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 int error = ofproto_run1(p);
1000 error = ofproto_run2(p, false);
1006 process_port_change(struct ofproto *ofproto, int error, char *devname)
1008 if (error == ENOBUFS) {
1009 reinit_ports(ofproto);
1010 } else if (!error) {
1011 update_port(ofproto, devname);
1016 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1017 * means that 'ofconn' is more interesting for monitoring than a lower return
1020 snoop_preference(const struct ofconn *ofconn)
1022 switch (ofconn->role) {
1023 case NX_ROLE_MASTER:
1030 /* Shouldn't happen. */
1035 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1036 * Connects this vconn to a controller. */
1038 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1040 struct ofconn *ofconn, *best;
1042 /* Pick a controller for monitoring. */
1044 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1045 if (ofconn->type == OFCONN_PRIMARY
1046 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1052 rconn_add_monitor(best->rconn, vconn);
1054 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1060 ofproto_run1(struct ofproto *p)
1062 struct ofconn *ofconn, *next_ofconn;
1063 struct ofservice *ofservice;
1068 if (shash_is_empty(&p->port_by_name)) {
1072 for (i = 0; i < 50; i++) {
1075 error = dpif_recv(p->dpif, &buf);
1077 if (error == ENODEV) {
1078 /* Someone destroyed the datapath behind our back. The caller
1079 * better destroy us and give up, because we're just going to
1080 * spin from here on out. */
1081 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1082 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1083 dpif_name(p->dpif));
1089 handle_odp_msg(p, buf);
1092 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1093 process_port_change(p, error, devname);
1095 while ((error = netdev_monitor_poll(p->netdev_monitor,
1096 &devname)) != EAGAIN) {
1097 process_port_change(p, error, devname);
1101 if (time_msec() >= p->next_in_band_update) {
1102 update_in_band_remotes(p);
1104 in_band_run(p->in_band);
1107 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1109 ofconn_run(ofconn, p);
1112 /* Fail-open maintenance. Do this after processing the ofconns since
1113 * fail-open checks the status of the controller rconn. */
1115 fail_open_run(p->fail_open);
1118 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1119 struct vconn *vconn;
1122 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1124 struct rconn *rconn;
1127 rconn = rconn_create(ofservice->probe_interval, 0);
1128 name = ofconn_make_name(p, vconn_get_name(vconn));
1129 rconn_connect_unreliably(rconn, vconn, name);
1132 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1133 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1134 ofservice->burst_limit);
1135 } else if (retval != EAGAIN) {
1136 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1140 for (i = 0; i < p->n_snoops; i++) {
1141 struct vconn *vconn;
1144 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1146 add_snooper(p, vconn);
1147 } else if (retval != EAGAIN) {
1148 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1152 if (time_msec() >= p->next_expiration) {
1153 COVERAGE_INC(ofproto_expiration);
1154 p->next_expiration = time_msec() + 1000;
1157 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1159 /* Let the hook know that we're at a stable point: all outstanding data
1160 * in existing flows has been accounted to the account_cb. Thus, the
1161 * hook can now reasonably do operations that depend on having accurate
1162 * flow volume accounting (currently, that's just bond rebalancing). */
1163 if (p->ofhooks->account_checkpoint_cb) {
1164 p->ofhooks->account_checkpoint_cb(p->aux);
1169 netflow_run(p->netflow);
1172 ofproto_sflow_run(p->sflow);
1178 struct revalidate_cbdata {
1179 struct ofproto *ofproto;
1180 bool revalidate_all; /* Revalidate all exact-match rules? */
1181 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1182 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1186 ofproto_run2(struct ofproto *p, bool revalidate_all)
1188 if (p->need_revalidate || revalidate_all
1189 || !tag_set_is_empty(&p->revalidate_set)) {
1190 struct revalidate_cbdata cbdata;
1192 cbdata.revalidate_all = revalidate_all;
1193 cbdata.revalidate_subrules = p->need_revalidate;
1194 cbdata.revalidate_set = p->revalidate_set;
1195 tag_set_init(&p->revalidate_set);
1196 COVERAGE_INC(ofproto_revalidate);
1197 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1198 p->need_revalidate = false;
1205 ofproto_wait(struct ofproto *p)
1207 struct ofservice *ofservice;
1208 struct ofconn *ofconn;
1211 dpif_recv_wait(p->dpif);
1212 dpif_port_poll_wait(p->dpif);
1213 netdev_monitor_poll_wait(p->netdev_monitor);
1214 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1215 ofconn_wait(ofconn);
1218 poll_timer_wait_until(p->next_in_band_update);
1219 in_band_wait(p->in_band);
1222 fail_open_wait(p->fail_open);
1225 ofproto_sflow_wait(p->sflow);
1227 if (!tag_set_is_empty(&p->revalidate_set)) {
1228 poll_immediate_wake();
1230 if (p->need_revalidate) {
1231 /* Shouldn't happen, but if it does just go around again. */
1232 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1233 poll_immediate_wake();
1234 } else if (p->next_expiration != LLONG_MAX) {
1235 poll_timer_wait_until(p->next_expiration);
1237 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1238 pvconn_wait(ofservice->pvconn);
1240 for (i = 0; i < p->n_snoops; i++) {
1241 pvconn_wait(p->snoops[i]);
1246 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1248 tag_set_add(&ofproto->revalidate_set, tag);
1252 ofproto_get_revalidate_set(struct ofproto *ofproto)
1254 return &ofproto->revalidate_set;
1258 ofproto_is_alive(const struct ofproto *p)
1260 return !hmap_is_empty(&p->controllers);
1264 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1265 const union ofp_action *actions, size_t n_actions,
1266 const struct ofpbuf *packet)
1268 struct odp_actions odp_actions;
1271 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1277 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1279 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1280 odp_actions.n_actions, packet);
1285 ofproto_add_flow(struct ofproto *p,
1286 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1287 const union ofp_action *actions, size_t n_actions,
1291 rule = rule_create(p, NULL, actions, n_actions,
1292 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1294 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1295 rule_insert(p, rule, NULL, 0);
1299 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1300 uint32_t wildcards, unsigned int priority)
1304 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1308 rule_remove(ofproto, rule);
1313 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1315 struct rule *rule = rule_from_cls_rule(rule_);
1316 struct ofproto *ofproto = ofproto_;
1318 /* Mark the flow as not installed, even though it might really be
1319 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1320 * There is no point in uninstalling it individually since we are about to
1321 * blow away all the flows with dpif_flow_flush(). */
1322 rule->installed = false;
1324 rule_remove(ofproto, rule);
1328 ofproto_flush_flows(struct ofproto *ofproto)
1330 COVERAGE_INC(ofproto_flush);
1331 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1332 dpif_flow_flush(ofproto->dpif);
1333 if (ofproto->in_band) {
1334 in_band_flushed(ofproto->in_band);
1336 if (ofproto->fail_open) {
1337 fail_open_flushed(ofproto->fail_open);
1342 reinit_ports(struct ofproto *p)
1344 struct svec devnames;
1345 struct ofport *ofport;
1346 unsigned int port_no;
1347 struct odp_port *odp_ports;
1351 svec_init(&devnames);
1352 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1353 svec_add (&devnames, (char *) ofport->opp.name);
1355 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1356 for (i = 0; i < n_odp_ports; i++) {
1357 svec_add (&devnames, odp_ports[i].devname);
1361 svec_sort_unique(&devnames);
1362 for (i = 0; i < devnames.n; i++) {
1363 update_port(p, devnames.names[i]);
1365 svec_destroy(&devnames);
1369 refresh_port_group(struct ofproto *p, unsigned int group)
1373 struct ofport *port;
1374 unsigned int port_no;
1376 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1378 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1380 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1381 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1382 ports[n_ports++] = port_no;
1385 dpif_port_group_set(p->dpif, group, ports, n_ports);
1392 refresh_port_groups(struct ofproto *p)
1394 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1395 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1397 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1401 static struct ofport *
1402 make_ofport(const struct odp_port *odp_port)
1404 struct netdev_options netdev_options;
1405 enum netdev_flags flags;
1406 struct ofport *ofport;
1407 struct netdev *netdev;
1411 memset(&netdev_options, 0, sizeof netdev_options);
1412 netdev_options.name = odp_port->devname;
1413 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1415 error = netdev_open(&netdev_options, &netdev);
1417 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1418 "cannot be opened (%s)",
1419 odp_port->devname, odp_port->port,
1420 odp_port->devname, strerror(error));
1424 ofport = xmalloc(sizeof *ofport);
1425 ofport->netdev = netdev;
1426 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1427 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1428 memcpy(ofport->opp.name, odp_port->devname,
1429 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1430 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1432 netdev_get_flags(netdev, &flags);
1433 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1435 netdev_get_carrier(netdev, &carrier);
1436 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1438 netdev_get_features(netdev,
1439 &ofport->opp.curr, &ofport->opp.advertised,
1440 &ofport->opp.supported, &ofport->opp.peer);
1445 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1447 if (port_array_get(&p->ports, odp_port->port)) {
1448 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1451 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1452 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1461 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1463 const struct ofp_phy_port *a = &a_->opp;
1464 const struct ofp_phy_port *b = &b_->opp;
1466 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1467 return (a->port_no == b->port_no
1468 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1469 && !strcmp((char *) a->name, (char *) b->name)
1470 && a->state == b->state
1471 && a->config == b->config
1472 && a->curr == b->curr
1473 && a->advertised == b->advertised
1474 && a->supported == b->supported
1475 && a->peer == b->peer);
1479 send_port_status(struct ofproto *p, const struct ofport *ofport,
1482 /* XXX Should limit the number of queued port status change messages. */
1483 struct ofconn *ofconn;
1484 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1485 struct ofp_port_status *ops;
1488 if (!ofconn_receives_async_msgs(ofconn)) {
1492 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1493 ops->reason = reason;
1494 ops->desc = ofport->opp;
1495 hton_ofp_phy_port(&ops->desc);
1496 queue_tx(b, ofconn, NULL);
1498 if (p->ofhooks->port_changed_cb) {
1499 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1504 ofport_install(struct ofproto *p, struct ofport *ofport)
1506 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1507 const char *netdev_name = (const char *) ofport->opp.name;
1509 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1510 port_array_set(&p->ports, odp_port, ofport);
1511 shash_add(&p->port_by_name, netdev_name, ofport);
1513 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1518 ofport_remove(struct ofproto *p, struct ofport *ofport)
1520 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1522 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1523 port_array_delete(&p->ports, odp_port);
1524 shash_delete(&p->port_by_name,
1525 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1527 ofproto_sflow_del_port(p->sflow, odp_port);
1532 ofport_free(struct ofport *ofport)
1535 netdev_close(ofport->netdev);
1541 update_port(struct ofproto *p, const char *devname)
1543 struct odp_port odp_port;
1544 struct ofport *old_ofport;
1545 struct ofport *new_ofport;
1548 COVERAGE_INC(ofproto_update_port);
1550 /* Query the datapath for port information. */
1551 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1553 /* Find the old ofport. */
1554 old_ofport = shash_find_data(&p->port_by_name, devname);
1557 /* There's no port named 'devname' but there might be a port with
1558 * the same port number. This could happen if a port is deleted
1559 * and then a new one added in its place very quickly, or if a port
1560 * is renamed. In the former case we want to send an OFPPR_DELETE
1561 * and an OFPPR_ADD, and in the latter case we want to send a
1562 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1563 * the old port's ifindex against the new port, or perhaps less
1564 * reliably but more portably by comparing the old port's MAC
1565 * against the new port's MAC. However, this code isn't that smart
1566 * and always sends an OFPPR_MODIFY (XXX). */
1567 old_ofport = port_array_get(&p->ports, odp_port.port);
1569 } else if (error != ENOENT && error != ENODEV) {
1570 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1571 "%s", strerror(error));
1575 /* Create a new ofport. */
1576 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1578 /* Eliminate a few pathological cases. */
1579 if (!old_ofport && !new_ofport) {
1581 } else if (old_ofport && new_ofport) {
1582 /* Most of the 'config' bits are OpenFlow soft state, but
1583 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1584 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1585 * leaves the other bits 0.) */
1586 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1588 if (ofport_equal(old_ofport, new_ofport)) {
1589 /* False alarm--no change. */
1590 ofport_free(new_ofport);
1595 /* Now deal with the normal cases. */
1597 ofport_remove(p, old_ofport);
1600 ofport_install(p, new_ofport);
1602 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1603 (!old_ofport ? OFPPR_ADD
1604 : !new_ofport ? OFPPR_DELETE
1606 ofport_free(old_ofport);
1608 /* Update port groups. */
1609 refresh_port_groups(p);
1613 init_ports(struct ofproto *p)
1615 struct odp_port *ports;
1620 error = dpif_port_list(p->dpif, &ports, &n_ports);
1625 for (i = 0; i < n_ports; i++) {
1626 const struct odp_port *odp_port = &ports[i];
1627 if (!ofport_conflicts(p, odp_port)) {
1628 struct ofport *ofport = make_ofport(odp_port);
1630 ofport_install(p, ofport);
1635 refresh_port_groups(p);
1639 static struct ofconn *
1640 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1642 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1643 ofconn->ofproto = p;
1644 list_push_back(&p->all_conns, &ofconn->node);
1645 ofconn->rconn = rconn;
1646 ofconn->type = type;
1647 ofconn->role = NX_ROLE_OTHER;
1648 ofconn->packet_in_counter = rconn_packet_counter_create ();
1649 ofconn->pktbuf = NULL;
1650 ofconn->miss_send_len = 0;
1651 ofconn->reply_counter = rconn_packet_counter_create ();
1656 ofconn_destroy(struct ofconn *ofconn)
1658 if (ofconn->type == OFCONN_PRIMARY) {
1659 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1661 discovery_destroy(ofconn->discovery);
1663 list_remove(&ofconn->node);
1664 switch_status_unregister(ofconn->ss);
1665 rconn_destroy(ofconn->rconn);
1666 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1667 rconn_packet_counter_destroy(ofconn->reply_counter);
1668 pktbuf_destroy(ofconn->pktbuf);
1673 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1678 if (ofconn->discovery) {
1679 char *controller_name;
1680 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1681 discovery_question_connectivity(ofconn->discovery);
1683 if (discovery_run(ofconn->discovery, &controller_name)) {
1684 if (controller_name) {
1685 char *ofconn_name = ofconn_make_name(p, controller_name);
1686 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1689 rconn_disconnect(ofconn->rconn);
1694 for (i = 0; i < N_SCHEDULERS; i++) {
1695 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1698 rconn_run(ofconn->rconn);
1700 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1701 /* Limit the number of iterations to prevent other tasks from
1703 for (iteration = 0; iteration < 50; iteration++) {
1704 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1709 fail_open_maybe_recover(p->fail_open);
1711 handle_openflow(ofconn, p, of_msg);
1712 ofpbuf_delete(of_msg);
1716 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1717 ofconn_destroy(ofconn);
1722 ofconn_wait(struct ofconn *ofconn)
1726 if (ofconn->discovery) {
1727 discovery_wait(ofconn->discovery);
1729 for (i = 0; i < N_SCHEDULERS; i++) {
1730 pinsched_wait(ofconn->schedulers[i]);
1732 rconn_run_wait(ofconn->rconn);
1733 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1734 rconn_recv_wait(ofconn->rconn);
1736 COVERAGE_INC(ofproto_ofconn_stuck);
1740 /* Returns true if 'ofconn' should receive asynchronous messages. */
1742 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1744 if (ofconn->type == OFCONN_PRIMARY) {
1745 /* Primary controllers always get asynchronous messages unless they
1746 * have configured themselves as "slaves". */
1747 return ofconn->role != NX_ROLE_SLAVE;
1749 /* Service connections don't get asynchronous messages unless they have
1750 * explicitly asked for them by setting a nonzero miss send length. */
1751 return ofconn->miss_send_len > 0;
1755 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1756 * and 'target', suitable for use in log messages for identifying the
1759 * The name is dynamically allocated. The caller should free it (with free())
1760 * when it is no longer needed. */
1762 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1764 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1768 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1772 for (i = 0; i < N_SCHEDULERS; i++) {
1773 struct pinsched **s = &ofconn->schedulers[i];
1777 *s = pinsched_create(rate, burst,
1778 ofconn->ofproto->switch_status);
1780 pinsched_set_limits(*s, rate, burst);
1783 pinsched_destroy(*s);
1790 ofservice_reconfigure(struct ofservice *ofservice,
1791 const struct ofproto_controller *c)
1793 ofservice->probe_interval = c->probe_interval;
1794 ofservice->rate_limit = c->rate_limit;
1795 ofservice->burst_limit = c->burst_limit;
1798 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1799 * positive errno value. */
1801 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1803 struct ofservice *ofservice;
1804 struct pvconn *pvconn;
1807 error = pvconn_open(c->target, &pvconn);
1812 ofservice = xzalloc(sizeof *ofservice);
1813 hmap_insert(&ofproto->services, &ofservice->node,
1814 hash_string(c->target, 0));
1815 ofservice->pvconn = pvconn;
1817 ofservice_reconfigure(ofservice, c);
1823 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1825 hmap_remove(&ofproto->services, &ofservice->node);
1826 pvconn_close(ofservice->pvconn);
1830 /* Finds and returns the ofservice within 'ofproto' that has the given
1831 * 'target', or a null pointer if none exists. */
1832 static struct ofservice *
1833 ofservice_lookup(struct ofproto *ofproto, const char *target)
1835 struct ofservice *ofservice;
1837 HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node,
1838 hash_string(target, 0), &ofproto->services) {
1839 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1846 /* Caller is responsible for initializing the 'cr' member of the returned
1848 static struct rule *
1849 rule_create(struct ofproto *ofproto, struct rule *super,
1850 const union ofp_action *actions, size_t n_actions,
1851 uint16_t idle_timeout, uint16_t hard_timeout,
1852 uint64_t flow_cookie, bool send_flow_removed)
1854 struct rule *rule = xzalloc(sizeof *rule);
1855 rule->idle_timeout = idle_timeout;
1856 rule->hard_timeout = hard_timeout;
1857 rule->flow_cookie = flow_cookie;
1858 rule->used = rule->created = time_msec();
1859 rule->send_flow_removed = send_flow_removed;
1860 rule->super = super;
1862 list_push_back(&super->list, &rule->list);
1864 list_init(&rule->list);
1866 rule->n_actions = n_actions;
1867 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1868 netflow_flow_clear(&rule->nf_flow);
1869 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1874 static struct rule *
1875 rule_from_cls_rule(const struct cls_rule *cls_rule)
1877 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1881 rule_free(struct rule *rule)
1883 free(rule->actions);
1884 free(rule->odp_actions);
1888 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1889 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1890 * through all of its subrules and revalidates them, destroying any that no
1891 * longer has a super-rule (which is probably all of them).
1893 * Before calling this function, the caller must make have removed 'rule' from
1894 * the classifier. If 'rule' is an exact-match rule, the caller is also
1895 * responsible for ensuring that it has been uninstalled from the datapath. */
1897 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1900 struct rule *subrule, *next;
1901 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1902 revalidate_rule(ofproto, subrule);
1905 list_remove(&rule->list);
1911 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1913 const union ofp_action *oa;
1914 struct actions_iterator i;
1916 if (out_port == htons(OFPP_NONE)) {
1919 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1920 oa = actions_next(&i)) {
1921 if (action_outputs_to_port(oa, out_port)) {
1928 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1929 * 'packet', which arrived on 'in_port'.
1931 * Takes ownership of 'packet'. */
1933 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1934 const union odp_action *actions, size_t n_actions,
1935 struct ofpbuf *packet)
1937 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1938 /* As an optimization, avoid a round-trip from userspace to kernel to
1939 * userspace. This also avoids possibly filling up kernel packet
1940 * buffers along the way. */
1941 struct odp_msg *msg;
1943 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1944 msg->type = _ODPL_ACTION_NR;
1945 msg->length = sizeof(struct odp_msg) + packet->size;
1946 msg->port = in_port;
1948 msg->arg = actions[0].controller.arg;
1950 send_packet_in(ofproto, packet);
1956 error = dpif_execute(ofproto->dpif, in_port,
1957 actions, n_actions, packet);
1958 ofpbuf_delete(packet);
1963 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1964 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1965 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1967 * The flow that 'packet' actually contains does not need to actually match
1968 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1969 * the packet and byte counters for 'rule' will be credited for the packet sent
1970 * out whether or not the packet actually matches 'rule'.
1972 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1973 * the caller must already have accurately composed ODP actions for it given
1974 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1975 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1976 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1977 * actions and apply them to 'packet'.
1979 * Takes ownership of 'packet'. */
1981 rule_execute(struct ofproto *ofproto, struct rule *rule,
1982 struct ofpbuf *packet, const flow_t *flow)
1984 const union odp_action *actions;
1985 struct odp_flow_stats stats;
1987 struct odp_actions a;
1989 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1991 /* Grab or compose the ODP actions.
1993 * The special case for an exact-match 'rule' where 'flow' is not the
1994 * rule's flow is important to avoid, e.g., sending a packet out its input
1995 * port simply because the ODP actions were composed for the wrong
1997 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1998 struct rule *super = rule->super ? rule->super : rule;
1999 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
2000 packet, &a, NULL, 0, NULL)) {
2001 ofpbuf_delete(packet);
2004 actions = a.actions;
2005 n_actions = a.n_actions;
2007 actions = rule->odp_actions;
2008 n_actions = rule->n_odp_actions;
2011 /* Execute the ODP actions. */
2012 flow_extract_stats(flow, packet, &stats);
2013 if (execute_odp_actions(ofproto, flow->in_port,
2014 actions, n_actions, packet)) {
2015 update_stats(ofproto, rule, &stats);
2016 rule->used = time_msec();
2017 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2021 /* Inserts 'rule' into 'p''s flow table.
2023 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2024 * actions on it and credits the statistics for sending the packet to 'rule'.
2025 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2028 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2031 struct rule *displaced_rule;
2033 /* Insert the rule in the classifier. */
2034 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2035 if (!rule->cr.wc.wildcards) {
2036 rule_make_actions(p, rule, packet);
2039 /* Send the packet and credit it to the rule. */
2042 flow_extract(packet, 0, in_port, &flow);
2043 rule_execute(p, rule, packet, &flow);
2046 /* Install the rule in the datapath only after sending the packet, to
2047 * avoid packet reordering. */
2048 if (rule->cr.wc.wildcards) {
2049 COVERAGE_INC(ofproto_add_wc_flow);
2050 p->need_revalidate = true;
2052 rule_install(p, rule, displaced_rule);
2055 /* Free the rule that was displaced, if any. */
2056 if (displaced_rule) {
2057 rule_destroy(p, displaced_rule);
2061 static struct rule *
2062 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2065 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2066 rule->idle_timeout, rule->hard_timeout,
2068 COVERAGE_INC(ofproto_subrule_create);
2069 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2070 : rule->cr.priority), &subrule->cr);
2071 classifier_insert_exact(&ofproto->cls, &subrule->cr);
2077 rule_remove(struct ofproto *ofproto, struct rule *rule)
2079 if (rule->cr.wc.wildcards) {
2080 COVERAGE_INC(ofproto_del_wc_flow);
2081 ofproto->need_revalidate = true;
2083 rule_uninstall(ofproto, rule);
2085 classifier_remove(&ofproto->cls, &rule->cr);
2086 rule_destroy(ofproto, rule);
2089 /* Returns true if the actions changed, false otherwise. */
2091 rule_make_actions(struct ofproto *p, struct rule *rule,
2092 const struct ofpbuf *packet)
2094 const struct rule *super;
2095 struct odp_actions a;
2098 assert(!rule->cr.wc.wildcards);
2100 super = rule->super ? rule->super : rule;
2102 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2103 packet, &a, &rule->tags, &rule->may_install,
2104 &rule->nf_flow.output_iface);
2106 actions_len = a.n_actions * sizeof *a.actions;
2107 if (rule->n_odp_actions != a.n_actions
2108 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2109 COVERAGE_INC(ofproto_odp_unchanged);
2110 free(rule->odp_actions);
2111 rule->n_odp_actions = a.n_actions;
2112 rule->odp_actions = xmemdup(a.actions, actions_len);
2120 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2121 struct odp_flow_put *put)
2123 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2124 put->flow.key = rule->cr.flow;
2125 put->flow.actions = rule->odp_actions;
2126 put->flow.n_actions = rule->n_odp_actions;
2127 put->flow.flags = 0;
2129 return dpif_flow_put(ofproto->dpif, put);
2133 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2135 assert(!rule->cr.wc.wildcards);
2137 if (rule->may_install) {
2138 struct odp_flow_put put;
2139 if (!do_put_flow(p, rule,
2140 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2142 rule->installed = true;
2143 if (displaced_rule) {
2144 update_stats(p, displaced_rule, &put.flow.stats);
2145 rule_post_uninstall(p, displaced_rule);
2148 } else if (displaced_rule) {
2149 rule_uninstall(p, displaced_rule);
2154 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2156 if (rule->installed) {
2157 struct odp_flow_put put;
2158 COVERAGE_INC(ofproto_dp_missed);
2159 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2161 rule_install(ofproto, rule, NULL);
2166 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2168 bool actions_changed;
2169 uint16_t new_out_iface, old_out_iface;
2171 old_out_iface = rule->nf_flow.output_iface;
2172 actions_changed = rule_make_actions(ofproto, rule, NULL);
2174 if (rule->may_install) {
2175 if (rule->installed) {
2176 if (actions_changed) {
2177 struct odp_flow_put put;
2178 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2179 | ODPPF_ZERO_STATS, &put);
2180 update_stats(ofproto, rule, &put.flow.stats);
2182 /* Temporarily set the old output iface so that NetFlow
2183 * messages have the correct output interface for the old
2185 new_out_iface = rule->nf_flow.output_iface;
2186 rule->nf_flow.output_iface = old_out_iface;
2187 rule_post_uninstall(ofproto, rule);
2188 rule->nf_flow.output_iface = new_out_iface;
2191 rule_install(ofproto, rule, NULL);
2194 rule_uninstall(ofproto, rule);
2199 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2201 uint64_t total_bytes = rule->byte_count + extra_bytes;
2203 if (ofproto->ofhooks->account_flow_cb
2204 && total_bytes > rule->accounted_bytes)
2206 ofproto->ofhooks->account_flow_cb(
2207 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2208 total_bytes - rule->accounted_bytes, ofproto->aux);
2209 rule->accounted_bytes = total_bytes;
2214 rule_uninstall(struct ofproto *p, struct rule *rule)
2216 assert(!rule->cr.wc.wildcards);
2217 if (rule->installed) {
2218 struct odp_flow odp_flow;
2220 odp_flow.key = rule->cr.flow;
2221 odp_flow.actions = NULL;
2222 odp_flow.n_actions = 0;
2224 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2225 update_stats(p, rule, &odp_flow.stats);
2227 rule->installed = false;
2229 rule_post_uninstall(p, rule);
2234 is_controller_rule(struct rule *rule)
2236 /* If the only action is send to the controller then don't report
2237 * NetFlow expiration messages since it is just part of the control
2238 * logic for the network and not real traffic. */
2242 && rule->super->n_actions == 1
2243 && action_outputs_to_port(&rule->super->actions[0],
2244 htons(OFPP_CONTROLLER)));
2248 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2250 struct rule *super = rule->super;
2252 rule_account(ofproto, rule, 0);
2254 if (ofproto->netflow && !is_controller_rule(rule)) {
2255 struct ofexpired expired;
2256 expired.flow = rule->cr.flow;
2257 expired.packet_count = rule->packet_count;
2258 expired.byte_count = rule->byte_count;
2259 expired.used = rule->used;
2260 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2263 super->packet_count += rule->packet_count;
2264 super->byte_count += rule->byte_count;
2266 /* Reset counters to prevent double counting if the rule ever gets
2268 rule->packet_count = 0;
2269 rule->byte_count = 0;
2270 rule->accounted_bytes = 0;
2272 netflow_flow_clear(&rule->nf_flow);
2277 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2278 struct rconn_packet_counter *counter)
2280 update_openflow_length(msg);
2281 if (rconn_send(ofconn->rconn, msg, counter)) {
2287 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2288 int error, const void *data, size_t len)
2291 struct ofp_error_msg *oem;
2293 if (!(error >> 16)) {
2294 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2299 COVERAGE_INC(ofproto_error);
2300 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2301 oh ? oh->xid : 0, &buf);
2302 oem->type = htons((unsigned int) error >> 16);
2303 oem->code = htons(error & 0xffff);
2304 memcpy(oem->data, data, len);
2305 queue_tx(buf, ofconn, ofconn->reply_counter);
2309 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2312 size_t oh_length = ntohs(oh->length);
2313 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2317 hton_ofp_phy_port(struct ofp_phy_port *opp)
2319 opp->port_no = htons(opp->port_no);
2320 opp->config = htonl(opp->config);
2321 opp->state = htonl(opp->state);
2322 opp->curr = htonl(opp->curr);
2323 opp->advertised = htonl(opp->advertised);
2324 opp->supported = htonl(opp->supported);
2325 opp->peer = htonl(opp->peer);
2329 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2331 struct ofp_header *rq = oh;
2332 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2337 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2338 struct ofp_header *oh)
2340 struct ofp_switch_features *osf;
2342 unsigned int port_no;
2343 struct ofport *port;
2345 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2346 osf->datapath_id = htonll(p->datapath_id);
2347 osf->n_buffers = htonl(pktbuf_capacity());
2349 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2350 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2351 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2352 (1u << OFPAT_SET_VLAN_VID) |
2353 (1u << OFPAT_SET_VLAN_PCP) |
2354 (1u << OFPAT_STRIP_VLAN) |
2355 (1u << OFPAT_SET_DL_SRC) |
2356 (1u << OFPAT_SET_DL_DST) |
2357 (1u << OFPAT_SET_NW_SRC) |
2358 (1u << OFPAT_SET_NW_DST) |
2359 (1u << OFPAT_SET_NW_TOS) |
2360 (1u << OFPAT_SET_TP_SRC) |
2361 (1u << OFPAT_SET_TP_DST) |
2362 (1u << OFPAT_ENQUEUE));
2364 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2365 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2368 queue_tx(buf, ofconn, ofconn->reply_counter);
2373 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2374 struct ofp_header *oh)
2377 struct ofp_switch_config *osc;
2381 /* Figure out flags. */
2382 dpif_get_drop_frags(p->dpif, &drop_frags);
2383 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2386 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2387 osc->flags = htons(flags);
2388 osc->miss_send_len = htons(ofconn->miss_send_len);
2389 queue_tx(buf, ofconn, ofconn->reply_counter);
2395 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2396 struct ofp_switch_config *osc)
2401 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2405 flags = ntohs(osc->flags);
2407 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2408 switch (flags & OFPC_FRAG_MASK) {
2409 case OFPC_FRAG_NORMAL:
2410 dpif_set_drop_frags(p->dpif, false);
2412 case OFPC_FRAG_DROP:
2413 dpif_set_drop_frags(p->dpif, true);
2416 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2422 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2428 add_output_group_action(struct odp_actions *actions, uint16_t group,
2429 uint16_t *nf_output_iface)
2431 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2433 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2434 *nf_output_iface = NF_OUT_FLOOD;
2439 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2441 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2442 a->controller.arg = max_len;
2445 struct action_xlate_ctx {
2447 flow_t flow; /* Flow to which these actions correspond. */
2448 int recurse; /* Recursion level, via xlate_table_action. */
2449 struct ofproto *ofproto;
2450 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2451 * null pointer if we are revalidating
2452 * without a packet to refer to. */
2455 struct odp_actions *out; /* Datapath actions. */
2456 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2457 bool may_set_up_flow; /* True ordinarily; false if the actions must
2458 * be reassessed for every packet. */
2459 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2462 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2463 * flow translation. */
2464 #define MAX_RESUBMIT_RECURSION 8
2466 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2467 struct action_xlate_ctx *ctx);
2470 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2472 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2475 if (ofport->opp.config & OFPPC_NO_FWD) {
2476 /* Forwarding disabled on port. */
2481 * We don't have an ofport record for this port, but it doesn't hurt to
2482 * allow forwarding to it anyhow. Maybe such a port will appear later
2483 * and we're pre-populating the flow table.
2487 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2488 ctx->nf_output_iface = port;
2491 static struct rule *
2492 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2495 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2497 /* The rule we found might not be valid, since we could be in need of
2498 * revalidation. If it is not valid, don't return it. */
2501 && ofproto->need_revalidate
2502 && !revalidate_rule(ofproto, rule)) {
2503 COVERAGE_INC(ofproto_invalidated);
2511 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2513 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2514 uint16_t old_in_port;
2517 /* Look up a flow with 'in_port' as the input port. Then restore the
2518 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2519 * have surprising behavior). */
2520 old_in_port = ctx->flow.in_port;
2521 ctx->flow.in_port = in_port;
2522 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2523 ctx->flow.in_port = old_in_port;
2531 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2535 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2537 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2538 MAX_RESUBMIT_RECURSION);
2543 xlate_output_action__(struct action_xlate_ctx *ctx,
2544 uint16_t port, uint16_t max_len)
2547 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2549 ctx->nf_output_iface = NF_OUT_DROP;
2553 add_output_action(ctx, ctx->flow.in_port);
2556 xlate_table_action(ctx, ctx->flow.in_port);
2559 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2560 ctx->out, ctx->tags,
2561 &ctx->nf_output_iface,
2562 ctx->ofproto->aux)) {
2563 COVERAGE_INC(ofproto_uninstallable);
2564 ctx->may_set_up_flow = false;
2568 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2569 &ctx->nf_output_iface);
2572 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2574 case OFPP_CONTROLLER:
2575 add_controller_action(ctx->out, max_len);
2578 add_output_action(ctx, ODPP_LOCAL);
2581 odp_port = ofp_port_to_odp_port(port);
2582 if (odp_port != ctx->flow.in_port) {
2583 add_output_action(ctx, odp_port);
2588 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2589 ctx->nf_output_iface = NF_OUT_FLOOD;
2590 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2591 ctx->nf_output_iface = prev_nf_output_iface;
2592 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2593 ctx->nf_output_iface != NF_OUT_FLOOD) {
2594 ctx->nf_output_iface = NF_OUT_MULTI;
2599 xlate_output_action(struct action_xlate_ctx *ctx,
2600 const struct ofp_action_output *oao)
2602 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2605 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2606 * optimization, because we're going to add another action that sets the
2607 * priority immediately after, or because there are no actions following the
2610 remove_pop_action(struct action_xlate_ctx *ctx)
2612 size_t n = ctx->out->n_actions;
2613 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2614 ctx->out->n_actions--;
2619 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2620 const struct ofp_action_enqueue *oae)
2622 uint16_t ofp_port, odp_port;
2626 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2629 /* Fall back to ordinary output action. */
2630 xlate_output_action__(ctx, ntohs(oae->port), 0);
2634 /* Figure out ODP output port. */
2635 ofp_port = ntohs(oae->port);
2636 if (ofp_port != OFPP_IN_PORT) {
2637 odp_port = ofp_port_to_odp_port(ofp_port);
2639 odp_port = ctx->flow.in_port;
2642 /* Add ODP actions. */
2643 remove_pop_action(ctx);
2644 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2646 add_output_action(ctx, odp_port);
2647 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2649 /* Update NetFlow output port. */
2650 if (ctx->nf_output_iface == NF_OUT_DROP) {
2651 ctx->nf_output_iface = odp_port;
2652 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2653 ctx->nf_output_iface = NF_OUT_MULTI;
2658 xlate_nicira_action(struct action_xlate_ctx *ctx,
2659 const struct nx_action_header *nah)
2661 const struct nx_action_resubmit *nar;
2662 const struct nx_action_set_tunnel *nast;
2663 union odp_action *oa;
2664 int subtype = ntohs(nah->subtype);
2666 assert(nah->vendor == htonl(NX_VENDOR_ID));
2668 case NXAST_RESUBMIT:
2669 nar = (const struct nx_action_resubmit *) nah;
2670 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2673 case NXAST_SET_TUNNEL:
2674 nast = (const struct nx_action_set_tunnel *) nah;
2675 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2676 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2679 case NXAST_DROP_SPOOFED_ARP:
2680 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2681 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2685 /* If you add a new action here that modifies flow data, don't forget to
2686 * update the flow key in ctx->flow at the same time. */
2689 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2695 do_xlate_actions(const union ofp_action *in, size_t n_in,
2696 struct action_xlate_ctx *ctx)
2698 struct actions_iterator iter;
2699 const union ofp_action *ia;
2700 const struct ofport *port;
2702 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2703 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2704 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2705 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2706 /* Drop this flow. */
2710 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2711 uint16_t type = ntohs(ia->type);
2712 union odp_action *oa;
2716 xlate_output_action(ctx, &ia->output);
2719 case OFPAT_SET_VLAN_VID:
2720 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2721 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2724 case OFPAT_SET_VLAN_PCP:
2725 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2726 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2729 case OFPAT_STRIP_VLAN:
2730 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2731 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2732 ctx->flow.dl_vlan_pcp = 0;
2735 case OFPAT_SET_DL_SRC:
2736 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2737 memcpy(oa->dl_addr.dl_addr,
2738 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2739 memcpy(ctx->flow.dl_src,
2740 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2743 case OFPAT_SET_DL_DST:
2744 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2745 memcpy(oa->dl_addr.dl_addr,
2746 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2747 memcpy(ctx->flow.dl_dst,
2748 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2751 case OFPAT_SET_NW_SRC:
2752 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2753 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2756 case OFPAT_SET_NW_DST:
2757 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2758 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2761 case OFPAT_SET_NW_TOS:
2762 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2763 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2766 case OFPAT_SET_TP_SRC:
2767 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2768 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2771 case OFPAT_SET_TP_DST:
2772 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2773 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2777 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2781 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2785 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2792 xlate_actions(const union ofp_action *in, size_t n_in,
2793 const flow_t *flow, struct ofproto *ofproto,
2794 const struct ofpbuf *packet,
2795 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2796 uint16_t *nf_output_iface)
2798 tag_type no_tags = 0;
2799 struct action_xlate_ctx ctx;
2800 COVERAGE_INC(ofproto_ofp2odp);
2801 odp_actions_init(out);
2804 ctx.ofproto = ofproto;
2805 ctx.packet = packet;
2807 ctx.tags = tags ? tags : &no_tags;
2808 ctx.may_set_up_flow = true;
2809 ctx.nf_output_iface = NF_OUT_DROP;
2810 do_xlate_actions(in, n_in, &ctx);
2811 remove_pop_action(&ctx);
2813 /* Check with in-band control to see if we're allowed to set up this
2815 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2816 ctx.may_set_up_flow = false;
2819 if (may_set_up_flow) {
2820 *may_set_up_flow = ctx.may_set_up_flow;
2822 if (nf_output_iface) {
2823 *nf_output_iface = ctx.nf_output_iface;
2825 if (odp_actions_overflow(out)) {
2826 COVERAGE_INC(odp_overflow);
2827 odp_actions_init(out);
2828 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2833 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2834 * error message code (composed with ofp_mkerr()) for the caller to propagate
2835 * upward. Otherwise, returns 0.
2837 * 'oh' is used to make log messages more informative. */
2839 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2841 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2842 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2845 type_name = ofp_message_type_to_string(oh->type);
2846 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2850 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2857 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2858 struct ofp_header *oh)
2860 struct ofp_packet_out *opo;
2861 struct ofpbuf payload, *buffer;
2862 struct odp_actions actions;
2868 error = reject_slave_controller(ofconn, oh);
2873 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2877 opo = (struct ofp_packet_out *) oh;
2879 COVERAGE_INC(ofproto_packet_out);
2880 if (opo->buffer_id != htonl(UINT32_MAX)) {
2881 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2883 if (error || !buffer) {
2891 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2892 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2893 &flow, p, &payload, &actions, NULL, NULL, NULL);
2898 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2900 ofpbuf_delete(buffer);
2906 update_port_config(struct ofproto *p, struct ofport *port,
2907 uint32_t config, uint32_t mask)
2909 mask &= config ^ port->opp.config;
2910 if (mask & OFPPC_PORT_DOWN) {
2911 if (config & OFPPC_PORT_DOWN) {
2912 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2914 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2917 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2918 if (mask & REVALIDATE_BITS) {
2919 COVERAGE_INC(ofproto_costly_flags);
2920 port->opp.config ^= mask & REVALIDATE_BITS;
2921 p->need_revalidate = true;
2923 #undef REVALIDATE_BITS
2924 if (mask & OFPPC_NO_FLOOD) {
2925 port->opp.config ^= OFPPC_NO_FLOOD;
2926 refresh_port_groups(p);
2928 if (mask & OFPPC_NO_PACKET_IN) {
2929 port->opp.config ^= OFPPC_NO_PACKET_IN;
2934 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2935 struct ofp_header *oh)
2937 const struct ofp_port_mod *opm;
2938 struct ofport *port;
2941 error = reject_slave_controller(ofconn, oh);
2945 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2949 opm = (struct ofp_port_mod *) oh;
2951 port = port_array_get(&p->ports,
2952 ofp_port_to_odp_port(ntohs(opm->port_no)));
2954 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2955 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2956 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2958 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2959 if (opm->advertise) {
2960 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2966 static struct ofpbuf *
2967 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2969 struct ofp_stats_reply *osr;
2972 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2973 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2975 osr->flags = htons(0);
2979 static struct ofpbuf *
2980 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2982 return make_stats_reply(request->header.xid, request->type, body_len);
2986 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2988 struct ofpbuf *msg = *msgp;
2989 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2990 if (nbytes + msg->size > UINT16_MAX) {
2991 struct ofp_stats_reply *reply = msg->data;
2992 reply->flags = htons(OFPSF_REPLY_MORE);
2993 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2994 queue_tx(msg, ofconn, ofconn->reply_counter);
2996 return ofpbuf_put_uninit(*msgp, nbytes);
3000 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3001 struct ofp_stats_request *request)
3003 struct ofp_desc_stats *ods;
3006 msg = start_stats_reply(request, sizeof *ods);
3007 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3008 memset(ods, 0, sizeof *ods);
3009 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3010 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3011 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3012 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3013 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3014 queue_tx(msg, ofconn, ofconn->reply_counter);
3020 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
3022 struct rule *rule = rule_from_cls_rule(cls_rule);
3023 int *n_subrules = n_subrules_;
3031 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3032 struct ofp_stats_request *request)
3034 struct ofp_table_stats *ots;
3036 struct odp_stats dpstats;
3037 int n_exact, n_subrules, n_wild;
3039 msg = start_stats_reply(request, sizeof *ots * 2);
3041 /* Count rules of various kinds. */
3043 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
3044 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3045 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3048 dpif_get_dp_stats(p->dpif, &dpstats);
3049 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3050 memset(ots, 0, sizeof *ots);
3051 ots->table_id = TABLEID_HASH;
3052 strcpy(ots->name, "hash");
3053 ots->wildcards = htonl(0);
3054 ots->max_entries = htonl(dpstats.max_capacity);
3055 ots->active_count = htonl(n_exact);
3056 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3058 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3060 /* Classifier table. */
3061 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3062 memset(ots, 0, sizeof *ots);
3063 ots->table_id = TABLEID_CLASSIFIER;
3064 strcpy(ots->name, "classifier");
3065 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3067 ots->max_entries = htonl(65536);
3068 ots->active_count = htonl(n_wild);
3069 ots->lookup_count = htonll(0); /* XXX */
3070 ots->matched_count = htonll(0); /* XXX */
3072 queue_tx(msg, ofconn, ofconn->reply_counter);
3077 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
3078 struct ofpbuf **msgp)
3080 struct netdev_stats stats;
3081 struct ofp_port_stats *ops;
3083 /* Intentionally ignore return value, since errors will set
3084 * 'stats' to all-1s, which is correct for OpenFlow, and
3085 * netdev_get_stats() will log errors. */
3086 netdev_get_stats(port->netdev, &stats);
3088 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3089 ops->port_no = htons(odp_port_to_ofp_port(port_no));
3090 memset(ops->pad, 0, sizeof ops->pad);
3091 ops->rx_packets = htonll(stats.rx_packets);
3092 ops->tx_packets = htonll(stats.tx_packets);
3093 ops->rx_bytes = htonll(stats.rx_bytes);
3094 ops->tx_bytes = htonll(stats.tx_bytes);
3095 ops->rx_dropped = htonll(stats.rx_dropped);
3096 ops->tx_dropped = htonll(stats.tx_dropped);
3097 ops->rx_errors = htonll(stats.rx_errors);
3098 ops->tx_errors = htonll(stats.tx_errors);
3099 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3100 ops->rx_over_err = htonll(stats.rx_over_errors);
3101 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3102 ops->collisions = htonll(stats.collisions);
3106 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3107 struct ofp_stats_request *osr,
3110 struct ofp_port_stats_request *psr;
3111 struct ofp_port_stats *ops;
3113 struct ofport *port;
3114 unsigned int port_no;
3116 if (arg_size != sizeof *psr) {
3117 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3119 psr = (struct ofp_port_stats_request *) osr->body;
3121 msg = start_stats_reply(osr, sizeof *ops * 16);
3122 if (psr->port_no != htons(OFPP_NONE)) {
3123 port = port_array_get(&p->ports,
3124 ofp_port_to_odp_port(ntohs(psr->port_no)));
3126 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
3129 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
3130 append_port_stat(port, port_no, ofconn, &msg);
3134 queue_tx(msg, ofconn, ofconn->reply_counter);
3138 struct flow_stats_cbdata {
3139 struct ofproto *ofproto;
3140 struct ofconn *ofconn;
3145 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3146 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3147 * returned statistic include statistics for all of 'rule''s subrules. */
3149 query_stats(struct ofproto *p, struct rule *rule,
3150 uint64_t *packet_countp, uint64_t *byte_countp)
3152 uint64_t packet_count, byte_count;
3153 struct rule *subrule;
3154 struct odp_flow *odp_flows;
3157 /* Start from historical data for 'rule' itself that are no longer tracked
3158 * by the datapath. This counts, for example, subrules that have
3160 packet_count = rule->packet_count;
3161 byte_count = rule->byte_count;
3163 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3164 * wildcarded then on all of its subrules.
3166 * Also, add any statistics that are not tracked by the datapath for each
3167 * subrule. This includes, for example, statistics for packets that were
3168 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3170 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3171 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3172 if (rule->cr.wc.wildcards) {
3174 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
3175 odp_flows[i++].key = subrule->cr.flow;
3176 packet_count += subrule->packet_count;
3177 byte_count += subrule->byte_count;
3180 odp_flows[0].key = rule->cr.flow;
3183 /* Fetch up-to-date statistics from the datapath and add them in. */
3184 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3186 for (i = 0; i < n_odp_flows; i++) {
3187 struct odp_flow *odp_flow = &odp_flows[i];
3188 packet_count += odp_flow->stats.n_packets;
3189 byte_count += odp_flow->stats.n_bytes;
3194 /* Return the stats to the caller. */
3195 *packet_countp = packet_count;
3196 *byte_countp = byte_count;
3200 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3202 struct rule *rule = rule_from_cls_rule(rule_);
3203 struct flow_stats_cbdata *cbdata = cbdata_;
3204 struct ofp_flow_stats *ofs;
3205 uint64_t packet_count, byte_count;
3206 size_t act_len, len;
3207 long long int tdiff = time_msec() - rule->created;
3208 uint32_t sec = tdiff / 1000;
3209 uint32_t msec = tdiff - (sec * 1000);
3211 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3215 act_len = sizeof *rule->actions * rule->n_actions;
3216 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3218 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3220 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3221 ofs->length = htons(len);
3222 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3224 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3225 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3226 ofs->duration_sec = htonl(sec);
3227 ofs->duration_nsec = htonl(msec * 1000000);
3228 ofs->cookie = rule->flow_cookie;
3229 ofs->priority = htons(rule->cr.priority);
3230 ofs->idle_timeout = htons(rule->idle_timeout);
3231 ofs->hard_timeout = htons(rule->hard_timeout);
3232 memset(ofs->pad2, 0, sizeof ofs->pad2);
3233 ofs->packet_count = htonll(packet_count);
3234 ofs->byte_count = htonll(byte_count);
3235 memcpy(ofs->actions, rule->actions, act_len);
3239 table_id_to_include(uint8_t table_id)
3241 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3242 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3243 : table_id == 0xff ? CLS_INC_ALL
3248 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3249 const struct ofp_stats_request *osr,
3252 struct ofp_flow_stats_request *fsr;
3253 struct flow_stats_cbdata cbdata;
3254 struct cls_rule target;
3256 if (arg_size != sizeof *fsr) {
3257 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3259 fsr = (struct ofp_flow_stats_request *) osr->body;
3261 COVERAGE_INC(ofproto_flows_req);
3263 cbdata.ofconn = ofconn;
3264 cbdata.out_port = fsr->out_port;
3265 cbdata.msg = start_stats_reply(osr, 1024);
3266 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3267 classifier_for_each_match(&p->cls, &target,
3268 table_id_to_include(fsr->table_id),
3269 flow_stats_cb, &cbdata);
3270 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3274 struct flow_stats_ds_cbdata {
3275 struct ofproto *ofproto;
3280 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3282 struct rule *rule = rule_from_cls_rule(rule_);
3283 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3284 struct ds *results = cbdata->results;
3285 struct ofp_match match;
3286 uint64_t packet_count, byte_count;
3287 size_t act_len = sizeof *rule->actions * rule->n_actions;
3289 /* Don't report on subrules. */
3290 if (rule->super != NULL) {
3294 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3295 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3296 cbdata->ofproto->tun_id_from_cookie, &match);
3298 ds_put_format(results, "duration=%llds, ",
3299 (time_msec() - rule->created) / 1000);
3300 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3301 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3302 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3303 ofp_print_match(results, &match, true);
3304 ofp_print_actions(results, &rule->actions->header, act_len);
3305 ds_put_cstr(results, "\n");
3308 /* Adds a pretty-printed description of all flows to 'results', including
3309 * those marked hidden by secchan (e.g., by in-band control). */
3311 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3313 struct ofp_match match;
3314 struct cls_rule target;
3315 struct flow_stats_ds_cbdata cbdata;
3317 memset(&match, 0, sizeof match);
3318 match.wildcards = htonl(OVSFW_ALL);
3321 cbdata.results = results;
3323 cls_rule_from_match(&match, 0, false, 0, &target);
3324 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3325 flow_stats_ds_cb, &cbdata);
3328 struct aggregate_stats_cbdata {
3329 struct ofproto *ofproto;
3331 uint64_t packet_count;
3332 uint64_t byte_count;
3337 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3339 struct rule *rule = rule_from_cls_rule(rule_);
3340 struct aggregate_stats_cbdata *cbdata = cbdata_;
3341 uint64_t packet_count, byte_count;
3343 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3347 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3349 cbdata->packet_count += packet_count;
3350 cbdata->byte_count += byte_count;
3355 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3356 const struct ofp_stats_request *osr,
3359 struct ofp_aggregate_stats_request *asr;
3360 struct ofp_aggregate_stats_reply *reply;
3361 struct aggregate_stats_cbdata cbdata;
3362 struct cls_rule target;
3365 if (arg_size != sizeof *asr) {
3366 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3368 asr = (struct ofp_aggregate_stats_request *) osr->body;
3370 COVERAGE_INC(ofproto_agg_request);
3372 cbdata.out_port = asr->out_port;
3373 cbdata.packet_count = 0;
3374 cbdata.byte_count = 0;
3376 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3377 classifier_for_each_match(&p->cls, &target,
3378 table_id_to_include(asr->table_id),
3379 aggregate_stats_cb, &cbdata);
3381 msg = start_stats_reply(osr, sizeof *reply);
3382 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3383 reply->flow_count = htonl(cbdata.n_flows);
3384 reply->packet_count = htonll(cbdata.packet_count);
3385 reply->byte_count = htonll(cbdata.byte_count);
3386 queue_tx(msg, ofconn, ofconn->reply_counter);
3390 struct queue_stats_cbdata {
3391 struct ofconn *ofconn;
3397 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3398 const struct netdev_queue_stats *stats)
3400 struct ofp_queue_stats *reply;
3402 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3403 reply->port_no = htons(cbdata->port_no);
3404 memset(reply->pad, 0, sizeof reply->pad);
3405 reply->queue_id = htonl(queue_id);
3406 reply->tx_bytes = htonll(stats->tx_bytes);
3407 reply->tx_packets = htonll(stats->tx_packets);
3408 reply->tx_errors = htonll(stats->tx_errors);
3412 handle_queue_stats_dump_cb(uint32_t queue_id,
3413 struct netdev_queue_stats *stats,
3416 struct queue_stats_cbdata *cbdata = cbdata_;
3418 put_queue_stats(cbdata, queue_id, stats);
3422 handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
3424 struct queue_stats_cbdata *cbdata)
3426 cbdata->port_no = port_no;
3427 if (queue_id == OFPQ_ALL) {
3428 netdev_dump_queue_stats(port->netdev,
3429 handle_queue_stats_dump_cb, cbdata);
3431 struct netdev_queue_stats stats;
3433 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3434 put_queue_stats(cbdata, queue_id, &stats);
3439 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3440 const struct ofp_stats_request *osr,
3443 struct ofp_queue_stats_request *qsr;
3444 struct queue_stats_cbdata cbdata;
3445 struct ofport *port;
3446 unsigned int port_no;
3449 if (arg_size != sizeof *qsr) {
3450 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3452 qsr = (struct ofp_queue_stats_request *) osr->body;
3454 COVERAGE_INC(ofproto_queue_req);
3456 cbdata.ofconn = ofconn;
3457 cbdata.msg = start_stats_reply(osr, 128);
3459 port_no = ntohs(qsr->port_no);
3460 queue_id = ntohl(qsr->queue_id);
3461 if (port_no == OFPP_ALL) {
3462 PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
3463 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3465 } else if (port_no < ofproto->max_ports) {
3466 port = port_array_get(&ofproto->ports, port_no);
3468 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3471 ofpbuf_delete(cbdata.msg);
3472 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3474 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3480 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3481 struct ofp_header *oh)
3483 struct ofp_stats_request *osr;
3487 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3492 osr = (struct ofp_stats_request *) oh;
3494 switch (ntohs(osr->type)) {
3496 return handle_desc_stats_request(p, ofconn, osr);
3499 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3501 case OFPST_AGGREGATE:
3502 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3505 return handle_table_stats_request(p, ofconn, osr);
3508 return handle_port_stats_request(p, ofconn, osr, arg_size);
3511 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3514 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3517 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3521 static long long int
3522 msec_from_nsec(uint64_t sec, uint32_t nsec)
3524 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3528 update_time(struct ofproto *ofproto, struct rule *rule,
3529 const struct odp_flow_stats *stats)
3531 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3532 if (used > rule->used) {
3534 if (rule->super && used > rule->super->used) {
3535 rule->super->used = used;
3537 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3542 update_stats(struct ofproto *ofproto, struct rule *rule,
3543 const struct odp_flow_stats *stats)
3545 if (stats->n_packets) {
3546 update_time(ofproto, rule, stats);
3547 rule->packet_count += stats->n_packets;
3548 rule->byte_count += stats->n_bytes;
3549 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3553 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3554 * in which no matching flow already exists in the flow table.
3556 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3557 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3558 * code as encoded by ofp_mkerr() on failure.
3560 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3563 add_flow(struct ofproto *p, struct ofconn *ofconn,
3564 const struct ofp_flow_mod *ofm, size_t n_actions)
3566 struct ofpbuf *packet;
3571 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3575 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3577 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3578 ntohs(ofm->priority))) {
3579 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3583 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3584 n_actions, ntohs(ofm->idle_timeout),
3585 ntohs(ofm->hard_timeout), ofm->cookie,
3586 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3587 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3588 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3591 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3592 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3596 in_port = UINT16_MAX;
3599 rule_insert(p, rule, packet, in_port);
3603 static struct rule *
3604 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3609 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3611 return rule_from_cls_rule(classifier_find_rule_exactly(
3612 &p->cls, &flow, wildcards,
3613 ntohs(ofm->priority)));
3617 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3618 struct rule *rule, const struct ofp_flow_mod *ofm)
3620 struct ofpbuf *packet;
3625 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3629 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3635 flow_extract(packet, 0, in_port, &flow);
3636 rule_execute(ofproto, rule, packet, &flow);
3641 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3643 struct modify_flows_cbdata {
3644 struct ofproto *ofproto;
3645 const struct ofp_flow_mod *ofm;
3650 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3651 size_t n_actions, struct rule *);
3652 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3654 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3655 * encoded by ofp_mkerr() on failure.
3657 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3660 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3661 const struct ofp_flow_mod *ofm, size_t n_actions)
3663 struct modify_flows_cbdata cbdata;
3664 struct cls_rule target;
3668 cbdata.n_actions = n_actions;
3669 cbdata.match = NULL;
3671 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3674 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3675 modify_flows_cb, &cbdata);
3677 /* This credits the packet to whichever flow happened to happened to
3678 * match last. That's weird. Maybe we should do a lookup for the
3679 * flow that actually matches the packet? Who knows. */
3680 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3683 return add_flow(p, ofconn, ofm, n_actions);
3687 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3688 * code as encoded by ofp_mkerr() on failure.
3690 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3693 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3694 struct ofp_flow_mod *ofm, size_t n_actions)
3696 struct rule *rule = find_flow_strict(p, ofm);
3697 if (rule && !rule_is_hidden(rule)) {
3698 modify_flow(p, ofm, n_actions, rule);
3699 return send_buffered_packet(p, ofconn, rule, ofm);
3701 return add_flow(p, ofconn, ofm, n_actions);
3705 /* Callback for modify_flows_loose(). */
3707 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3709 struct rule *rule = rule_from_cls_rule(rule_);
3710 struct modify_flows_cbdata *cbdata = cbdata_;
3712 if (!rule_is_hidden(rule)) {
3713 cbdata->match = rule;
3714 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3718 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3719 * been identified as a flow in 'p''s flow table to be modified, by changing
3720 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3721 * ofp_action[] structures). */
3723 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3724 size_t n_actions, struct rule *rule)
3726 size_t actions_len = n_actions * sizeof *rule->actions;
3728 rule->flow_cookie = ofm->cookie;
3730 /* If the actions are the same, do nothing. */
3731 if (n_actions == rule->n_actions
3732 && !memcmp(ofm->actions, rule->actions, actions_len))
3737 /* Replace actions. */
3738 free(rule->actions);
3739 rule->actions = xmemdup(ofm->actions, actions_len);
3740 rule->n_actions = n_actions;
3742 /* Make sure that the datapath gets updated properly. */
3743 if (rule->cr.wc.wildcards) {
3744 COVERAGE_INC(ofproto_mod_wc_flow);
3745 p->need_revalidate = true;
3747 rule_update_actions(p, rule);
3753 /* OFPFC_DELETE implementation. */
3755 struct delete_flows_cbdata {
3756 struct ofproto *ofproto;
3760 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3761 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3763 /* Implements OFPFC_DELETE. */
3765 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3767 struct delete_flows_cbdata cbdata;
3768 struct cls_rule target;
3771 cbdata.out_port = ofm->out_port;
3773 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3776 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3777 delete_flows_cb, &cbdata);
3780 /* Implements OFPFC_DELETE_STRICT. */
3782 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3784 struct rule *rule = find_flow_strict(p, ofm);
3786 delete_flow(p, rule, ofm->out_port);
3790 /* Callback for delete_flows_loose(). */
3792 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3794 struct rule *rule = rule_from_cls_rule(rule_);
3795 struct delete_flows_cbdata *cbdata = cbdata_;
3797 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3800 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3801 * been identified as a flow to delete from 'p''s flow table, by deleting the
3802 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3805 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3806 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3807 * specified 'out_port'. */
3809 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3811 if (rule_is_hidden(rule)) {
3815 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3819 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3820 rule_remove(p, rule);
3824 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3825 struct ofp_flow_mod *ofm)
3827 struct ofp_match orig_match;
3831 error = reject_slave_controller(ofconn, &ofm->header);
3835 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3836 sizeof *ofm->actions, &n_actions);
3841 /* We do not support the emergency flow cache. It will hopefully
3842 * get dropped from OpenFlow in the near future. */
3843 if (ofm->flags & htons(OFPFF_EMERG)) {
3844 /* There isn't a good fit for an error code, so just state that the
3845 * flow table is full. */
3846 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3849 /* Normalize ofp->match. If normalization actually changes anything, then
3850 * log the differences. */
3851 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3852 orig_match = ofm->match;
3853 normalize_match(&ofm->match);
3854 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3855 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3856 if (!VLOG_DROP_INFO(&normal_rl)) {
3857 char *old = ofp_match_to_literal_string(&orig_match);
3858 char *new = ofp_match_to_literal_string(&ofm->match);
3859 VLOG_INFO("%s: normalization changed ofp_match, details:",
3860 rconn_get_name(ofconn->rconn));
3861 VLOG_INFO(" pre: %s", old);
3862 VLOG_INFO("post: %s", new);
3868 if (!ofm->match.wildcards) {
3869 ofm->priority = htons(UINT16_MAX);
3872 error = validate_actions((const union ofp_action *) ofm->actions,
3873 n_actions, p->max_ports);
3878 switch (ntohs(ofm->command)) {
3880 return add_flow(p, ofconn, ofm, n_actions);
3883 return modify_flows_loose(p, ofconn, ofm, n_actions);
3885 case OFPFC_MODIFY_STRICT:
3886 return modify_flow_strict(p, ofconn, ofm, n_actions);
3889 delete_flows_loose(p, ofm);
3892 case OFPFC_DELETE_STRICT:
3893 delete_flow_strict(p, ofm);
3897 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3902 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3906 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3911 p->tun_id_from_cookie = !!msg->set;
3916 handle_role_request(struct ofproto *ofproto,
3917 struct ofconn *ofconn, struct nicira_header *msg)
3919 struct nx_role_request *nrr;
3920 struct nx_role_request *reply;
3924 if (ntohs(msg->header.length) != sizeof *nrr) {
3925 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3926 ntohs(msg->header.length), sizeof *nrr);
3927 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3929 nrr = (struct nx_role_request *) msg;
3931 if (ofconn->type != OFCONN_PRIMARY) {
3932 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3934 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3937 role = ntohl(nrr->role);
3938 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3939 && role != NX_ROLE_SLAVE) {
3940 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3942 /* There's no good error code for this. */
3943 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3946 if (role == NX_ROLE_MASTER) {
3947 struct ofconn *other;
3949 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3950 &ofproto->controllers) {
3951 if (other->role == NX_ROLE_MASTER) {
3952 other->role = NX_ROLE_SLAVE;
3956 ofconn->role = role;
3958 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3960 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3961 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3962 reply->role = htonl(role);
3963 queue_tx(buf, ofconn, ofconn->reply_counter);
3969 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3971 struct ofp_vendor_header *ovh = msg;
3972 struct nicira_header *nh;
3974 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3975 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3976 "(expected at least %zu)",
3977 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3978 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3980 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3981 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3983 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3984 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3985 "(expected at least %zu)",
3986 ntohs(ovh->header.length), sizeof(struct nicira_header));
3987 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3991 switch (ntohl(nh->subtype)) {
3992 case NXT_STATUS_REQUEST:
3993 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3996 case NXT_TUN_ID_FROM_COOKIE:
3997 return handle_tun_id_from_cookie(p, msg);
3999 case NXT_ROLE_REQUEST:
4000 return handle_role_request(p, ofconn, msg);
4003 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4007 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4009 struct ofp_header *ob;
4012 /* Currently, everything executes synchronously, so we can just
4013 * immediately send the barrier reply. */
4014 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4015 queue_tx(buf, ofconn, ofconn->reply_counter);
4020 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4021 struct ofpbuf *ofp_msg)
4023 struct ofp_header *oh = ofp_msg->data;
4026 COVERAGE_INC(ofproto_recv_openflow);
4028 case OFPT_ECHO_REQUEST:
4029 error = handle_echo_request(ofconn, oh);
4032 case OFPT_ECHO_REPLY:
4036 case OFPT_FEATURES_REQUEST:
4037 error = handle_features_request(p, ofconn, oh);
4040 case OFPT_GET_CONFIG_REQUEST:
4041 error = handle_get_config_request(p, ofconn, oh);
4044 case OFPT_SET_CONFIG:
4045 error = handle_set_config(p, ofconn, ofp_msg->data);
4048 case OFPT_PACKET_OUT:
4049 error = handle_packet_out(p, ofconn, ofp_msg->data);
4053 error = handle_port_mod(p, ofconn, oh);
4057 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4060 case OFPT_STATS_REQUEST:
4061 error = handle_stats_request(p, ofconn, oh);
4065 error = handle_vendor(p, ofconn, ofp_msg->data);
4068 case OFPT_BARRIER_REQUEST:
4069 error = handle_barrier_request(ofconn, oh);
4073 if (VLOG_IS_WARN_ENABLED()) {
4074 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4075 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4078 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4083 send_error_oh(ofconn, ofp_msg->data, error);
4088 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4090 struct odp_msg *msg = packet->data;
4092 struct ofpbuf payload;
4095 payload.data = msg + 1;
4096 payload.size = msg->length - sizeof *msg;
4097 flow_extract(&payload, msg->arg, msg->port, &flow);
4099 /* Check with in-band control to see if this packet should be sent
4100 * to the local port regardless of the flow table. */
4101 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4102 union odp_action action;
4104 memset(&action, 0, sizeof(action));
4105 action.output.type = ODPAT_OUTPUT;
4106 action.output.port = ODPP_LOCAL;
4107 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
4110 rule = lookup_valid_rule(p, &flow);
4112 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4113 struct ofport *port = port_array_get(&p->ports, msg->port);
4115 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4116 COVERAGE_INC(ofproto_no_packet_in);
4117 /* XXX install 'drop' flow entry */
4118 ofpbuf_delete(packet);
4122 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4125 COVERAGE_INC(ofproto_packet_in);
4126 send_packet_in(p, packet);
4130 if (rule->cr.wc.wildcards) {
4131 rule = rule_create_subrule(p, rule, &flow);
4132 rule_make_actions(p, rule, packet);
4134 if (!rule->may_install) {
4135 /* The rule is not installable, that is, we need to process every
4136 * packet, so process the current packet and set its actions into
4138 rule_make_actions(p, rule, packet);
4140 /* XXX revalidate rule if it needs it */
4144 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4146 * Extra-special case for fail-open mode.
4148 * We are in fail-open mode and the packet matched the fail-open rule,
4149 * but we are connected to a controller too. We should send the packet
4150 * up to the controller in the hope that it will try to set up a flow
4151 * and thereby allow us to exit fail-open.
4153 * See the top-level comment in fail-open.c for more information.
4155 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4156 DPIF_RECV_MSG_PADDING));
4159 ofpbuf_pull(packet, sizeof *msg);
4160 rule_execute(p, rule, packet, &flow);
4161 rule_reinstall(p, rule);
4165 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4167 struct odp_msg *msg = packet->data;
4169 switch (msg->type) {
4170 case _ODPL_ACTION_NR:
4171 COVERAGE_INC(ofproto_ctlr_action);
4172 send_packet_in(p, packet);
4175 case _ODPL_SFLOW_NR:
4177 ofproto_sflow_received(p->sflow, msg);
4179 ofpbuf_delete(packet);
4183 handle_odp_miss_msg(p, packet);
4187 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4194 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4196 struct rule *sub = rule_from_cls_rule(sub_);
4197 struct revalidate_cbdata *cbdata = cbdata_;
4199 if (cbdata->revalidate_all
4200 || (cbdata->revalidate_subrules && sub->super)
4201 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4202 revalidate_rule(cbdata->ofproto, sub);
4207 revalidate_rule(struct ofproto *p, struct rule *rule)
4209 const flow_t *flow = &rule->cr.flow;
4211 COVERAGE_INC(ofproto_revalidate_rule);
4214 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4216 rule_remove(p, rule);
4218 } else if (super != rule->super) {
4219 COVERAGE_INC(ofproto_revalidate_moved);
4220 list_remove(&rule->list);
4221 list_push_back(&super->list, &rule->list);
4222 rule->super = super;
4223 rule->hard_timeout = super->hard_timeout;
4224 rule->idle_timeout = super->idle_timeout;
4225 rule->created = super->created;
4230 rule_update_actions(p, rule);
4234 static struct ofpbuf *
4235 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4236 long long int now, uint8_t reason)
4238 struct ofp_flow_removed *ofr;
4240 long long int tdiff = now - rule->created;
4241 uint32_t sec = tdiff / 1000;
4242 uint32_t msec = tdiff - (sec * 1000);
4244 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4245 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4247 ofr->cookie = rule->flow_cookie;
4248 ofr->priority = htons(rule->cr.priority);
4249 ofr->reason = reason;
4250 ofr->duration_sec = htonl(sec);
4251 ofr->duration_nsec = htonl(msec * 1000000);
4252 ofr->idle_timeout = htons(rule->idle_timeout);
4253 ofr->packet_count = htonll(rule->packet_count);
4254 ofr->byte_count = htonll(rule->byte_count);
4260 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
4262 assert(rule->installed);
4263 assert(!rule->cr.wc.wildcards);
4266 rule_remove(ofproto, rule);
4268 rule_uninstall(ofproto, rule);
4273 send_flow_removed(struct ofproto *p, struct rule *rule,
4274 long long int now, uint8_t reason)
4276 struct ofconn *ofconn;
4277 struct ofconn *prev;
4278 struct ofpbuf *buf = NULL;
4280 /* We limit the maximum number of queued flow expirations it by accounting
4281 * them under the counter for replies. That works because preventing
4282 * OpenFlow requests from being processed also prevents new flows from
4283 * being added (and expiring). (It also prevents processing OpenFlow
4284 * requests that would not add new flows, so it is imperfect.) */
4287 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4288 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
4289 && ofconn_receives_async_msgs(ofconn)) {
4291 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4293 buf = compose_flow_removed(p, rule, now, reason);
4299 queue_tx(buf, prev, prev->reply_counter);
4305 expire_rule(struct cls_rule *cls_rule, void *p_)
4307 struct ofproto *p = p_;
4308 struct rule *rule = rule_from_cls_rule(cls_rule);
4309 long long int hard_expire, idle_expire, expire, now;
4311 hard_expire = (rule->hard_timeout
4312 ? rule->created + rule->hard_timeout * 1000
4314 idle_expire = (rule->idle_timeout
4315 && (rule->super || list_is_empty(&rule->list))
4316 ? rule->used + rule->idle_timeout * 1000
4318 expire = MIN(hard_expire, idle_expire);
4322 if (rule->installed && now >= rule->used + 5000) {
4323 uninstall_idle_flow(p, rule);
4324 } else if (!rule->cr.wc.wildcards) {
4325 active_timeout(p, rule);
4331 COVERAGE_INC(ofproto_expired);
4333 /* Update stats. This code will be a no-op if the rule expired
4334 * due to an idle timeout. */
4335 if (rule->cr.wc.wildcards) {
4336 struct rule *subrule, *next;
4337 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
4338 rule_remove(p, subrule);
4341 rule_uninstall(p, rule);
4344 if (!rule_is_hidden(rule)) {
4345 send_flow_removed(p, rule, now,
4347 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4349 rule_remove(p, rule);
4353 active_timeout(struct ofproto *ofproto, struct rule *rule)
4355 if (ofproto->netflow && !is_controller_rule(rule) &&
4356 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4357 struct ofexpired expired;
4358 struct odp_flow odp_flow;
4360 /* Get updated flow stats. */
4361 memset(&odp_flow, 0, sizeof odp_flow);
4362 if (rule->installed) {
4363 odp_flow.key = rule->cr.flow;
4364 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4365 dpif_flow_get(ofproto->dpif, &odp_flow);
4367 if (odp_flow.stats.n_packets) {
4368 update_time(ofproto, rule, &odp_flow.stats);
4369 netflow_flow_update_flags(&rule->nf_flow,
4370 odp_flow.stats.tcp_flags);
4374 expired.flow = rule->cr.flow;
4375 expired.packet_count = rule->packet_count +
4376 odp_flow.stats.n_packets;
4377 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4378 expired.used = rule->used;
4380 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4382 /* Schedule us to send the accumulated records once we have
4383 * collected all of them. */
4384 poll_immediate_wake();
4389 update_used(struct ofproto *p)
4391 struct odp_flow *flows;
4396 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4401 for (i = 0; i < n_flows; i++) {
4402 struct odp_flow *f = &flows[i];
4405 rule = rule_from_cls_rule(
4406 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4407 if (!rule || !rule->installed) {
4408 COVERAGE_INC(ofproto_unexpected_rule);
4409 dpif_flow_del(p->dpif, f);
4413 update_time(p, rule, &f->stats);
4414 rule_account(p, rule, f->stats.n_bytes);
4419 /* pinsched callback for sending 'packet' on 'ofconn'. */
4421 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4423 struct ofconn *ofconn = ofconn_;
4425 rconn_send_with_limit(ofconn->rconn, packet,
4426 ofconn->packet_in_counter, 100);
4429 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4430 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4431 * packet scheduler for sending.
4433 * 'max_len' specifies the maximum number of bytes of the packet to send on
4434 * 'ofconn' (INT_MAX specifies no limit).
4436 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4437 * ownership is transferred to this function. */
4439 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4442 struct ofproto *ofproto = ofconn->ofproto;
4443 struct ofp_packet_in *opi = packet->data;
4444 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4445 int send_len, trim_size;
4449 if (opi->reason == OFPR_ACTION) {
4450 buffer_id = UINT32_MAX;
4451 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4452 buffer_id = pktbuf_get_null();
4453 } else if (!ofconn->pktbuf) {
4454 buffer_id = UINT32_MAX;
4456 struct ofpbuf payload;
4457 payload.data = opi->data;
4458 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4459 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4462 /* Figure out how much of the packet to send. */
4463 send_len = ntohs(opi->total_len);
4464 if (buffer_id != UINT32_MAX) {
4465 send_len = MIN(send_len, ofconn->miss_send_len);
4467 send_len = MIN(send_len, max_len);
4469 /* Adjust packet length and clone if necessary. */
4470 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4472 packet = ofpbuf_clone_data(packet->data, trim_size);
4475 packet->size = trim_size;
4478 /* Update packet headers. */
4479 opi->buffer_id = htonl(buffer_id);
4480 update_openflow_length(packet);
4482 /* Hand over to packet scheduler. It might immediately call into
4483 * do_send_packet_in() or it might buffer it for a while (until a later
4484 * call to pinsched_run()). */
4485 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4486 packet, do_send_packet_in, ofconn);
4489 /* Replace struct odp_msg header in 'packet' by equivalent struct
4490 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4491 * returned by dpif_recv()).
4493 * The conversion is not complete: the caller still needs to trim any unneeded
4494 * payload off the end of the buffer, set the length in the OpenFlow header,
4495 * and set buffer_id. Those require us to know the controller settings and so
4496 * must be done on a per-controller basis.
4498 * Returns the maximum number of bytes of the packet that should be sent to
4499 * the controller (INT_MAX if no limit). */
4501 do_convert_to_packet_in(struct ofpbuf *packet)
4503 struct odp_msg *msg = packet->data;
4504 struct ofp_packet_in *opi;
4510 /* Extract relevant header fields */
4511 if (msg->type == _ODPL_ACTION_NR) {
4512 reason = OFPR_ACTION;
4515 reason = OFPR_NO_MATCH;
4518 total_len = msg->length - sizeof *msg;
4519 in_port = odp_port_to_ofp_port(msg->port);
4521 /* Repurpose packet buffer by overwriting header. */
4522 ofpbuf_pull(packet, sizeof(struct odp_msg));
4523 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4524 opi->header.version = OFP_VERSION;
4525 opi->header.type = OFPT_PACKET_IN;
4526 opi->total_len = htons(total_len);
4527 opi->in_port = htons(in_port);
4528 opi->reason = reason;
4533 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4534 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4535 * as necessary according to their individual configurations.
4537 * 'packet' must have sufficient headroom to convert it into a struct
4538 * ofp_packet_in (e.g. as returned by dpif_recv()).
4540 * Takes ownership of 'packet'. */
4542 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4544 struct ofconn *ofconn, *prev;
4547 max_len = do_convert_to_packet_in(packet);
4550 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4551 if (ofconn_receives_async_msgs(ofconn)) {
4553 schedule_packet_in(prev, packet, max_len, true);
4559 schedule_packet_in(prev, packet, max_len, false);
4561 ofpbuf_delete(packet);
4566 pick_datapath_id(const struct ofproto *ofproto)
4568 const struct ofport *port;
4570 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4572 uint8_t ea[ETH_ADDR_LEN];
4575 error = netdev_get_etheraddr(port->netdev, ea);
4577 return eth_addr_to_uint64(ea);
4579 VLOG_WARN("could not get MAC address for %s (%s)",
4580 netdev_get_name(port->netdev), strerror(error));
4582 return ofproto->fallback_dpid;
4586 pick_fallback_dpid(void)
4588 uint8_t ea[ETH_ADDR_LEN];
4589 eth_addr_nicira_random(ea);
4590 return eth_addr_to_uint64(ea);
4594 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4595 struct odp_actions *actions, tag_type *tags,
4596 uint16_t *nf_output_iface, void *ofproto_)
4598 struct ofproto *ofproto = ofproto_;
4601 /* Drop frames for reserved multicast addresses. */
4602 if (eth_addr_is_reserved(flow->dl_dst)) {
4606 /* Learn source MAC (but don't try to learn from revalidation). */
4607 if (packet != NULL) {
4608 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4610 GRAT_ARP_LOCK_NONE);
4612 /* The log messages here could actually be useful in debugging,
4613 * so keep the rate limit relatively high. */
4614 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4615 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4616 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4617 ofproto_revalidate(ofproto, rev_tag);
4621 /* Determine output port. */
4622 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4625 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4626 } else if (out_port != flow->in_port) {
4627 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4628 *nf_output_iface = out_port;
4636 static const struct ofhooks default_ofhooks = {
4638 default_normal_ofhook_cb,