buf, prev, prev->reply_counter);
3930 expire_rule(struct cls_rule *cls_rule, void *p_)
3932 struct ofproto *p = p_;
3933 struct rule *rule = rule_from_cls_rule(cls_rule);
3934 long long int hard_expire, idle_expire, expire, now;
3936 hard_expire = (rule->hard_timeout
3937 ? rule->created + rule->hard_timeout * 1000
3939 idle_expire = (rule->idle_timeout
3940 && (rule->super || list_is_empty(&rule->list))
3941 ? rule->used + rule->idle_timeout * 1000
3943 expire = MIN(hard_expire, idle_expire);
3947 if (rule->installed && now >= rule->used + 5000) {
3948 uninstall_idle_flow(p, rule);
3949 } else if (!rule->cr.wc.wildcards) {
3950 active_timeout(p, rule);
3956 COVERAGE_INC(ofproto_expired);
3958 /* Update stats. This code will be a no-op if the rule expired
3959 * due to an idle timeout. */
3960 if (rule->cr.wc.wildcards) {
3961 struct rule *subrule, *next;
3962 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3963 rule_remove(p, subrule);
3966 rule_uninstall(p, rule);
3969 if (!rule_is_hidden(rule)) {
3970 send_flow_removed(p, rule, now,
3972 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3974 rule_remove(p, rule);
3978 active_timeout(struct ofproto *ofproto, struct rule *rule)
3980 if (ofproto->netflow && !is_controller_rule(rule) &&
3981 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3982 struct ofexpired expired;
3983 struct odp_flow odp_flow;
3985 /* Get updated flow stats. */
3986 memset(&odp_flow, 0, sizeof odp_flow);
3987 if (rule->installed) {
3988 odp_flow.key = rule->cr.flow;
3989 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3990 dpif_flow_get(ofproto->dpif, &odp_flow);
3992 if (odp_flow.stats.n_packets) {
3993 update_time(ofproto, rule, &odp_flow.stats);
3994 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3995 odp_flow.stats.tcp_flags);
3999 expired.flow = rule->cr.flow;
4000 expired.packet_count = rule->packet_count +
4001 odp_flow.stats.n_packets;
4002 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4003 expired.used = rule->used;
4005 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4007 /* Schedule us to send the accumulated records once we have
4008 * collected all of them. */
4009 poll_immediate_wake();
4014 update_used(struct ofproto *p)
4016 struct odp_flow *flows;
4021 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4026 for (i = 0; i < n_flows; i++) {
4027 struct odp_flow *f = &flows[i];
4030 rule = rule_from_cls_rule(
4031 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4032 if (!rule || !rule->installed) {
4033 COVERAGE_INC(ofproto_unexpected_rule);
4034 dpif_flow_del(p->dpif, f);
4038 update_time(p, rule, &f->stats);
4039 rule_account(p, rule, f->stats.n_bytes);
4044 /* pinsched callback for sending 'packet' on 'ofconn'. */
4046 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4048 struct ofconn *ofconn = ofconn_;
4050 rconn_send_with_limit(ofconn->rconn, packet,
4051 ofconn->packet_in_counter, 100);
4054 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4055 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4056 * packet scheduler for sending.
4058 * 'max_len' specifies the maximum number of bytes of the packet to send on
4059 * 'ofconn' (INT_MAX specifies no limit).
4061 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4062 * ownership is transferred to this function. */
4064 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4067 struct ofproto *ofproto = ofconn->ofproto;
4068 struct ofp_packet_in *opi = packet->data;
4069 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4070 int send_len, trim_size;
4074 if (opi->reason == OFPR_ACTION) {
4075 buffer_id = UINT32_MAX;
4076 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4077 buffer_id = pktbuf_get_null();
4078 } else if (!ofconn->pktbuf) {
4079 buffer_id = UINT32_MAX;
4081 struct ofpbuf payload;
4082 payload.data = opi->data;
4083 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4084 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4087 /* Figure out how much of the packet to send. */
4088 send_len = ntohs(opi->total_len);
4089 if (buffer_id != UINT32_MAX) {
4090 send_len = MIN(send_len, ofconn->miss_send_len);
4092 send_len = MIN(send_len, max_len);
4094 /* Adjust packet length and clone if necessary. */
4095 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4097 packet = ofpbuf_clone_data(packet->data, trim_size);
4100 packet->size = trim_size;
4103 /* Update packet headers. */
4104 opi->buffer_id = htonl(buffer_id);
4105 update_openflow_length(packet);
4107 /* Hand over to packet scheduler. It might immediately call into
4108 * do_send_packet_in() or it might buffer it for a while (until a later
4109 * call to pinsched_run()). */
4110 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4111 packet, do_send_packet_in, ofconn);
4114 /* Replace struct odp_msg header in 'packet' by equivalent struct
4115 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4116 * returned by dpif_recv()).
4118 * The conversion is not complete: the caller still needs to trim any unneeded
4119 * payload off the end of the buffer, set the length in the OpenFlow header,
4120 * and set buffer_id. Those require us to know the controller settings and so
4121 * must be done on a per-controller basis.
4123 * Returns the maximum number of bytes of the packet that should be sent to
4124 * the controller (INT_MAX if no limit). */
4126 do_convert_to_packet_in(struct ofpbuf *packet)
4128 struct odp_msg *msg = packet->data;
4129 struct ofp_packet_in *opi;
4135 /* Extract relevant header fields */
4136 if (msg->type == _ODPL_ACTION_NR) {
4137 reason = OFPR_ACTION;
4140 reason = OFPR_NO_MATCH;
4143 total_len = msg->length - sizeof *msg;
4144 in_port = odp_port_to_ofp_port(msg->port);
4146 /* Repurpose packet buffer by overwriting header. */
4147 ofpbuf_pull(packet, sizeof(struct odp_msg));
4148 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4149 opi->header.version = OFP_VERSION;
4150 opi->header.type = OFPT_PACKET_IN;
4151 opi->total_len = htons(total_len);
4152 opi->in_port = htons(in_port);
4153 opi->reason = reason;
4158 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4159 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4160 * as necessary according to their individual configurations.
4162 * 'packet' must have sufficient headroom to convert it into a struct
4163 * ofp_packet_in (e.g. as returned by dpif_recv()).
4165 * Takes ownership of 'packet'. */
4167 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4169 struct ofconn *ofconn, *prev;
4172 max_len = do_convert_to_packet_in(packet);
4175 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4176 if (ofconn_receives_async_msgs(ofconn)) {
4178 schedule_packet_in(prev, packet, max_len, true);
4184 schedule_packet_in(prev, packet, max_len, false);
4186 ofpbuf_delete(packet);
4191 pick_datapath_id(const struct ofproto *ofproto)
4193 const struct ofport *port;
4195 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4197 uint8_t ea[ETH_ADDR_LEN];
4200 error = netdev_get_etheraddr(port->netdev, ea);
4202 return eth_addr_to_uint64(ea);
4204 VLOG_WARN("could not get MAC address for %s (%s)",
4205 netdev_get_name(port->netdev), strerror(error));
4207 return ofproto->fallback_dpid;
4211 pick_fallback_dpid(void)
4213 uint8_t ea[ETH_ADDR_LEN];
4214 eth_addr_nicira_random(ea);
4215 return eth_addr_to_uint64(ea);
4219 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4220 struct odp_actions *actions, tag_type *tags,
4221 uint16_t *nf_output_iface, void *ofproto_)
4223 struct ofproto *ofproto = ofproto_;
4226 /* Drop frames for reserved multicast addresses. */
4227 if (eth_addr_is_reserved(flow->dl_dst)) {
4231 /* Learn source MAC (but don't try to learn from revalidation). */
4232 if (packet != NULL) {
4233 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4236 /* The log messages here could actually be useful in debugging,
4237 * so keep the rate limit relatively high. */
4238 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4239 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4240 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4241 ofproto_revalidate(ofproto, rev_tag);
4245 /* Determine output port. */
4246 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
4248 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4249 } else if (out_port != flow->in_port) {
4250 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4251 *nf_output_iface = out_port;
4259 static const struct ofhooks default_ofhooks = {
4261 default_normal_ofhook_cb,