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"
40 #include "ofproto-sflow.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
44 #include "openvswitch/datapath-protocol.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
53 #include "stream-ssl.h"
62 VLOG_DEFINE_THIS_MODULE(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 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
173 * maintains persistent connections to these controllers and by default
174 * sends them asynchronous messages such as packet-ins.
176 * - "Service" 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.
180 * Currently, active (tcp, ssl, unix) connections are always "primary"
181 * connections and passive (ptcp, pssl, punix) connections are always "service"
182 * connections. There is no inherent reason for this, but it reflects the
186 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
187 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
190 /* A listener for incoming OpenFlow "service" connections. */
192 struct hmap_node node; /* In struct ofproto's "services" hmap. */
193 struct pvconn *pvconn; /* OpenFlow connection listener. */
195 /* These are not used by ofservice directly. They are settings for
196 * accepted "struct ofconn"s from the pvconn. */
197 int probe_interval; /* Max idle time before probing, in seconds. */
198 int rate_limit; /* Max packet-in rate in packets per second. */
199 int burst_limit; /* Limit on accumulating packet credits. */
202 static struct ofservice *ofservice_lookup(struct ofproto *,
204 static int ofservice_create(struct ofproto *,
205 const struct ofproto_controller *);
206 static void ofservice_reconfigure(struct ofservice *,
207 const struct ofproto_controller *);
208 static void ofservice_destroy(struct ofproto *, struct ofservice *);
210 /* An OpenFlow connection. */
212 struct ofproto *ofproto; /* The ofproto that owns this connection. */
213 struct list node; /* In struct ofproto's "all_conns" list. */
214 struct rconn *rconn; /* OpenFlow connection. */
215 enum ofconn_type type; /* Type. */
217 /* OFPT_PACKET_IN related data. */
218 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
219 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
220 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
221 int miss_send_len; /* Bytes to send of buffered packets. */
223 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
224 * requests, and the maximum number before we stop reading OpenFlow
226 #define OFCONN_REPLY_MAX 100
227 struct rconn_packet_counter *reply_counter;
229 /* type == OFCONN_PRIMARY only. */
230 enum nx_role role; /* Role. */
231 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
232 struct discovery *discovery; /* Controller discovery object, if enabled. */
233 struct status_category *ss; /* Switch status category. */
234 enum ofproto_band band; /* In-band or out-of-band? */
237 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
238 * "schedulers" array. Their values are 0 and 1, and their meanings and values
239 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
240 * case anything ever changes, check their values here. */
241 #define N_SCHEDULERS 2
242 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
243 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
244 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
245 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
247 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
249 static void ofconn_destroy(struct ofconn *);
250 static void ofconn_run(struct ofconn *, struct ofproto *);
251 static void ofconn_wait(struct ofconn *);
252 static bool ofconn_receives_async_msgs(const struct ofconn *);
253 static char *ofconn_make_name(const struct ofproto *, const char *target);
254 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
256 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
257 struct rconn_packet_counter *counter);
259 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
260 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
264 uint64_t datapath_id; /* Datapath ID. */
265 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
266 char *mfr_desc; /* Manufacturer. */
267 char *hw_desc; /* Hardware. */
268 char *sw_desc; /* Software version. */
269 char *serial_desc; /* Serial number. */
270 char *dp_desc; /* Datapath description. */
274 struct netdev_monitor *netdev_monitor;
275 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
277 struct shash port_by_name;
281 struct switch_status *switch_status;
282 struct fail_open *fail_open;
283 struct netflow *netflow;
284 struct ofproto_sflow *sflow;
286 /* In-band control. */
287 struct in_band *in_band;
288 long long int next_in_band_update;
289 struct sockaddr_in *extra_in_band_remotes;
290 size_t n_extra_remotes;
293 struct classifier cls;
294 bool need_revalidate;
295 long long int next_expiration;
296 struct tag_set revalidate_set;
297 bool tun_id_from_cookie;
299 /* OpenFlow connections. */
300 struct hmap controllers; /* Controller "struct ofconn"s. */
301 struct list all_conns; /* Contains "struct ofconn"s. */
302 enum ofproto_fail_mode fail_mode;
304 /* OpenFlow listeners. */
305 struct hmap services; /* Contains "struct ofservice"s. */
306 struct pvconn **snoops;
309 /* Hooks for ovs-vswitchd. */
310 const struct ofhooks *ofhooks;
313 /* Used by default ofhooks. */
314 struct mac_learning *ml;
317 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
319 static const struct ofhooks default_ofhooks;
321 static uint64_t pick_datapath_id(const struct ofproto *);
322 static uint64_t pick_fallback_dpid(void);
324 static void update_used(struct ofproto *);
325 static void update_stats(struct ofproto *, struct rule *,
326 const struct odp_flow_stats *);
327 static void expire_rule(struct cls_rule *, void *ofproto);
328 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
329 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
330 static void revalidate_cb(struct cls_rule *rule_, void *p_);
332 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
334 static void handle_openflow(struct ofconn *, struct ofproto *,
337 static void refresh_port_groups(struct ofproto *);
339 static void update_port(struct ofproto *, const char *devname);
340 static int init_ports(struct ofproto *);
341 static void reinit_ports(struct ofproto *);
344 ofproto_create(const char *datapath, const char *datapath_type,
345 const struct ofhooks *ofhooks, void *aux,
346 struct ofproto **ofprotop)
348 struct odp_stats stats;
355 /* Connect to datapath and start listening for messages. */
356 error = dpif_open(datapath, datapath_type, &dpif);
358 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
361 error = dpif_get_dp_stats(dpif, &stats);
363 VLOG_ERR("failed to obtain stats for datapath %s: %s",
364 datapath, strerror(error));
368 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
370 VLOG_ERR("failed to listen on datapath %s: %s",
371 datapath, strerror(error));
375 dpif_flow_flush(dpif);
376 dpif_recv_purge(dpif);
378 /* Initialize settings. */
379 p = xzalloc(sizeof *p);
380 p->fallback_dpid = pick_fallback_dpid();
381 p->datapath_id = p->fallback_dpid;
382 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
383 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
384 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
385 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
386 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
388 /* Initialize datapath. */
390 p->netdev_monitor = netdev_monitor_create();
391 port_array_init(&p->ports);
392 shash_init(&p->port_by_name);
393 p->max_ports = stats.max_ports;
395 /* Initialize submodules. */
396 p->switch_status = switch_status_create(p);
402 /* Initialize flow table. */
403 classifier_init(&p->cls);
404 p->need_revalidate = false;
405 p->next_expiration = time_msec() + 1000;
406 tag_set_init(&p->revalidate_set);
408 /* Initialize OpenFlow connections. */
409 list_init(&p->all_conns);
410 hmap_init(&p->controllers);
411 hmap_init(&p->services);
415 /* Initialize hooks. */
417 p->ofhooks = ofhooks;
421 p->ofhooks = &default_ofhooks;
423 p->ml = mac_learning_create();
426 /* Pick final datapath ID. */
427 p->datapath_id = pick_datapath_id(p);
428 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
435 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
437 uint64_t old_dpid = p->datapath_id;
438 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
439 if (p->datapath_id != old_dpid) {
440 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
442 /* Force all active connections to reconnect, since there is no way to
443 * notify a controller that the datapath ID has changed. */
444 ofproto_reconnect_controllers(p);
449 is_discovery_controller(const struct ofproto_controller *c)
451 return !strcmp(c->target, "discover");
455 is_in_band_controller(const struct ofproto_controller *c)
457 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
460 /* Creates a new controller in 'ofproto'. Some of the settings are initially
461 * drawn from 'c', but update_controller() needs to be called later to finish
462 * the new ofconn's configuration. */
464 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
466 struct discovery *discovery;
467 struct ofconn *ofconn;
469 if (is_discovery_controller(c)) {
470 int error = discovery_create(c->accept_re, c->update_resolv_conf,
471 ofproto->dpif, ofproto->switch_status,
480 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
481 ofconn->pktbuf = pktbuf_create();
482 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
484 ofconn->discovery = discovery;
486 char *name = ofconn_make_name(ofproto, c->target);
487 rconn_connect(ofconn->rconn, c->target, name);
490 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
491 hash_string(c->target, 0));
494 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
495 * target or turn discovery on or off (these are done by creating new ofconns
496 * and deleting old ones), but it can update the rest of an ofconn's
499 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
503 ofconn->band = (is_in_band_controller(c)
504 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
506 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
508 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
509 rconn_set_probe_interval(ofconn->rconn, probe_interval);
511 if (ofconn->discovery) {
512 discovery_set_update_resolv_conf(ofconn->discovery,
513 c->update_resolv_conf);
514 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
517 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
521 ofconn_get_target(const struct ofconn *ofconn)
523 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
526 static struct ofconn *
527 find_controller_by_target(struct ofproto *ofproto, const char *target)
529 struct ofconn *ofconn;
531 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
532 hash_string(target, 0), &ofproto->controllers) {
533 if (!strcmp(ofconn_get_target(ofconn), target)) {
541 update_in_band_remotes(struct ofproto *ofproto)
543 const struct ofconn *ofconn;
544 struct sockaddr_in *addrs;
545 size_t max_addrs, n_addrs;
549 /* Allocate enough memory for as many remotes as we could possibly have. */
550 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
551 addrs = xmalloc(max_addrs * sizeof *addrs);
554 /* Add all the remotes. */
556 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
557 struct sockaddr_in *sin = &addrs[n_addrs];
559 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
563 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
564 if (sin->sin_addr.s_addr) {
565 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
568 if (ofconn->discovery) {
572 for (i = 0; i < ofproto->n_extra_remotes; i++) {
573 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
576 /* Create or update or destroy in-band.
578 * Ordinarily we only enable in-band if there's at least one remote
579 * address, but discovery needs the in-band rules for DHCP to be installed
580 * even before we know any remote addresses. */
581 if (n_addrs || discovery) {
582 if (!ofproto->in_band) {
583 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
586 if (ofproto->in_band) {
587 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
589 ofproto->next_in_band_update = time_msec() + 1000;
591 in_band_destroy(ofproto->in_band);
592 ofproto->in_band = NULL;
600 update_fail_open(struct ofproto *p)
602 struct ofconn *ofconn;
604 if (!hmap_is_empty(&p->controllers)
605 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
606 struct rconn **rconns;
610 p->fail_open = fail_open_create(p, p->switch_status);
614 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
615 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
616 rconns[n++] = ofconn->rconn;
619 fail_open_set_controllers(p->fail_open, rconns, n);
620 /* p->fail_open takes ownership of 'rconns'. */
622 fail_open_destroy(p->fail_open);
628 ofproto_set_controllers(struct ofproto *p,
629 const struct ofproto_controller *controllers,
630 size_t n_controllers)
632 struct shash new_controllers;
633 struct ofconn *ofconn, *next_ofconn;
634 struct ofservice *ofservice, *next_ofservice;
638 /* Create newly configured controllers and services.
639 * Create a name to ofproto_controller mapping in 'new_controllers'. */
640 shash_init(&new_controllers);
641 for (i = 0; i < n_controllers; i++) {
642 const struct ofproto_controller *c = &controllers[i];
644 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
645 if (!find_controller_by_target(p, c->target)) {
646 add_controller(p, c);
648 } else if (!pvconn_verify_name(c->target)) {
649 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
653 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
654 dpif_name(p->dpif), c->target);
658 shash_add_once(&new_controllers, c->target, &controllers[i]);
661 /* Delete controllers that are no longer configured.
662 * Update configuration of all now-existing controllers. */
664 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node,
666 struct ofproto_controller *c;
668 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
670 ofconn_destroy(ofconn);
672 update_controller(ofconn, c);
679 /* Delete services that are no longer configured.
680 * Update configuration of all now-existing services. */
681 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
683 struct ofproto_controller *c;
685 c = shash_find_data(&new_controllers,
686 pvconn_get_name(ofservice->pvconn));
688 ofservice_destroy(p, ofservice);
690 ofservice_reconfigure(ofservice, c);
694 shash_destroy(&new_controllers);
696 update_in_band_remotes(p);
699 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
700 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
701 struct ofconn, hmap_node);
702 ofconn->ss = switch_status_register(p->switch_status, "remote",
703 rconn_status_cb, ofconn->rconn);
708 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
710 p->fail_mode = fail_mode;
714 /* Drops the connections between 'ofproto' and all of its controllers, forcing
715 * them to reconnect. */
717 ofproto_reconnect_controllers(struct ofproto *ofproto)
719 struct ofconn *ofconn;
721 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
722 rconn_reconnect(ofconn->rconn);
727 any_extras_changed(const struct ofproto *ofproto,
728 const struct sockaddr_in *extras, size_t n)
732 if (n != ofproto->n_extra_remotes) {
736 for (i = 0; i < n; i++) {
737 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
738 const struct sockaddr_in *new = &extras[i];
740 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
741 old->sin_port != new->sin_port) {
749 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
750 * in-band control should guarantee access, in the same way that in-band
751 * control guarantees access to OpenFlow controllers. */
753 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
754 const struct sockaddr_in *extras, size_t n)
756 if (!any_extras_changed(ofproto, extras, n)) {
760 free(ofproto->extra_in_band_remotes);
761 ofproto->n_extra_remotes = n;
762 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
764 update_in_band_remotes(ofproto);
768 ofproto_set_desc(struct ofproto *p,
769 const char *mfr_desc, const char *hw_desc,
770 const char *sw_desc, const char *serial_desc,
773 struct ofp_desc_stats *ods;
776 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
777 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
778 sizeof ods->mfr_desc);
781 p->mfr_desc = xstrdup(mfr_desc);
784 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
785 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
786 sizeof ods->hw_desc);
789 p->hw_desc = xstrdup(hw_desc);
792 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
793 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
794 sizeof ods->sw_desc);
797 p->sw_desc = xstrdup(sw_desc);
800 if (strlen(serial_desc) >= sizeof ods->serial_num) {
801 VLOG_WARN("truncating serial_desc, must be less than %zu "
803 sizeof ods->serial_num);
805 free(p->serial_desc);
806 p->serial_desc = xstrdup(serial_desc);
809 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
810 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
811 sizeof ods->dp_desc);
814 p->dp_desc = xstrdup(dp_desc);
819 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
820 const struct svec *svec)
822 struct pvconn **pvconns = *pvconnsp;
823 size_t n_pvconns = *n_pvconnsp;
827 for (i = 0; i < n_pvconns; i++) {
828 pvconn_close(pvconns[i]);
832 pvconns = xmalloc(svec->n * sizeof *pvconns);
834 for (i = 0; i < svec->n; i++) {
835 const char *name = svec->names[i];
836 struct pvconn *pvconn;
839 error = pvconn_open(name, &pvconn);
841 pvconns[n_pvconns++] = pvconn;
843 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
851 *n_pvconnsp = n_pvconns;
857 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
859 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
863 ofproto_set_netflow(struct ofproto *ofproto,
864 const struct netflow_options *nf_options)
866 if (nf_options && nf_options->collectors.n) {
867 if (!ofproto->netflow) {
868 ofproto->netflow = netflow_create();
870 return netflow_set_options(ofproto->netflow, nf_options);
872 netflow_destroy(ofproto->netflow);
873 ofproto->netflow = NULL;
879 ofproto_set_sflow(struct ofproto *ofproto,
880 const struct ofproto_sflow_options *oso)
882 struct ofproto_sflow *os = ofproto->sflow;
885 struct ofport *ofport;
886 unsigned int odp_port;
888 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
889 refresh_port_groups(ofproto);
890 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
891 ofproto_sflow_add_port(os, odp_port,
892 netdev_get_name(ofport->netdev));
895 ofproto_sflow_set_options(os, oso);
897 ofproto_sflow_destroy(os);
898 ofproto->sflow = NULL;
903 ofproto_get_datapath_id(const struct ofproto *ofproto)
905 return ofproto->datapath_id;
909 ofproto_has_primary_controller(const struct ofproto *ofproto)
911 return !hmap_is_empty(&ofproto->controllers);
914 enum ofproto_fail_mode
915 ofproto_get_fail_mode(const struct ofproto *p)
921 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
925 for (i = 0; i < ofproto->n_snoops; i++) {
926 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
931 ofproto_destroy(struct ofproto *p)
933 struct ofservice *ofservice, *next_ofservice;
934 struct ofconn *ofconn, *next_ofconn;
935 struct ofport *ofport;
936 unsigned int port_no;
943 /* Destroy fail-open and in-band early, since they touch the classifier. */
944 fail_open_destroy(p->fail_open);
947 in_band_destroy(p->in_band);
949 free(p->extra_in_band_remotes);
951 ofproto_flush_flows(p);
952 classifier_destroy(&p->cls);
954 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
956 ofconn_destroy(ofconn);
958 hmap_destroy(&p->controllers);
961 netdev_monitor_destroy(p->netdev_monitor);
962 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
965 shash_destroy(&p->port_by_name);
967 switch_status_destroy(p->switch_status);
968 netflow_destroy(p->netflow);
969 ofproto_sflow_destroy(p->sflow);
971 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
973 ofservice_destroy(p, ofservice);
975 hmap_destroy(&p->services);
977 for (i = 0; i < p->n_snoops; i++) {
978 pvconn_close(p->snoops[i]);
982 mac_learning_destroy(p->ml);
987 free(p->serial_desc);
990 port_array_destroy(&p->ports);
996 ofproto_run(struct ofproto *p)
998 /* handle protocol messages coming from the datapath */
999 int error = ofproto_run1(p);
1001 error = ofproto_run2(p, false);
1007 process_port_change(struct ofproto *ofproto, int error, char *devname)
1009 if (error == ENOBUFS) {
1010 reinit_ports(ofproto);
1011 } else if (!error) {
1012 update_port(ofproto, devname);
1017 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1018 * means that 'ofconn' is more interesting for monitoring than a lower return
1021 snoop_preference(const struct ofconn *ofconn)
1023 switch (ofconn->role) {
1024 case NX_ROLE_MASTER:
1031 /* Shouldn't happen. */
1036 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1037 * Connects this vconn to a controller. */
1039 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1041 struct ofconn *ofconn, *best;
1043 /* Pick a controller for monitoring. */
1045 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1046 if (ofconn->type == OFCONN_PRIMARY
1047 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1053 rconn_add_monitor(best->rconn, vconn);
1055 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1061 * Calls the netdevice dpif_netdev_recv() callback,
1062 * that read a protocol packet from the dpif queue
1063 * and handle the message
1066 ofproto_run1(struct ofproto *p)
1068 struct ofconn *ofconn, *next_ofconn;
1069 struct ofservice *ofservice;
1074 if (shash_is_empty(&p->port_by_name)) {
1078 for (i = 0; i < 50; i++) {
1082 error = dpif_recv(p->dpif, &buf);
1084 if (error == ENODEV) {
1085 /* Someone destroyed the datapath behind our back. The caller
1086 * better destroy us and give up, because we're just going to
1087 * spin from here on out. */
1088 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1089 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1090 dpif_name(p->dpif));
1096 handle_odp_msg(p, buf);
1099 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1100 process_port_change(p, error, devname);
1102 while ((error = netdev_monitor_poll(p->netdev_monitor,
1103 &devname)) != EAGAIN) {
1104 process_port_change(p, error, devname);
1108 if (time_msec() >= p->next_in_band_update) {
1109 update_in_band_remotes(p);
1111 in_band_run(p->in_band);
1114 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1116 ofconn_run(ofconn, p);
1119 /* Fail-open maintenance. Do this after processing the ofconns since
1120 * fail-open checks the status of the controller rconn. */
1122 fail_open_run(p->fail_open);
1125 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1126 struct vconn *vconn;
1129 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1131 struct ofconn *ofconn;
1132 struct rconn *rconn;
1135 rconn = rconn_create(ofservice->probe_interval, 0);
1136 name = ofconn_make_name(p, vconn_get_name(vconn));
1137 rconn_connect_unreliably(rconn, vconn, name);
1140 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1141 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1142 ofservice->burst_limit);
1143 } else if (retval != EAGAIN) {
1144 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1148 for (i = 0; i < p->n_snoops; i++) {
1149 struct vconn *vconn;
1152 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1154 add_snooper(p, vconn);
1155 } else if (retval != EAGAIN) {
1156 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1160 if (time_msec() >= p->next_expiration) {
1161 COVERAGE_INC(ofproto_expiration);
1162 p->next_expiration = time_msec() + 1000;
1165 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1167 /* Let the hook know that we're at a stable point: all outstanding data
1168 * in existing flows has been accounted to the account_cb. Thus, the
1169 * hook can now reasonably do operations that depend on having accurate
1170 * flow volume accounting (currently, that's just bond rebalancing). */
1171 if (p->ofhooks->account_checkpoint_cb) {
1172 p->ofhooks->account_checkpoint_cb(p->aux);
1177 netflow_run(p->netflow);
1180 ofproto_sflow_run(p->sflow);
1186 struct revalidate_cbdata {
1187 struct ofproto *ofproto;
1188 bool revalidate_all; /* Revalidate all exact-match rules? */
1189 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1190 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1194 ofproto_run2(struct ofproto *p, bool revalidate_all)
1196 if (p->need_revalidate || revalidate_all
1197 || !tag_set_is_empty(&p->revalidate_set)) {
1198 struct revalidate_cbdata cbdata;
1200 cbdata.revalidate_all = revalidate_all;
1201 cbdata.revalidate_subrules = p->need_revalidate;
1202 cbdata.revalidate_set = p->revalidate_set;
1203 tag_set_init(&p->revalidate_set);
1204 COVERAGE_INC(ofproto_revalidate);
1205 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1206 p->need_revalidate = false;
1213 ofproto_wait(struct ofproto *p)
1215 struct ofservice *ofservice;
1216 struct ofconn *ofconn;
1219 dpif_recv_wait(p->dpif);
1220 dpif_port_poll_wait(p->dpif);
1221 netdev_monitor_poll_wait(p->netdev_monitor);
1222 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1223 ofconn_wait(ofconn);
1226 poll_timer_wait_until(p->next_in_band_update);
1227 in_band_wait(p->in_band);
1230 fail_open_wait(p->fail_open);
1233 ofproto_sflow_wait(p->sflow);
1235 if (!tag_set_is_empty(&p->revalidate_set)) {
1236 poll_immediate_wake();
1238 if (p->need_revalidate) {
1239 /* Shouldn't happen, but if it does just go around again. */
1240 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1241 poll_immediate_wake();
1242 } else if (p->next_expiration != LLONG_MAX) {
1243 poll_timer_wait_until(p->next_expiration);
1245 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1246 pvconn_wait(ofservice->pvconn);
1248 for (i = 0; i < p->n_snoops; i++) {
1249 pvconn_wait(p->snoops[i]);
1254 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1256 tag_set_add(&ofproto->revalidate_set, tag);
1260 ofproto_get_revalidate_set(struct ofproto *ofproto)
1262 return &ofproto->revalidate_set;
1266 ofproto_is_alive(const struct ofproto *p)
1268 return !hmap_is_empty(&p->controllers);
1272 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1273 const union ofp_action *actions, size_t n_actions,
1274 const struct ofpbuf *packet)
1276 struct odp_actions odp_actions;
1279 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1285 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1287 fprintf(stderr, "OFPROTO EXECUTE\n");
1288 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1289 odp_actions.n_actions, packet);
1294 ofproto_add_flow(struct ofproto *p,
1295 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1296 const union ofp_action *actions, size_t n_actions,
1300 rule = rule_create(p, NULL, actions, n_actions,
1301 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1303 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1304 rule_insert(p, rule, NULL, 0);
1308 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1309 uint32_t wildcards, unsigned int priority)
1313 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1317 rule_remove(ofproto, rule);
1322 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1324 struct rule *rule = rule_from_cls_rule(rule_);
1325 struct ofproto *ofproto = ofproto_;
1327 /* Mark the flow as not installed, even though it might really be
1328 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1329 * There is no point in uninstalling it individually since we are about to
1330 * blow away all the flows with dpif_flow_flush(). */
1331 rule->installed = false;
1333 rule_remove(ofproto, rule);
1337 ofproto_flush_flows(struct ofproto *ofproto)
1339 COVERAGE_INC(ofproto_flush);
1340 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1341 dpif_flow_flush(ofproto->dpif);
1342 if (ofproto->in_band) {
1343 in_band_flushed(ofproto->in_band);
1345 if (ofproto->fail_open) {
1346 fail_open_flushed(ofproto->fail_open);
1351 reinit_ports(struct ofproto *p)
1353 struct svec devnames;
1354 struct ofport *ofport;
1355 unsigned int port_no;
1356 struct odp_port *odp_ports;
1360 svec_init(&devnames);
1361 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1362 svec_add (&devnames, (char *) ofport->opp.name);
1364 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1365 for (i = 0; i < n_odp_ports; i++) {
1366 svec_add (&devnames, odp_ports[i].devname);
1370 svec_sort_unique(&devnames);
1371 for (i = 0; i < devnames.n; i++) {
1372 update_port(p, devnames.names[i]);
1374 svec_destroy(&devnames);
1378 refresh_port_group(struct ofproto *p, unsigned int group)
1382 struct ofport *port;
1383 unsigned int port_no;
1385 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1387 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1389 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1390 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1391 ports[n_ports++] = port_no;
1394 dpif_port_group_set(p->dpif, group, ports, n_ports);
1401 refresh_port_groups(struct ofproto *p)
1403 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1404 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1406 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1410 static struct ofport *
1411 make_ofport(const struct odp_port *odp_port)
1413 struct netdev_options netdev_options;
1414 enum netdev_flags flags;
1415 struct ofport *ofport;
1416 struct netdev *netdev;
1420 memset(&netdev_options, 0, sizeof netdev_options);
1421 netdev_options.name = odp_port->devname;
1422 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1424 error = netdev_open(&netdev_options, &netdev);
1426 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1427 "cannot be opened (%s)",
1428 odp_port->devname, odp_port->port,
1429 odp_port->devname, strerror(error));
1433 ofport = xmalloc(sizeof *ofport);
1434 ofport->netdev = netdev;
1435 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1436 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1437 memcpy(ofport->opp.name, odp_port->devname,
1438 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1439 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1441 netdev_get_flags(netdev, &flags);
1442 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1444 netdev_get_carrier(netdev, &carrier);
1445 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1447 netdev_get_features(netdev,
1448 &ofport->opp.curr, &ofport->opp.advertised,
1449 &ofport->opp.supported, &ofport->opp.peer);
1454 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1456 if (port_array_get(&p->ports, odp_port->port)) {
1457 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1460 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1461 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1470 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1472 const struct ofp_phy_port *a = &a_->opp;
1473 const struct ofp_phy_port *b = &b_->opp;
1475 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1476 return (a->port_no == b->port_no
1477 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1478 && !strcmp((char *) a->name, (char *) b->name)
1479 && a->state == b->state
1480 && a->config == b->config
1481 && a->curr == b->curr
1482 && a->advertised == b->advertised
1483 && a->supported == b->supported
1484 && a->peer == b->peer);
1488 send_port_status(struct ofproto *p, const struct ofport *ofport,
1491 /* XXX Should limit the number of queued port status change messages. */
1492 struct ofconn *ofconn;
1493 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1494 struct ofp_port_status *ops;
1497 if (!ofconn_receives_async_msgs(ofconn)) {
1501 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1502 ops->reason = reason;
1503 ops->desc = ofport->opp;
1504 hton_ofp_phy_port(&ops->desc);
1505 queue_tx(b, ofconn, NULL);
1507 if (p->ofhooks->port_changed_cb) {
1508 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1513 ofport_install(struct ofproto *p, struct ofport *ofport)
1515 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1516 const char *netdev_name = (const char *) ofport->opp.name;
1518 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1519 port_array_set(&p->ports, odp_port, ofport);
1520 shash_add(&p->port_by_name, netdev_name, ofport);
1522 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1527 ofport_remove(struct ofproto *p, struct ofport *ofport)
1529 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1531 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1532 port_array_delete(&p->ports, odp_port);
1533 shash_delete(&p->port_by_name,
1534 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1536 ofproto_sflow_del_port(p->sflow, odp_port);
1541 ofport_free(struct ofport *ofport)
1544 netdev_close(ofport->netdev);
1550 update_port(struct ofproto *p, const char *devname)
1552 struct odp_port odp_port;
1553 struct ofport *old_ofport;
1554 struct ofport *new_ofport;
1557 COVERAGE_INC(ofproto_update_port);
1559 /* Query the datapath for port information. */
1560 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1562 /* Find the old ofport. */
1563 old_ofport = shash_find_data(&p->port_by_name, devname);
1566 /* There's no port named 'devname' but there might be a port with
1567 * the same port number. This could happen if a port is deleted
1568 * and then a new one added in its place very quickly, or if a port
1569 * is renamed. In the former case we want to send an OFPPR_DELETE
1570 * and an OFPPR_ADD, and in the latter case we want to send a
1571 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1572 * the old port's ifindex against the new port, or perhaps less
1573 * reliably but more portably by comparing the old port's MAC
1574 * against the new port's MAC. However, this code isn't that smart
1575 * and always sends an OFPPR_MODIFY (XXX). */
1576 old_ofport = port_array_get(&p->ports, odp_port.port);
1578 } else if (error != ENOENT && error != ENODEV) {
1579 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1580 "%s", strerror(error));
1584 /* Create a new ofport. */
1585 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1587 /* Eliminate a few pathological cases. */
1588 if (!old_ofport && !new_ofport) {
1590 } else if (old_ofport && new_ofport) {
1591 /* Most of the 'config' bits are OpenFlow soft state, but
1592 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1593 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1594 * leaves the other bits 0.) */
1595 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1597 if (ofport_equal(old_ofport, new_ofport)) {
1598 /* False alarm--no change. */
1599 ofport_free(new_ofport);
1604 /* Now deal with the normal cases. */
1606 ofport_remove(p, old_ofport);
1609 ofport_install(p, new_ofport);
1611 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1612 (!old_ofport ? OFPPR_ADD
1613 : !new_ofport ? OFPPR_DELETE
1615 ofport_free(old_ofport);
1617 /* Update port groups. */
1618 refresh_port_groups(p);
1622 init_ports(struct ofproto *p)
1624 struct odp_port *ports;
1629 error = dpif_port_list(p->dpif, &ports, &n_ports);
1634 for (i = 0; i < n_ports; i++) {
1635 const struct odp_port *odp_port = &ports[i];
1636 if (!ofport_conflicts(p, odp_port)) {
1637 struct ofport *ofport = make_ofport(odp_port);
1639 ofport_install(p, ofport);
1644 refresh_port_groups(p);
1648 static struct ofconn *
1649 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1651 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1652 ofconn->ofproto = p;
1653 list_push_back(&p->all_conns, &ofconn->node);
1654 ofconn->rconn = rconn;
1655 ofconn->type = type;
1656 ofconn->role = NX_ROLE_OTHER;
1657 ofconn->packet_in_counter = rconn_packet_counter_create ();
1658 ofconn->pktbuf = NULL;
1659 ofconn->miss_send_len = 0;
1660 ofconn->reply_counter = rconn_packet_counter_create ();
1665 ofconn_destroy(struct ofconn *ofconn)
1667 if (ofconn->type == OFCONN_PRIMARY) {
1668 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1670 discovery_destroy(ofconn->discovery);
1672 list_remove(&ofconn->node);
1673 switch_status_unregister(ofconn->ss);
1674 rconn_destroy(ofconn->rconn);
1675 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1676 rconn_packet_counter_destroy(ofconn->reply_counter);
1677 pktbuf_destroy(ofconn->pktbuf);
1682 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1687 if (ofconn->discovery) {
1688 char *controller_name;
1689 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1690 discovery_question_connectivity(ofconn->discovery);
1692 if (discovery_run(ofconn->discovery, &controller_name)) {
1693 if (controller_name) {
1694 char *ofconn_name = ofconn_make_name(p, controller_name);
1695 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1698 rconn_disconnect(ofconn->rconn);
1703 for (i = 0; i < N_SCHEDULERS; i++) {
1704 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1707 rconn_run(ofconn->rconn);
1709 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1710 /* Limit the number of iterations to prevent other tasks from
1712 for (iteration = 0; iteration < 50; iteration++) {
1713 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1718 fail_open_maybe_recover(p->fail_open);
1720 handle_openflow(ofconn, p, of_msg);
1721 ofpbuf_delete(of_msg);
1725 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1726 ofconn_destroy(ofconn);
1731 ofconn_wait(struct ofconn *ofconn)
1735 if (ofconn->discovery) {
1736 discovery_wait(ofconn->discovery);
1738 for (i = 0; i < N_SCHEDULERS; i++) {
1739 pinsched_wait(ofconn->schedulers[i]);
1741 rconn_run_wait(ofconn->rconn);
1742 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1743 rconn_recv_wait(ofconn->rconn);
1745 COVERAGE_INC(ofproto_ofconn_stuck);
1749 /* Returns true if 'ofconn' should receive asynchronous messages. */
1751 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1753 if (ofconn->type == OFCONN_PRIMARY) {
1754 /* Primary controllers always get asynchronous messages unless they
1755 * have configured themselves as "slaves". */
1756 return ofconn->role != NX_ROLE_SLAVE;
1758 /* Service connections don't get asynchronous messages unless they have
1759 * explicitly asked for them by setting a nonzero miss send length. */
1760 return ofconn->miss_send_len > 0;
1764 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1765 * and 'target', suitable for use in log messages for identifying the
1768 * The name is dynamically allocated. The caller should free it (with free())
1769 * when it is no longer needed. */
1771 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1773 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1777 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1781 for (i = 0; i < N_SCHEDULERS; i++) {
1782 struct pinsched **s = &ofconn->schedulers[i];
1786 *s = pinsched_create(rate, burst,
1787 ofconn->ofproto->switch_status);
1789 pinsched_set_limits(*s, rate, burst);
1792 pinsched_destroy(*s);
1799 ofservice_reconfigure(struct ofservice *ofservice,
1800 const struct ofproto_controller *c)
1802 ofservice->probe_interval = c->probe_interval;
1803 ofservice->rate_limit = c->rate_limit;
1804 ofservice->burst_limit = c->burst_limit;
1807 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1808 * positive errno value. */
1810 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1812 struct ofservice *ofservice;
1813 struct pvconn *pvconn;
1816 error = pvconn_open(c->target, &pvconn);
1821 ofservice = xzalloc(sizeof *ofservice);
1822 hmap_insert(&ofproto->services, &ofservice->node,
1823 hash_string(c->target, 0));
1824 ofservice->pvconn = pvconn;
1826 ofservice_reconfigure(ofservice, c);
1832 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1834 hmap_remove(&ofproto->services, &ofservice->node);
1835 pvconn_close(ofservice->pvconn);
1839 /* Finds and returns the ofservice within 'ofproto' that has the given
1840 * 'target', or a null pointer if none exists. */
1841 static struct ofservice *
1842 ofservice_lookup(struct ofproto *ofproto, const char *target)
1844 struct ofservice *ofservice;
1846 HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node,
1847 hash_string(target, 0), &ofproto->services) {
1848 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1855 /* Caller is responsible for initializing the 'cr' member of the returned
1857 static struct rule *
1858 rule_create(struct ofproto *ofproto, struct rule *super,
1859 const union ofp_action *actions, size_t n_actions,
1860 uint16_t idle_timeout, uint16_t hard_timeout,
1861 uint64_t flow_cookie, bool send_flow_removed)
1863 struct rule *rule = xzalloc(sizeof *rule);
1864 rule->idle_timeout = idle_timeout;
1865 rule->hard_timeout = hard_timeout;
1866 rule->flow_cookie = flow_cookie;
1867 rule->used = rule->created = time_msec();
1868 rule->send_flow_removed = send_flow_removed;
1869 rule->super = super;
1871 list_push_back(&super->list, &rule->list);
1873 list_init(&rule->list);
1875 rule->n_actions = n_actions;
1876 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1877 netflow_flow_clear(&rule->nf_flow);
1878 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1883 static struct rule *
1884 rule_from_cls_rule(const struct cls_rule *cls_rule)
1886 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1890 rule_free(struct rule *rule)
1892 free(rule->actions);
1893 free(rule->odp_actions);
1897 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1898 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1899 * through all of its subrules and revalidates them, destroying any that no
1900 * longer has a super-rule (which is probably all of them).
1902 * Before calling this function, the caller must make have removed 'rule' from
1903 * the classifier. If 'rule' is an exact-match rule, the caller is also
1904 * responsible for ensuring that it has been uninstalled from the datapath. */
1906 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1909 struct rule *subrule, *next;
1910 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1911 revalidate_rule(ofproto, subrule);
1914 list_remove(&rule->list);
1920 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1922 const union ofp_action *oa;
1923 struct actions_iterator i;
1925 if (out_port == htons(OFPP_NONE)) {
1928 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1929 oa = actions_next(&i)) {
1930 if (action_outputs_to_port(oa, out_port)) {
1937 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1938 * 'packet', which arrived on 'in_port'.
1940 * Takes ownership of 'packet'. */
1942 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1943 const union odp_action *actions, size_t n_actions,
1944 struct ofpbuf *packet)
1946 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1947 /* As an optimization, avoid a round-trip from userspace to kernel to
1948 * userspace. This also avoids possibly filling up kernel packet
1949 * buffers along the way. */
1950 struct odp_msg *msg;
1952 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1953 msg->type = _ODPL_ACTION_NR;
1954 msg->length = sizeof(struct odp_msg) + packet->size;
1955 msg->port = in_port;
1957 msg->arg = actions[0].controller.arg;
1959 send_packet_in(ofproto, packet);
1965 error = dpif_execute(ofproto->dpif, in_port,
1966 actions, n_actions, packet);
1967 ofpbuf_delete(packet);
1972 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1973 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1974 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1976 * The flow that 'packet' actually contains does not need to actually match
1977 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1978 * the packet and byte counters for 'rule' will be credited for the packet sent
1979 * out whether or not the packet actually matches 'rule'.
1981 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1982 * the caller must already have accurately composed ODP actions for it given
1983 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1984 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1985 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1986 * actions and apply them to 'packet'.
1988 * Takes ownership of 'packet'. */
1990 rule_execute(struct ofproto *ofproto, struct rule *rule,
1991 struct ofpbuf *packet, const flow_t *flow)
1993 const union odp_action *actions;
1994 struct odp_flow_stats stats;
1996 struct odp_actions a;
1998 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2000 /* Grab or compose the ODP actions.
2002 * The special case for an exact-match 'rule' where 'flow' is not the
2003 * rule's flow is important to avoid, e.g., sending a packet out its input
2004 * port simply because the ODP actions were composed for the wrong
2006 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
2007 struct rule *super = rule->super ? rule->super : rule;
2008 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
2009 packet, &a, NULL, 0, NULL)) {
2010 ofpbuf_delete(packet);
2013 actions = a.actions;
2014 n_actions = a.n_actions;
2016 actions = rule->odp_actions;
2017 n_actions = rule->n_odp_actions;
2020 /* Execute the ODP actions. */
2021 flow_extract_stats(flow, packet, &stats);
2022 if (execute_odp_actions(ofproto, flow->in_port,
2023 actions, n_actions, packet)) {
2024 update_stats(ofproto, rule, &stats);
2025 rule->used = time_msec();
2026 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2030 /* Inserts 'rule' into 'p''s flow table.
2032 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2033 * actions on it and credits the statistics for sending the packet to 'rule'.
2034 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2037 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2040 struct rule *displaced_rule;
2042 /* Insert the rule in the classifier. */
2043 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2044 if (!rule->cr.wc.wildcards) {
2045 rule_make_actions(p, rule, packet);
2048 /* Send the packet and credit it to the rule. */
2051 flow_extract(packet, 0, in_port, &flow);
2052 rule_execute(p, rule, packet, &flow);
2055 /* Install the rule in the datapath only after sending the packet, to
2056 * avoid packet reordering. */
2057 if (rule->cr.wc.wildcards) {
2058 COVERAGE_INC(ofproto_add_wc_flow);
2059 p->need_revalidate = true;
2061 rule_install(p, rule, displaced_rule);
2064 /* Free the rule that was displaced, if any. */
2065 if (displaced_rule) {
2066 rule_destroy(p, displaced_rule);
2070 static struct rule *
2071 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2074 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2075 rule->idle_timeout, rule->hard_timeout,
2077 COVERAGE_INC(ofproto_subrule_create);
2078 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2079 : rule->cr.priority), &subrule->cr);
2080 classifier_insert_exact(&ofproto->cls, &subrule->cr);
2086 rule_remove(struct ofproto *ofproto, struct rule *rule)
2088 if (rule->cr.wc.wildcards) {
2089 COVERAGE_INC(ofproto_del_wc_flow);
2090 ofproto->need_revalidate = true;
2092 rule_uninstall(ofproto, rule);
2094 classifier_remove(&ofproto->cls, &rule->cr);
2095 rule_destroy(ofproto, rule);
2098 /* Returns true if the actions changed, false otherwise. */
2100 rule_make_actions(struct ofproto *p, struct rule *rule,
2101 const struct ofpbuf *packet)
2103 const struct rule *super;
2104 struct odp_actions a;
2107 assert(!rule->cr.wc.wildcards);
2109 super = rule->super ? rule->super : rule;
2111 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2112 packet, &a, &rule->tags, &rule->may_install,
2113 &rule->nf_flow.output_iface);
2115 actions_len = a.n_actions * sizeof *a.actions;
2116 if (rule->n_odp_actions != a.n_actions
2117 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2118 COVERAGE_INC(ofproto_odp_unchanged);
2119 free(rule->odp_actions);
2120 rule->n_odp_actions = a.n_actions;
2121 rule->odp_actions = xmemdup(a.actions, actions_len);
2129 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2130 struct odp_flow_put *put)
2132 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2133 put->flow.key = rule->cr.flow;
2134 put->flow.actions = rule->odp_actions;
2135 put->flow.n_actions = rule->n_odp_actions;
2136 put->flow.flags = 0;
2138 return dpif_flow_put(ofproto->dpif, put);
2142 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2144 assert(!rule->cr.wc.wildcards);
2146 if (rule->may_install) {
2147 struct odp_flow_put put;
2148 if (!do_put_flow(p, rule,
2149 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2151 rule->installed = true;
2152 if (displaced_rule) {
2153 update_stats(p, displaced_rule, &put.flow.stats);
2154 rule_post_uninstall(p, displaced_rule);
2157 } else if (displaced_rule) {
2158 rule_uninstall(p, displaced_rule);
2163 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2165 if (rule->installed) {
2166 struct odp_flow_put put;
2167 COVERAGE_INC(ofproto_dp_missed);
2168 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2170 rule_install(ofproto, rule, NULL);
2175 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2177 bool actions_changed;
2178 uint16_t new_out_iface, old_out_iface;
2180 old_out_iface = rule->nf_flow.output_iface;
2181 actions_changed = rule_make_actions(ofproto, rule, NULL);
2183 if (rule->may_install) {
2184 if (rule->installed) {
2185 if (actions_changed) {
2186 struct odp_flow_put put;
2187 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2188 | ODPPF_ZERO_STATS, &put);
2189 update_stats(ofproto, rule, &put.flow.stats);
2191 /* Temporarily set the old output iface so that NetFlow
2192 * messages have the correct output interface for the old
2194 new_out_iface = rule->nf_flow.output_iface;
2195 rule->nf_flow.output_iface = old_out_iface;
2196 rule_post_uninstall(ofproto, rule);
2197 rule->nf_flow.output_iface = new_out_iface;
2200 rule_install(ofproto, rule, NULL);
2203 rule_uninstall(ofproto, rule);
2208 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2210 uint64_t total_bytes = rule->byte_count + extra_bytes;
2212 if (ofproto->ofhooks->account_flow_cb
2213 && total_bytes > rule->accounted_bytes)
2215 ofproto->ofhooks->account_flow_cb(
2216 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2217 total_bytes - rule->accounted_bytes, ofproto->aux);
2218 rule->accounted_bytes = total_bytes;
2223 rule_uninstall(struct ofproto *p, struct rule *rule)
2225 assert(!rule->cr.wc.wildcards);
2226 if (rule->installed) {
2227 struct odp_flow odp_flow;
2229 odp_flow.key = rule->cr.flow;
2230 odp_flow.actions = NULL;
2231 odp_flow.n_actions = 0;
2233 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2234 update_stats(p, rule, &odp_flow.stats);
2236 rule->installed = false;
2238 rule_post_uninstall(p, rule);
2243 is_controller_rule(struct rule *rule)
2245 /* If the only action is send to the controller then don't report
2246 * NetFlow expiration messages since it is just part of the control
2247 * logic for the network and not real traffic. */
2251 && rule->super->n_actions == 1
2252 && action_outputs_to_port(&rule->super->actions[0],
2253 htons(OFPP_CONTROLLER)));
2257 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2259 struct rule *super = rule->super;
2261 rule_account(ofproto, rule, 0);
2263 if (ofproto->netflow && !is_controller_rule(rule)) {
2264 struct ofexpired expired;
2265 expired.flow = rule->cr.flow;
2266 expired.packet_count = rule->packet_count;
2267 expired.byte_count = rule->byte_count;
2268 expired.used = rule->used;
2269 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2272 super->packet_count += rule->packet_count;
2273 super->byte_count += rule->byte_count;
2275 /* Reset counters to prevent double counting if the rule ever gets
2277 rule->packet_count = 0;
2278 rule->byte_count = 0;
2279 rule->accounted_bytes = 0;
2281 netflow_flow_clear(&rule->nf_flow);
2286 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2287 struct rconn_packet_counter *counter)
2289 update_openflow_length(msg);
2290 if (rconn_send(ofconn->rconn, msg, counter)) {
2296 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2297 int error, const void *data, size_t len)
2300 struct ofp_error_msg *oem;
2302 if (!(error >> 16)) {
2303 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2308 COVERAGE_INC(ofproto_error);
2309 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2310 oh ? oh->xid : 0, &buf);
2311 oem->type = htons((unsigned int) error >> 16);
2312 oem->code = htons(error & 0xffff);
2313 memcpy(oem->data, data, len);
2314 queue_tx(buf, ofconn, ofconn->reply_counter);
2318 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2321 size_t oh_length = ntohs(oh->length);
2322 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2326 hton_ofp_phy_port(struct ofp_phy_port *opp)
2328 opp->port_no = htons(opp->port_no);
2329 opp->config = htonl(opp->config);
2330 opp->state = htonl(opp->state);
2331 opp->curr = htonl(opp->curr);
2332 opp->advertised = htonl(opp->advertised);
2333 opp->supported = htonl(opp->supported);
2334 opp->peer = htonl(opp->peer);
2338 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2340 struct ofp_header *rq = oh;
2341 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2346 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2347 struct ofp_header *oh)
2349 struct ofp_switch_features *osf;
2351 unsigned int port_no;
2352 struct ofport *port;
2354 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2355 osf->datapath_id = htonll(p->datapath_id);
2356 osf->n_buffers = htonl(pktbuf_capacity());
2358 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2359 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2360 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2361 (1u << OFPAT_SET_VLAN_VID) |
2362 (1u << OFPAT_SET_VLAN_PCP) |
2363 (1u << OFPAT_STRIP_VLAN) |
2364 (1u << OFPAT_SET_DL_SRC) |
2365 (1u << OFPAT_SET_DL_DST) |
2366 (1u << OFPAT_SET_NW_SRC) |
2367 (1u << OFPAT_SET_NW_DST) |
2368 (1u << OFPAT_SET_NW_TOS) |
2369 (1u << OFPAT_SET_TP_SRC) |
2370 (1u << OFPAT_SET_TP_DST) |
2371 (1u << OFPAT_ENQUEUE));
2373 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2374 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2377 queue_tx(buf, ofconn, ofconn->reply_counter);
2382 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2383 struct ofp_header *oh)
2386 struct ofp_switch_config *osc;
2390 /* Figure out flags. */
2391 dpif_get_drop_frags(p->dpif, &drop_frags);
2392 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2395 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2396 osc->flags = htons(flags);
2397 osc->miss_send_len = htons(ofconn->miss_send_len);
2398 queue_tx(buf, ofconn, ofconn->reply_counter);
2404 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2405 struct ofp_switch_config *osc)
2410 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2414 flags = ntohs(osc->flags);
2416 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2417 switch (flags & OFPC_FRAG_MASK) {
2418 case OFPC_FRAG_NORMAL:
2419 dpif_set_drop_frags(p->dpif, false);
2421 case OFPC_FRAG_DROP:
2422 dpif_set_drop_frags(p->dpif, true);
2425 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2431 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2437 add_output_group_action(struct odp_actions *actions, uint16_t group,
2438 uint16_t *nf_output_iface)
2440 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2442 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2443 *nf_output_iface = NF_OUT_FLOOD;
2448 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2450 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2451 a->controller.arg = max_len;
2454 struct action_xlate_ctx {
2456 flow_t flow; /* Flow to which these actions correspond. */
2457 int recurse; /* Recursion level, via xlate_table_action. */
2458 struct ofproto *ofproto;
2459 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2460 * null pointer if we are revalidating
2461 * without a packet to refer to. */
2464 struct odp_actions *out; /* Datapath actions. */
2465 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2466 bool may_set_up_flow; /* True ordinarily; false if the actions must
2467 * be reassessed for every packet. */
2468 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2471 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2472 * flow translation. */
2473 #define MAX_RESUBMIT_RECURSION 8
2475 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2476 struct action_xlate_ctx *ctx);
2479 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2481 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2484 if (ofport->opp.config & OFPPC_NO_FWD) {
2485 /* Forwarding disabled on port. */
2490 * We don't have an ofport record for this port, but it doesn't hurt to
2491 * allow forwarding to it anyhow. Maybe such a port will appear later
2492 * and we're pre-populating the flow table.
2496 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2497 ctx->nf_output_iface = port;
2500 static struct rule *
2501 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2504 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2506 /* The rule we found might not be valid, since we could be in need of
2507 * revalidation. If it is not valid, don't return it. */
2510 && ofproto->need_revalidate
2511 && !revalidate_rule(ofproto, rule)) {
2512 COVERAGE_INC(ofproto_invalidated);
2520 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2522 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2523 uint16_t old_in_port;
2526 /* Look up a flow with 'in_port' as the input port. Then restore the
2527 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2528 * have surprising behavior). */
2529 old_in_port = ctx->flow.in_port;
2530 ctx->flow.in_port = in_port;
2531 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2532 ctx->flow.in_port = old_in_port;
2540 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2544 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2546 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2547 MAX_RESUBMIT_RECURSION);
2552 xlate_output_action__(struct action_xlate_ctx *ctx,
2553 uint16_t port, uint16_t max_len)
2556 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2558 ctx->nf_output_iface = NF_OUT_DROP;
2562 add_output_action(ctx, ctx->flow.in_port);
2565 xlate_table_action(ctx, ctx->flow.in_port);
2568 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2569 ctx->out, ctx->tags,
2570 &ctx->nf_output_iface,
2571 ctx->ofproto->aux)) {
2572 COVERAGE_INC(ofproto_uninstallable);
2573 ctx->may_set_up_flow = false;
2577 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2578 &ctx->nf_output_iface);
2581 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2583 case OFPP_CONTROLLER:
2584 add_controller_action(ctx->out, max_len);
2587 add_output_action(ctx, ODPP_LOCAL);
2590 odp_port = ofp_port_to_odp_port(port);
2591 if (odp_port != ctx->flow.in_port) {
2592 add_output_action(ctx, odp_port);
2597 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2598 ctx->nf_output_iface = NF_OUT_FLOOD;
2599 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2600 ctx->nf_output_iface = prev_nf_output_iface;
2601 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2602 ctx->nf_output_iface != NF_OUT_FLOOD) {
2603 ctx->nf_output_iface = NF_OUT_MULTI;
2608 xlate_output_action(struct action_xlate_ctx *ctx,
2609 const struct ofp_action_output *oao)
2611 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2614 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2615 * optimization, because we're going to add another action that sets the
2616 * priority immediately after, or because there are no actions following the
2619 remove_pop_action(struct action_xlate_ctx *ctx)
2621 size_t n = ctx->out->n_actions;
2622 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2623 ctx->out->n_actions--;
2628 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2629 const struct ofp_action_enqueue *oae)
2631 uint16_t ofp_port, odp_port;
2635 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2638 /* Fall back to ordinary output action. */
2639 xlate_output_action__(ctx, ntohs(oae->port), 0);
2643 /* Figure out ODP output port. */
2644 ofp_port = ntohs(oae->port);
2645 if (ofp_port != OFPP_IN_PORT) {
2646 odp_port = ofp_port_to_odp_port(ofp_port);
2648 odp_port = ctx->flow.in_port;
2651 /* Add ODP actions. */
2652 remove_pop_action(ctx);
2653 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2655 add_output_action(ctx, odp_port);
2656 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2658 /* Update NetFlow output port. */
2659 if (ctx->nf_output_iface == NF_OUT_DROP) {
2660 ctx->nf_output_iface = odp_port;
2661 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2662 ctx->nf_output_iface = NF_OUT_MULTI;
2667 xlate_nicira_action(struct action_xlate_ctx *ctx,
2668 const struct nx_action_header *nah)
2670 const struct nx_action_resubmit *nar;
2671 const struct nx_action_set_tunnel *nast;
2672 union odp_action *oa;
2673 int subtype = ntohs(nah->subtype);
2675 assert(nah->vendor == htonl(NX_VENDOR_ID));
2677 case NXAST_RESUBMIT:
2678 nar = (const struct nx_action_resubmit *) nah;
2679 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2682 case NXAST_SET_TUNNEL:
2683 nast = (const struct nx_action_set_tunnel *) nah;
2684 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2685 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2688 case NXAST_DROP_SPOOFED_ARP:
2689 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2690 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2694 /* If you add a new action here that modifies flow data, don't forget to
2695 * update the flow key in ctx->flow at the same time. */
2698 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2704 do_xlate_actions(const union ofp_action *in, size_t n_in,
2705 struct action_xlate_ctx *ctx)
2707 struct actions_iterator iter;
2708 const union ofp_action *ia;
2709 const struct ofport *port;
2711 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2712 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2713 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2714 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2715 /* Drop this flow. */
2719 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2720 uint16_t type = ntohs(ia->type);
2721 union odp_action *oa;
2725 xlate_output_action(ctx, &ia->output);
2728 case OFPAT_SET_VLAN_VID:
2729 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2730 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2733 case OFPAT_SET_VLAN_PCP:
2734 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2735 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2738 case OFPAT_STRIP_VLAN:
2739 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2740 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2741 ctx->flow.dl_vlan_pcp = 0;
2744 case OFPAT_SET_DL_SRC:
2745 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2746 memcpy(oa->dl_addr.dl_addr,
2747 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2748 memcpy(ctx->flow.dl_src,
2749 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2752 case OFPAT_SET_DL_DST:
2753 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2754 memcpy(oa->dl_addr.dl_addr,
2755 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2756 memcpy(ctx->flow.dl_dst,
2757 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2760 case OFPAT_SET_NW_SRC:
2761 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2762 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2765 case OFPAT_SET_NW_DST:
2766 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2767 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2770 case OFPAT_SET_NW_TOS:
2771 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2772 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2775 case OFPAT_SET_TP_SRC:
2776 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2777 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2780 case OFPAT_SET_TP_DST:
2781 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2782 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2786 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2790 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2794 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2801 xlate_actions(const union ofp_action *in, size_t n_in,
2802 const flow_t *flow, struct ofproto *ofproto,
2803 const struct ofpbuf *packet,
2804 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2805 uint16_t *nf_output_iface)
2807 tag_type no_tags = 0;
2808 struct action_xlate_ctx ctx;
2809 COVERAGE_INC(ofproto_ofp2odp);
2810 odp_actions_init(out);
2813 ctx.ofproto = ofproto;
2814 ctx.packet = packet;
2816 ctx.tags = tags ? tags : &no_tags;
2817 ctx.may_set_up_flow = true;
2818 ctx.nf_output_iface = NF_OUT_DROP;
2819 do_xlate_actions(in, n_in, &ctx);
2820 remove_pop_action(&ctx);
2822 /* Check with in-band control to see if we're allowed to set up this
2824 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2825 ctx.may_set_up_flow = false;
2828 if (may_set_up_flow) {
2829 *may_set_up_flow = ctx.may_set_up_flow;
2831 if (nf_output_iface) {
2832 *nf_output_iface = ctx.nf_output_iface;
2834 if (odp_actions_overflow(out)) {
2835 COVERAGE_INC(odp_overflow);
2836 odp_actions_init(out);
2837 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2842 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2843 * error message code (composed with ofp_mkerr()) for the caller to propagate
2844 * upward. Otherwise, returns 0.
2846 * 'oh' is used to make log messages more informative. */
2848 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2850 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2851 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2854 type_name = ofp_message_type_to_string(oh->type);
2855 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2859 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2866 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2867 struct ofp_header *oh)
2869 struct ofp_packet_out *opo;
2870 struct ofpbuf payload, *buffer;
2871 struct odp_actions actions;
2877 error = reject_slave_controller(ofconn, oh);
2882 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2886 opo = (struct ofp_packet_out *) oh;
2888 COVERAGE_INC(ofproto_packet_out);
2889 if (opo->buffer_id != htonl(UINT32_MAX)) {
2890 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2892 if (error || !buffer) {
2900 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2901 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2902 &flow, p, &payload, &actions, NULL, NULL, NULL);
2907 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2909 ofpbuf_delete(buffer);
2915 update_port_config(struct ofproto *p, struct ofport *port,
2916 uint32_t config, uint32_t mask)
2918 mask &= config ^ port->opp.config;
2919 if (mask & OFPPC_PORT_DOWN) {
2920 if (config & OFPPC_PORT_DOWN) {
2921 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2923 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2926 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2927 if (mask & REVALIDATE_BITS) {
2928 COVERAGE_INC(ofproto_costly_flags);
2929 port->opp.config ^= mask & REVALIDATE_BITS;
2930 p->need_revalidate = true;
2932 #undef REVALIDATE_BITS
2933 if (mask & OFPPC_NO_FLOOD) {
2934 port->opp.config ^= OFPPC_NO_FLOOD;
2935 refresh_port_groups(p);
2937 if (mask & OFPPC_NO_PACKET_IN) {
2938 port->opp.config ^= OFPPC_NO_PACKET_IN;
2943 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2944 struct ofp_header *oh)
2946 const struct ofp_port_mod *opm;
2947 struct ofport *port;
2950 error = reject_slave_controller(ofconn, oh);
2954 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2958 opm = (struct ofp_port_mod *) oh;
2960 port = port_array_get(&p->ports,
2961 ofp_port_to_odp_port(ntohs(opm->port_no)));
2963 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2964 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2965 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2967 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2968 if (opm->advertise) {
2969 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2975 static struct ofpbuf *
2976 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2978 struct ofp_stats_reply *osr;
2981 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2982 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2984 osr->flags = htons(0);
2988 static struct ofpbuf *
2989 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2991 return make_stats_reply(request->header.xid, request->type, body_len);
2995 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2997 struct ofpbuf *msg = *msgp;
2998 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2999 if (nbytes + msg->size > UINT16_MAX) {
3000 struct ofp_stats_reply *reply = msg->data;
3001 reply->flags = htons(OFPSF_REPLY_MORE);
3002 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
3003 queue_tx(msg, ofconn, ofconn->reply_counter);
3005 return ofpbuf_put_uninit(*msgp, nbytes);
3009 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3010 struct ofp_stats_request *request)
3012 struct ofp_desc_stats *ods;
3015 msg = start_stats_reply(request, sizeof *ods);
3016 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3017 memset(ods, 0, sizeof *ods);
3018 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3019 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3020 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3021 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3022 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3023 queue_tx(msg, ofconn, ofconn->reply_counter);
3029 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
3031 struct rule *rule = rule_from_cls_rule(cls_rule);
3032 int *n_subrules = n_subrules_;
3040 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3041 struct ofp_stats_request *request)
3043 struct ofp_table_stats *ots;
3045 struct odp_stats dpstats;
3046 int n_exact, n_subrules, n_wild;
3048 msg = start_stats_reply(request, sizeof *ots * 2);
3050 /* Count rules of various kinds. */
3052 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
3053 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3054 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3057 dpif_get_dp_stats(p->dpif, &dpstats);
3058 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3059 memset(ots, 0, sizeof *ots);
3060 ots->table_id = TABLEID_HASH;
3061 strcpy(ots->name, "hash");
3062 ots->wildcards = htonl(0);
3063 ots->max_entries = htonl(dpstats.max_capacity);
3064 ots->active_count = htonl(n_exact);
3065 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3067 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3069 /* Classifier table. */
3070 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3071 memset(ots, 0, sizeof *ots);
3072 ots->table_id = TABLEID_CLASSIFIER;
3073 strcpy(ots->name, "classifier");
3074 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3076 ots->max_entries = htonl(65536);
3077 ots->active_count = htonl(n_wild);
3078 ots->lookup_count = htonll(0); /* XXX */
3079 ots->matched_count = htonll(0); /* XXX */
3081 queue_tx(msg, ofconn, ofconn->reply_counter);
3086 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
3087 struct ofpbuf **msgp)
3089 struct netdev_stats stats;
3090 struct ofp_port_stats *ops;
3092 /* Intentionally ignore return value, since errors will set
3093 * 'stats' to all-1s, which is correct for OpenFlow, and
3094 * netdev_get_stats() will log errors. */
3095 netdev_get_stats(port->netdev, &stats);
3097 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3098 ops->port_no = htons(odp_port_to_ofp_port(port_no));
3099 memset(ops->pad, 0, sizeof ops->pad);
3100 ops->rx_packets = htonll(stats.rx_packets);
3101 ops->tx_packets = htonll(stats.tx_packets);
3102 ops->rx_bytes = htonll(stats.rx_bytes);
3103 ops->tx_bytes = htonll(stats.tx_bytes);
3104 ops->rx_dropped = htonll(stats.rx_dropped);
3105 ops->tx_dropped = htonll(stats.tx_dropped);
3106 ops->rx_errors = htonll(stats.rx_errors);
3107 ops->tx_errors = htonll(stats.tx_errors);
3108 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3109 ops->rx_over_err = htonll(stats.rx_over_errors);
3110 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3111 ops->collisions = htonll(stats.collisions);
3115 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3116 struct ofp_stats_request *osr,
3119 struct ofp_port_stats_request *psr;
3120 struct ofp_port_stats *ops;
3122 struct ofport *port;
3123 unsigned int port_no;
3125 if (arg_size != sizeof *psr) {
3126 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3128 psr = (struct ofp_port_stats_request *) osr->body;
3130 msg = start_stats_reply(osr, sizeof *ops * 16);
3131 if (psr->port_no != htons(OFPP_NONE)) {
3132 port = port_array_get(&p->ports,
3133 ofp_port_to_odp_port(ntohs(psr->port_no)));
3135 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
3138 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
3139 append_port_stat(port, port_no, ofconn, &msg);
3143 queue_tx(msg, ofconn, ofconn->reply_counter);
3147 struct flow_stats_cbdata {
3148 struct ofproto *ofproto;
3149 struct ofconn *ofconn;
3154 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3155 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3156 * returned statistic include statistics for all of 'rule''s subrules. */
3158 query_stats(struct ofproto *p, struct rule *rule,
3159 uint64_t *packet_countp, uint64_t *byte_countp)
3161 uint64_t packet_count, byte_count;
3162 struct rule *subrule;
3163 struct odp_flow *odp_flows;
3166 /* Start from historical data for 'rule' itself that are no longer tracked
3167 * by the datapath. This counts, for example, subrules that have
3169 packet_count = rule->packet_count;
3170 byte_count = rule->byte_count;
3172 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3173 * wildcarded then on all of its subrules.
3175 * Also, add any statistics that are not tracked by the datapath for each
3176 * subrule. This includes, for example, statistics for packets that were
3177 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3179 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3180 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3181 if (rule->cr.wc.wildcards) {
3183 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
3184 odp_flows[i++].key = subrule->cr.flow;
3185 packet_count += subrule->packet_count;
3186 byte_count += subrule->byte_count;
3189 odp_flows[0].key = rule->cr.flow;
3192 /* Fetch up-to-date statistics from the datapath and add them in. */
3193 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3195 for (i = 0; i < n_odp_flows; i++) {
3196 struct odp_flow *odp_flow = &odp_flows[i];
3197 packet_count += odp_flow->stats.n_packets;
3198 byte_count += odp_flow->stats.n_bytes;
3203 /* Return the stats to the caller. */
3204 *packet_countp = packet_count;
3205 *byte_countp = byte_count;
3209 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3211 struct rule *rule = rule_from_cls_rule(rule_);
3212 struct flow_stats_cbdata *cbdata = cbdata_;
3213 struct ofp_flow_stats *ofs;
3214 uint64_t packet_count, byte_count;
3215 size_t act_len, len;
3216 long long int tdiff = time_msec() - rule->created;
3217 uint32_t sec = tdiff / 1000;
3218 uint32_t msec = tdiff - (sec * 1000);
3220 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3224 act_len = sizeof *rule->actions * rule->n_actions;
3225 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3227 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3229 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3230 ofs->length = htons(len);
3231 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3233 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3234 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3235 ofs->duration_sec = htonl(sec);
3236 ofs->duration_nsec = htonl(msec * 1000000);
3237 ofs->cookie = rule->flow_cookie;
3238 ofs->priority = htons(rule->cr.priority);
3239 ofs->idle_timeout = htons(rule->idle_timeout);
3240 ofs->hard_timeout = htons(rule->hard_timeout);
3241 memset(ofs->pad2, 0, sizeof ofs->pad2);
3242 ofs->packet_count = htonll(packet_count);
3243 ofs->byte_count = htonll(byte_count);
3244 memcpy(ofs->actions, rule->actions, act_len);
3248 table_id_to_include(uint8_t table_id)
3250 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3251 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3252 : table_id == 0xff ? CLS_INC_ALL
3257 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3258 const struct ofp_stats_request *osr,
3261 struct ofp_flow_stats_request *fsr;
3262 struct flow_stats_cbdata cbdata;
3263 struct cls_rule target;
3265 if (arg_size != sizeof *fsr) {
3266 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3268 fsr = (struct ofp_flow_stats_request *) osr->body;
3270 COVERAGE_INC(ofproto_flows_req);
3272 cbdata.ofconn = ofconn;
3273 cbdata.out_port = fsr->out_port;
3274 cbdata.msg = start_stats_reply(osr, 1024);
3275 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3276 classifier_for_each_match(&p->cls, &target,
3277 table_id_to_include(fsr->table_id),
3278 flow_stats_cb, &cbdata);
3279 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3283 struct flow_stats_ds_cbdata {
3284 struct ofproto *ofproto;
3289 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3291 struct rule *rule = rule_from_cls_rule(rule_);
3292 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3293 struct ds *results = cbdata->results;
3294 struct ofp_match match;
3295 uint64_t packet_count, byte_count;
3296 size_t act_len = sizeof *rule->actions * rule->n_actions;
3298 /* Don't report on subrules. */
3299 if (rule->super != NULL) {
3303 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3304 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3305 cbdata->ofproto->tun_id_from_cookie, &match);
3307 ds_put_format(results, "duration=%llds, ",
3308 (time_msec() - rule->created) / 1000);
3309 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3310 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3311 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3312 ofp_print_match(results, &match, true);
3313 ofp_print_actions(results, &rule->actions->header, act_len);
3314 ds_put_cstr(results, "\n");
3317 /* Adds a pretty-printed description of all flows to 'results', including
3318 * those marked hidden by secchan (e.g., by in-band control). */
3320 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3322 struct ofp_match match;
3323 struct cls_rule target;
3324 struct flow_stats_ds_cbdata cbdata;
3326 memset(&match, 0, sizeof match);
3327 match.wildcards = htonl(OVSFW_ALL);
3330 cbdata.results = results;
3332 cls_rule_from_match(&match, 0, false, 0, &target);
3333 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3334 flow_stats_ds_cb, &cbdata);
3337 struct aggregate_stats_cbdata {
3338 struct ofproto *ofproto;
3340 uint64_t packet_count;
3341 uint64_t byte_count;
3346 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3348 struct rule *rule = rule_from_cls_rule(rule_);
3349 struct aggregate_stats_cbdata *cbdata = cbdata_;
3350 uint64_t packet_count, byte_count;
3352 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3356 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3358 cbdata->packet_count += packet_count;
3359 cbdata->byte_count += byte_count;
3364 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3365 const struct ofp_stats_request *osr,
3368 struct ofp_aggregate_stats_request *asr;
3369 struct ofp_aggregate_stats_reply *reply;
3370 struct aggregate_stats_cbdata cbdata;
3371 struct cls_rule target;
3374 if (arg_size != sizeof *asr) {
3375 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3377 asr = (struct ofp_aggregate_stats_request *) osr->body;
3379 COVERAGE_INC(ofproto_agg_request);
3381 cbdata.out_port = asr->out_port;
3382 cbdata.packet_count = 0;
3383 cbdata.byte_count = 0;
3385 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3386 classifier_for_each_match(&p->cls, &target,
3387 table_id_to_include(asr->table_id),
3388 aggregate_stats_cb, &cbdata);
3390 msg = start_stats_reply(osr, sizeof *reply);
3391 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3392 reply->flow_count = htonl(cbdata.n_flows);
3393 reply->packet_count = htonll(cbdata.packet_count);
3394 reply->byte_count = htonll(cbdata.byte_count);
3395 queue_tx(msg, ofconn, ofconn->reply_counter);
3399 struct queue_stats_cbdata {
3400 struct ofconn *ofconn;
3406 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3407 const struct netdev_queue_stats *stats)
3409 struct ofp_queue_stats *reply;
3411 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3412 reply->port_no = htons(cbdata->port_no);
3413 memset(reply->pad, 0, sizeof reply->pad);
3414 reply->queue_id = htonl(queue_id);
3415 reply->tx_bytes = htonll(stats->tx_bytes);
3416 reply->tx_packets = htonll(stats->tx_packets);
3417 reply->tx_errors = htonll(stats->tx_errors);
3421 handle_queue_stats_dump_cb(uint32_t queue_id,
3422 struct netdev_queue_stats *stats,
3425 struct queue_stats_cbdata *cbdata = cbdata_;
3427 put_queue_stats(cbdata, queue_id, stats);
3431 handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
3433 struct queue_stats_cbdata *cbdata)
3435 cbdata->port_no = port_no;
3436 if (queue_id == OFPQ_ALL) {
3437 netdev_dump_queue_stats(port->netdev,
3438 handle_queue_stats_dump_cb, cbdata);
3440 struct netdev_queue_stats stats;
3442 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3443 put_queue_stats(cbdata, queue_id, &stats);
3448 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3449 const struct ofp_stats_request *osr,
3452 struct ofp_queue_stats_request *qsr;
3453 struct queue_stats_cbdata cbdata;
3454 struct ofport *port;
3455 unsigned int port_no;
3458 if (arg_size != sizeof *qsr) {
3459 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3461 qsr = (struct ofp_queue_stats_request *) osr->body;
3463 COVERAGE_INC(ofproto_queue_req);
3465 cbdata.ofconn = ofconn;
3466 cbdata.msg = start_stats_reply(osr, 128);
3468 port_no = ntohs(qsr->port_no);
3469 queue_id = ntohl(qsr->queue_id);
3470 if (port_no == OFPP_ALL) {
3471 PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
3472 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3474 } else if (port_no < ofproto->max_ports) {
3475 port = port_array_get(&ofproto->ports, port_no);
3477 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3480 ofpbuf_delete(cbdata.msg);
3481 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3483 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3489 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3490 struct ofp_header *oh)
3492 struct ofp_stats_request *osr;
3496 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3501 osr = (struct ofp_stats_request *) oh;
3503 switch (ntohs(osr->type)) {
3505 return handle_desc_stats_request(p, ofconn, osr);
3508 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3510 case OFPST_AGGREGATE:
3511 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3514 return handle_table_stats_request(p, ofconn, osr);
3517 return handle_port_stats_request(p, ofconn, osr, arg_size);
3520 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3523 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3526 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3530 static long long int
3531 msec_from_nsec(uint64_t sec, uint32_t nsec)
3533 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3537 update_time(struct ofproto *ofproto, struct rule *rule,
3538 const struct odp_flow_stats *stats)
3540 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3541 if (used > rule->used) {
3543 if (rule->super && used > rule->super->used) {
3544 rule->super->used = used;
3546 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3551 update_stats(struct ofproto *ofproto, struct rule *rule,
3552 const struct odp_flow_stats *stats)
3554 if (stats->n_packets) {
3555 update_time(ofproto, rule, stats);
3556 rule->packet_count += stats->n_packets;
3557 rule->byte_count += stats->n_bytes;
3558 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3562 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3563 * in which no matching flow already exists in the flow table.
3565 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3566 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3567 * code as encoded by ofp_mkerr() on failure.
3569 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3572 add_flow(struct ofproto *p, struct ofconn *ofconn,
3573 const struct ofp_flow_mod *ofm, size_t n_actions)
3575 struct ofpbuf *packet;
3580 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3584 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3586 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3587 ntohs(ofm->priority))) {
3588 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3592 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3593 n_actions, ntohs(ofm->idle_timeout),
3594 ntohs(ofm->hard_timeout), ofm->cookie,
3595 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3596 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3597 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3600 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3601 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3605 in_port = UINT16_MAX;
3608 rule_insert(p, rule, packet, in_port);
3612 static struct rule *
3613 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3618 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3620 return rule_from_cls_rule(classifier_find_rule_exactly(
3621 &p->cls, &flow, wildcards,
3622 ntohs(ofm->priority)));
3626 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3627 struct rule *rule, const struct ofp_flow_mod *ofm)
3629 struct ofpbuf *packet;
3634 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3638 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3644 flow_extract(packet, 0, in_port, &flow);
3645 rule_execute(ofproto, rule, packet, &flow);
3650 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3652 struct modify_flows_cbdata {
3653 struct ofproto *ofproto;
3654 const struct ofp_flow_mod *ofm;
3659 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3660 size_t n_actions, struct rule *);
3661 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3663 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3664 * encoded by ofp_mkerr() on failure.
3666 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3669 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3670 const struct ofp_flow_mod *ofm, size_t n_actions)
3672 struct modify_flows_cbdata cbdata;
3673 struct cls_rule target;
3677 cbdata.n_actions = n_actions;
3678 cbdata.match = NULL;
3680 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3683 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3684 modify_flows_cb, &cbdata);
3686 /* This credits the packet to whichever flow happened to happened to
3687 * match last. That's weird. Maybe we should do a lookup for the
3688 * flow that actually matches the packet? Who knows. */
3689 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3692 return add_flow(p, ofconn, ofm, n_actions);
3696 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3697 * code as encoded by ofp_mkerr() on failure.
3699 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3702 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3703 struct ofp_flow_mod *ofm, size_t n_actions)
3705 struct rule *rule = find_flow_strict(p, ofm);
3706 if (rule && !rule_is_hidden(rule)) {
3707 modify_flow(p, ofm, n_actions, rule);
3708 return send_buffered_packet(p, ofconn, rule, ofm);
3710 return add_flow(p, ofconn, ofm, n_actions);
3714 /* Callback for modify_flows_loose(). */
3716 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3718 struct rule *rule = rule_from_cls_rule(rule_);
3719 struct modify_flows_cbdata *cbdata = cbdata_;
3721 if (!rule_is_hidden(rule)) {
3722 cbdata->match = rule;
3723 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3727 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3728 * been identified as a flow in 'p''s flow table to be modified, by changing
3729 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3730 * ofp_action[] structures). */
3732 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3733 size_t n_actions, struct rule *rule)
3735 size_t actions_len = n_actions * sizeof *rule->actions;
3737 rule->flow_cookie = ofm->cookie;
3739 /* If the actions are the same, do nothing. */
3740 if (n_actions == rule->n_actions
3741 && !memcmp(ofm->actions, rule->actions, actions_len))
3746 /* Replace actions. */
3747 free(rule->actions);
3748 rule->actions = xmemdup(ofm->actions, actions_len);
3749 rule->n_actions = n_actions;
3751 /* Make sure that the datapath gets updated properly. */
3752 if (rule->cr.wc.wildcards) {
3753 COVERAGE_INC(ofproto_mod_wc_flow);
3754 p->need_revalidate = true;
3756 rule_update_actions(p, rule);
3762 /* OFPFC_DELETE implementation. */
3764 struct delete_flows_cbdata {
3765 struct ofproto *ofproto;
3769 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3770 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3772 /* Implements OFPFC_DELETE. */
3774 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3776 struct delete_flows_cbdata cbdata;
3777 struct cls_rule target;
3780 cbdata.out_port = ofm->out_port;
3782 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3785 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3786 delete_flows_cb, &cbdata);
3789 /* Implements OFPFC_DELETE_STRICT. */
3791 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3793 struct rule *rule = find_flow_strict(p, ofm);
3795 delete_flow(p, rule, ofm->out_port);
3799 /* Callback for delete_flows_loose(). */
3801 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3803 struct rule *rule = rule_from_cls_rule(rule_);
3804 struct delete_flows_cbdata *cbdata = cbdata_;
3806 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3809 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3810 * been identified as a flow to delete from 'p''s flow table, by deleting the
3811 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3814 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3815 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3816 * specified 'out_port'. */
3818 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3820 if (rule_is_hidden(rule)) {
3824 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3828 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3829 rule_remove(p, rule);
3833 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3834 struct ofp_flow_mod *ofm)
3836 struct ofp_match orig_match;
3840 error = reject_slave_controller(ofconn, &ofm->header);
3844 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3845 sizeof *ofm->actions, &n_actions);
3850 /* We do not support the emergency flow cache. It will hopefully
3851 * get dropped from OpenFlow in the near future. */
3852 if (ofm->flags & htons(OFPFF_EMERG)) {
3853 /* There isn't a good fit for an error code, so just state that the
3854 * flow table is full. */
3855 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3858 /* Normalize ofp->match. If normalization actually changes anything, then
3859 * log the differences. */
3860 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3861 orig_match = ofm->match;
3862 normalize_match(&ofm->match);
3863 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3864 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3865 if (!VLOG_DROP_INFO(&normal_rl)) {
3866 char *old = ofp_match_to_literal_string(&orig_match);
3867 char *new = ofp_match_to_literal_string(&ofm->match);
3868 VLOG_INFO("%s: normalization changed ofp_match, details:",
3869 rconn_get_name(ofconn->rconn));
3870 VLOG_INFO(" pre: %s", old);
3871 VLOG_INFO("post: %s", new);
3877 if (!ofm->match.wildcards) {
3878 ofm->priority = htons(UINT16_MAX);
3881 error = validate_actions((const union ofp_action *) ofm->actions,
3882 n_actions, p->max_ports);
3887 switch (ntohs(ofm->command)) {
3889 return add_flow(p, ofconn, ofm, n_actions);
3892 return modify_flows_loose(p, ofconn, ofm, n_actions);
3894 case OFPFC_MODIFY_STRICT:
3895 return modify_flow_strict(p, ofconn, ofm, n_actions);
3898 delete_flows_loose(p, ofm);
3901 case OFPFC_DELETE_STRICT:
3902 delete_flow_strict(p, ofm);
3906 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3911 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3915 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3920 p->tun_id_from_cookie = !!msg->set;
3925 handle_role_request(struct ofproto *ofproto,
3926 struct ofconn *ofconn, struct nicira_header *msg)
3928 struct nx_role_request *nrr;
3929 struct nx_role_request *reply;
3933 if (ntohs(msg->header.length) != sizeof *nrr) {
3934 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3935 ntohs(msg->header.length), sizeof *nrr);
3936 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3938 nrr = (struct nx_role_request *) msg;
3940 if (ofconn->type != OFCONN_PRIMARY) {
3941 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3943 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3946 role = ntohl(nrr->role);
3947 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3948 && role != NX_ROLE_SLAVE) {
3949 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3951 /* There's no good error code for this. */
3952 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3955 if (role == NX_ROLE_MASTER) {
3956 struct ofconn *other;
3958 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3959 &ofproto->controllers) {
3960 if (other->role == NX_ROLE_MASTER) {
3961 other->role = NX_ROLE_SLAVE;
3965 ofconn->role = role;
3967 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3969 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3970 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3971 reply->role = htonl(role);
3972 queue_tx(buf, ofconn, ofconn->reply_counter);
3978 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3980 struct ofp_vendor_header *ovh = msg;
3981 struct nicira_header *nh;
3983 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3984 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3985 "(expected at least %zu)",
3986 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3987 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3989 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3990 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3992 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3993 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3994 "(expected at least %zu)",
3995 ntohs(ovh->header.length), sizeof(struct nicira_header));
3996 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4000 switch (ntohl(nh->subtype)) {
4001 case NXT_STATUS_REQUEST:
4002 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4005 case NXT_TUN_ID_FROM_COOKIE:
4006 return handle_tun_id_from_cookie(p, msg);
4008 case NXT_ROLE_REQUEST:
4009 return handle_role_request(p, ofconn, msg);
4012 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4016 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4018 struct ofp_header *ob;
4021 /* Currently, everything executes synchronously, so we can just
4022 * immediately send the barrier reply. */
4023 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4024 queue_tx(buf, ofconn, ofconn->reply_counter);
4029 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4030 struct ofpbuf *ofp_msg)
4032 struct ofp_header *oh = ofp_msg->data;
4035 COVERAGE_INC(ofproto_recv_openflow);
4037 case OFPT_ECHO_REQUEST:
4038 error = handle_echo_request(ofconn, oh);
4041 case OFPT_ECHO_REPLY:
4045 case OFPT_FEATURES_REQUEST:
4046 error = handle_features_request(p, ofconn, oh);
4049 case OFPT_GET_CONFIG_REQUEST:
4050 error = handle_get_config_request(p, ofconn, oh);
4053 case OFPT_SET_CONFIG:
4054 error = handle_set_config(p, ofconn, ofp_msg->data);
4057 case OFPT_PACKET_OUT:
4058 error = handle_packet_out(p, ofconn, ofp_msg->data);
4062 error = handle_port_mod(p, ofconn, oh);
4066 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4069 case OFPT_STATS_REQUEST:
4070 error = handle_stats_request(p, ofconn, oh);
4074 error = handle_vendor(p, ofconn, ofp_msg->data);
4077 case OFPT_BARRIER_REQUEST:
4078 error = handle_barrier_request(ofconn, oh);
4082 if (VLOG_IS_WARN_ENABLED()) {
4083 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4084 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4087 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4092 send_error_oh(ofconn, ofp_msg->data, error);
4097 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4099 struct odp_msg *msg = packet->data;
4101 struct ofpbuf payload;
4104 payload.data = msg + 1;
4105 payload.size = msg->length - sizeof *msg;
4106 flow_extract(&payload, msg->arg, msg->port, &flow);
4108 /* Check with in-band control to see if this packet should be sent
4109 * to the local port regardless of the flow table. */
4110 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4111 union odp_action action;
4113 memset(&action, 0, sizeof(action));
4114 action.output.type = ODPAT_OUTPUT;
4115 action.output.port = ODPP_LOCAL;
4116 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
4119 rule = lookup_valid_rule(p, &flow);
4121 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4122 struct ofport *port = port_array_get(&p->ports, msg->port);
4124 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4125 COVERAGE_INC(ofproto_no_packet_in);
4126 /* XXX install 'drop' flow entry */
4127 ofpbuf_delete(packet);
4131 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4134 COVERAGE_INC(ofproto_packet_in);
4135 send_packet_in(p, packet);
4139 if (rule->cr.wc.wildcards) {
4140 rule = rule_create_subrule(p, rule, &flow);
4141 rule_make_actions(p, rule, packet);
4143 if (!rule->may_install) {
4144 /* The rule is not installable, that is, we need to process every
4145 * packet, so process the current packet and set its actions into
4147 rule_make_actions(p, rule, packet);
4149 /* XXX revalidate rule if it needs it */
4153 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4155 * Extra-special case for fail-open mode.
4157 * We are in fail-open mode and the packet matched the fail-open rule,
4158 * but we are connected to a controller too. We should send the packet
4159 * up to the controller in the hope that it will try to set up a flow
4160 * and thereby allow us to exit fail-open.
4162 * See the top-level comment in fail-open.c for more information.
4164 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4165 DPIF_RECV_MSG_PADDING));
4168 ofpbuf_pull(packet, sizeof *msg);
4169 rule_execute(p, rule, packet, &flow);
4170 rule_reinstall(p, rule);
4174 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4176 struct odp_msg *msg = packet->data;
4178 switch (msg->type) {
4179 case _ODPL_ACTION_NR:
4180 COVERAGE_INC(ofproto_ctlr_action);
4181 send_packet_in(p, packet);
4184 case _ODPL_SFLOW_NR:
4186 ofproto_sflow_received(p->sflow, msg);
4188 ofpbuf_delete(packet);
4192 handle_odp_miss_msg(p, packet);
4196 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4203 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4205 struct rule *sub = rule_from_cls_rule(sub_);
4206 struct revalidate_cbdata *cbdata = cbdata_;
4208 if (cbdata->revalidate_all
4209 || (cbdata->revalidate_subrules && sub->super)
4210 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4211 revalidate_rule(cbdata->ofproto, sub);
4216 revalidate_rule(struct ofproto *p, struct rule *rule)
4218 const flow_t *flow = &rule->cr.flow;
4220 COVERAGE_INC(ofproto_revalidate_rule);
4223 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4225 rule_remove(p, rule);
4227 } else if (super != rule->super) {
4228 COVERAGE_INC(ofproto_revalidate_moved);
4229 list_remove(&rule->list);
4230 list_push_back(&super->list, &rule->list);
4231 rule->super = super;
4232 rule->hard_timeout = super->hard_timeout;
4233 rule->idle_timeout = super->idle_timeout;
4234 rule->created = super->created;
4239 rule_update_actions(p, rule);
4243 static struct ofpbuf *
4244 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4245 long long int now, uint8_t reason)
4247 struct ofp_flow_removed *ofr;
4249 long long int tdiff = now - rule->created;
4250 uint32_t sec = tdiff / 1000;
4251 uint32_t msec = tdiff - (sec * 1000);
4253 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4254 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4256 ofr->cookie = rule->flow_cookie;
4257 ofr->priority = htons(rule->cr.priority);
4258 ofr->reason = reason;
4259 ofr->duration_sec = htonl(sec);
4260 ofr->duration_nsec = htonl(msec * 1000000);
4261 ofr->idle_timeout = htons(rule->idle_timeout);
4262 ofr->packet_count = htonll(rule->packet_count);
4263 ofr->byte_count = htonll(rule->byte_count);
4269 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
4271 assert(rule->installed);
4272 assert(!rule->cr.wc.wildcards);
4275 rule_remove(ofproto, rule);
4277 rule_uninstall(ofproto, rule);
4282 send_flow_removed(struct ofproto *p, struct rule *rule,
4283 long long int now, uint8_t reason)
4285 struct ofconn *ofconn;
4286 struct ofconn *prev;
4287 struct ofpbuf *buf = NULL;
4289 /* We limit the maximum number of queued flow expirations it by accounting
4290 * them under the counter for replies. That works because preventing
4291 * OpenFlow requests from being processed also prevents new flows from
4292 * being added (and expiring). (It also prevents processing OpenFlow
4293 * requests that would not add new flows, so it is imperfect.) */
4296 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4297 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
4298 && ofconn_receives_async_msgs(ofconn)) {
4300 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4302 buf = compose_flow_removed(p, rule, now, reason);
4308 queue_tx(buf, prev, prev->reply_counter);
4314 expire_rule(struct cls_rule *cls_rule, void *p_)
4316 struct ofproto *p = p_;
4317 struct rule *rule = rule_from_cls_rule(cls_rule);
4318 long long int hard_expire, idle_expire, expire, now;
4320 hard_expire = (rule->hard_timeout
4321 ? rule->created + rule->hard_timeout * 1000
4323 idle_expire = (rule->idle_timeout
4324 && (rule->super || list_is_empty(&rule->list))
4325 ? rule->used + rule->idle_timeout * 1000
4327 expire = MIN(hard_expire, idle_expire);
4331 if (rule->installed && now >= rule->used + 5000) {
4332 uninstall_idle_flow(p, rule);
4333 } else if (!rule->cr.wc.wildcards) {
4334 active_timeout(p, rule);
4340 COVERAGE_INC(ofproto_expired);
4342 /* Update stats. This code will be a no-op if the rule expired
4343 * due to an idle timeout. */
4344 if (rule->cr.wc.wildcards) {
4345 struct rule *subrule, *next;
4346 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
4347 rule_remove(p, subrule);
4350 rule_uninstall(p, rule);
4353 if (!rule_is_hidden(rule)) {
4354 send_flow_removed(p, rule, now,
4356 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4358 rule_remove(p, rule);
4362 active_timeout(struct ofproto *ofproto, struct rule *rule)
4364 if (ofproto->netflow && !is_controller_rule(rule) &&
4365 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4366 struct ofexpired expired;
4367 struct odp_flow odp_flow;
4369 /* Get updated flow stats. */
4370 memset(&odp_flow, 0, sizeof odp_flow);
4371 if (rule->installed) {
4372 odp_flow.key = rule->cr.flow;
4373 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4374 dpif_flow_get(ofproto->dpif, &odp_flow);
4376 if (odp_flow.stats.n_packets) {
4377 update_time(ofproto, rule, &odp_flow.stats);
4378 netflow_flow_update_flags(&rule->nf_flow,
4379 odp_flow.stats.tcp_flags);
4383 expired.flow = rule->cr.flow;
4384 expired.packet_count = rule->packet_count +
4385 odp_flow.stats.n_packets;
4386 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4387 expired.used = rule->used;
4389 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4391 /* Schedule us to send the accumulated records once we have
4392 * collected all of them. */
4393 poll_immediate_wake();
4398 update_used(struct ofproto *p)
4400 struct odp_flow *flows;
4405 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4410 for (i = 0; i < n_flows; i++) {
4411 struct odp_flow *f = &flows[i];
4414 rule = rule_from_cls_rule(
4415 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4416 if (!rule || !rule->installed) {
4417 COVERAGE_INC(ofproto_unexpected_rule);
4418 dpif_flow_del(p->dpif, f);
4422 update_time(p, rule, &f->stats);
4423 rule_account(p, rule, f->stats.n_bytes);
4428 /* pinsched callback for sending 'packet' on 'ofconn'. */
4430 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4432 struct ofconn *ofconn = ofconn_;
4434 rconn_send_with_limit(ofconn->rconn, packet,
4435 ofconn->packet_in_counter, 100);
4438 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4439 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4440 * packet scheduler for sending.
4442 * 'max_len' specifies the maximum number of bytes of the packet to send on
4443 * 'ofconn' (INT_MAX specifies no limit).
4445 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4446 * ownership is transferred to this function. */
4448 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4451 struct ofproto *ofproto = ofconn->ofproto;
4452 struct ofp_packet_in *opi = packet->data;
4453 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4454 int send_len, trim_size;
4458 if (opi->reason == OFPR_ACTION) {
4459 buffer_id = UINT32_MAX;
4460 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4461 buffer_id = pktbuf_get_null();
4462 } else if (!ofconn->pktbuf) {
4463 buffer_id = UINT32_MAX;
4465 struct ofpbuf payload;
4466 payload.data = opi->data;
4467 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4468 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4471 /* Figure out how much of the packet to send. */
4472 send_len = ntohs(opi->total_len);
4473 if (buffer_id != UINT32_MAX) {
4474 send_len = MIN(send_len, ofconn->miss_send_len);
4476 send_len = MIN(send_len, max_len);
4478 /* Adjust packet length and clone if necessary. */
4479 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4481 packet = ofpbuf_clone_data(packet->data, trim_size);
4484 packet->size = trim_size;
4487 /* Update packet headers. */
4488 opi->buffer_id = htonl(buffer_id);
4489 update_openflow_length(packet);
4491 /* Hand over to packet scheduler. It might immediately call into
4492 * do_send_packet_in() or it might buffer it for a while (until a later
4493 * call to pinsched_run()). */
4494 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4495 packet, do_send_packet_in, ofconn);
4498 /* Replace struct odp_msg header in 'packet' by equivalent struct
4499 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4500 * returned by dpif_recv()).
4502 * The conversion is not complete: the caller still needs to trim any unneeded
4503 * payload off the end of the buffer, set the length in the OpenFlow header,
4504 * and set buffer_id. Those require us to know the controller settings and so
4505 * must be done on a per-controller basis.
4507 * Returns the maximum number of bytes of the packet that should be sent to
4508 * the controller (INT_MAX if no limit). */
4510 do_convert_to_packet_in(struct ofpbuf *packet)
4512 struct odp_msg *msg = packet->data;
4513 struct ofp_packet_in *opi;
4519 /* Extract relevant header fields */
4520 if (msg->type == _ODPL_ACTION_NR) {
4521 reason = OFPR_ACTION;
4524 reason = OFPR_NO_MATCH;
4527 total_len = msg->length - sizeof *msg;
4528 in_port = odp_port_to_ofp_port(msg->port);
4530 /* Repurpose packet buffer by overwriting header. */
4531 ofpbuf_pull(packet, sizeof(struct odp_msg));
4532 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4533 opi->header.version = OFP_VERSION;
4534 opi->header.type = OFPT_PACKET_IN;
4535 opi->total_len = htons(total_len);
4536 opi->in_port = htons(in_port);
4537 opi->reason = reason;
4542 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4543 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4544 * as necessary according to their individual configurations.
4546 * 'packet' must have sufficient headroom to convert it into a struct
4547 * ofp_packet_in (e.g. as returned by dpif_recv()).
4549 * Takes ownership of 'packet'. */
4551 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4553 struct ofconn *ofconn, *prev;
4556 max_len = do_convert_to_packet_in(packet);
4559 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4560 if (ofconn_receives_async_msgs(ofconn)) {
4562 schedule_packet_in(prev, packet, max_len, true);
4568 schedule_packet_in(prev, packet, max_len, false);
4570 ofpbuf_delete(packet);
4575 pick_datapath_id(const struct ofproto *ofproto)
4577 const struct ofport *port;
4579 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4581 uint8_t ea[ETH_ADDR_LEN];
4584 error = netdev_get_etheraddr(port->netdev, ea);
4586 return eth_addr_to_uint64(ea);
4588 VLOG_WARN("could not get MAC address for %s (%s)",
4589 netdev_get_name(port->netdev), strerror(error));
4591 return ofproto->fallback_dpid;
4595 pick_fallback_dpid(void)
4597 uint8_t ea[ETH_ADDR_LEN];
4598 eth_addr_nicira_random(ea);
4599 return eth_addr_to_uint64(ea);
4603 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4604 struct odp_actions *actions, tag_type *tags,
4605 uint16_t *nf_output_iface, void *ofproto_)
4607 struct ofproto *ofproto = ofproto_;
4610 /* Drop frames for reserved multicast addresses. */
4611 if (eth_addr_is_reserved(flow->dl_dst)) {
4615 /* Learn source MAC (but don't try to learn from revalidation). */
4616 if (packet != NULL) {
4617 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4619 GRAT_ARP_LOCK_NONE);
4621 /* The log messages here could actually be useful in debugging,
4622 * so keep the rate limit relatively high. */
4623 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4624 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4625 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4626 ofproto_revalidate(ofproto, rev_tag);
4630 /* Determine output port. */
4631 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4634 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4635 } else if (out_port != flow->in_port) {
4636 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4637 *nf_output_iface = out_port;
4645 static const struct ofhooks default_ofhooks = {
4647 default_normal_ofhook_cb,