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 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
550 if (sin->sin_addr.s_addr) {
551 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
554 if (ofconn->discovery) {
558 for (i = 0; i < ofproto->n_extra_remotes; i++) {
559 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
562 /* Create or update or destroy in-band.
564 * Ordinarily we only enable in-band if there's at least one remote
565 * address, but discovery needs the in-band rules for DHCP to be installed
566 * even before we know any remote addresses. */
567 if (n_addrs || discovery) {
568 if (!ofproto->in_band) {
569 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
572 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
573 ofproto->next_in_band_update = time_msec() + 1000;
575 in_band_destroy(ofproto->in_band);
576 ofproto->in_band = NULL;
584 ofproto_set_controllers(struct ofproto *p,
585 const struct ofproto_controller *controllers,
586 size_t n_controllers)
588 struct shash new_controllers;
589 enum ofproto_fail_mode fail_mode;
590 struct ofconn *ofconn, *next;
594 shash_init(&new_controllers);
595 for (i = 0; i < n_controllers; i++) {
596 const struct ofproto_controller *c = &controllers[i];
598 shash_add_once(&new_controllers, c->target, &controllers[i]);
599 if (!find_controller_by_target(p, c->target)) {
600 add_controller(p, c);
604 fail_mode = OFPROTO_FAIL_STANDALONE;
606 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
608 struct ofproto_controller *c;
610 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
612 ofconn_destroy(ofconn);
614 update_controller(ofconn, c);
618 if (c->fail == OFPROTO_FAIL_SECURE) {
619 fail_mode = OFPROTO_FAIL_SECURE;
623 shash_destroy(&new_controllers);
625 update_in_band_remotes(p);
627 if (!hmap_is_empty(&p->controllers)
628 && fail_mode == OFPROTO_FAIL_STANDALONE) {
629 struct rconn **rconns;
633 p->fail_open = fail_open_create(p, p->switch_status);
637 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
638 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
639 rconns[n++] = ofconn->rconn;
642 fail_open_set_controllers(p->fail_open, rconns, n);
643 /* p->fail_open takes ownership of 'rconns'. */
645 fail_open_destroy(p->fail_open);
649 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
650 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
651 struct ofconn, hmap_node);
652 ofconn->ss = switch_status_register(p->switch_status, "remote",
653 rconn_status_cb, ofconn->rconn);
658 any_extras_changed(const struct ofproto *ofproto,
659 const struct sockaddr_in *extras, size_t n)
663 if (n != ofproto->n_extra_remotes) {
667 for (i = 0; i < n; i++) {
668 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
669 const struct sockaddr_in *new = &extras[i];
671 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
672 old->sin_port != new->sin_port) {
680 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
681 * in-band control should guarantee access, in the same way that in-band
682 * control guarantees access to OpenFlow controllers. */
684 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
685 const struct sockaddr_in *extras, size_t n)
687 if (!any_extras_changed(ofproto, extras, n)) {
691 free(ofproto->extra_in_band_remotes);
692 ofproto->n_extra_remotes = n;
693 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
695 update_in_band_remotes(ofproto);
699 ofproto_set_desc(struct ofproto *p,
700 const char *mfr_desc, const char *hw_desc,
701 const char *sw_desc, const char *serial_desc,
704 struct ofp_desc_stats *ods;
707 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
708 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
709 sizeof ods->mfr_desc);
712 p->mfr_desc = xstrdup(mfr_desc);
715 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
716 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
717 sizeof ods->hw_desc);
720 p->hw_desc = xstrdup(hw_desc);
723 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
724 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
725 sizeof ods->sw_desc);
728 p->sw_desc = xstrdup(sw_desc);
731 if (strlen(serial_desc) >= sizeof ods->serial_num) {
732 VLOG_WARN("truncating serial_desc, must be less than %zu "
734 sizeof ods->serial_num);
736 free(p->serial_desc);
737 p->serial_desc = xstrdup(serial_desc);
740 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
741 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
742 sizeof ods->dp_desc);
745 p->dp_desc = xstrdup(dp_desc);
750 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
751 const struct svec *svec)
753 struct pvconn **pvconns = *pvconnsp;
754 size_t n_pvconns = *n_pvconnsp;
758 for (i = 0; i < n_pvconns; i++) {
759 pvconn_close(pvconns[i]);
763 pvconns = xmalloc(svec->n * sizeof *pvconns);
765 for (i = 0; i < svec->n; i++) {
766 const char *name = svec->names[i];
767 struct pvconn *pvconn;
770 error = pvconn_open(name, &pvconn);
772 pvconns[n_pvconns++] = pvconn;
774 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
782 *n_pvconnsp = n_pvconns;
788 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
790 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
794 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
796 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
800 ofproto_set_netflow(struct ofproto *ofproto,
801 const struct netflow_options *nf_options)
803 if (nf_options && nf_options->collectors.n) {
804 if (!ofproto->netflow) {
805 ofproto->netflow = netflow_create();
807 return netflow_set_options(ofproto->netflow, nf_options);
809 netflow_destroy(ofproto->netflow);
810 ofproto->netflow = NULL;
816 ofproto_set_sflow(struct ofproto *ofproto,
817 const struct ofproto_sflow_options *oso)
819 struct ofproto_sflow *os = ofproto->sflow;
822 struct ofport *ofport;
823 unsigned int odp_port;
825 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
826 refresh_port_groups(ofproto);
827 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
828 ofproto_sflow_add_port(os, odp_port,
829 netdev_get_name(ofport->netdev));
832 ofproto_sflow_set_options(os, oso);
834 ofproto_sflow_destroy(os);
835 ofproto->sflow = NULL;
840 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
844 VLOG_WARN("STP is not yet implemented");
852 ofproto_get_datapath_id(const struct ofproto *ofproto)
854 return ofproto->datapath_id;
858 ofproto_has_controller(const struct ofproto *ofproto)
860 return !hmap_is_empty(&ofproto->controllers);
864 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
868 for (i = 0; i < ofproto->n_listeners; i++) {
869 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
874 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
878 for (i = 0; i < ofproto->n_snoops; i++) {
879 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
884 ofproto_destroy(struct ofproto *p)
886 struct ofconn *ofconn, *next_ofconn;
887 struct ofport *ofport;
888 unsigned int port_no;
895 /* Destroy fail-open and in-band early, since they touch the classifier. */
896 fail_open_destroy(p->fail_open);
899 in_band_destroy(p->in_band);
901 free(p->extra_in_band_remotes);
903 ofproto_flush_flows(p);
904 classifier_destroy(&p->cls);
906 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
908 ofconn_destroy(ofconn);
910 hmap_destroy(&p->controllers);
913 netdev_monitor_destroy(p->netdev_monitor);
914 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
917 shash_destroy(&p->port_by_name);
919 switch_status_destroy(p->switch_status);
920 netflow_destroy(p->netflow);
921 ofproto_sflow_destroy(p->sflow);
923 for (i = 0; i < p->n_listeners; i++) {
924 pvconn_close(p->listeners[i]);
928 for (i = 0; i < p->n_snoops; i++) {
929 pvconn_close(p->snoops[i]);
933 mac_learning_destroy(p->ml);
938 free(p->serial_desc);
941 port_array_destroy(&p->ports);
947 ofproto_run(struct ofproto *p)
949 int error = ofproto_run1(p);
951 error = ofproto_run2(p, false);
957 process_port_change(struct ofproto *ofproto, int error, char *devname)
959 if (error == ENOBUFS) {
960 reinit_ports(ofproto);
962 update_port(ofproto, devname);
967 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
968 * means that 'ofconn' is more interesting for monitoring than a lower return
971 snoop_preference(const struct ofconn *ofconn)
973 switch (ofconn->role) {
981 /* Shouldn't happen. */
986 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
987 * Connects this vconn to a controller. */
989 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
991 struct ofconn *ofconn, *best;
993 /* Pick a controller for monitoring. */
995 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
996 if (ofconn->type == OFCONN_CONTROLLER
997 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1003 rconn_add_monitor(best->rconn, vconn);
1005 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1011 ofproto_run1(struct ofproto *p)
1013 struct ofconn *ofconn, *next_ofconn;
1018 if (shash_is_empty(&p->port_by_name)) {
1022 for (i = 0; i < 50; i++) {
1026 error = dpif_recv(p->dpif, &buf);
1028 if (error == ENODEV) {
1029 /* Someone destroyed the datapath behind our back. The caller
1030 * better destroy us and give up, because we're just going to
1031 * spin from here on out. */
1032 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1033 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1034 dpif_name(p->dpif));
1040 handle_odp_msg(p, buf);
1043 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1044 process_port_change(p, error, devname);
1046 while ((error = netdev_monitor_poll(p->netdev_monitor,
1047 &devname)) != EAGAIN) {
1048 process_port_change(p, error, devname);
1052 if (time_msec() >= p->next_in_band_update) {
1053 update_in_band_remotes(p);
1055 in_band_run(p->in_band);
1058 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1060 ofconn_run(ofconn, p);
1063 /* Fail-open maintenance. Do this after processing the ofconns since
1064 * fail-open checks the status of the controller rconn. */
1066 fail_open_run(p->fail_open);
1069 for (i = 0; i < p->n_listeners; i++) {
1070 struct vconn *vconn;
1073 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
1075 ofconn_create(p, rconn_new_from_vconn("passive", vconn),
1077 } else if (retval != EAGAIN) {
1078 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1082 for (i = 0; i < p->n_snoops; i++) {
1083 struct vconn *vconn;
1086 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1088 add_snooper(p, vconn);
1089 } else if (retval != EAGAIN) {
1090 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1094 if (time_msec() >= p->next_expiration) {
1095 COVERAGE_INC(ofproto_expiration);
1096 p->next_expiration = time_msec() + 1000;
1099 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1101 /* Let the hook know that we're at a stable point: all outstanding data
1102 * in existing flows has been accounted to the account_cb. Thus, the
1103 * hook can now reasonably do operations that depend on having accurate
1104 * flow volume accounting (currently, that's just bond rebalancing). */
1105 if (p->ofhooks->account_checkpoint_cb) {
1106 p->ofhooks->account_checkpoint_cb(p->aux);
1111 netflow_run(p->netflow);
1114 ofproto_sflow_run(p->sflow);
1120 struct revalidate_cbdata {
1121 struct ofproto *ofproto;
1122 bool revalidate_all; /* Revalidate all exact-match rules? */
1123 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1124 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1128 ofproto_run2(struct ofproto *p, bool revalidate_all)
1130 if (p->need_revalidate || revalidate_all
1131 || !tag_set_is_empty(&p->revalidate_set)) {
1132 struct revalidate_cbdata cbdata;
1134 cbdata.revalidate_all = revalidate_all;
1135 cbdata.revalidate_subrules = p->need_revalidate;
1136 cbdata.revalidate_set = p->revalidate_set;
1137 tag_set_init(&p->revalidate_set);
1138 COVERAGE_INC(ofproto_revalidate);
1139 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1140 p->need_revalidate = false;
1147 ofproto_wait(struct ofproto *p)
1149 struct ofconn *ofconn;
1152 dpif_recv_wait(p->dpif);
1153 dpif_port_poll_wait(p->dpif);
1154 netdev_monitor_poll_wait(p->netdev_monitor);
1155 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1156 ofconn_wait(ofconn);
1159 poll_timer_wait_until(p->next_in_band_update);
1160 in_band_wait(p->in_band);
1163 fail_open_wait(p->fail_open);
1166 ofproto_sflow_wait(p->sflow);
1168 if (!tag_set_is_empty(&p->revalidate_set)) {
1169 poll_immediate_wake();
1171 if (p->need_revalidate) {
1172 /* Shouldn't happen, but if it does just go around again. */
1173 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1174 poll_immediate_wake();
1175 } else if (p->next_expiration != LLONG_MAX) {
1176 poll_timer_wait_until(p->next_expiration);
1178 for (i = 0; i < p->n_listeners; i++) {
1179 pvconn_wait(p->listeners[i]);
1181 for (i = 0; i < p->n_snoops; i++) {
1182 pvconn_wait(p->snoops[i]);
1187 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1189 tag_set_add(&ofproto->revalidate_set, tag);
1193 ofproto_get_revalidate_set(struct ofproto *ofproto)
1195 return &ofproto->revalidate_set;
1199 ofproto_is_alive(const struct ofproto *p)
1201 return !hmap_is_empty(&p->controllers);
1205 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1206 const union ofp_action *actions, size_t n_actions,
1207 const struct ofpbuf *packet)
1209 struct odp_actions odp_actions;
1212 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1218 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1220 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1221 odp_actions.n_actions, packet);
1226 ofproto_add_flow(struct ofproto *p,
1227 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1228 const union ofp_action *actions, size_t n_actions,
1232 rule = rule_create(p, NULL, actions, n_actions,
1233 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1235 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1236 rule_insert(p, rule, NULL, 0);
1240 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1241 uint32_t wildcards, unsigned int priority)
1245 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1249 rule_remove(ofproto, rule);
1254 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1256 struct rule *rule = rule_from_cls_rule(rule_);
1257 struct ofproto *ofproto = ofproto_;
1259 /* Mark the flow as not installed, even though it might really be
1260 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1261 * There is no point in uninstalling it individually since we are about to
1262 * blow away all the flows with dpif_flow_flush(). */
1263 rule->installed = false;
1265 rule_remove(ofproto, rule);
1269 ofproto_flush_flows(struct ofproto *ofproto)
1271 COVERAGE_INC(ofproto_flush);
1272 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1273 dpif_flow_flush(ofproto->dpif);
1274 if (ofproto->in_band) {
1275 in_band_flushed(ofproto->in_band);
1277 if (ofproto->fail_open) {
1278 fail_open_flushed(ofproto->fail_open);
1283 reinit_ports(struct ofproto *p)
1285 struct svec devnames;
1286 struct ofport *ofport;
1287 unsigned int port_no;
1288 struct odp_port *odp_ports;
1292 svec_init(&devnames);
1293 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1294 svec_add (&devnames, (char *) ofport->opp.name);
1296 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1297 for (i = 0; i < n_odp_ports; i++) {
1298 svec_add (&devnames, odp_ports[i].devname);
1302 svec_sort_unique(&devnames);
1303 for (i = 0; i < devnames.n; i++) {
1304 update_port(p, devnames.names[i]);
1306 svec_destroy(&devnames);
1310 refresh_port_group(struct ofproto *p, unsigned int group)
1314 struct ofport *port;
1315 unsigned int port_no;
1317 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1319 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1321 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1322 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1323 ports[n_ports++] = port_no;
1326 dpif_port_group_set(p->dpif, group, ports, n_ports);
1333 refresh_port_groups(struct ofproto *p)
1335 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1336 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1338 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1342 static struct ofport *
1343 make_ofport(const struct odp_port *odp_port)
1345 struct netdev_options netdev_options;
1346 enum netdev_flags flags;
1347 struct ofport *ofport;
1348 struct netdev *netdev;
1352 memset(&netdev_options, 0, sizeof netdev_options);
1353 netdev_options.name = odp_port->devname;
1354 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1355 netdev_options.may_open = true;
1357 error = netdev_open(&netdev_options, &netdev);
1359 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1360 "cannot be opened (%s)",
1361 odp_port->devname, odp_port->port,
1362 odp_port->devname, strerror(error));
1366 ofport = xmalloc(sizeof *ofport);
1367 ofport->netdev = netdev;
1368 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1369 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1370 memcpy(ofport->opp.name, odp_port->devname,
1371 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1372 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1374 netdev_get_flags(netdev, &flags);
1375 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1377 netdev_get_carrier(netdev, &carrier);
1378 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1380 netdev_get_features(netdev,
1381 &ofport->opp.curr, &ofport->opp.advertised,
1382 &ofport->opp.supported, &ofport->opp.peer);
1387 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1389 if (port_array_get(&p->ports, odp_port->port)) {
1390 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1393 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1394 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1403 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1405 const struct ofp_phy_port *a = &a_->opp;
1406 const struct ofp_phy_port *b = &b_->opp;
1408 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1409 return (a->port_no == b->port_no
1410 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1411 && !strcmp((char *) a->name, (char *) b->name)
1412 && a->state == b->state
1413 && a->config == b->config
1414 && a->curr == b->curr
1415 && a->advertised == b->advertised
1416 && a->supported == b->supported
1417 && a->peer == b->peer);
1421 send_port_status(struct ofproto *p, const struct ofport *ofport,
1424 /* XXX Should limit the number of queued port status change messages. */
1425 struct ofconn *ofconn;
1426 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1427 struct ofp_port_status *ops;
1430 if (!ofconn_receives_async_msgs(ofconn)) {
1434 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1435 ops->reason = reason;
1436 ops->desc = ofport->opp;
1437 hton_ofp_phy_port(&ops->desc);
1438 queue_tx(b, ofconn, NULL);
1440 if (p->ofhooks->port_changed_cb) {
1441 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1446 ofport_install(struct ofproto *p, struct ofport *ofport)
1448 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1449 const char *netdev_name = (const char *) ofport->opp.name;
1451 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1452 port_array_set(&p->ports, odp_port, ofport);
1453 shash_add(&p->port_by_name, netdev_name, ofport);
1455 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1460 ofport_remove(struct ofproto *p, struct ofport *ofport)
1462 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1464 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1465 port_array_set(&p->ports, odp_port, NULL);
1466 shash_delete(&p->port_by_name,
1467 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1469 ofproto_sflow_del_port(p->sflow, odp_port);
1474 ofport_free(struct ofport *ofport)
1477 netdev_close(ofport->netdev);
1483 update_port(struct ofproto *p, const char *devname)
1485 struct odp_port odp_port;
1486 struct ofport *old_ofport;
1487 struct ofport *new_ofport;
1490 COVERAGE_INC(ofproto_update_port);
1492 /* Query the datapath for port information. */
1493 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1495 /* Find the old ofport. */
1496 old_ofport = shash_find_data(&p->port_by_name, devname);
1499 /* There's no port named 'devname' but there might be a port with
1500 * the same port number. This could happen if a port is deleted
1501 * and then a new one added in its place very quickly, or if a port
1502 * is renamed. In the former case we want to send an OFPPR_DELETE
1503 * and an OFPPR_ADD, and in the latter case we want to send a
1504 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1505 * the old port's ifindex against the new port, or perhaps less
1506 * reliably but more portably by comparing the old port's MAC
1507 * against the new port's MAC. However, this code isn't that smart
1508 * and always sends an OFPPR_MODIFY (XXX). */
1509 old_ofport = port_array_get(&p->ports, odp_port.port);
1511 } else if (error != ENOENT && error != ENODEV) {
1512 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1513 "%s", strerror(error));
1517 /* Create a new ofport. */
1518 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1520 /* Eliminate a few pathological cases. */
1521 if (!old_ofport && !new_ofport) {
1523 } else if (old_ofport && new_ofport) {
1524 /* Most of the 'config' bits are OpenFlow soft state, but
1525 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1526 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1527 * leaves the other bits 0.) */
1528 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1530 if (ofport_equal(old_ofport, new_ofport)) {
1531 /* False alarm--no change. */
1532 ofport_free(new_ofport);
1537 /* Now deal with the normal cases. */
1539 ofport_remove(p, old_ofport);
1542 ofport_install(p, new_ofport);
1544 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1545 (!old_ofport ? OFPPR_ADD
1546 : !new_ofport ? OFPPR_DELETE
1548 ofport_free(old_ofport);
1550 /* Update port groups. */
1551 refresh_port_groups(p);
1555 init_ports(struct ofproto *p)
1557 struct odp_port *ports;
1562 error = dpif_port_list(p->dpif, &ports, &n_ports);
1567 for (i = 0; i < n_ports; i++) {
1568 const struct odp_port *odp_port = &ports[i];
1569 if (!ofport_conflicts(p, odp_port)) {
1570 struct ofport *ofport = make_ofport(odp_port);
1572 ofport_install(p, ofport);
1577 refresh_port_groups(p);
1581 static struct ofconn *
1582 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1584 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1585 ofconn->ofproto = p;
1586 list_push_back(&p->all_conns, &ofconn->node);
1587 ofconn->rconn = rconn;
1588 ofconn->type = type;
1589 ofconn->role = NX_ROLE_OTHER;
1590 ofconn->packet_in_counter = rconn_packet_counter_create ();
1591 ofconn->pktbuf = NULL;
1592 ofconn->miss_send_len = 0;
1593 ofconn->reply_counter = rconn_packet_counter_create ();
1598 ofconn_destroy(struct ofconn *ofconn)
1600 if (ofconn->type == OFCONN_CONTROLLER) {
1601 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1603 discovery_destroy(ofconn->discovery);
1605 list_remove(&ofconn->node);
1606 switch_status_unregister(ofconn->ss);
1607 rconn_destroy(ofconn->rconn);
1608 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1609 rconn_packet_counter_destroy(ofconn->reply_counter);
1610 pktbuf_destroy(ofconn->pktbuf);
1615 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1620 if (ofconn->discovery) {
1621 char *controller_name;
1622 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1623 discovery_question_connectivity(ofconn->discovery);
1625 if (discovery_run(ofconn->discovery, &controller_name)) {
1626 if (controller_name) {
1627 rconn_connect(ofconn->rconn, controller_name);
1629 rconn_disconnect(ofconn->rconn);
1634 for (i = 0; i < N_SCHEDULERS; i++) {
1635 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1638 rconn_run(ofconn->rconn);
1640 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1641 /* Limit the number of iterations to prevent other tasks from
1643 for (iteration = 0; iteration < 50; iteration++) {
1644 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1649 fail_open_maybe_recover(p->fail_open);
1651 handle_openflow(ofconn, p, of_msg);
1652 ofpbuf_delete(of_msg);
1656 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1657 ofconn_destroy(ofconn);
1662 ofconn_wait(struct ofconn *ofconn)
1666 if (ofconn->discovery) {
1667 discovery_wait(ofconn->discovery);
1669 for (i = 0; i < N_SCHEDULERS; i++) {
1670 pinsched_wait(ofconn->schedulers[i]);
1672 rconn_run_wait(ofconn->rconn);
1673 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1674 rconn_recv_wait(ofconn->rconn);
1676 COVERAGE_INC(ofproto_ofconn_stuck);
1680 /* Returns true if 'ofconn' should receive asynchronous messages. */
1682 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1684 if (ofconn->type == OFCONN_CONTROLLER) {
1685 /* Ordinary controllers always get asynchronous messages unless they
1686 * have configured themselves as "slaves". */
1687 return ofconn->role != NX_ROLE_SLAVE;
1689 /* Transient connections don't get asynchronous messages unless they
1690 * have explicitly asked for them by setting a nonzero miss send
1692 return ofconn->miss_send_len > 0;
1696 /* Caller is responsible for initializing the 'cr' member of the returned
1698 static struct rule *
1699 rule_create(struct ofproto *ofproto, struct rule *super,
1700 const union ofp_action *actions, size_t n_actions,
1701 uint16_t idle_timeout, uint16_t hard_timeout,
1702 uint64_t flow_cookie, bool send_flow_removed)
1704 struct rule *rule = xzalloc(sizeof *rule);
1705 rule->idle_timeout = idle_timeout;
1706 rule->hard_timeout = hard_timeout;
1707 rule->flow_cookie = flow_cookie;
1708 rule->used = rule->created = time_msec();
1709 rule->send_flow_removed = send_flow_removed;
1710 rule->super = super;
1712 list_push_back(&super->list, &rule->list);
1714 list_init(&rule->list);
1716 rule->n_actions = n_actions;
1717 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1718 netflow_flow_clear(&rule->nf_flow);
1719 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1724 static struct rule *
1725 rule_from_cls_rule(const struct cls_rule *cls_rule)
1727 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1731 rule_free(struct rule *rule)
1733 free(rule->actions);
1734 free(rule->odp_actions);
1738 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1739 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1740 * through all of its subrules and revalidates them, destroying any that no
1741 * longer has a super-rule (which is probably all of them).
1743 * Before calling this function, the caller must make have removed 'rule' from
1744 * the classifier. If 'rule' is an exact-match rule, the caller is also
1745 * responsible for ensuring that it has been uninstalled from the datapath. */
1747 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1750 struct rule *subrule, *next;
1751 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1752 revalidate_rule(ofproto, subrule);
1755 list_remove(&rule->list);
1761 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1763 const union ofp_action *oa;
1764 struct actions_iterator i;
1766 if (out_port == htons(OFPP_NONE)) {
1769 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1770 oa = actions_next(&i)) {
1771 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1778 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1779 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1781 * The flow that 'packet' actually contains does not need to actually match
1782 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1783 * the packet and byte counters for 'rule' will be credited for the packet sent
1784 * out whether or not the packet actually matches 'rule'.
1786 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1787 * the caller must already have accurately composed ODP actions for it given
1788 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1789 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1790 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1791 * actions and apply them to 'packet'. */
1793 rule_execute(struct ofproto *ofproto, struct rule *rule,
1794 struct ofpbuf *packet, const flow_t *flow)
1796 const union odp_action *actions;
1798 struct odp_actions a;
1800 /* Grab or compose the ODP actions.
1802 * The special case for an exact-match 'rule' where 'flow' is not the
1803 * rule's flow is important to avoid, e.g., sending a packet out its input
1804 * port simply because the ODP actions were composed for the wrong
1806 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1807 struct rule *super = rule->super ? rule->super : rule;
1808 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1809 packet, &a, NULL, 0, NULL)) {
1812 actions = a.actions;
1813 n_actions = a.n_actions;
1815 actions = rule->odp_actions;
1816 n_actions = rule->n_odp_actions;
1819 /* Execute the ODP actions. */
1820 if (!dpif_execute(ofproto->dpif, flow->in_port,
1821 actions, n_actions, packet)) {
1822 struct odp_flow_stats stats;
1823 flow_extract_stats(flow, packet, &stats);
1824 update_stats(ofproto, rule, &stats);
1825 rule->used = time_msec();
1826 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1831 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1834 struct rule *displaced_rule;
1836 /* Insert the rule in the classifier. */
1837 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1838 if (!rule->cr.wc.wildcards) {
1839 rule_make_actions(p, rule, packet);
1842 /* Send the packet and credit it to the rule. */
1845 flow_extract(packet, 0, in_port, &flow);
1846 rule_execute(p, rule, packet, &flow);
1849 /* Install the rule in the datapath only after sending the packet, to
1850 * avoid packet reordering. */
1851 if (rule->cr.wc.wildcards) {
1852 COVERAGE_INC(ofproto_add_wc_flow);
1853 p->need_revalidate = true;
1855 rule_install(p, rule, displaced_rule);
1858 /* Free the rule that was displaced, if any. */
1859 if (displaced_rule) {
1860 rule_destroy(p, displaced_rule);
1864 static struct rule *
1865 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1868 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1869 rule->idle_timeout, rule->hard_timeout,
1871 COVERAGE_INC(ofproto_subrule_create);
1872 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1873 : rule->cr.priority), &subrule->cr);
1874 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1880 rule_remove(struct ofproto *ofproto, struct rule *rule)
1882 if (rule->cr.wc.wildcards) {
1883 COVERAGE_INC(ofproto_del_wc_flow);
1884 ofproto->need_revalidate = true;
1886 rule_uninstall(ofproto, rule);
1888 classifier_remove(&ofproto->cls, &rule->cr);
1889 rule_destroy(ofproto, rule);
1892 /* Returns true if the actions changed, false otherwise. */
1894 rule_make_actions(struct ofproto *p, struct rule *rule,
1895 const struct ofpbuf *packet)
1897 const struct rule *super;
1898 struct odp_actions a;
1901 assert(!rule->cr.wc.wildcards);
1903 super = rule->super ? rule->super : rule;
1905 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1906 packet, &a, &rule->tags, &rule->may_install,
1907 &rule->nf_flow.output_iface);
1909 actions_len = a.n_actions * sizeof *a.actions;
1910 if (rule->n_odp_actions != a.n_actions
1911 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1912 COVERAGE_INC(ofproto_odp_unchanged);
1913 free(rule->odp_actions);
1914 rule->n_odp_actions = a.n_actions;
1915 rule->odp_actions = xmemdup(a.actions, actions_len);
1923 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1924 struct odp_flow_put *put)
1926 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1927 put->flow.key = rule->cr.flow;
1928 put->flow.actions = rule->odp_actions;
1929 put->flow.n_actions = rule->n_odp_actions;
1930 put->flow.flags = 0;
1932 return dpif_flow_put(ofproto->dpif, put);
1936 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1938 assert(!rule->cr.wc.wildcards);
1940 if (rule->may_install) {
1941 struct odp_flow_put put;
1942 if (!do_put_flow(p, rule,
1943 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1945 rule->installed = true;
1946 if (displaced_rule) {
1947 update_stats(p, displaced_rule, &put.flow.stats);
1948 rule_post_uninstall(p, displaced_rule);
1951 } else if (displaced_rule) {
1952 rule_uninstall(p, displaced_rule);
1957 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1959 if (rule->installed) {
1960 struct odp_flow_put put;
1961 COVERAGE_INC(ofproto_dp_missed);
1962 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1964 rule_install(ofproto, rule, NULL);
1969 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1971 bool actions_changed;
1972 uint16_t new_out_iface, old_out_iface;
1974 old_out_iface = rule->nf_flow.output_iface;
1975 actions_changed = rule_make_actions(ofproto, rule, NULL);
1977 if (rule->may_install) {
1978 if (rule->installed) {
1979 if (actions_changed) {
1980 struct odp_flow_put put;
1981 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1982 | ODPPF_ZERO_STATS, &put);
1983 update_stats(ofproto, rule, &put.flow.stats);
1985 /* Temporarily set the old output iface so that NetFlow
1986 * messages have the correct output interface for the old
1988 new_out_iface = rule->nf_flow.output_iface;
1989 rule->nf_flow.output_iface = old_out_iface;
1990 rule_post_uninstall(ofproto, rule);
1991 rule->nf_flow.output_iface = new_out_iface;
1994 rule_install(ofproto, rule, NULL);
1997 rule_uninstall(ofproto, rule);
2002 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2004 uint64_t total_bytes = rule->byte_count + extra_bytes;
2006 if (ofproto->ofhooks->account_flow_cb
2007 && total_bytes > rule->accounted_bytes)
2009 ofproto->ofhooks->account_flow_cb(
2010 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
2011 total_bytes - rule->accounted_bytes, ofproto->aux);
2012 rule->accounted_bytes = total_bytes;
2017 rule_uninstall(struct ofproto *p, struct rule *rule)
2019 assert(!rule->cr.wc.wildcards);
2020 if (rule->installed) {
2021 struct odp_flow odp_flow;
2023 odp_flow.key = rule->cr.flow;
2024 odp_flow.actions = NULL;
2025 odp_flow.n_actions = 0;
2027 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2028 update_stats(p, rule, &odp_flow.stats);
2030 rule->installed = false;
2032 rule_post_uninstall(p, rule);
2037 is_controller_rule(struct rule *rule)
2039 /* If the only action is send to the controller then don't report
2040 * NetFlow expiration messages since it is just part of the control
2041 * logic for the network and not real traffic. */
2043 if (rule && rule->super) {
2044 struct rule *super = rule->super;
2046 return super->n_actions == 1 &&
2047 super->actions[0].type == htons(OFPAT_OUTPUT) &&
2048 super->actions[0].output.port == htons(OFPP_CONTROLLER);
2055 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2057 struct rule *super = rule->super;
2059 rule_account(ofproto, rule, 0);
2061 if (ofproto->netflow && !is_controller_rule(rule)) {
2062 struct ofexpired expired;
2063 expired.flow = rule->cr.flow;
2064 expired.packet_count = rule->packet_count;
2065 expired.byte_count = rule->byte_count;
2066 expired.used = rule->used;
2067 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2070 super->packet_count += rule->packet_count;
2071 super->byte_count += rule->byte_count;
2073 /* Reset counters to prevent double counting if the rule ever gets
2075 rule->packet_count = 0;
2076 rule->byte_count = 0;
2077 rule->accounted_bytes = 0;
2079 netflow_flow_clear(&rule->nf_flow);
2084 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2085 struct rconn_packet_counter *counter)
2087 update_openflow_length(msg);
2088 if (rconn_send(ofconn->rconn, msg, counter)) {
2094 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2095 int error, const void *data, size_t len)
2098 struct ofp_error_msg *oem;
2100 if (!(error >> 16)) {
2101 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2106 COVERAGE_INC(ofproto_error);
2107 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2108 oh ? oh->xid : 0, &buf);
2109 oem->type = htons((unsigned int) error >> 16);
2110 oem->code = htons(error & 0xffff);
2111 memcpy(oem->data, data, len);
2112 queue_tx(buf, ofconn, ofconn->reply_counter);
2116 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2119 size_t oh_length = ntohs(oh->length);
2120 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2124 hton_ofp_phy_port(struct ofp_phy_port *opp)
2126 opp->port_no = htons(opp->port_no);
2127 opp->config = htonl(opp->config);
2128 opp->state = htonl(opp->state);
2129 opp->curr = htonl(opp->curr);
2130 opp->advertised = htonl(opp->advertised);
2131 opp->supported = htonl(opp->supported);
2132 opp->peer = htonl(opp->peer);
2136 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2138 struct ofp_header *rq = oh;
2139 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2144 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2145 struct ofp_header *oh)
2147 struct ofp_switch_features *osf;
2149 unsigned int port_no;
2150 struct ofport *port;
2152 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2153 osf->datapath_id = htonll(p->datapath_id);
2154 osf->n_buffers = htonl(pktbuf_capacity());
2156 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2157 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2158 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2159 (1u << OFPAT_SET_VLAN_VID) |
2160 (1u << OFPAT_SET_VLAN_PCP) |
2161 (1u << OFPAT_STRIP_VLAN) |
2162 (1u << OFPAT_SET_DL_SRC) |
2163 (1u << OFPAT_SET_DL_DST) |
2164 (1u << OFPAT_SET_NW_SRC) |
2165 (1u << OFPAT_SET_NW_DST) |
2166 (1u << OFPAT_SET_NW_TOS) |
2167 (1u << OFPAT_SET_TP_SRC) |
2168 (1u << OFPAT_SET_TP_DST));
2170 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2171 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2174 queue_tx(buf, ofconn, ofconn->reply_counter);
2179 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2180 struct ofp_header *oh)
2183 struct ofp_switch_config *osc;
2187 /* Figure out flags. */
2188 dpif_get_drop_frags(p->dpif, &drop_frags);
2189 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2192 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2193 osc->flags = htons(flags);
2194 osc->miss_send_len = htons(ofconn->miss_send_len);
2195 queue_tx(buf, ofconn, ofconn->reply_counter);
2201 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2202 struct ofp_switch_config *osc)
2207 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2211 flags = ntohs(osc->flags);
2213 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2214 switch (flags & OFPC_FRAG_MASK) {
2215 case OFPC_FRAG_NORMAL:
2216 dpif_set_drop_frags(p->dpif, false);
2218 case OFPC_FRAG_DROP:
2219 dpif_set_drop_frags(p->dpif, true);
2222 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2228 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2234 add_output_group_action(struct odp_actions *actions, uint16_t group,
2235 uint16_t *nf_output_iface)
2237 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2239 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2240 *nf_output_iface = NF_OUT_FLOOD;
2245 add_controller_action(struct odp_actions *actions,
2246 const struct ofp_action_output *oao)
2248 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2249 a->controller.arg = ntohs(oao->max_len);
2252 struct action_xlate_ctx {
2254 flow_t flow; /* Flow to which these actions correspond. */
2255 int recurse; /* Recursion level, via xlate_table_action. */
2256 struct ofproto *ofproto;
2257 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2258 * null pointer if we are revalidating
2259 * without a packet to refer to. */
2262 struct odp_actions *out; /* Datapath actions. */
2263 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2264 bool may_set_up_flow; /* True ordinarily; false if the actions must
2265 * be reassessed for every packet. */
2266 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2269 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2270 struct action_xlate_ctx *ctx);
2273 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2275 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2278 if (ofport->opp.config & OFPPC_NO_FWD) {
2279 /* Forwarding disabled on port. */
2284 * We don't have an ofport record for this port, but it doesn't hurt to
2285 * allow forwarding to it anyhow. Maybe such a port will appear later
2286 * and we're pre-populating the flow table.
2290 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2291 ctx->nf_output_iface = port;
2294 static struct rule *
2295 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2298 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2300 /* The rule we found might not be valid, since we could be in need of
2301 * revalidation. If it is not valid, don't return it. */
2304 && ofproto->need_revalidate
2305 && !revalidate_rule(ofproto, rule)) {
2306 COVERAGE_INC(ofproto_invalidated);
2314 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2316 if (!ctx->recurse) {
2317 uint16_t old_in_port;
2320 /* Look up a flow with 'in_port' as the input port. Then restore the
2321 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2322 * have surprising behavior). */
2323 old_in_port = ctx->flow.in_port;
2324 ctx->flow.in_port = in_port;
2325 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2326 ctx->flow.in_port = old_in_port;
2334 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2341 xlate_output_action(struct action_xlate_ctx *ctx,
2342 const struct ofp_action_output *oao)
2345 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2347 ctx->nf_output_iface = NF_OUT_DROP;
2349 switch (ntohs(oao->port)) {
2351 add_output_action(ctx, ctx->flow.in_port);
2354 xlate_table_action(ctx, ctx->flow.in_port);
2357 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2358 ctx->out, ctx->tags,
2359 &ctx->nf_output_iface,
2360 ctx->ofproto->aux)) {
2361 COVERAGE_INC(ofproto_uninstallable);
2362 ctx->may_set_up_flow = false;
2366 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2367 &ctx->nf_output_iface);
2370 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2372 case OFPP_CONTROLLER:
2373 add_controller_action(ctx->out, oao);
2376 add_output_action(ctx, ODPP_LOCAL);
2379 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2380 if (odp_port != ctx->flow.in_port) {
2381 add_output_action(ctx, odp_port);
2386 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2387 ctx->nf_output_iface = NF_OUT_FLOOD;
2388 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2389 ctx->nf_output_iface = prev_nf_output_iface;
2390 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2391 ctx->nf_output_iface != NF_OUT_FLOOD) {
2392 ctx->nf_output_iface = NF_OUT_MULTI;
2397 xlate_nicira_action(struct action_xlate_ctx *ctx,
2398 const struct nx_action_header *nah)
2400 const struct nx_action_resubmit *nar;
2401 const struct nx_action_set_tunnel *nast;
2402 union odp_action *oa;
2403 int subtype = ntohs(nah->subtype);
2405 assert(nah->vendor == htonl(NX_VENDOR_ID));
2407 case NXAST_RESUBMIT:
2408 nar = (const struct nx_action_resubmit *) nah;
2409 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2412 case NXAST_SET_TUNNEL:
2413 nast = (const struct nx_action_set_tunnel *) nah;
2414 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2415 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2418 /* If you add a new action here that modifies flow data, don't forget to
2419 * update the flow key in ctx->flow in the same key. */
2422 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2428 do_xlate_actions(const union ofp_action *in, size_t n_in,
2429 struct action_xlate_ctx *ctx)
2431 struct actions_iterator iter;
2432 const union ofp_action *ia;
2433 const struct ofport *port;
2435 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2436 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2437 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2438 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2439 /* Drop this flow. */
2443 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2444 uint16_t type = ntohs(ia->type);
2445 union odp_action *oa;
2449 xlate_output_action(ctx, &ia->output);
2452 case OFPAT_SET_VLAN_VID:
2453 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2454 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2457 case OFPAT_SET_VLAN_PCP:
2458 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2459 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2462 case OFPAT_STRIP_VLAN:
2463 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2464 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2465 ctx->flow.dl_vlan_pcp = 0;
2468 case OFPAT_SET_DL_SRC:
2469 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2470 memcpy(oa->dl_addr.dl_addr,
2471 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2472 memcpy(ctx->flow.dl_src,
2473 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2476 case OFPAT_SET_DL_DST:
2477 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2478 memcpy(oa->dl_addr.dl_addr,
2479 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2480 memcpy(ctx->flow.dl_dst,
2481 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2484 case OFPAT_SET_NW_SRC:
2485 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2486 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2489 case OFPAT_SET_NW_DST:
2490 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2491 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2494 case OFPAT_SET_NW_TOS:
2495 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2496 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2499 case OFPAT_SET_TP_SRC:
2500 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2501 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2504 case OFPAT_SET_TP_DST:
2505 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2506 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2510 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2514 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2521 xlate_actions(const union ofp_action *in, size_t n_in,
2522 const flow_t *flow, struct ofproto *ofproto,
2523 const struct ofpbuf *packet,
2524 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2525 uint16_t *nf_output_iface)
2527 tag_type no_tags = 0;
2528 struct action_xlate_ctx ctx;
2529 COVERAGE_INC(ofproto_ofp2odp);
2530 odp_actions_init(out);
2533 ctx.ofproto = ofproto;
2534 ctx.packet = packet;
2536 ctx.tags = tags ? tags : &no_tags;
2537 ctx.may_set_up_flow = true;
2538 ctx.nf_output_iface = NF_OUT_DROP;
2539 do_xlate_actions(in, n_in, &ctx);
2541 /* Check with in-band control to see if we're allowed to set up this
2543 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2544 ctx.may_set_up_flow = false;
2547 if (may_set_up_flow) {
2548 *may_set_up_flow = ctx.may_set_up_flow;
2550 if (nf_output_iface) {
2551 *nf_output_iface = ctx.nf_output_iface;
2553 if (odp_actions_overflow(out)) {
2554 odp_actions_init(out);
2555 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2560 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2561 * error message code (composed with ofp_mkerr()) for the caller to propagate
2562 * upward. Otherwise, returns 0.
2564 * 'oh' is used to make log messages more informative. */
2566 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2568 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2569 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2572 type_name = ofp_message_type_to_string(oh->type);
2573 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2577 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2584 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2585 struct ofp_header *oh)
2587 struct ofp_packet_out *opo;
2588 struct ofpbuf payload, *buffer;
2589 struct odp_actions actions;
2595 error = reject_slave_controller(ofconn, oh);
2600 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2604 opo = (struct ofp_packet_out *) oh;
2606 COVERAGE_INC(ofproto_packet_out);
2607 if (opo->buffer_id != htonl(UINT32_MAX)) {
2608 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2610 if (error || !buffer) {
2618 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2619 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2620 &flow, p, &payload, &actions, NULL, NULL, NULL);
2625 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2627 ofpbuf_delete(buffer);
2633 update_port_config(struct ofproto *p, struct ofport *port,
2634 uint32_t config, uint32_t mask)
2636 mask &= config ^ port->opp.config;
2637 if (mask & OFPPC_PORT_DOWN) {
2638 if (config & OFPPC_PORT_DOWN) {
2639 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2641 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2644 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2645 if (mask & REVALIDATE_BITS) {
2646 COVERAGE_INC(ofproto_costly_flags);
2647 port->opp.config ^= mask & REVALIDATE_BITS;
2648 p->need_revalidate = true;
2650 #undef REVALIDATE_BITS
2651 if (mask & OFPPC_NO_FLOOD) {
2652 port->opp.config ^= OFPPC_NO_FLOOD;
2653 refresh_port_groups(p);
2655 if (mask & OFPPC_NO_PACKET_IN) {
2656 port->opp.config ^= OFPPC_NO_PACKET_IN;
2661 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2662 struct ofp_header *oh)
2664 const struct ofp_port_mod *opm;
2665 struct ofport *port;
2668 error = reject_slave_controller(ofconn, oh);
2672 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2676 opm = (struct ofp_port_mod *) oh;
2678 port = port_array_get(&p->ports,
2679 ofp_port_to_odp_port(ntohs(opm->port_no)));
2681 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2682 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2683 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2685 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2686 if (opm->advertise) {
2687 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2693 static struct ofpbuf *
2694 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2696 struct ofp_stats_reply *osr;
2699 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2700 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2702 osr->flags = htons(0);
2706 static struct ofpbuf *
2707 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2709 return make_stats_reply(request->header.xid, request->type, body_len);
2713 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2715 struct ofpbuf *msg = *msgp;
2716 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2717 if (nbytes + msg->size > UINT16_MAX) {
2718 struct ofp_stats_reply *reply = msg->data;
2719 reply->flags = htons(OFPSF_REPLY_MORE);
2720 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2721 queue_tx(msg, ofconn, ofconn->reply_counter);
2723 return ofpbuf_put_uninit(*msgp, nbytes);
2727 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2728 struct ofp_stats_request *request)
2730 struct ofp_desc_stats *ods;
2733 msg = start_stats_reply(request, sizeof *ods);
2734 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2735 memset(ods, 0, sizeof *ods);
2736 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2737 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2738 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2739 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2740 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2741 queue_tx(msg, ofconn, ofconn->reply_counter);
2747 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2749 struct rule *rule = rule_from_cls_rule(cls_rule);
2750 int *n_subrules = n_subrules_;
2758 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2759 struct ofp_stats_request *request)
2761 struct ofp_table_stats *ots;
2763 struct odp_stats dpstats;
2764 int n_exact, n_subrules, n_wild;
2766 msg = start_stats_reply(request, sizeof *ots * 2);
2768 /* Count rules of various kinds. */
2770 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2771 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2772 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2775 dpif_get_dp_stats(p->dpif, &dpstats);
2776 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2777 memset(ots, 0, sizeof *ots);
2778 ots->table_id = TABLEID_HASH;
2779 strcpy(ots->name, "hash");
2780 ots->wildcards = htonl(0);
2781 ots->max_entries = htonl(dpstats.max_capacity);
2782 ots->active_count = htonl(n_exact);
2783 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2785 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2787 /* Classifier table. */
2788 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2789 memset(ots, 0, sizeof *ots);
2790 ots->table_id = TABLEID_CLASSIFIER;
2791 strcpy(ots->name, "classifier");
2792 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2794 ots->max_entries = htonl(65536);
2795 ots->active_count = htonl(n_wild);
2796 ots->lookup_count = htonll(0); /* XXX */
2797 ots->matched_count = htonll(0); /* XXX */
2799 queue_tx(msg, ofconn, ofconn->reply_counter);
2804 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2805 struct ofpbuf **msgp)
2807 struct netdev_stats stats;
2808 struct ofp_port_stats *ops;
2810 /* Intentionally ignore return value, since errors will set
2811 * 'stats' to all-1s, which is correct for OpenFlow, and
2812 * netdev_get_stats() will log errors. */
2813 netdev_get_stats(port->netdev, &stats);
2815 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2816 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2817 memset(ops->pad, 0, sizeof ops->pad);
2818 ops->rx_packets = htonll(stats.rx_packets);
2819 ops->tx_packets = htonll(stats.tx_packets);
2820 ops->rx_bytes = htonll(stats.rx_bytes);
2821 ops->tx_bytes = htonll(stats.tx_bytes);
2822 ops->rx_dropped = htonll(stats.rx_dropped);
2823 ops->tx_dropped = htonll(stats.tx_dropped);
2824 ops->rx_errors = htonll(stats.rx_errors);
2825 ops->tx_errors = htonll(stats.tx_errors);
2826 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2827 ops->rx_over_err = htonll(stats.rx_over_errors);
2828 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2829 ops->collisions = htonll(stats.collisions);
2833 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2834 struct ofp_stats_request *osr,
2837 struct ofp_port_stats_request *psr;
2838 struct ofp_port_stats *ops;
2840 struct ofport *port;
2841 unsigned int port_no;
2843 if (arg_size != sizeof *psr) {
2844 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2846 psr = (struct ofp_port_stats_request *) osr->body;
2848 msg = start_stats_reply(osr, sizeof *ops * 16);
2849 if (psr->port_no != htons(OFPP_NONE)) {
2850 port = port_array_get(&p->ports,
2851 ofp_port_to_odp_port(ntohs(psr->port_no)));
2853 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2856 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2857 append_port_stat(port, port_no, ofconn, &msg);
2861 queue_tx(msg, ofconn, ofconn->reply_counter);
2865 struct flow_stats_cbdata {
2866 struct ofproto *ofproto;
2867 struct ofconn *ofconn;
2872 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2873 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2874 * returned statistic include statistics for all of 'rule''s subrules. */
2876 query_stats(struct ofproto *p, struct rule *rule,
2877 uint64_t *packet_countp, uint64_t *byte_countp)
2879 uint64_t packet_count, byte_count;
2880 struct rule *subrule;
2881 struct odp_flow *odp_flows;
2884 /* Start from historical data for 'rule' itself that are no longer tracked
2885 * by the datapath. This counts, for example, subrules that have
2887 packet_count = rule->packet_count;
2888 byte_count = rule->byte_count;
2890 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2891 * wildcarded then on all of its subrules.
2893 * Also, add any statistics that are not tracked by the datapath for each
2894 * subrule. This includes, for example, statistics for packets that were
2895 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2897 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2898 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2899 if (rule->cr.wc.wildcards) {
2901 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2902 odp_flows[i++].key = subrule->cr.flow;
2903 packet_count += subrule->packet_count;
2904 byte_count += subrule->byte_count;
2907 odp_flows[0].key = rule->cr.flow;
2910 /* Fetch up-to-date statistics from the datapath and add them in. */
2911 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2913 for (i = 0; i < n_odp_flows; i++) {
2914 struct odp_flow *odp_flow = &odp_flows[i];
2915 packet_count += odp_flow->stats.n_packets;
2916 byte_count += odp_flow->stats.n_bytes;
2921 /* Return the stats to the caller. */
2922 *packet_countp = packet_count;
2923 *byte_countp = byte_count;
2927 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2929 struct rule *rule = rule_from_cls_rule(rule_);
2930 struct flow_stats_cbdata *cbdata = cbdata_;
2931 struct ofp_flow_stats *ofs;
2932 uint64_t packet_count, byte_count;
2933 size_t act_len, len;
2934 long long int tdiff = time_msec() - rule->created;
2935 uint32_t sec = tdiff / 1000;
2936 uint32_t msec = tdiff - (sec * 1000);
2938 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2942 act_len = sizeof *rule->actions * rule->n_actions;
2943 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2945 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2947 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2948 ofs->length = htons(len);
2949 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2951 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2952 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2953 ofs->duration_sec = htonl(sec);
2954 ofs->duration_nsec = htonl(msec * 1000000);
2955 ofs->cookie = rule->flow_cookie;
2956 ofs->priority = htons(rule->cr.priority);
2957 ofs->idle_timeout = htons(rule->idle_timeout);
2958 ofs->hard_timeout = htons(rule->hard_timeout);
2959 memset(ofs->pad2, 0, sizeof ofs->pad2);
2960 ofs->packet_count = htonll(packet_count);
2961 ofs->byte_count = htonll(byte_count);
2962 memcpy(ofs->actions, rule->actions, act_len);
2966 table_id_to_include(uint8_t table_id)
2968 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2969 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2970 : table_id == 0xff ? CLS_INC_ALL
2975 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2976 const struct ofp_stats_request *osr,
2979 struct ofp_flow_stats_request *fsr;
2980 struct flow_stats_cbdata cbdata;
2981 struct cls_rule target;
2983 if (arg_size != sizeof *fsr) {
2984 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2986 fsr = (struct ofp_flow_stats_request *) osr->body;
2988 COVERAGE_INC(ofproto_flows_req);
2990 cbdata.ofconn = ofconn;
2991 cbdata.out_port = fsr->out_port;
2992 cbdata.msg = start_stats_reply(osr, 1024);
2993 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
2994 classifier_for_each_match(&p->cls, &target,
2995 table_id_to_include(fsr->table_id),
2996 flow_stats_cb, &cbdata);
2997 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3001 struct flow_stats_ds_cbdata {
3002 struct ofproto *ofproto;
3007 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3009 struct rule *rule = rule_from_cls_rule(rule_);
3010 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3011 struct ds *results = cbdata->results;
3012 struct ofp_match match;
3013 uint64_t packet_count, byte_count;
3014 size_t act_len = sizeof *rule->actions * rule->n_actions;
3016 /* Don't report on subrules. */
3017 if (rule->super != NULL) {
3021 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3022 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3023 cbdata->ofproto->tun_id_from_cookie, &match);
3025 ds_put_format(results, "duration=%llds, ",
3026 (time_msec() - rule->created) / 1000);
3027 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3028 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3029 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3030 ofp_print_match(results, &match, true);
3031 ofp_print_actions(results, &rule->actions->header, act_len);
3032 ds_put_cstr(results, "\n");
3035 /* Adds a pretty-printed description of all flows to 'results', including
3036 * those marked hidden by secchan (e.g., by in-band control). */
3038 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3040 struct ofp_match match;
3041 struct cls_rule target;
3042 struct flow_stats_ds_cbdata cbdata;
3044 memset(&match, 0, sizeof match);
3045 match.wildcards = htonl(OVSFW_ALL);
3048 cbdata.results = results;
3050 cls_rule_from_match(&match, 0, false, 0, &target);
3051 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3052 flow_stats_ds_cb, &cbdata);
3055 struct aggregate_stats_cbdata {
3056 struct ofproto *ofproto;
3058 uint64_t packet_count;
3059 uint64_t byte_count;
3064 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3066 struct rule *rule = rule_from_cls_rule(rule_);
3067 struct aggregate_stats_cbdata *cbdata = cbdata_;
3068 uint64_t packet_count, byte_count;
3070 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3074 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3076 cbdata->packet_count += packet_count;
3077 cbdata->byte_count += byte_count;
3082 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3083 const struct ofp_stats_request *osr,
3086 struct ofp_aggregate_stats_request *asr;
3087 struct ofp_aggregate_stats_reply *reply;
3088 struct aggregate_stats_cbdata cbdata;
3089 struct cls_rule target;
3092 if (arg_size != sizeof *asr) {
3093 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3095 asr = (struct ofp_aggregate_stats_request *) osr->body;
3097 COVERAGE_INC(ofproto_agg_request);
3099 cbdata.out_port = asr->out_port;
3100 cbdata.packet_count = 0;
3101 cbdata.byte_count = 0;
3103 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3104 classifier_for_each_match(&p->cls, &target,
3105 table_id_to_include(asr->table_id),
3106 aggregate_stats_cb, &cbdata);
3108 msg = start_stats_reply(osr, sizeof *reply);
3109 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3110 reply->flow_count = htonl(cbdata.n_flows);
3111 reply->packet_count = htonll(cbdata.packet_count);
3112 reply->byte_count = htonll(cbdata.byte_count);
3113 queue_tx(msg, ofconn, ofconn->reply_counter);
3118 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3119 struct ofp_header *oh)
3121 struct ofp_stats_request *osr;
3125 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3130 osr = (struct ofp_stats_request *) oh;
3132 switch (ntohs(osr->type)) {
3134 return handle_desc_stats_request(p, ofconn, osr);
3137 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3139 case OFPST_AGGREGATE:
3140 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3143 return handle_table_stats_request(p, ofconn, osr);
3146 return handle_port_stats_request(p, ofconn, osr, arg_size);
3149 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3152 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3156 static long long int
3157 msec_from_nsec(uint64_t sec, uint32_t nsec)
3159 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3163 update_time(struct ofproto *ofproto, struct rule *rule,
3164 const struct odp_flow_stats *stats)
3166 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3167 if (used > rule->used) {
3169 if (rule->super && used > rule->super->used) {
3170 rule->super->used = used;
3172 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3177 update_stats(struct ofproto *ofproto, struct rule *rule,
3178 const struct odp_flow_stats *stats)
3180 if (stats->n_packets) {
3181 update_time(ofproto, rule, stats);
3182 rule->packet_count += stats->n_packets;
3183 rule->byte_count += stats->n_bytes;
3184 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3189 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3190 * in which no matching flow already exists in the flow table.
3192 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3193 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3194 * code as encoded by ofp_mkerr() on failure.
3196 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3199 add_flow(struct ofproto *p, struct ofconn *ofconn,
3200 const struct ofp_flow_mod *ofm, size_t n_actions)
3202 struct ofpbuf *packet;
3207 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3211 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3213 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3214 ntohs(ofm->priority))) {
3215 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3219 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3220 n_actions, ntohs(ofm->idle_timeout),
3221 ntohs(ofm->hard_timeout), ofm->cookie,
3222 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3223 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3224 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3227 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3228 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3232 in_port = UINT16_MAX;
3235 rule_insert(p, rule, packet, in_port);
3236 ofpbuf_delete(packet);
3240 static struct rule *
3241 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3246 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3248 return rule_from_cls_rule(classifier_find_rule_exactly(
3249 &p->cls, &flow, wildcards,
3250 ntohs(ofm->priority)));
3254 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3255 struct rule *rule, const struct ofp_flow_mod *ofm)
3257 struct ofpbuf *packet;
3262 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3266 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3272 flow_extract(packet, 0, in_port, &flow);
3273 rule_execute(ofproto, rule, packet, &flow);
3274 ofpbuf_delete(packet);
3279 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3281 struct modify_flows_cbdata {
3282 struct ofproto *ofproto;
3283 const struct ofp_flow_mod *ofm;
3288 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3289 size_t n_actions, struct rule *);
3290 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3292 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3293 * encoded by ofp_mkerr() on failure.
3295 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3298 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3299 const struct ofp_flow_mod *ofm, size_t n_actions)
3301 struct modify_flows_cbdata cbdata;
3302 struct cls_rule target;
3306 cbdata.n_actions = n_actions;
3307 cbdata.match = NULL;
3309 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3312 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3313 modify_flows_cb, &cbdata);
3315 /* This credits the packet to whichever flow happened to happened to
3316 * match last. That's weird. Maybe we should do a lookup for the
3317 * flow that actually matches the packet? Who knows. */
3318 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3321 return add_flow(p, ofconn, ofm, n_actions);
3325 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3326 * code as encoded by ofp_mkerr() on failure.
3328 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3331 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3332 struct ofp_flow_mod *ofm, size_t n_actions)
3334 struct rule *rule = find_flow_strict(p, ofm);
3335 if (rule && !rule_is_hidden(rule)) {
3336 modify_flow(p, ofm, n_actions, rule);
3337 return send_buffered_packet(p, ofconn, rule, ofm);
3339 return add_flow(p, ofconn, ofm, n_actions);
3343 /* Callback for modify_flows_loose(). */
3345 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3347 struct rule *rule = rule_from_cls_rule(rule_);
3348 struct modify_flows_cbdata *cbdata = cbdata_;
3350 if (!rule_is_hidden(rule)) {
3351 cbdata->match = rule;
3352 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3356 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3357 * been identified as a flow in 'p''s flow table to be modified, by changing
3358 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3359 * ofp_action[] structures). */
3361 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3362 size_t n_actions, struct rule *rule)
3364 size_t actions_len = n_actions * sizeof *rule->actions;
3366 rule->flow_cookie = ofm->cookie;
3368 /* If the actions are the same, do nothing. */
3369 if (n_actions == rule->n_actions
3370 && !memcmp(ofm->actions, rule->actions, actions_len))
3375 /* Replace actions. */
3376 free(rule->actions);
3377 rule->actions = xmemdup(ofm->actions, actions_len);
3378 rule->n_actions = n_actions;
3380 /* Make sure that the datapath gets updated properly. */
3381 if (rule->cr.wc.wildcards) {
3382 COVERAGE_INC(ofproto_mod_wc_flow);
3383 p->need_revalidate = true;
3385 rule_update_actions(p, rule);
3391 /* OFPFC_DELETE implementation. */
3393 struct delete_flows_cbdata {
3394 struct ofproto *ofproto;
3398 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3399 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3401 /* Implements OFPFC_DELETE. */
3403 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3405 struct delete_flows_cbdata cbdata;
3406 struct cls_rule target;
3409 cbdata.out_port = ofm->out_port;
3411 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3414 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3415 delete_flows_cb, &cbdata);
3418 /* Implements OFPFC_DELETE_STRICT. */
3420 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3422 struct rule *rule = find_flow_strict(p, ofm);
3424 delete_flow(p, rule, ofm->out_port);
3428 /* Callback for delete_flows_loose(). */
3430 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3432 struct rule *rule = rule_from_cls_rule(rule_);
3433 struct delete_flows_cbdata *cbdata = cbdata_;
3435 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3438 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3439 * been identified as a flow to delete from 'p''s flow table, by deleting the
3440 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3443 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3444 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3445 * specified 'out_port'. */
3447 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3449 if (rule_is_hidden(rule)) {
3453 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3457 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3458 rule_remove(p, rule);
3462 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3463 struct ofp_flow_mod *ofm)
3468 error = reject_slave_controller(ofconn, &ofm->header);
3472 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3473 sizeof *ofm->actions, &n_actions);
3478 /* We do not support the emergency flow cache. It will hopefully
3479 * get dropped from OpenFlow in the near future. */
3480 if (ofm->flags & htons(OFPFF_EMERG)) {
3481 /* There isn't a good fit for an error code, so just state that the
3482 * flow table is full. */
3483 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3486 normalize_match(&ofm->match);
3487 if (!ofm->match.wildcards) {
3488 ofm->priority = htons(UINT16_MAX);
3491 error = validate_actions((const union ofp_action *) ofm->actions,
3492 n_actions, p->max_ports);
3497 switch (ntohs(ofm->command)) {
3499 return add_flow(p, ofconn, ofm, n_actions);
3502 return modify_flows_loose(p, ofconn, ofm, n_actions);
3504 case OFPFC_MODIFY_STRICT:
3505 return modify_flow_strict(p, ofconn, ofm, n_actions);
3508 delete_flows_loose(p, ofm);
3511 case OFPFC_DELETE_STRICT:
3512 delete_flow_strict(p, ofm);
3516 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3521 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3525 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3530 p->tun_id_from_cookie = !!msg->set;
3535 handle_role_request(struct ofproto *ofproto,
3536 struct ofconn *ofconn, struct nicira_header *msg)
3538 struct nx_role_request *nrr;
3539 struct nx_role_request *reply;
3543 if (ntohs(msg->header.length) != sizeof *nrr) {
3544 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3545 ntohs(msg->header.length), sizeof *nrr);
3546 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3548 nrr = (struct nx_role_request *) msg;
3550 if (ofconn->type != OFCONN_CONTROLLER) {
3551 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3553 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3556 role = ntohl(nrr->role);
3557 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3558 && role != NX_ROLE_SLAVE) {
3559 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3561 /* There's no good error code for this. */
3562 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3565 if (role == NX_ROLE_MASTER) {
3566 struct ofconn *other;
3568 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3569 &ofproto->controllers) {
3570 if (other->role == NX_ROLE_MASTER) {
3571 other->role = NX_ROLE_SLAVE;
3575 ofconn->role = role;
3577 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3579 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3580 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3581 reply->role = htonl(role);
3582 queue_tx(buf, ofconn, ofconn->reply_counter);
3588 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3590 struct ofp_vendor_header *ovh = msg;
3591 struct nicira_header *nh;
3593 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3594 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3595 "(expected at least %zu)",
3596 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3597 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3599 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3600 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3602 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3603 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3604 "(expected at least %zu)",
3605 ntohs(ovh->header.length), sizeof(struct nicira_header));
3606 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3610 switch (ntohl(nh->subtype)) {
3611 case NXT_STATUS_REQUEST:
3612 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3615 case NXT_TUN_ID_FROM_COOKIE:
3616 return handle_tun_id_from_cookie(p, msg);
3618 case NXT_ROLE_REQUEST:
3619 return handle_role_request(p, ofconn, msg);
3622 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3626 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3628 struct ofp_header *ob;
3631 /* Currently, everything executes synchronously, so we can just
3632 * immediately send the barrier reply. */
3633 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3634 queue_tx(buf, ofconn, ofconn->reply_counter);
3639 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3640 struct ofpbuf *ofp_msg)
3642 struct ofp_header *oh = ofp_msg->data;
3645 COVERAGE_INC(ofproto_recv_openflow);
3647 case OFPT_ECHO_REQUEST:
3648 error = handle_echo_request(ofconn, oh);
3651 case OFPT_ECHO_REPLY:
3655 case OFPT_FEATURES_REQUEST:
3656 error = handle_features_request(p, ofconn, oh);
3659 case OFPT_GET_CONFIG_REQUEST:
3660 error = handle_get_config_request(p, ofconn, oh);
3663 case OFPT_SET_CONFIG:
3664 error = handle_set_config(p, ofconn, ofp_msg->data);
3667 case OFPT_PACKET_OUT:
3668 error = handle_packet_out(p, ofconn, ofp_msg->data);
3672 error = handle_port_mod(p, ofconn, oh);
3676 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3679 case OFPT_STATS_REQUEST:
3680 error = handle_stats_request(p, ofconn, oh);
3684 error = handle_vendor(p, ofconn, ofp_msg->data);
3687 case OFPT_BARRIER_REQUEST:
3688 error = handle_barrier_request(ofconn, oh);
3692 if (VLOG_IS_WARN_ENABLED()) {
3693 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3694 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3697 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3702 send_error_oh(ofconn, ofp_msg->data, error);
3707 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3709 struct odp_msg *msg = packet->data;
3711 struct ofpbuf payload;
3714 payload.data = msg + 1;
3715 payload.size = msg->length - sizeof *msg;
3716 flow_extract(&payload, msg->arg, msg->port, &flow);
3718 /* Check with in-band control to see if this packet should be sent
3719 * to the local port regardless of the flow table. */
3720 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3721 union odp_action action;
3723 memset(&action, 0, sizeof(action));
3724 action.output.type = ODPAT_OUTPUT;
3725 action.output.port = ODPP_LOCAL;
3726 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3729 rule = lookup_valid_rule(p, &flow);
3731 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3732 struct ofport *port = port_array_get(&p->ports, msg->port);
3734 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3735 COVERAGE_INC(ofproto_no_packet_in);
3736 /* XXX install 'drop' flow entry */
3737 ofpbuf_delete(packet);
3741 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3744 COVERAGE_INC(ofproto_packet_in);
3745 send_packet_in(p, packet);
3749 if (rule->cr.wc.wildcards) {
3750 rule = rule_create_subrule(p, rule, &flow);
3751 rule_make_actions(p, rule, packet);
3753 if (!rule->may_install) {
3754 /* The rule is not installable, that is, we need to process every
3755 * packet, so process the current packet and set its actions into
3757 rule_make_actions(p, rule, packet);
3759 /* XXX revalidate rule if it needs it */
3763 rule_execute(p, rule, &payload, &flow);
3764 rule_reinstall(p, rule);
3766 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3768 * Extra-special case for fail-open mode.
3770 * We are in fail-open mode and the packet matched the fail-open rule,
3771 * but we are connected to a controller too. We should send the packet
3772 * up to the controller in the hope that it will try to set up a flow
3773 * and thereby allow us to exit fail-open.
3775 * See the top-level comment in fail-open.c for more information.
3777 send_packet_in(p, packet);
3779 ofpbuf_delete(packet);
3784 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3786 struct odp_msg *msg = packet->data;
3788 switch (msg->type) {
3789 case _ODPL_ACTION_NR:
3790 COVERAGE_INC(ofproto_ctlr_action);
3791 send_packet_in(p, packet);
3794 case _ODPL_SFLOW_NR:
3796 ofproto_sflow_received(p->sflow, msg);
3798 ofpbuf_delete(packet);
3802 handle_odp_miss_msg(p, packet);
3806 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3813 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3815 struct rule *sub = rule_from_cls_rule(sub_);
3816 struct revalidate_cbdata *cbdata = cbdata_;
3818 if (cbdata->revalidate_all
3819 || (cbdata->revalidate_subrules && sub->super)
3820 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3821 revalidate_rule(cbdata->ofproto, sub);
3826 revalidate_rule(struct ofproto *p, struct rule *rule)
3828 const flow_t *flow = &rule->cr.flow;
3830 COVERAGE_INC(ofproto_revalidate_rule);
3833 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3835 rule_remove(p, rule);
3837 } else if (super != rule->super) {
3838 COVERAGE_INC(ofproto_revalidate_moved);
3839 list_remove(&rule->list);
3840 list_push_back(&super->list, &rule->list);
3841 rule->super = super;
3842 rule->hard_timeout = super->hard_timeout;
3843 rule->idle_timeout = super->idle_timeout;
3844 rule->created = super->created;
3849 rule_update_actions(p, rule);
3853 static struct ofpbuf *
3854 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3855 long long int now, uint8_t reason)
3857 struct ofp_flow_removed *ofr;
3859 long long int tdiff = now - rule->created;
3860 uint32_t sec = tdiff / 1000;
3861 uint32_t msec = tdiff - (sec * 1000);
3863 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3864 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3866 ofr->cookie = rule->flow_cookie;
3867 ofr->priority = htons(rule->cr.priority);
3868 ofr->reason = reason;
3869 ofr->duration_sec = htonl(sec);
3870 ofr->duration_nsec = htonl(msec * 1000000);
3871 ofr->idle_timeout = htons(rule->idle_timeout);
3872 ofr->packet_count = htonll(rule->packet_count);
3873 ofr->byte_count = htonll(rule->byte_count);
3879 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3881 assert(rule->installed);
3882 assert(!rule->cr.wc.wildcards);
3885 rule_remove(ofproto, rule);
3887 rule_uninstall(ofproto, rule);
3892 send_flow_removed(struct ofproto *p, struct rule *rule,
3893 long long int now, uint8_t reason)
3895 struct ofconn *ofconn;
3896 struct ofconn *prev;
3897 struct ofpbuf *buf = NULL;
3899 /* We limit the maximum number of queued flow expirations it by accounting
3900 * them under the counter for replies. That works because preventing
3901 * OpenFlow requests from being processed also prevents new flows from
3902 * being added (and expiring). (It also prevents processing OpenFlow
3903 * requests that would not add new flows, so it is imperfect.) */
3906 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3907 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
3908 && ofconn_receives_async_msgs(ofconn)) {
3910 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3912 buf = compose_flow_removed(p, rule, now, reason);
3918 queue_tx(buf, prev, prev->reply_counter);
3924 expire_rule(struct cls_rule *cls_rule, void *p_)
3926 struct ofproto *p = p_;
3927 struct rule *rule = rule_from_cls_rule(cls_rule);
3928 long long int hard_expire, idle_expire, expire, now;
3930 hard_expire = (rule->hard_timeout
3931 ? rule->created + rule->hard_timeout * 1000
3933 idle_expire = (rule->idle_timeout
3934 && (rule->super || list_is_empty(&rule->list))
3935 ? rule->used + rule->idle_timeout * 1000
3937 expire = MIN(hard_expire, idle_expire);
3941 if (rule->installed && now >= rule->used + 5000) {
3942 uninstall_idle_flow(p, rule);
3943 } else if (!rule->cr.wc.wildcards) {
3944 active_timeout(p, rule);
3950 COVERAGE_INC(ofproto_expired);
3952 /* Update stats. This code will be a no-op if the rule expired
3953 * due to an idle timeout. */
3954 if (rule->cr.wc.wildcards) {
3955 struct rule *subrule, *next;
3956 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3957 rule_remove(p, subrule);
3960 rule_uninstall(p, rule);
3963 if (!rule_is_hidden(rule)) {
3964 send_flow_removed(p, rule, now,
3966 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3968 rule_remove(p, rule);
3972 active_timeout(struct ofproto *ofproto, struct rule *rule)
3974 if (ofproto->netflow && !is_controller_rule(rule) &&
3975 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3976 struct ofexpired expired;
3977 struct odp_flow odp_flow;
3979 /* Get updated flow stats. */
3980 memset(&odp_flow, 0, sizeof odp_flow);
3981 if (rule->installed) {
3982 odp_flow.key = rule->cr.flow;
3983 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3984 dpif_flow_get(ofproto->dpif, &odp_flow);
3986 if (odp_flow.stats.n_packets) {
3987 update_time(ofproto, rule, &odp_flow.stats);
3988 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3989 odp_flow.stats.tcp_flags);
3993 expired.flow = rule->cr.flow;
3994 expired.packet_count = rule->packet_count +
3995 odp_flow.stats.n_packets;
3996 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3997 expired.used = rule->used;
3999 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4001 /* Schedule us to send the accumulated records once we have
4002 * collected all of them. */
4003 poll_immediate_wake();
4008 update_used(struct ofproto *p)
4010 struct odp_flow *flows;
4015 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4020 for (i = 0; i < n_flows; i++) {
4021 struct odp_flow *f = &flows[i];
4024 rule = rule_from_cls_rule(
4025 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4026 if (!rule || !rule->installed) {
4027 COVERAGE_INC(ofproto_unexpected_rule);
4028 dpif_flow_del(p->dpif, f);
4032 update_time(p, rule, &f->stats);
4033 rule_account(p, rule, f->stats.n_bytes);
4038 /* pinsched callback for sending 'packet' on 'ofconn'. */
4040 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4042 struct ofconn *ofconn = ofconn_;
4044 rconn_send_with_limit(ofconn->rconn, packet,
4045 ofconn->packet_in_counter, 100);
4048 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4049 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4050 * packet scheduler for sending.
4052 * 'max_len' specifies the maximum number of bytes of the packet to send on
4053 * 'ofconn' (INT_MAX specifies no limit).
4055 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4056 * ownership is transferred to this function. */
4058 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4061 struct ofproto *ofproto = ofconn->ofproto;
4062 struct ofp_packet_in *opi = packet->data;
4063 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4064 int send_len, trim_size;
4068 if (opi->reason == OFPR_ACTION) {
4069 buffer_id = UINT32_MAX;
4070 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4071 buffer_id = pktbuf_get_null();
4072 } else if (!ofconn->pktbuf) {
4073 buffer_id = UINT32_MAX;
4075 struct ofpbuf payload;
4076 payload.data = opi->data;
4077 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4078 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4081 /* Figure out how much of the packet to send. */
4082 send_len = ntohs(opi->total_len);
4083 if (buffer_id != UINT32_MAX) {
4084 send_len = MIN(send_len, ofconn->miss_send_len);
4086 send_len = MIN(send_len, max_len);
4088 /* Adjust packet length and clone if necessary. */
4089 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4091 packet = ofpbuf_clone_data(packet->data, trim_size);
4094 packet->size = trim_size;
4097 /* Update packet headers. */
4098 opi->buffer_id = htonl(buffer_id);
4099 update_openflow_length(packet);
4101 /* Hand over to packet scheduler. It might immediately call into
4102 * do_send_packet_in() or it might buffer it for a while (until a later
4103 * call to pinsched_run()). */
4104 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4105 packet, do_send_packet_in, ofconn);
4108 /* Replace struct odp_msg header in 'packet' by equivalent struct
4109 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4110 * returned by dpif_recv()).
4112 * The conversion is not complete: the caller still needs to trim any unneeded
4113 * payload off the end of the buffer, set the length in the OpenFlow header,
4114 * and set buffer_id. Those require us to know the controller settings and so
4115 * must be done on a per-controller basis.
4117 * Returns the maximum number of bytes of the packet that should be sent to
4118 * the controller (INT_MAX if no limit). */
4120 do_convert_to_packet_in(struct ofpbuf *packet)
4122 struct odp_msg *msg = packet->data;
4123 struct ofp_packet_in *opi;
4129 /* Extract relevant header fields */
4130 if (msg->type == _ODPL_ACTION_NR) {
4131 reason = OFPR_ACTION;
4134 reason = OFPR_NO_MATCH;
4137 total_len = msg->length - sizeof *msg;
4138 in_port = odp_port_to_ofp_port(msg->port);
4140 /* Repurpose packet buffer by overwriting header. */
4141 ofpbuf_pull(packet, sizeof(struct odp_msg));
4142 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4143 opi->header.version = OFP_VERSION;
4144 opi->header.type = OFPT_PACKET_IN;
4145 opi->total_len = htons(total_len);
4146 opi->in_port = htons(in_port);
4147 opi->reason = reason;
4152 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4153 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4154 * as necessary according to their individual configurations.
4156 * 'packet' must have sufficient headroom to convert it into a struct
4157 * ofp_packet_in (e.g. as returned by dpif_recv()).
4159 * Takes ownership of 'packet'. */
4161 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4163 struct ofconn *ofconn, *prev;
4166 max_len = do_convert_to_packet_in(packet);
4169 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4170 if (ofconn_receives_async_msgs(ofconn)) {
4172 schedule_packet_in(prev, packet, max_len, true);
4178 schedule_packet_in(prev, packet, max_len, false);
4180 ofpbuf_delete(packet);
4185 pick_datapath_id(const struct ofproto *ofproto)
4187 const struct ofport *port;
4189 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4191 uint8_t ea[ETH_ADDR_LEN];
4194 error = netdev_get_etheraddr(port->netdev, ea);
4196 return eth_addr_to_uint64(ea);
4198 VLOG_WARN("could not get MAC address for %s (%s)",
4199 netdev_get_name(port->netdev), strerror(error));
4201 return ofproto->fallback_dpid;
4205 pick_fallback_dpid(void)
4207 uint8_t ea[ETH_ADDR_LEN];
4208 eth_addr_nicira_random(ea);
4209 return eth_addr_to_uint64(ea);
4213 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4214 struct odp_actions *actions, tag_type *tags,
4215 uint16_t *nf_output_iface, void *ofproto_)
4217 struct ofproto *ofproto = ofproto_;
4220 /* Drop frames for reserved multicast addresses. */
4221 if (eth_addr_is_reserved(flow->dl_dst)) {
4225 /* Learn source MAC (but don't try to learn from revalidation). */
4226 if (packet != NULL) {
4227 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4230 /* The log messages here could actually be useful in debugging,
4231 * so keep the rate limit relatively high. */
4232 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4233 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4234 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4235 ofproto_revalidate(ofproto, rev_tag);
4239 /* Determine output port. */
4240 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
4242 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4243 } else if (out_port != flow->in_port) {
4244 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4245 *nf_output_iface = out_port;
4253 static const struct ofhooks default_ofhooks = {
4255 default_normal_ofhook_cb,