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(vconn), OFCONN_TRANSIENT);
1082 } else if (retval != EAGAIN) {
1083 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1087 for (i = 0; i < p->n_snoops; i++) {
1088 struct vconn *vconn;
1091 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1093 add_snooper(p, vconn);
1094 } else if (retval != EAGAIN) {
1095 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1099 if (time_msec() >= p->next_expiration) {
1100 COVERAGE_INC(ofproto_expiration);
1101 p->next_expiration = time_msec() + 1000;
1104 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1106 /* Let the hook know that we're at a stable point: all outstanding data
1107 * in existing flows has been accounted to the account_cb. Thus, the
1108 * hook can now reasonably do operations that depend on having accurate
1109 * flow volume accounting (currently, that's just bond rebalancing). */
1110 if (p->ofhooks->account_checkpoint_cb) {
1111 p->ofhooks->account_checkpoint_cb(p->aux);
1116 netflow_run(p->netflow);
1119 ofproto_sflow_run(p->sflow);
1125 struct revalidate_cbdata {
1126 struct ofproto *ofproto;
1127 bool revalidate_all; /* Revalidate all exact-match rules? */
1128 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1129 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1133 ofproto_run2(struct ofproto *p, bool revalidate_all)
1135 if (p->need_revalidate || revalidate_all
1136 || !tag_set_is_empty(&p->revalidate_set)) {
1137 struct revalidate_cbdata cbdata;
1139 cbdata.revalidate_all = revalidate_all;
1140 cbdata.revalidate_subrules = p->need_revalidate;
1141 cbdata.revalidate_set = p->revalidate_set;
1142 tag_set_init(&p->revalidate_set);
1143 COVERAGE_INC(ofproto_revalidate);
1144 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1145 p->need_revalidate = false;
1152 ofproto_wait(struct ofproto *p)
1154 struct ofconn *ofconn;
1157 dpif_recv_wait(p->dpif);
1158 dpif_port_poll_wait(p->dpif);
1159 netdev_monitor_poll_wait(p->netdev_monitor);
1160 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1161 ofconn_wait(ofconn);
1164 poll_timer_wait_until(p->next_in_band_update);
1165 in_band_wait(p->in_band);
1168 fail_open_wait(p->fail_open);
1171 ofproto_sflow_wait(p->sflow);
1173 if (!tag_set_is_empty(&p->revalidate_set)) {
1174 poll_immediate_wake();
1176 if (p->need_revalidate) {
1177 /* Shouldn't happen, but if it does just go around again. */
1178 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1179 poll_immediate_wake();
1180 } else if (p->next_expiration != LLONG_MAX) {
1181 poll_timer_wait_until(p->next_expiration);
1183 for (i = 0; i < p->n_listeners; i++) {
1184 pvconn_wait(p->listeners[i]);
1186 for (i = 0; i < p->n_snoops; i++) {
1187 pvconn_wait(p->snoops[i]);
1192 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1194 tag_set_add(&ofproto->revalidate_set, tag);
1198 ofproto_get_revalidate_set(struct ofproto *ofproto)
1200 return &ofproto->revalidate_set;
1204 ofproto_is_alive(const struct ofproto *p)
1206 return !hmap_is_empty(&p->controllers);
1210 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1211 const union ofp_action *actions, size_t n_actions,
1212 const struct ofpbuf *packet)
1214 struct odp_actions odp_actions;
1217 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1223 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1225 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1226 odp_actions.n_actions, packet);
1231 ofproto_add_flow(struct ofproto *p,
1232 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1233 const union ofp_action *actions, size_t n_actions,
1237 rule = rule_create(p, NULL, actions, n_actions,
1238 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1240 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1241 rule_insert(p, rule, NULL, 0);
1245 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1246 uint32_t wildcards, unsigned int priority)
1250 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1254 rule_remove(ofproto, rule);
1259 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1261 struct rule *rule = rule_from_cls_rule(rule_);
1262 struct ofproto *ofproto = ofproto_;
1264 /* Mark the flow as not installed, even though it might really be
1265 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1266 * There is no point in uninstalling it individually since we are about to
1267 * blow away all the flows with dpif_flow_flush(). */
1268 rule->installed = false;
1270 rule_remove(ofproto, rule);
1274 ofproto_flush_flows(struct ofproto *ofproto)
1276 COVERAGE_INC(ofproto_flush);
1277 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1278 dpif_flow_flush(ofproto->dpif);
1279 if (ofproto->in_band) {
1280 in_band_flushed(ofproto->in_band);
1282 if (ofproto->fail_open) {
1283 fail_open_flushed(ofproto->fail_open);
1288 reinit_ports(struct ofproto *p)
1290 struct svec devnames;
1291 struct ofport *ofport;
1292 unsigned int port_no;
1293 struct odp_port *odp_ports;
1297 svec_init(&devnames);
1298 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1299 svec_add (&devnames, (char *) ofport->opp.name);
1301 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1302 for (i = 0; i < n_odp_ports; i++) {
1303 svec_add (&devnames, odp_ports[i].devname);
1307 svec_sort_unique(&devnames);
1308 for (i = 0; i < devnames.n; i++) {
1309 update_port(p, devnames.names[i]);
1311 svec_destroy(&devnames);
1315 refresh_port_group(struct ofproto *p, unsigned int group)
1319 struct ofport *port;
1320 unsigned int port_no;
1322 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1324 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1326 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1327 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1328 ports[n_ports++] = port_no;
1331 dpif_port_group_set(p->dpif, group, ports, n_ports);
1338 refresh_port_groups(struct ofproto *p)
1340 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1341 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1343 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1347 static struct ofport *
1348 make_ofport(const struct odp_port *odp_port)
1350 struct netdev_options netdev_options;
1351 enum netdev_flags flags;
1352 struct ofport *ofport;
1353 struct netdev *netdev;
1357 memset(&netdev_options, 0, sizeof netdev_options);
1358 netdev_options.name = odp_port->devname;
1359 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1361 error = netdev_open(&netdev_options, &netdev);
1363 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1364 "cannot be opened (%s)",
1365 odp_port->devname, odp_port->port,
1366 odp_port->devname, strerror(error));
1370 ofport = xmalloc(sizeof *ofport);
1371 ofport->netdev = netdev;
1372 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1373 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1374 memcpy(ofport->opp.name, odp_port->devname,
1375 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1376 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1378 netdev_get_flags(netdev, &flags);
1379 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1381 netdev_get_carrier(netdev, &carrier);
1382 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1384 netdev_get_features(netdev,
1385 &ofport->opp.curr, &ofport->opp.advertised,
1386 &ofport->opp.supported, &ofport->opp.peer);
1391 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1393 if (port_array_get(&p->ports, odp_port->port)) {
1394 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1397 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1398 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1407 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1409 const struct ofp_phy_port *a = &a_->opp;
1410 const struct ofp_phy_port *b = &b_->opp;
1412 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1413 return (a->port_no == b->port_no
1414 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1415 && !strcmp((char *) a->name, (char *) b->name)
1416 && a->state == b->state
1417 && a->config == b->config
1418 && a->curr == b->curr
1419 && a->advertised == b->advertised
1420 && a->supported == b->supported
1421 && a->peer == b->peer);
1425 send_port_status(struct ofproto *p, const struct ofport *ofport,
1428 /* XXX Should limit the number of queued port status change messages. */
1429 struct ofconn *ofconn;
1430 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1431 struct ofp_port_status *ops;
1434 if (!ofconn_receives_async_msgs(ofconn)) {
1438 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1439 ops->reason = reason;
1440 ops->desc = ofport->opp;
1441 hton_ofp_phy_port(&ops->desc);
1442 queue_tx(b, ofconn, NULL);
1444 if (p->ofhooks->port_changed_cb) {
1445 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1450 ofport_install(struct ofproto *p, struct ofport *ofport)
1452 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1453 const char *netdev_name = (const char *) ofport->opp.name;
1455 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1456 port_array_set(&p->ports, odp_port, ofport);
1457 shash_add(&p->port_by_name, netdev_name, ofport);
1459 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1464 ofport_remove(struct ofproto *p, struct ofport *ofport)
1466 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1468 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1469 port_array_set(&p->ports, odp_port, NULL);
1470 shash_delete(&p->port_by_name,
1471 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1473 ofproto_sflow_del_port(p->sflow, odp_port);
1478 ofport_free(struct ofport *ofport)
1481 netdev_close(ofport->netdev);
1487 update_port(struct ofproto *p, const char *devname)
1489 struct odp_port odp_port;
1490 struct ofport *old_ofport;
1491 struct ofport *new_ofport;
1494 COVERAGE_INC(ofproto_update_port);
1496 /* Query the datapath for port information. */
1497 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1499 /* Find the old ofport. */
1500 old_ofport = shash_find_data(&p->port_by_name, devname);
1503 /* There's no port named 'devname' but there might be a port with
1504 * the same port number. This could happen if a port is deleted
1505 * and then a new one added in its place very quickly, or if a port
1506 * is renamed. In the former case we want to send an OFPPR_DELETE
1507 * and an OFPPR_ADD, and in the latter case we want to send a
1508 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1509 * the old port's ifindex against the new port, or perhaps less
1510 * reliably but more portably by comparing the old port's MAC
1511 * against the new port's MAC. However, this code isn't that smart
1512 * and always sends an OFPPR_MODIFY (XXX). */
1513 old_ofport = port_array_get(&p->ports, odp_port.port);
1515 } else if (error != ENOENT && error != ENODEV) {
1516 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1517 "%s", strerror(error));
1521 /* Create a new ofport. */
1522 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1524 /* Eliminate a few pathological cases. */
1525 if (!old_ofport && !new_ofport) {
1527 } else if (old_ofport && new_ofport) {
1528 /* Most of the 'config' bits are OpenFlow soft state, but
1529 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1530 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1531 * leaves the other bits 0.) */
1532 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1534 if (ofport_equal(old_ofport, new_ofport)) {
1535 /* False alarm--no change. */
1536 ofport_free(new_ofport);
1541 /* Now deal with the normal cases. */
1543 ofport_remove(p, old_ofport);
1546 ofport_install(p, new_ofport);
1548 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1549 (!old_ofport ? OFPPR_ADD
1550 : !new_ofport ? OFPPR_DELETE
1552 ofport_free(old_ofport);
1554 /* Update port groups. */
1555 refresh_port_groups(p);
1559 init_ports(struct ofproto *p)
1561 struct odp_port *ports;
1566 error = dpif_port_list(p->dpif, &ports, &n_ports);
1571 for (i = 0; i < n_ports; i++) {
1572 const struct odp_port *odp_port = &ports[i];
1573 if (!ofport_conflicts(p, odp_port)) {
1574 struct ofport *ofport = make_ofport(odp_port);
1576 ofport_install(p, ofport);
1581 refresh_port_groups(p);
1585 static struct ofconn *
1586 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1588 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1589 ofconn->ofproto = p;
1590 list_push_back(&p->all_conns, &ofconn->node);
1591 ofconn->rconn = rconn;
1592 ofconn->type = type;
1593 ofconn->role = NX_ROLE_OTHER;
1594 ofconn->packet_in_counter = rconn_packet_counter_create ();
1595 ofconn->pktbuf = NULL;
1596 ofconn->miss_send_len = 0;
1597 ofconn->reply_counter = rconn_packet_counter_create ();
1602 ofconn_destroy(struct ofconn *ofconn)
1604 if (ofconn->type == OFCONN_CONTROLLER) {
1605 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1607 discovery_destroy(ofconn->discovery);
1609 list_remove(&ofconn->node);
1610 switch_status_unregister(ofconn->ss);
1611 rconn_destroy(ofconn->rconn);
1612 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1613 rconn_packet_counter_destroy(ofconn->reply_counter);
1614 pktbuf_destroy(ofconn->pktbuf);
1619 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1624 if (ofconn->discovery) {
1625 char *controller_name;
1626 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1627 discovery_question_connectivity(ofconn->discovery);
1629 if (discovery_run(ofconn->discovery, &controller_name)) {
1630 if (controller_name) {
1631 rconn_connect(ofconn->rconn, controller_name);
1633 rconn_disconnect(ofconn->rconn);
1638 for (i = 0; i < N_SCHEDULERS; i++) {
1639 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1642 rconn_run(ofconn->rconn);
1644 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1645 /* Limit the number of iterations to prevent other tasks from
1647 for (iteration = 0; iteration < 50; iteration++) {
1648 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1653 fail_open_maybe_recover(p->fail_open);
1655 handle_openflow(ofconn, p, of_msg);
1656 ofpbuf_delete(of_msg);
1660 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1661 ofconn_destroy(ofconn);
1666 ofconn_wait(struct ofconn *ofconn)
1670 if (ofconn->discovery) {
1671 discovery_wait(ofconn->discovery);
1673 for (i = 0; i < N_SCHEDULERS; i++) {
1674 pinsched_wait(ofconn->schedulers[i]);
1676 rconn_run_wait(ofconn->rconn);
1677 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1678 rconn_recv_wait(ofconn->rconn);
1680 COVERAGE_INC(ofproto_ofconn_stuck);
1684 /* Returns true if 'ofconn' should receive asynchronous messages. */
1686 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1688 if (ofconn->type == OFCONN_CONTROLLER) {
1689 /* Ordinary controllers always get asynchronous messages unless they
1690 * have configured themselves as "slaves". */
1691 return ofconn->role != NX_ROLE_SLAVE;
1693 /* Transient connections don't get asynchronous messages unless they
1694 * have explicitly asked for them by setting a nonzero miss send
1696 return ofconn->miss_send_len > 0;
1700 /* Caller is responsible for initializing the 'cr' member of the returned
1702 static struct rule *
1703 rule_create(struct ofproto *ofproto, struct rule *super,
1704 const union ofp_action *actions, size_t n_actions,
1705 uint16_t idle_timeout, uint16_t hard_timeout,
1706 uint64_t flow_cookie, bool send_flow_removed)
1708 struct rule *rule = xzalloc(sizeof *rule);
1709 rule->idle_timeout = idle_timeout;
1710 rule->hard_timeout = hard_timeout;
1711 rule->flow_cookie = flow_cookie;
1712 rule->used = rule->created = time_msec();
1713 rule->send_flow_removed = send_flow_removed;
1714 rule->super = super;
1716 list_push_back(&super->list, &rule->list);
1718 list_init(&rule->list);
1720 rule->n_actions = n_actions;
1721 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1722 netflow_flow_clear(&rule->nf_flow);
1723 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1728 static struct rule *
1729 rule_from_cls_rule(const struct cls_rule *cls_rule)
1731 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1735 rule_free(struct rule *rule)
1737 free(rule->actions);
1738 free(rule->odp_actions);
1742 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1743 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1744 * through all of its subrules and revalidates them, destroying any that no
1745 * longer has a super-rule (which is probably all of them).
1747 * Before calling this function, the caller must make have removed 'rule' from
1748 * the classifier. If 'rule' is an exact-match rule, the caller is also
1749 * responsible for ensuring that it has been uninstalled from the datapath. */
1751 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1754 struct rule *subrule, *next;
1755 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1756 revalidate_rule(ofproto, subrule);
1759 list_remove(&rule->list);
1765 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1767 const union ofp_action *oa;
1768 struct actions_iterator i;
1770 if (out_port == htons(OFPP_NONE)) {
1773 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1774 oa = actions_next(&i)) {
1775 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1782 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1783 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1785 * The flow that 'packet' actually contains does not need to actually match
1786 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1787 * the packet and byte counters for 'rule' will be credited for the packet sent
1788 * out whether or not the packet actually matches 'rule'.
1790 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1791 * the caller must already have accurately composed ODP actions for it given
1792 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1793 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1794 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1795 * actions and apply them to 'packet'. */
1797 rule_execute(struct ofproto *ofproto, struct rule *rule,
1798 struct ofpbuf *packet, const flow_t *flow)
1800 const union odp_action *actions;
1802 struct odp_actions a;
1804 /* Grab or compose the ODP actions.
1806 * The special case for an exact-match 'rule' where 'flow' is not the
1807 * rule's flow is important to avoid, e.g., sending a packet out its input
1808 * port simply because the ODP actions were composed for the wrong
1810 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1811 struct rule *super = rule->super ? rule->super : rule;
1812 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1813 packet, &a, NULL, 0, NULL)) {
1816 actions = a.actions;
1817 n_actions = a.n_actions;
1819 actions = rule->odp_actions;
1820 n_actions = rule->n_odp_actions;
1823 /* Execute the ODP actions. */
1824 if (!dpif_execute(ofproto->dpif, flow->in_port,
1825 actions, n_actions, packet)) {
1826 struct odp_flow_stats stats;
1827 flow_extract_stats(flow, packet, &stats);
1828 update_stats(ofproto, rule, &stats);
1829 rule->used = time_msec();
1830 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1835 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1838 struct rule *displaced_rule;
1840 /* Insert the rule in the classifier. */
1841 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1842 if (!rule->cr.wc.wildcards) {
1843 rule_make_actions(p, rule, packet);
1846 /* Send the packet and credit it to the rule. */
1849 flow_extract(packet, 0, in_port, &flow);
1850 rule_execute(p, rule, packet, &flow);
1853 /* Install the rule in the datapath only after sending the packet, to
1854 * avoid packet reordering. */
1855 if (rule->cr.wc.wildcards) {
1856 COVERAGE_INC(ofproto_add_wc_flow);
1857 p->need_revalidate = true;
1859 rule_install(p, rule, displaced_rule);
1862 /* Free the rule that was displaced, if any. */
1863 if (displaced_rule) {
1864 rule_destroy(p, displaced_rule);
1868 static struct rule *
1869 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1872 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1873 rule->idle_timeout, rule->hard_timeout,
1875 COVERAGE_INC(ofproto_subrule_create);
1876 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1877 : rule->cr.priority), &subrule->cr);
1878 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1884 rule_remove(struct ofproto *ofproto, struct rule *rule)
1886 if (rule->cr.wc.wildcards) {
1887 COVERAGE_INC(ofproto_del_wc_flow);
1888 ofproto->need_revalidate = true;
1890 rule_uninstall(ofproto, rule);
1892 classifier_remove(&ofproto->cls, &rule->cr);
1893 rule_destroy(ofproto, rule);
1896 /* Returns true if the actions changed, false otherwise. */
1898 rule_make_actions(struct ofproto *p, struct rule *rule,
1899 const struct ofpbuf *packet)
1901 const struct rule *super;
1902 struct odp_actions a;
1905 assert(!rule->cr.wc.wildcards);
1907 super = rule->super ? rule->super : rule;
1909 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1910 packet, &a, &rule->tags, &rule->may_install,
1911 &rule->nf_flow.output_iface);
1913 actions_len = a.n_actions * sizeof *a.actions;
1914 if (rule->n_odp_actions != a.n_actions
1915 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1916 COVERAGE_INC(ofproto_odp_unchanged);
1917 free(rule->odp_actions);
1918 rule->n_odp_actions = a.n_actions;
1919 rule->odp_actions = xmemdup(a.actions, actions_len);
1927 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1928 struct odp_flow_put *put)
1930 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1931 put->flow.key = rule->cr.flow;
1932 put->flow.actions = rule->odp_actions;
1933 put->flow.n_actions = rule->n_odp_actions;
1934 put->flow.flags = 0;
1936 return dpif_flow_put(ofproto->dpif, put);
1940 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1942 assert(!rule->cr.wc.wildcards);
1944 if (rule->may_install) {
1945 struct odp_flow_put put;
1946 if (!do_put_flow(p, rule,
1947 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1949 rule->installed = true;
1950 if (displaced_rule) {
1951 update_stats(p, displaced_rule, &put.flow.stats);
1952 rule_post_uninstall(p, displaced_rule);
1955 } else if (displaced_rule) {
1956 rule_uninstall(p, displaced_rule);
1961 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1963 if (rule->installed) {
1964 struct odp_flow_put put;
1965 COVERAGE_INC(ofproto_dp_missed);
1966 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1968 rule_install(ofproto, rule, NULL);
1973 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1975 bool actions_changed;
1976 uint16_t new_out_iface, old_out_iface;
1978 old_out_iface = rule->nf_flow.output_iface;
1979 actions_changed = rule_make_actions(ofproto, rule, NULL);
1981 if (rule->may_install) {
1982 if (rule->installed) {
1983 if (actions_changed) {
1984 struct odp_flow_put put;
1985 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1986 | ODPPF_ZERO_STATS, &put);
1987 update_stats(ofproto, rule, &put.flow.stats);
1989 /* Temporarily set the old output iface so that NetFlow
1990 * messages have the correct output interface for the old
1992 new_out_iface = rule->nf_flow.output_iface;
1993 rule->nf_flow.output_iface = old_out_iface;
1994 rule_post_uninstall(ofproto, rule);
1995 rule->nf_flow.output_iface = new_out_iface;
1998 rule_install(ofproto, rule, NULL);
2001 rule_uninstall(ofproto, rule);
2006 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2008 uint64_t total_bytes = rule->byte_count + extra_bytes;
2010 if (ofproto->ofhooks->account_flow_cb
2011 && total_bytes > rule->accounted_bytes)
2013 ofproto->ofhooks->account_flow_cb(
2014 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
2015 total_bytes - rule->accounted_bytes, ofproto->aux);
2016 rule->accounted_bytes = total_bytes;
2021 rule_uninstall(struct ofproto *p, struct rule *rule)
2023 assert(!rule->cr.wc.wildcards);
2024 if (rule->installed) {
2025 struct odp_flow odp_flow;
2027 odp_flow.key = rule->cr.flow;
2028 odp_flow.actions = NULL;
2029 odp_flow.n_actions = 0;
2031 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2032 update_stats(p, rule, &odp_flow.stats);
2034 rule->installed = false;
2036 rule_post_uninstall(p, rule);
2041 is_controller_rule(struct rule *rule)
2043 /* If the only action is send to the controller then don't report
2044 * NetFlow expiration messages since it is just part of the control
2045 * logic for the network and not real traffic. */
2047 if (rule && rule->super) {
2048 struct rule *super = rule->super;
2050 return super->n_actions == 1 &&
2051 super->actions[0].type == htons(OFPAT_OUTPUT) &&
2052 super->actions[0].output.port == htons(OFPP_CONTROLLER);
2059 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2061 struct rule *super = rule->super;
2063 rule_account(ofproto, rule, 0);
2065 if (ofproto->netflow && !is_controller_rule(rule)) {
2066 struct ofexpired expired;
2067 expired.flow = rule->cr.flow;
2068 expired.packet_count = rule->packet_count;
2069 expired.byte_count = rule->byte_count;
2070 expired.used = rule->used;
2071 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2074 super->packet_count += rule->packet_count;
2075 super->byte_count += rule->byte_count;
2077 /* Reset counters to prevent double counting if the rule ever gets
2079 rule->packet_count = 0;
2080 rule->byte_count = 0;
2081 rule->accounted_bytes = 0;
2083 netflow_flow_clear(&rule->nf_flow);
2088 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2089 struct rconn_packet_counter *counter)
2091 update_openflow_length(msg);
2092 if (rconn_send(ofconn->rconn, msg, counter)) {
2098 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2099 int error, const void *data, size_t len)
2102 struct ofp_error_msg *oem;
2104 if (!(error >> 16)) {
2105 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2110 COVERAGE_INC(ofproto_error);
2111 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2112 oh ? oh->xid : 0, &buf);
2113 oem->type = htons((unsigned int) error >> 16);
2114 oem->code = htons(error & 0xffff);
2115 memcpy(oem->data, data, len);
2116 queue_tx(buf, ofconn, ofconn->reply_counter);
2120 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2123 size_t oh_length = ntohs(oh->length);
2124 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2128 hton_ofp_phy_port(struct ofp_phy_port *opp)
2130 opp->port_no = htons(opp->port_no);
2131 opp->config = htonl(opp->config);
2132 opp->state = htonl(opp->state);
2133 opp->curr = htonl(opp->curr);
2134 opp->advertised = htonl(opp->advertised);
2135 opp->supported = htonl(opp->supported);
2136 opp->peer = htonl(opp->peer);
2140 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2142 struct ofp_header *rq = oh;
2143 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2148 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2149 struct ofp_header *oh)
2151 struct ofp_switch_features *osf;
2153 unsigned int port_no;
2154 struct ofport *port;
2156 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2157 osf->datapath_id = htonll(p->datapath_id);
2158 osf->n_buffers = htonl(pktbuf_capacity());
2160 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2161 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2162 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2163 (1u << OFPAT_SET_VLAN_VID) |
2164 (1u << OFPAT_SET_VLAN_PCP) |
2165 (1u << OFPAT_STRIP_VLAN) |
2166 (1u << OFPAT_SET_DL_SRC) |
2167 (1u << OFPAT_SET_DL_DST) |
2168 (1u << OFPAT_SET_NW_SRC) |
2169 (1u << OFPAT_SET_NW_DST) |
2170 (1u << OFPAT_SET_NW_TOS) |
2171 (1u << OFPAT_SET_TP_SRC) |
2172 (1u << OFPAT_SET_TP_DST));
2174 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2175 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2178 queue_tx(buf, ofconn, ofconn->reply_counter);
2183 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2184 struct ofp_header *oh)
2187 struct ofp_switch_config *osc;
2191 /* Figure out flags. */
2192 dpif_get_drop_frags(p->dpif, &drop_frags);
2193 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2196 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2197 osc->flags = htons(flags);
2198 osc->miss_send_len = htons(ofconn->miss_send_len);
2199 queue_tx(buf, ofconn, ofconn->reply_counter);
2205 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2206 struct ofp_switch_config *osc)
2211 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2215 flags = ntohs(osc->flags);
2217 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2218 switch (flags & OFPC_FRAG_MASK) {
2219 case OFPC_FRAG_NORMAL:
2220 dpif_set_drop_frags(p->dpif, false);
2222 case OFPC_FRAG_DROP:
2223 dpif_set_drop_frags(p->dpif, true);
2226 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2232 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2238 add_output_group_action(struct odp_actions *actions, uint16_t group,
2239 uint16_t *nf_output_iface)
2241 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2243 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2244 *nf_output_iface = NF_OUT_FLOOD;
2249 add_controller_action(struct odp_actions *actions,
2250 const struct ofp_action_output *oao)
2252 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2253 a->controller.arg = ntohs(oao->max_len);
2256 struct action_xlate_ctx {
2258 flow_t flow; /* Flow to which these actions correspond. */
2259 int recurse; /* Recursion level, via xlate_table_action. */
2260 struct ofproto *ofproto;
2261 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2262 * null pointer if we are revalidating
2263 * without a packet to refer to. */
2266 struct odp_actions *out; /* Datapath actions. */
2267 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2268 bool may_set_up_flow; /* True ordinarily; false if the actions must
2269 * be reassessed for every packet. */
2270 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2273 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2274 struct action_xlate_ctx *ctx);
2277 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2279 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2282 if (ofport->opp.config & OFPPC_NO_FWD) {
2283 /* Forwarding disabled on port. */
2288 * We don't have an ofport record for this port, but it doesn't hurt to
2289 * allow forwarding to it anyhow. Maybe such a port will appear later
2290 * and we're pre-populating the flow table.
2294 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2295 ctx->nf_output_iface = port;
2298 static struct rule *
2299 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2302 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2304 /* The rule we found might not be valid, since we could be in need of
2305 * revalidation. If it is not valid, don't return it. */
2308 && ofproto->need_revalidate
2309 && !revalidate_rule(ofproto, rule)) {
2310 COVERAGE_INC(ofproto_invalidated);
2318 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2320 if (!ctx->recurse) {
2321 uint16_t old_in_port;
2324 /* Look up a flow with 'in_port' as the input port. Then restore the
2325 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2326 * have surprising behavior). */
2327 old_in_port = ctx->flow.in_port;
2328 ctx->flow.in_port = in_port;
2329 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2330 ctx->flow.in_port = old_in_port;
2338 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2345 xlate_output_action(struct action_xlate_ctx *ctx,
2346 const struct ofp_action_output *oao)
2349 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2351 ctx->nf_output_iface = NF_OUT_DROP;
2353 switch (ntohs(oao->port)) {
2355 add_output_action(ctx, ctx->flow.in_port);
2358 xlate_table_action(ctx, ctx->flow.in_port);
2361 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2362 ctx->out, ctx->tags,
2363 &ctx->nf_output_iface,
2364 ctx->ofproto->aux)) {
2365 COVERAGE_INC(ofproto_uninstallable);
2366 ctx->may_set_up_flow = false;
2370 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2371 &ctx->nf_output_iface);
2374 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2376 case OFPP_CONTROLLER:
2377 add_controller_action(ctx->out, oao);
2380 add_output_action(ctx, ODPP_LOCAL);
2383 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2384 if (odp_port != ctx->flow.in_port) {
2385 add_output_action(ctx, odp_port);
2390 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2391 ctx->nf_output_iface = NF_OUT_FLOOD;
2392 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2393 ctx->nf_output_iface = prev_nf_output_iface;
2394 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2395 ctx->nf_output_iface != NF_OUT_FLOOD) {
2396 ctx->nf_output_iface = NF_OUT_MULTI;
2401 xlate_nicira_action(struct action_xlate_ctx *ctx,
2402 const struct nx_action_header *nah)
2404 const struct nx_action_resubmit *nar;
2405 const struct nx_action_set_tunnel *nast;
2406 union odp_action *oa;
2407 int subtype = ntohs(nah->subtype);
2409 assert(nah->vendor == htonl(NX_VENDOR_ID));
2411 case NXAST_RESUBMIT:
2412 nar = (const struct nx_action_resubmit *) nah;
2413 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2416 case NXAST_SET_TUNNEL:
2417 nast = (const struct nx_action_set_tunnel *) nah;
2418 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2419 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2422 /* If you add a new action here that modifies flow data, don't forget to
2423 * update the flow key in ctx->flow in the same key. */
2426 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2432 do_xlate_actions(const union ofp_action *in, size_t n_in,
2433 struct action_xlate_ctx *ctx)
2435 struct actions_iterator iter;
2436 const union ofp_action *ia;
2437 const struct ofport *port;
2439 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2440 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2441 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2442 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2443 /* Drop this flow. */
2447 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2448 uint16_t type = ntohs(ia->type);
2449 union odp_action *oa;
2453 xlate_output_action(ctx, &ia->output);
2456 case OFPAT_SET_VLAN_VID:
2457 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2458 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2461 case OFPAT_SET_VLAN_PCP:
2462 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2463 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2466 case OFPAT_STRIP_VLAN:
2467 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2468 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2469 ctx->flow.dl_vlan_pcp = 0;
2472 case OFPAT_SET_DL_SRC:
2473 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2474 memcpy(oa->dl_addr.dl_addr,
2475 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2476 memcpy(ctx->flow.dl_src,
2477 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2480 case OFPAT_SET_DL_DST:
2481 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2482 memcpy(oa->dl_addr.dl_addr,
2483 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2484 memcpy(ctx->flow.dl_dst,
2485 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2488 case OFPAT_SET_NW_SRC:
2489 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2490 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2493 case OFPAT_SET_NW_DST:
2494 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2495 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2498 case OFPAT_SET_NW_TOS:
2499 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2500 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2503 case OFPAT_SET_TP_SRC:
2504 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2505 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2508 case OFPAT_SET_TP_DST:
2509 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2510 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2514 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2518 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2525 xlate_actions(const union ofp_action *in, size_t n_in,
2526 const flow_t *flow, struct ofproto *ofproto,
2527 const struct ofpbuf *packet,
2528 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2529 uint16_t *nf_output_iface)
2531 tag_type no_tags = 0;
2532 struct action_xlate_ctx ctx;
2533 COVERAGE_INC(ofproto_ofp2odp);
2534 odp_actions_init(out);
2537 ctx.ofproto = ofproto;
2538 ctx.packet = packet;
2540 ctx.tags = tags ? tags : &no_tags;
2541 ctx.may_set_up_flow = true;
2542 ctx.nf_output_iface = NF_OUT_DROP;
2543 do_xlate_actions(in, n_in, &ctx);
2545 /* Check with in-band control to see if we're allowed to set up this
2547 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2548 ctx.may_set_up_flow = false;
2551 if (may_set_up_flow) {
2552 *may_set_up_flow = ctx.may_set_up_flow;
2554 if (nf_output_iface) {
2555 *nf_output_iface = ctx.nf_output_iface;
2557 if (odp_actions_overflow(out)) {
2558 odp_actions_init(out);
2559 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2564 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2565 * error message code (composed with ofp_mkerr()) for the caller to propagate
2566 * upward. Otherwise, returns 0.
2568 * 'oh' is used to make log messages more informative. */
2570 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2572 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2573 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2576 type_name = ofp_message_type_to_string(oh->type);
2577 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2581 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2588 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2589 struct ofp_header *oh)
2591 struct ofp_packet_out *opo;
2592 struct ofpbuf payload, *buffer;
2593 struct odp_actions actions;
2599 error = reject_slave_controller(ofconn, oh);
2604 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2608 opo = (struct ofp_packet_out *) oh;
2610 COVERAGE_INC(ofproto_packet_out);
2611 if (opo->buffer_id != htonl(UINT32_MAX)) {
2612 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2614 if (error || !buffer) {
2622 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2623 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2624 &flow, p, &payload, &actions, NULL, NULL, NULL);
2629 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2631 ofpbuf_delete(buffer);
2637 update_port_config(struct ofproto *p, struct ofport *port,
2638 uint32_t config, uint32_t mask)
2640 mask &= config ^ port->opp.config;
2641 if (mask & OFPPC_PORT_DOWN) {
2642 if (config & OFPPC_PORT_DOWN) {
2643 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2645 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2648 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2649 if (mask & REVALIDATE_BITS) {
2650 COVERAGE_INC(ofproto_costly_flags);
2651 port->opp.config ^= mask & REVALIDATE_BITS;
2652 p->need_revalidate = true;
2654 #undef REVALIDATE_BITS
2655 if (mask & OFPPC_NO_FLOOD) {
2656 port->opp.config ^= OFPPC_NO_FLOOD;
2657 refresh_port_groups(p);
2659 if (mask & OFPPC_NO_PACKET_IN) {
2660 port->opp.config ^= OFPPC_NO_PACKET_IN;
2665 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2666 struct ofp_header *oh)
2668 const struct ofp_port_mod *opm;
2669 struct ofport *port;
2672 error = reject_slave_controller(ofconn, oh);
2676 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2680 opm = (struct ofp_port_mod *) oh;
2682 port = port_array_get(&p->ports,
2683 ofp_port_to_odp_port(ntohs(opm->port_no)));
2685 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2686 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2687 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2689 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2690 if (opm->advertise) {
2691 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2697 static struct ofpbuf *
2698 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2700 struct ofp_stats_reply *osr;
2703 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2704 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2706 osr->flags = htons(0);
2710 static struct ofpbuf *
2711 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2713 return make_stats_reply(request->header.xid, request->type, body_len);
2717 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2719 struct ofpbuf *msg = *msgp;
2720 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2721 if (nbytes + msg->size > UINT16_MAX) {
2722 struct ofp_stats_reply *reply = msg->data;
2723 reply->flags = htons(OFPSF_REPLY_MORE);
2724 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2725 queue_tx(msg, ofconn, ofconn->reply_counter);
2727 return ofpbuf_put_uninit(*msgp, nbytes);
2731 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2732 struct ofp_stats_request *request)
2734 struct ofp_desc_stats *ods;
2737 msg = start_stats_reply(request, sizeof *ods);
2738 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2739 memset(ods, 0, sizeof *ods);
2740 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2741 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2742 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2743 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2744 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2745 queue_tx(msg, ofconn, ofconn->reply_counter);
2751 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2753 struct rule *rule = rule_from_cls_rule(cls_rule);
2754 int *n_subrules = n_subrules_;
2762 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2763 struct ofp_stats_request *request)
2765 struct ofp_table_stats *ots;
2767 struct odp_stats dpstats;
2768 int n_exact, n_subrules, n_wild;
2770 msg = start_stats_reply(request, sizeof *ots * 2);
2772 /* Count rules of various kinds. */
2774 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2775 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2776 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2779 dpif_get_dp_stats(p->dpif, &dpstats);
2780 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2781 memset(ots, 0, sizeof *ots);
2782 ots->table_id = TABLEID_HASH;
2783 strcpy(ots->name, "hash");
2784 ots->wildcards = htonl(0);
2785 ots->max_entries = htonl(dpstats.max_capacity);
2786 ots->active_count = htonl(n_exact);
2787 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2789 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2791 /* Classifier table. */
2792 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2793 memset(ots, 0, sizeof *ots);
2794 ots->table_id = TABLEID_CLASSIFIER;
2795 strcpy(ots->name, "classifier");
2796 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2798 ots->max_entries = htonl(65536);
2799 ots->active_count = htonl(n_wild);
2800 ots->lookup_count = htonll(0); /* XXX */
2801 ots->matched_count = htonll(0); /* XXX */
2803 queue_tx(msg, ofconn, ofconn->reply_counter);
2808 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2809 struct ofpbuf **msgp)
2811 struct netdev_stats stats;
2812 struct ofp_port_stats *ops;
2814 /* Intentionally ignore return value, since errors will set
2815 * 'stats' to all-1s, which is correct for OpenFlow, and
2816 * netdev_get_stats() will log errors. */
2817 netdev_get_stats(port->netdev, &stats);
2819 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2820 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2821 memset(ops->pad, 0, sizeof ops->pad);
2822 ops->rx_packets = htonll(stats.rx_packets);
2823 ops->tx_packets = htonll(stats.tx_packets);
2824 ops->rx_bytes = htonll(stats.rx_bytes);
2825 ops->tx_bytes = htonll(stats.tx_bytes);
2826 ops->rx_dropped = htonll(stats.rx_dropped);
2827 ops->tx_dropped = htonll(stats.tx_dropped);
2828 ops->rx_errors = htonll(stats.rx_errors);
2829 ops->tx_errors = htonll(stats.tx_errors);
2830 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2831 ops->rx_over_err = htonll(stats.rx_over_errors);
2832 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2833 ops->collisions = htonll(stats.collisions);
2837 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2838 struct ofp_stats_request *osr,
2841 struct ofp_port_stats_request *psr;
2842 struct ofp_port_stats *ops;
2844 struct ofport *port;
2845 unsigned int port_no;
2847 if (arg_size != sizeof *psr) {
2848 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2850 psr = (struct ofp_port_stats_request *) osr->body;
2852 msg = start_stats_reply(osr, sizeof *ops * 16);
2853 if (psr->port_no != htons(OFPP_NONE)) {
2854 port = port_array_get(&p->ports,
2855 ofp_port_to_odp_port(ntohs(psr->port_no)));
2857 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2860 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2861 append_port_stat(port, port_no, ofconn, &msg);
2865 queue_tx(msg, ofconn, ofconn->reply_counter);
2869 struct flow_stats_cbdata {
2870 struct ofproto *ofproto;
2871 struct ofconn *ofconn;
2876 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2877 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2878 * returned statistic include statistics for all of 'rule''s subrules. */
2880 query_stats(struct ofproto *p, struct rule *rule,
2881 uint64_t *packet_countp, uint64_t *byte_countp)
2883 uint64_t packet_count, byte_count;
2884 struct rule *subrule;
2885 struct odp_flow *odp_flows;
2888 /* Start from historical data for 'rule' itself that are no longer tracked
2889 * by the datapath. This counts, for example, subrules that have
2891 packet_count = rule->packet_count;
2892 byte_count = rule->byte_count;
2894 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2895 * wildcarded then on all of its subrules.
2897 * Also, add any statistics that are not tracked by the datapath for each
2898 * subrule. This includes, for example, statistics for packets that were
2899 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2901 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2902 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2903 if (rule->cr.wc.wildcards) {
2905 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2906 odp_flows[i++].key = subrule->cr.flow;
2907 packet_count += subrule->packet_count;
2908 byte_count += subrule->byte_count;
2911 odp_flows[0].key = rule->cr.flow;
2914 /* Fetch up-to-date statistics from the datapath and add them in. */
2915 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2917 for (i = 0; i < n_odp_flows; i++) {
2918 struct odp_flow *odp_flow = &odp_flows[i];
2919 packet_count += odp_flow->stats.n_packets;
2920 byte_count += odp_flow->stats.n_bytes;
2925 /* Return the stats to the caller. */
2926 *packet_countp = packet_count;
2927 *byte_countp = byte_count;
2931 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2933 struct rule *rule = rule_from_cls_rule(rule_);
2934 struct flow_stats_cbdata *cbdata = cbdata_;
2935 struct ofp_flow_stats *ofs;
2936 uint64_t packet_count, byte_count;
2937 size_t act_len, len;
2938 long long int tdiff = time_msec() - rule->created;
2939 uint32_t sec = tdiff / 1000;
2940 uint32_t msec = tdiff - (sec * 1000);
2942 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2946 act_len = sizeof *rule->actions * rule->n_actions;
2947 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2949 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2951 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2952 ofs->length = htons(len);
2953 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2955 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2956 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2957 ofs->duration_sec = htonl(sec);
2958 ofs->duration_nsec = htonl(msec * 1000000);
2959 ofs->cookie = rule->flow_cookie;
2960 ofs->priority = htons(rule->cr.priority);
2961 ofs->idle_timeout = htons(rule->idle_timeout);
2962 ofs->hard_timeout = htons(rule->hard_timeout);
2963 memset(ofs->pad2, 0, sizeof ofs->pad2);
2964 ofs->packet_count = htonll(packet_count);
2965 ofs->byte_count = htonll(byte_count);
2966 memcpy(ofs->actions, rule->actions, act_len);
2970 table_id_to_include(uint8_t table_id)
2972 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2973 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2974 : table_id == 0xff ? CLS_INC_ALL
2979 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2980 const struct ofp_stats_request *osr,
2983 struct ofp_flow_stats_request *fsr;
2984 struct flow_stats_cbdata cbdata;
2985 struct cls_rule target;
2987 if (arg_size != sizeof *fsr) {
2988 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2990 fsr = (struct ofp_flow_stats_request *) osr->body;
2992 COVERAGE_INC(ofproto_flows_req);
2994 cbdata.ofconn = ofconn;
2995 cbdata.out_port = fsr->out_port;
2996 cbdata.msg = start_stats_reply(osr, 1024);
2997 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
2998 classifier_for_each_match(&p->cls, &target,
2999 table_id_to_include(fsr->table_id),
3000 flow_stats_cb, &cbdata);
3001 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3005 struct flow_stats_ds_cbdata {
3006 struct ofproto *ofproto;
3011 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3013 struct rule *rule = rule_from_cls_rule(rule_);
3014 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3015 struct ds *results = cbdata->results;
3016 struct ofp_match match;
3017 uint64_t packet_count, byte_count;
3018 size_t act_len = sizeof *rule->actions * rule->n_actions;
3020 /* Don't report on subrules. */
3021 if (rule->super != NULL) {
3025 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3026 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3027 cbdata->ofproto->tun_id_from_cookie, &match);
3029 ds_put_format(results, "duration=%llds, ",
3030 (time_msec() - rule->created) / 1000);
3031 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3032 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3033 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3034 ofp_print_match(results, &match, true);
3035 ofp_print_actions(results, &rule->actions->header, act_len);
3036 ds_put_cstr(results, "\n");
3039 /* Adds a pretty-printed description of all flows to 'results', including
3040 * those marked hidden by secchan (e.g., by in-band control). */
3042 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3044 struct ofp_match match;
3045 struct cls_rule target;
3046 struct flow_stats_ds_cbdata cbdata;
3048 memset(&match, 0, sizeof match);
3049 match.wildcards = htonl(OVSFW_ALL);
3052 cbdata.results = results;
3054 cls_rule_from_match(&match, 0, false, 0, &target);
3055 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3056 flow_stats_ds_cb, &cbdata);
3059 struct aggregate_stats_cbdata {
3060 struct ofproto *ofproto;
3062 uint64_t packet_count;
3063 uint64_t byte_count;
3068 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3070 struct rule *rule = rule_from_cls_rule(rule_);
3071 struct aggregate_stats_cbdata *cbdata = cbdata_;
3072 uint64_t packet_count, byte_count;
3074 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3078 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3080 cbdata->packet_count += packet_count;
3081 cbdata->byte_count += byte_count;
3086 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3087 const struct ofp_stats_request *osr,
3090 struct ofp_aggregate_stats_request *asr;
3091 struct ofp_aggregate_stats_reply *reply;
3092 struct aggregate_stats_cbdata cbdata;
3093 struct cls_rule target;
3096 if (arg_size != sizeof *asr) {
3097 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3099 asr = (struct ofp_aggregate_stats_request *) osr->body;
3101 COVERAGE_INC(ofproto_agg_request);
3103 cbdata.out_port = asr->out_port;
3104 cbdata.packet_count = 0;
3105 cbdata.byte_count = 0;
3107 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3108 classifier_for_each_match(&p->cls, &target,
3109 table_id_to_include(asr->table_id),
3110 aggregate_stats_cb, &cbdata);
3112 msg = start_stats_reply(osr, sizeof *reply);
3113 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3114 reply->flow_count = htonl(cbdata.n_flows);
3115 reply->packet_count = htonll(cbdata.packet_count);
3116 reply->byte_count = htonll(cbdata.byte_count);
3117 queue_tx(msg, ofconn, ofconn->reply_counter);
3122 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3123 struct ofp_header *oh)
3125 struct ofp_stats_request *osr;
3129 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3134 osr = (struct ofp_stats_request *) oh;
3136 switch (ntohs(osr->type)) {
3138 return handle_desc_stats_request(p, ofconn, osr);
3141 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3143 case OFPST_AGGREGATE:
3144 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3147 return handle_table_stats_request(p, ofconn, osr);
3150 return handle_port_stats_request(p, ofconn, osr, arg_size);
3153 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3156 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3160 static long long int
3161 msec_from_nsec(uint64_t sec, uint32_t nsec)
3163 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3167 update_time(struct ofproto *ofproto, struct rule *rule,
3168 const struct odp_flow_stats *stats)
3170 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3171 if (used > rule->used) {
3173 if (rule->super && used > rule->super->used) {
3174 rule->super->used = used;
3176 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3181 update_stats(struct ofproto *ofproto, struct rule *rule,
3182 const struct odp_flow_stats *stats)
3184 if (stats->n_packets) {
3185 update_time(ofproto, rule, stats);
3186 rule->packet_count += stats->n_packets;
3187 rule->byte_count += stats->n_bytes;
3188 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3193 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3194 * in which no matching flow already exists in the flow table.
3196 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3197 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3198 * code as encoded by ofp_mkerr() on failure.
3200 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3203 add_flow(struct ofproto *p, struct ofconn *ofconn,
3204 const struct ofp_flow_mod *ofm, size_t n_actions)
3206 struct ofpbuf *packet;
3211 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3215 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3217 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3218 ntohs(ofm->priority))) {
3219 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3223 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3224 n_actions, ntohs(ofm->idle_timeout),
3225 ntohs(ofm->hard_timeout), ofm->cookie,
3226 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3227 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3228 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3231 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3232 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3236 in_port = UINT16_MAX;
3239 rule_insert(p, rule, packet, in_port);
3240 ofpbuf_delete(packet);
3244 static struct rule *
3245 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3250 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3252 return rule_from_cls_rule(classifier_find_rule_exactly(
3253 &p->cls, &flow, wildcards,
3254 ntohs(ofm->priority)));
3258 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3259 struct rule *rule, const struct ofp_flow_mod *ofm)
3261 struct ofpbuf *packet;
3266 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3270 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3276 flow_extract(packet, 0, in_port, &flow);
3277 rule_execute(ofproto, rule, packet, &flow);
3278 ofpbuf_delete(packet);
3283 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3285 struct modify_flows_cbdata {
3286 struct ofproto *ofproto;
3287 const struct ofp_flow_mod *ofm;
3292 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3293 size_t n_actions, struct rule *);
3294 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3296 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3297 * encoded by ofp_mkerr() on failure.
3299 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3302 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3303 const struct ofp_flow_mod *ofm, size_t n_actions)
3305 struct modify_flows_cbdata cbdata;
3306 struct cls_rule target;
3310 cbdata.n_actions = n_actions;
3311 cbdata.match = NULL;
3313 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3316 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3317 modify_flows_cb, &cbdata);
3319 /* This credits the packet to whichever flow happened to happened to
3320 * match last. That's weird. Maybe we should do a lookup for the
3321 * flow that actually matches the packet? Who knows. */
3322 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3325 return add_flow(p, ofconn, ofm, n_actions);
3329 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3330 * code as encoded by ofp_mkerr() on failure.
3332 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3335 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3336 struct ofp_flow_mod *ofm, size_t n_actions)
3338 struct rule *rule = find_flow_strict(p, ofm);
3339 if (rule && !rule_is_hidden(rule)) {
3340 modify_flow(p, ofm, n_actions, rule);
3341 return send_buffered_packet(p, ofconn, rule, ofm);
3343 return add_flow(p, ofconn, ofm, n_actions);
3347 /* Callback for modify_flows_loose(). */
3349 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3351 struct rule *rule = rule_from_cls_rule(rule_);
3352 struct modify_flows_cbdata *cbdata = cbdata_;
3354 if (!rule_is_hidden(rule)) {
3355 cbdata->match = rule;
3356 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3360 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3361 * been identified as a flow in 'p''s flow table to be modified, by changing
3362 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3363 * ofp_action[] structures). */
3365 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3366 size_t n_actions, struct rule *rule)
3368 size_t actions_len = n_actions * sizeof *rule->actions;
3370 rule->flow_cookie = ofm->cookie;
3372 /* If the actions are the same, do nothing. */
3373 if (n_actions == rule->n_actions
3374 && !memcmp(ofm->actions, rule->actions, actions_len))
3379 /* Replace actions. */
3380 free(rule->actions);
3381 rule->actions = xmemdup(ofm->actions, actions_len);
3382 rule->n_actions = n_actions;
3384 /* Make sure that the datapath gets updated properly. */
3385 if (rule->cr.wc.wildcards) {
3386 COVERAGE_INC(ofproto_mod_wc_flow);
3387 p->need_revalidate = true;
3389 rule_update_actions(p, rule);
3395 /* OFPFC_DELETE implementation. */
3397 struct delete_flows_cbdata {
3398 struct ofproto *ofproto;
3402 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3403 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3405 /* Implements OFPFC_DELETE. */
3407 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3409 struct delete_flows_cbdata cbdata;
3410 struct cls_rule target;
3413 cbdata.out_port = ofm->out_port;
3415 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3418 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3419 delete_flows_cb, &cbdata);
3422 /* Implements OFPFC_DELETE_STRICT. */
3424 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3426 struct rule *rule = find_flow_strict(p, ofm);
3428 delete_flow(p, rule, ofm->out_port);
3432 /* Callback for delete_flows_loose(). */
3434 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3436 struct rule *rule = rule_from_cls_rule(rule_);
3437 struct delete_flows_cbdata *cbdata = cbdata_;
3439 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3442 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3443 * been identified as a flow to delete from 'p''s flow table, by deleting the
3444 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3447 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3448 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3449 * specified 'out_port'. */
3451 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3453 if (rule_is_hidden(rule)) {
3457 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3461 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3462 rule_remove(p, rule);
3466 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3467 struct ofp_flow_mod *ofm)
3472 error = reject_slave_controller(ofconn, &ofm->header);
3476 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3477 sizeof *ofm->actions, &n_actions);
3482 /* We do not support the emergency flow cache. It will hopefully
3483 * get dropped from OpenFlow in the near future. */
3484 if (ofm->flags & htons(OFPFF_EMERG)) {
3485 /* There isn't a good fit for an error code, so just state that the
3486 * flow table is full. */
3487 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3490 normalize_match(&ofm->match);
3491 if (!ofm->match.wildcards) {
3492 ofm->priority = htons(UINT16_MAX);
3495 error = validate_actions((const union ofp_action *) ofm->actions,
3496 n_actions, p->max_ports);
3501 switch (ntohs(ofm->command)) {
3503 return add_flow(p, ofconn, ofm, n_actions);
3506 return modify_flows_loose(p, ofconn, ofm, n_actions);
3508 case OFPFC_MODIFY_STRICT:
3509 return modify_flow_strict(p, ofconn, ofm, n_actions);
3512 delete_flows_loose(p, ofm);
3515 case OFPFC_DELETE_STRICT:
3516 delete_flow_strict(p, ofm);
3520 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3525 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3529 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3534 p->tun_id_from_cookie = !!msg->set;
3539 handle_role_request(struct ofproto *ofproto,
3540 struct ofconn *ofconn, struct nicira_header *msg)
3542 struct nx_role_request *nrr;
3543 struct nx_role_request *reply;
3547 if (ntohs(msg->header.length) != sizeof *nrr) {
3548 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3549 ntohs(msg->header.length), sizeof *nrr);
3550 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3552 nrr = (struct nx_role_request *) msg;
3554 if (ofconn->type != OFCONN_CONTROLLER) {
3555 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3557 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3560 role = ntohl(nrr->role);
3561 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3562 && role != NX_ROLE_SLAVE) {
3563 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3565 /* There's no good error code for this. */
3566 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3569 if (role == NX_ROLE_MASTER) {
3570 struct ofconn *other;
3572 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3573 &ofproto->controllers) {
3574 if (other->role == NX_ROLE_MASTER) {
3575 other->role = NX_ROLE_SLAVE;
3579 ofconn->role = role;
3581 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3583 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3584 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3585 reply->role = htonl(role);
3586 queue_tx(buf, ofconn, ofconn->reply_counter);
3592 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3594 struct ofp_vendor_header *ovh = msg;
3595 struct nicira_header *nh;
3597 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3598 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3599 "(expected at least %zu)",
3600 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3601 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3603 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3604 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3606 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3607 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3608 "(expected at least %zu)",
3609 ntohs(ovh->header.length), sizeof(struct nicira_header));
3610 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3614 switch (ntohl(nh->subtype)) {
3615 case NXT_STATUS_REQUEST:
3616 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3619 case NXT_TUN_ID_FROM_COOKIE:
3620 return handle_tun_id_from_cookie(p, msg);
3622 case NXT_ROLE_REQUEST:
3623 return handle_role_request(p, ofconn, msg);
3626 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3630 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3632 struct ofp_header *ob;
3635 /* Currently, everything executes synchronously, so we can just
3636 * immediately send the barrier reply. */
3637 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3638 queue_tx(buf, ofconn, ofconn->reply_counter);
3643 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3644 struct ofpbuf *ofp_msg)
3646 struct ofp_header *oh = ofp_msg->data;
3649 COVERAGE_INC(ofproto_recv_openflow);
3651 case OFPT_ECHO_REQUEST:
3652 error = handle_echo_request(ofconn, oh);
3655 case OFPT_ECHO_REPLY:
3659 case OFPT_FEATURES_REQUEST:
3660 error = handle_features_request(p, ofconn, oh);
3663 case OFPT_GET_CONFIG_REQUEST:
3664 error = handle_get_config_request(p, ofconn, oh);
3667 case OFPT_SET_CONFIG:
3668 error = handle_set_config(p, ofconn, ofp_msg->data);
3671 case OFPT_PACKET_OUT:
3672 error = handle_packet_out(p, ofconn, ofp_msg->data);
3676 error = handle_port_mod(p, ofconn, oh);
3680 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3683 case OFPT_STATS_REQUEST:
3684 error = handle_stats_request(p, ofconn, oh);
3688 error = handle_vendor(p, ofconn, ofp_msg->data);
3691 case OFPT_BARRIER_REQUEST:
3692 error = handle_barrier_request(ofconn, oh);
3696 if (VLOG_IS_WARN_ENABLED()) {
3697 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3698 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3701 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3706 send_error_oh(ofconn, ofp_msg->data, error);
3711 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3713 struct odp_msg *msg = packet->data;
3715 struct ofpbuf payload;
3718 payload.data = msg + 1;
3719 payload.size = msg->length - sizeof *msg;
3720 flow_extract(&payload, msg->arg, msg->port, &flow);
3722 /* Check with in-band control to see if this packet should be sent
3723 * to the local port regardless of the flow table. */
3724 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3725 union odp_action action;
3727 memset(&action, 0, sizeof(action));
3728 action.output.type = ODPAT_OUTPUT;
3729 action.output.port = ODPP_LOCAL;
3730 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3733 rule = lookup_valid_rule(p, &flow);
3735 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3736 struct ofport *port = port_array_get(&p->ports, msg->port);
3738 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3739 COVERAGE_INC(ofproto_no_packet_in);
3740 /* XXX install 'drop' flow entry */
3741 ofpbuf_delete(packet);
3745 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3748 COVERAGE_INC(ofproto_packet_in);
3749 send_packet_in(p, packet);
3753 if (rule->cr.wc.wildcards) {
3754 rule = rule_create_subrule(p, rule, &flow);
3755 rule_make_actions(p, rule, packet);
3757 if (!rule->may_install) {
3758 /* The rule is not installable, that is, we need to process every
3759 * packet, so process the current packet and set its actions into
3761 rule_make_actions(p, rule, packet);
3763 /* XXX revalidate rule if it needs it */
3767 rule_execute(p, rule, &payload, &flow);
3768 rule_reinstall(p, rule);
3770 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3772 * Extra-special case for fail-open mode.
3774 * We are in fail-open mode and the packet matched the fail-open rule,
3775 * but we are connected to a controller too. We should send the packet
3776 * up to the controller in the hope that it will try to set up a flow
3777 * and thereby allow us to exit fail-open.
3779 * See the top-level comment in fail-open.c for more information.
3781 send_packet_in(p, packet);
3783 ofpbuf_delete(packet);
3788 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3790 struct odp_msg *msg = packet->data;
3792 switch (msg->type) {
3793 case _ODPL_ACTION_NR:
3794 COVERAGE_INC(ofproto_ctlr_action);
3795 send_packet_in(p, packet);
3798 case _ODPL_SFLOW_NR:
3800 ofproto_sflow_received(p->sflow, msg);
3802 ofpbuf_delete(packet);
3806 handle_odp_miss_msg(p, packet);
3810 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3817 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3819 struct rule *sub = rule_from_cls_rule(sub_);
3820 struct revalidate_cbdata *cbdata = cbdata_;
3822 if (cbdata->revalidate_all
3823 || (cbdata->revalidate_subrules && sub->super)
3824 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3825 revalidate_rule(cbdata->ofproto, sub);
3830 revalidate_rule(struct ofproto *p, struct rule *rule)
3832 const flow_t *flow = &rule->cr.flow;
3834 COVERAGE_INC(ofproto_revalidate_rule);
3837 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3839 rule_remove(p, rule);
3841 } else if (super != rule->super) {
3842 COVERAGE_INC(ofproto_revalidate_moved);
3843 list_remove(&rule->list);
3844 list_push_back(&super->list, &rule->list);
3845 rule->super = super;
3846 rule->hard_timeout = super->hard_timeout;
3847 rule->idle_timeout = super->idle_timeout;
3848 rule->created = super->created;
3853 rule_update_actions(p, rule);
3857 static struct ofpbuf *
3858 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3859 long long int now, uint8_t reason)
3861 struct ofp_flow_removed *ofr;
3863 long long int tdiff = now - rule->created;
3864 uint32_t sec = tdiff / 1000;
3865 uint32_t msec = tdiff - (sec * 1000);
3867 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3868 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3870 ofr->cookie = rule->flow_cookie;
3871 ofr->priority = htons(rule->cr.priority);
3872 ofr->reason = reason;
3873 ofr->duration_sec = htonl(sec);
3874 ofr->duration_nsec = htonl(msec * 1000000);
3875 ofr->idle_timeout = htons(rule->idle_timeout);
3876 ofr->packet_count = htonll(rule->packet_count);
3877 ofr->byte_count = htonll(rule->byte_count);
3883 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3885 assert(rule->installed);
3886 assert(!rule->cr.wc.wildcards);
3889 rule_remove(ofproto, rule);
3891 rule_uninstall(ofproto, rule);
3896 send_flow_removed(struct ofproto *p, struct rule *rule,
3897 long long int now, uint8_t reason)
3899 struct ofconn *ofconn;
3900 struct ofconn *prev;
3901 struct ofpbuf *buf = NULL;
3903 /* We limit the maximum number of queued flow expirations it by accounting
3904 * them under the counter for replies. That works because preventing
3905 * OpenFlow requests from being processed also prevents new flows from
3906 * being added (and expiring). (It also prevents processing OpenFlow
3907 * requests that would not add new flows, so it is imperfect.) */
3910 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3911 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
3912 && ofconn_receives_async_msgs(ofconn)) {
3914 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3916 buf = compose_flow_removed(p, rule, now, reason);
3922 queue_tx(buf, prev, prev->reply_counter);
3928 expire_rule(struct cls_rule *cls_rule, void *p_)
3930 struct ofproto *p = p_;
3931 struct rule *rule = rule_from_cls_rule(cls_rule);
3932 long long int hard_expire, idle_expire, expire, now;
3934 hard_expire = (rule->hard_timeout
3935 ? rule->created + rule->hard_timeout * 1000
3937 idle_expire = (rule->idle_timeout
3938 && (rule->super || list_is_empty(&rule->list))
3939 ? rule->used + rule->idle_timeout * 1000
3941 expire = MIN(hard_expire, idle_expire);
3945 if (rule->installed && now >= rule->used + 5000) {
3946 uninstall_idle_flow(p, rule);
3947 } else if (!rule->cr.wc.wildcards) {
3948 active_timeout(p, rule);
3954 COVERAGE_INC(ofproto_expired);
3956 /* Update stats. This code will be a no-op if the rule expired
3957 * due to an idle timeout. */
3958 if (rule->cr.wc.wildcards) {
3959 struct rule *subrule, *next;
3960 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3961 rule_remove(p, subrule);
3964 rule_uninstall(p, rule);
3967 if (!rule_is_hidden(rule)) {
3968 send_flow_removed(p, rule, now,
3970 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3972 rule_remove(p, rule);
3976 active_timeout(struct ofproto *ofproto, struct rule *rule)
3978 if (ofproto->netflow && !is_controller_rule(rule) &&
3979 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3980 struct ofexpired expired;
3981 struct odp_flow odp_flow;
3983 /* Get updated flow stats. */
3984 memset(&odp_flow, 0, sizeof odp_flow);
3985 if (rule->installed) {
3986 odp_flow.key = rule->cr.flow;
3987 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3988 dpif_flow_get(ofproto->dpif, &odp_flow);
3990 if (odp_flow.stats.n_packets) {
3991 update_time(ofproto, rule, &odp_flow.stats);
3992 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3993 odp_flow.stats.tcp_flags);
3997 expired.flow = rule->cr.flow;
3998 expired.packet_count = rule->packet_count +
3999 odp_flow.stats.n_packets;
4000 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4001 expired.used = rule->used;
4003 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4005 /* Schedule us to send the accumulated records once we have
4006 * collected all of them. */
4007 poll_immediate_wake();
4012 update_used(struct ofproto *p)
4014 struct odp_flow *flows;
4019 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4024 for (i = 0; i < n_flows; i++) {
4025 struct odp_flow *f = &flows[i];
4028 rule = rule_from_cls_rule(
4029 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4030 if (!rule || !rule->installed) {
4031 COVERAGE_INC(ofproto_unexpected_rule);
4032 dpif_flow_del(p->dpif, f);
4036 update_time(p, rule, &f->stats);
4037 rule_account(p, rule, f->stats.n_bytes);
4042 /* pinsched callback for sending 'packet' on 'ofconn'. */
4044 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4046 struct ofconn *ofconn = ofconn_;
4048 rconn_send_with_limit(ofconn->rconn, packet,
4049 ofconn->packet_in_counter, 100);
4052 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4053 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4054 * packet scheduler for sending.
4056 * 'max_len' specifies the maximum number of bytes of the packet to send on
4057 * 'ofconn' (INT_MAX specifies no limit).
4059 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4060 * ownership is transferred to this function. */
4062 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4065 struct ofproto *ofproto = ofconn->ofproto;
4066 struct ofp_packet_in *opi = packet->data;
4067 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4068 int send_len, trim_size;
4072 if (opi->reason == OFPR_ACTION) {
4073 buffer_id = UINT32_MAX;
4074 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4075 buffer_id = pktbuf_get_null();
4076 } else if (!ofconn->pktbuf) {
4077 buffer_id = UINT32_MAX;
4079 struct ofpbuf payload;
4080 payload.data = opi->data;
4081 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4082 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4085 /* Figure out how much of the packet to send. */
4086 send_len = ntohs(opi->total_len);
4087 if (buffer_id != UINT32_MAX) {
4088 send_len = MIN(send_len, ofconn->miss_send_len);
4090 send_len = MIN(send_len, max_len);
4092 /* Adjust packet length and clone if necessary. */
4093 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4095 packet = ofpbuf_clone_data(packet->data, trim_size);
4098 packet->size = trim_size;
4101 /* Update packet headers. */
4102 opi->buffer_id = htonl(buffer_id);
4103 update_openflow_length(packet);
4105 /* Hand over to packet scheduler. It might immediately call into
4106 * do_send_packet_in() or it might buffer it for a while (until a later
4107 * call to pinsched_run()). */
4108 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4109 packet, do_send_packet_in, ofconn);
4112 /* Replace struct odp_msg header in 'packet' by equivalent struct
4113 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4114 * returned by dpif_recv()).
4116 * The conversion is not complete: the caller still needs to trim any unneeded
4117 * payload off the end of the buffer, set the length in the OpenFlow header,
4118 * and set buffer_id. Those require us to know the controller settings and so
4119 * must be done on a per-controller basis.
4121 * Returns the maximum number of bytes of the packet that should be sent to
4122 * the controller (INT_MAX if no limit). */
4124 do_convert_to_packet_in(struct ofpbuf *packet)
4126 struct odp_msg *msg = packet->data;
4127 struct ofp_packet_in *opi;
4133 /* Extract relevant header fields */
4134 if (msg->type == _ODPL_ACTION_NR) {
4135 reason = OFPR_ACTION;
4138 reason = OFPR_NO_MATCH;
4141 total_len = msg->length - sizeof *msg;
4142 in_port = odp_port_to_ofp_port(msg->port);
4144 /* Repurpose packet buffer by overwriting header. */
4145 ofpbuf_pull(packet, sizeof(struct odp_msg));
4146 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4147 opi->header.version = OFP_VERSION;
4148 opi->header.type = OFPT_PACKET_IN;
4149 opi->total_len = htons(total_len);
4150 opi->in_port = htons(in_port);
4151 opi->reason = reason;
4156 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4157 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4158 * as necessary according to their individual configurations.
4160 * 'packet' must have sufficient headroom to convert it into a struct
4161 * ofp_packet_in (e.g. as returned by dpif_recv()).
4163 * Takes ownership of 'packet'. */
4165 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4167 struct ofconn *ofconn, *prev;
4170 max_len = do_convert_to_packet_in(packet);
4173 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4174 if (ofconn_receives_async_msgs(ofconn)) {
4176 schedule_packet_in(prev, packet, max_len, true);
4182 schedule_packet_in(prev, packet, max_len, false);
4184 ofpbuf_delete(packet);
4189 pick_datapath_id(const struct ofproto *ofproto)
4191 const struct ofport *port;
4193 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4195 uint8_t ea[ETH_ADDR_LEN];
4198 error = netdev_get_etheraddr(port->netdev, ea);
4200 return eth_addr_to_uint64(ea);
4202 VLOG_WARN("could not get MAC address for %s (%s)",
4203 netdev_get_name(port->netdev), strerror(error));
4205 return ofproto->fallback_dpid;
4209 pick_fallback_dpid(void)
4211 uint8_t ea[ETH_ADDR_LEN];
4212 eth_addr_nicira_random(ea);
4213 return eth_addr_to_uint64(ea);
4217 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4218 struct odp_actions *actions, tag_type *tags,
4219 uint16_t *nf_output_iface, void *ofproto_)
4221 struct ofproto *ofproto = ofproto_;
4224 /* Drop frames for reserved multicast addresses. */
4225 if (eth_addr_is_reserved(flow->dl_dst)) {
4229 /* Learn source MAC (but don't try to learn from revalidation). */
4230 if (packet != NULL) {
4231 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4233 GRAT_ARP_LOCK_NONE);
4235 /* The log messages here could actually be useful in debugging,
4236 * so keep the rate limit relatively high. */
4237 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4238 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4239 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4240 ofproto_revalidate(ofproto, rev_tag);
4244 /* Determine output port. */
4245 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,