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++) {
1076 error = dpif_recv(p->dpif, &buf);
1078 if (error == ENODEV) {
1079 /* Someone destroyed the datapath behind our back. The caller
1080 * better destroy us and give up, because we're just going to
1081 * spin from here on out. */
1082 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1083 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1084 dpif_name(p->dpif));
1090 handle_odp_msg(p, buf);
1093 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1094 process_port_change(p, error, devname);
1096 while ((error = netdev_monitor_poll(p->netdev_monitor,
1097 &devname)) != EAGAIN) {
1098 process_port_change(p, error, devname);
1102 if (time_msec() >= p->next_in_band_update) {
1103 update_in_band_remotes(p);
1105 in_band_run(p->in_band);
1108 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1110 ofconn_run(ofconn, p);
1113 /* Fail-open maintenance. Do this after processing the ofconns since
1114 * fail-open checks the status of the controller rconn. */
1116 fail_open_run(p->fail_open);
1119 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1120 struct vconn *vconn;
1123 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1125 struct ofconn *ofconn;
1126 struct rconn *rconn;
1129 rconn = rconn_create(ofservice->probe_interval, 0);
1130 name = ofconn_make_name(p, vconn_get_name(vconn));
1131 rconn_connect_unreliably(rconn, vconn, name);
1134 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1135 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1136 ofservice->burst_limit);
1137 } else if (retval != EAGAIN) {
1138 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1142 for (i = 0; i < p->n_snoops; i++) {
1143 struct vconn *vconn;
1146 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1148 add_snooper(p, vconn);
1149 } else if (retval != EAGAIN) {
1150 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1154 if (time_msec() >= p->next_expiration) {
1155 COVERAGE_INC(ofproto_expiration);
1156 p->next_expiration = time_msec() + 1000;
1159 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1161 /* Let the hook know that we're at a stable point: all outstanding data
1162 * in existing flows has been accounted to the account_cb. Thus, the
1163 * hook can now reasonably do operations that depend on having accurate
1164 * flow volume accounting (currently, that's just bond rebalancing). */
1165 if (p->ofhooks->account_checkpoint_cb) {
1166 p->ofhooks->account_checkpoint_cb(p->aux);
1171 netflow_run(p->netflow);
1174 ofproto_sflow_run(p->sflow);
1180 struct revalidate_cbdata {
1181 struct ofproto *ofproto;
1182 bool revalidate_all; /* Revalidate all exact-match rules? */
1183 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1184 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1188 ofproto_run2(struct ofproto *p, bool revalidate_all)
1190 if (p->need_revalidate || revalidate_all
1191 || !tag_set_is_empty(&p->revalidate_set)) {
1192 struct revalidate_cbdata cbdata;
1194 cbdata.revalidate_all = revalidate_all;
1195 cbdata.revalidate_subrules = p->need_revalidate;
1196 cbdata.revalidate_set = p->revalidate_set;
1197 tag_set_init(&p->revalidate_set);
1198 COVERAGE_INC(ofproto_revalidate);
1199 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1200 p->need_revalidate = false;
1207 ofproto_wait(struct ofproto *p)
1209 struct ofservice *ofservice;
1210 struct ofconn *ofconn;
1213 dpif_recv_wait(p->dpif);
1214 dpif_port_poll_wait(p->dpif);
1215 netdev_monitor_poll_wait(p->netdev_monitor);
1216 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1217 ofconn_wait(ofconn);
1220 poll_timer_wait_until(p->next_in_band_update);
1221 in_band_wait(p->in_band);
1224 fail_open_wait(p->fail_open);
1227 ofproto_sflow_wait(p->sflow);
1229 if (!tag_set_is_empty(&p->revalidate_set)) {
1230 poll_immediate_wake();
1232 if (p->need_revalidate) {
1233 /* Shouldn't happen, but if it does just go around again. */
1234 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1235 poll_immediate_wake();
1236 } else if (p->next_expiration != LLONG_MAX) {
1237 poll_timer_wait_until(p->next_expiration);
1239 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1240 pvconn_wait(ofservice->pvconn);
1242 for (i = 0; i < p->n_snoops; i++) {
1243 pvconn_wait(p->snoops[i]);
1248 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1250 tag_set_add(&ofproto->revalidate_set, tag);
1254 ofproto_get_revalidate_set(struct ofproto *ofproto)
1256 return &ofproto->revalidate_set;
1260 ofproto_is_alive(const struct ofproto *p)
1262 return !hmap_is_empty(&p->controllers);
1266 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1267 const union ofp_action *actions, size_t n_actions,
1268 const struct ofpbuf *packet)
1270 struct odp_actions odp_actions;
1273 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1279 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1281 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1282 odp_actions.n_actions, packet);
1287 ofproto_add_flow(struct ofproto *p,
1288 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1289 const union ofp_action *actions, size_t n_actions,
1293 rule = rule_create(p, NULL, actions, n_actions,
1294 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1296 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1297 rule_insert(p, rule, NULL, 0);
1301 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1302 uint32_t wildcards, unsigned int priority)
1306 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1310 rule_remove(ofproto, rule);
1315 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1317 struct rule *rule = rule_from_cls_rule(rule_);
1318 struct ofproto *ofproto = ofproto_;
1320 /* Mark the flow as not installed, even though it might really be
1321 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1322 * There is no point in uninstalling it individually since we are about to
1323 * blow away all the flows with dpif_flow_flush(). */
1324 rule->installed = false;
1326 rule_remove(ofproto, rule);
1330 ofproto_flush_flows(struct ofproto *ofproto)
1332 COVERAGE_INC(ofproto_flush);
1333 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1334 dpif_flow_flush(ofproto->dpif);
1335 if (ofproto->in_band) {
1336 in_band_flushed(ofproto->in_band);
1338 if (ofproto->fail_open) {
1339 fail_open_flushed(ofproto->fail_open);
1344 reinit_ports(struct ofproto *p)
1346 struct svec devnames;
1347 struct ofport *ofport;
1348 unsigned int port_no;
1349 struct odp_port *odp_ports;
1353 svec_init(&devnames);
1354 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1355 svec_add (&devnames, (char *) ofport->opp.name);
1357 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1358 for (i = 0; i < n_odp_ports; i++) {
1359 svec_add (&devnames, odp_ports[i].devname);
1363 svec_sort_unique(&devnames);
1364 for (i = 0; i < devnames.n; i++) {
1365 update_port(p, devnames.names[i]);
1367 svec_destroy(&devnames);
1371 refresh_port_group(struct ofproto *p, unsigned int group)
1375 struct ofport *port;
1376 unsigned int port_no;
1378 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1380 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1382 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1383 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1384 ports[n_ports++] = port_no;
1387 dpif_port_group_set(p->dpif, group, ports, n_ports);
1394 refresh_port_groups(struct ofproto *p)
1396 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1397 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1399 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1403 static struct ofport *
1404 make_ofport(const struct odp_port *odp_port)
1406 struct netdev_options netdev_options;
1407 enum netdev_flags flags;
1408 struct ofport *ofport;
1409 struct netdev *netdev;
1413 memset(&netdev_options, 0, sizeof netdev_options);
1414 netdev_options.name = odp_port->devname;
1415 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1417 error = netdev_open(&netdev_options, &netdev);
1419 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1420 "cannot be opened (%s)",
1421 odp_port->devname, odp_port->port,
1422 odp_port->devname, strerror(error));
1426 ofport = xmalloc(sizeof *ofport);
1427 ofport->netdev = netdev;
1428 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1429 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1430 memcpy(ofport->opp.name, odp_port->devname,
1431 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1432 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1434 netdev_get_flags(netdev, &flags);
1435 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1437 netdev_get_carrier(netdev, &carrier);
1438 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1440 netdev_get_features(netdev,
1441 &ofport->opp.curr, &ofport->opp.advertised,
1442 &ofport->opp.supported, &ofport->opp.peer);
1447 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1449 if (port_array_get(&p->ports, odp_port->port)) {
1450 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1453 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1454 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1463 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1465 const struct ofp_phy_port *a = &a_->opp;
1466 const struct ofp_phy_port *b = &b_->opp;
1468 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1469 return (a->port_no == b->port_no
1470 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1471 && !strcmp((char *) a->name, (char *) b->name)
1472 && a->state == b->state
1473 && a->config == b->config
1474 && a->curr == b->curr
1475 && a->advertised == b->advertised
1476 && a->supported == b->supported
1477 && a->peer == b->peer);
1481 send_port_status(struct ofproto *p, const struct ofport *ofport,
1484 /* XXX Should limit the number of queued port status change messages. */
1485 struct ofconn *ofconn;
1486 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1487 struct ofp_port_status *ops;
1490 if (!ofconn_receives_async_msgs(ofconn)) {
1494 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1495 ops->reason = reason;
1496 ops->desc = ofport->opp;
1497 hton_ofp_phy_port(&ops->desc);
1498 queue_tx(b, ofconn, NULL);
1500 if (p->ofhooks->port_changed_cb) {
1501 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1506 ofport_install(struct ofproto *p, struct ofport *ofport)
1508 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1509 const char *netdev_name = (const char *) ofport->opp.name;
1511 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1512 port_array_set(&p->ports, odp_port, ofport);
1513 shash_add(&p->port_by_name, netdev_name, ofport);
1515 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1520 ofport_remove(struct ofproto *p, struct ofport *ofport)
1522 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1524 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1525 port_array_delete(&p->ports, odp_port);
1526 shash_delete(&p->port_by_name,
1527 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1529 ofproto_sflow_del_port(p->sflow, odp_port);
1534 ofport_free(struct ofport *ofport)
1537 netdev_close(ofport->netdev);
1543 update_port(struct ofproto *p, const char *devname)
1545 struct odp_port odp_port;
1546 struct ofport *old_ofport;
1547 struct ofport *new_ofport;
1550 COVERAGE_INC(ofproto_update_port);
1552 /* Query the datapath for port information. */
1553 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1555 /* Find the old ofport. */
1556 old_ofport = shash_find_data(&p->port_by_name, devname);
1559 /* There's no port named 'devname' but there might be a port with
1560 * the same port number. This could happen if a port is deleted
1561 * and then a new one added in its place very quickly, or if a port
1562 * is renamed. In the former case we want to send an OFPPR_DELETE
1563 * and an OFPPR_ADD, and in the latter case we want to send a
1564 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1565 * the old port's ifindex against the new port, or perhaps less
1566 * reliably but more portably by comparing the old port's MAC
1567 * against the new port's MAC. However, this code isn't that smart
1568 * and always sends an OFPPR_MODIFY (XXX). */
1569 old_ofport = port_array_get(&p->ports, odp_port.port);
1571 } else if (error != ENOENT && error != ENODEV) {
1572 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1573 "%s", strerror(error));
1577 /* Create a new ofport. */
1578 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1580 /* Eliminate a few pathological cases. */
1581 if (!old_ofport && !new_ofport) {
1583 } else if (old_ofport && new_ofport) {
1584 /* Most of the 'config' bits are OpenFlow soft state, but
1585 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1586 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1587 * leaves the other bits 0.) */
1588 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1590 if (ofport_equal(old_ofport, new_ofport)) {
1591 /* False alarm--no change. */
1592 ofport_free(new_ofport);
1597 /* Now deal with the normal cases. */
1599 ofport_remove(p, old_ofport);
1602 ofport_install(p, new_ofport);
1604 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1605 (!old_ofport ? OFPPR_ADD
1606 : !new_ofport ? OFPPR_DELETE
1608 ofport_free(old_ofport);
1610 /* Update port groups. */
1611 refresh_port_groups(p);
1615 init_ports(struct ofproto *p)
1617 struct odp_port *ports;
1622 error = dpif_port_list(p->dpif, &ports, &n_ports);
1627 for (i = 0; i < n_ports; i++) {
1628 const struct odp_port *odp_port = &ports[i];
1629 if (!ofport_conflicts(p, odp_port)) {
1630 struct ofport *ofport = make_ofport(odp_port);
1632 ofport_install(p, ofport);
1637 refresh_port_groups(p);
1641 static struct ofconn *
1642 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1644 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1645 ofconn->ofproto = p;
1646 list_push_back(&p->all_conns, &ofconn->node);
1647 ofconn->rconn = rconn;
1648 ofconn->type = type;
1649 ofconn->role = NX_ROLE_OTHER;
1650 ofconn->packet_in_counter = rconn_packet_counter_create ();
1651 ofconn->pktbuf = NULL;
1652 ofconn->miss_send_len = 0;
1653 ofconn->reply_counter = rconn_packet_counter_create ();
1658 ofconn_destroy(struct ofconn *ofconn)
1660 if (ofconn->type == OFCONN_PRIMARY) {
1661 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1663 discovery_destroy(ofconn->discovery);
1665 list_remove(&ofconn->node);
1666 switch_status_unregister(ofconn->ss);
1667 rconn_destroy(ofconn->rconn);
1668 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1669 rconn_packet_counter_destroy(ofconn->reply_counter);
1670 pktbuf_destroy(ofconn->pktbuf);
1675 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1680 if (ofconn->discovery) {
1681 char *controller_name;
1682 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1683 discovery_question_connectivity(ofconn->discovery);
1685 if (discovery_run(ofconn->discovery, &controller_name)) {
1686 if (controller_name) {
1687 char *ofconn_name = ofconn_make_name(p, controller_name);
1688 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1691 rconn_disconnect(ofconn->rconn);
1696 for (i = 0; i < N_SCHEDULERS; i++) {
1697 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1700 rconn_run(ofconn->rconn);
1702 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1703 /* Limit the number of iterations to prevent other tasks from
1705 for (iteration = 0; iteration < 50; iteration++) {
1706 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1711 fail_open_maybe_recover(p->fail_open);
1713 handle_openflow(ofconn, p, of_msg);
1714 ofpbuf_delete(of_msg);
1718 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1719 ofconn_destroy(ofconn);
1724 ofconn_wait(struct ofconn *ofconn)
1728 if (ofconn->discovery) {
1729 discovery_wait(ofconn->discovery);
1731 for (i = 0; i < N_SCHEDULERS; i++) {
1732 pinsched_wait(ofconn->schedulers[i]);
1734 rconn_run_wait(ofconn->rconn);
1735 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1736 rconn_recv_wait(ofconn->rconn);
1738 COVERAGE_INC(ofproto_ofconn_stuck);
1742 /* Returns true if 'ofconn' should receive asynchronous messages. */
1744 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1746 if (ofconn->type == OFCONN_PRIMARY) {
1747 /* Primary controllers always get asynchronous messages unless they
1748 * have configured themselves as "slaves". */
1749 return ofconn->role != NX_ROLE_SLAVE;
1751 /* Service connections don't get asynchronous messages unless they have
1752 * explicitly asked for them by setting a nonzero miss send length. */
1753 return ofconn->miss_send_len > 0;
1757 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1758 * and 'target', suitable for use in log messages for identifying the
1761 * The name is dynamically allocated. The caller should free it (with free())
1762 * when it is no longer needed. */
1764 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1766 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1770 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1774 for (i = 0; i < N_SCHEDULERS; i++) {
1775 struct pinsched **s = &ofconn->schedulers[i];
1779 *s = pinsched_create(rate, burst,
1780 ofconn->ofproto->switch_status);
1782 pinsched_set_limits(*s, rate, burst);
1785 pinsched_destroy(*s);
1792 ofservice_reconfigure(struct ofservice *ofservice,
1793 const struct ofproto_controller *c)
1795 ofservice->probe_interval = c->probe_interval;
1796 ofservice->rate_limit = c->rate_limit;
1797 ofservice->burst_limit = c->burst_limit;
1800 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1801 * positive errno value. */
1803 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1805 struct ofservice *ofservice;
1806 struct pvconn *pvconn;
1809 error = pvconn_open(c->target, &pvconn);
1814 ofservice = xzalloc(sizeof *ofservice);
1815 hmap_insert(&ofproto->services, &ofservice->node,
1816 hash_string(c->target, 0));
1817 ofservice->pvconn = pvconn;
1819 ofservice_reconfigure(ofservice, c);
1825 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1827 hmap_remove(&ofproto->services, &ofservice->node);
1828 pvconn_close(ofservice->pvconn);
1832 /* Finds and returns the ofservice within 'ofproto' that has the given
1833 * 'target', or a null pointer if none exists. */
1834 static struct ofservice *
1835 ofservice_lookup(struct ofproto *ofproto, const char *target)
1837 struct ofservice *ofservice;
1839 HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node,
1840 hash_string(target, 0), &ofproto->services) {
1841 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1848 /* Caller is responsible for initializing the 'cr' member of the returned
1850 static struct rule *
1851 rule_create(struct ofproto *ofproto, struct rule *super,
1852 const union ofp_action *actions, size_t n_actions,
1853 uint16_t idle_timeout, uint16_t hard_timeout,
1854 uint64_t flow_cookie, bool send_flow_removed)
1856 struct rule *rule = xzalloc(sizeof *rule);
1857 rule->idle_timeout = idle_timeout;
1858 rule->hard_timeout = hard_timeout;
1859 rule->flow_cookie = flow_cookie;
1860 rule->used = rule->created = time_msec();
1861 rule->send_flow_removed = send_flow_removed;
1862 rule->super = super;
1864 list_push_back(&super->list, &rule->list);
1866 list_init(&rule->list);
1868 rule->n_actions = n_actions;
1869 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1870 netflow_flow_clear(&rule->nf_flow);
1871 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1876 static struct rule *
1877 rule_from_cls_rule(const struct cls_rule *cls_rule)
1879 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1883 rule_free(struct rule *rule)
1885 free(rule->actions);
1886 free(rule->odp_actions);
1890 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1891 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1892 * through all of its subrules and revalidates them, destroying any that no
1893 * longer has a super-rule (which is probably all of them).
1895 * Before calling this function, the caller must make have removed 'rule' from
1896 * the classifier. If 'rule' is an exact-match rule, the caller is also
1897 * responsible for ensuring that it has been uninstalled from the datapath. */
1899 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1902 struct rule *subrule, *next;
1903 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1904 revalidate_rule(ofproto, subrule);
1907 list_remove(&rule->list);
1913 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1915 const union ofp_action *oa;
1916 struct actions_iterator i;
1918 if (out_port == htons(OFPP_NONE)) {
1921 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1922 oa = actions_next(&i)) {
1923 if (action_outputs_to_port(oa, out_port)) {
1931 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1932 const union odp_action *actions, size_t n_actions,
1933 const struct ofpbuf *packet)
1935 if (n_actions > 0 && actions[0].type == ODPAT_CONTROLLER) {
1936 /* As an optimization, avoid a round-trip from userspace to kernel to
1937 * userspace. This also avoids possibly filling up kernel packet
1938 * buffers along the way. */
1939 struct ofpbuf *copy;
1940 struct odp_msg *msg;
1942 copy = ofpbuf_new(DPIF_RECV_MSG_PADDING + sizeof(struct odp_msg)
1944 ofpbuf_reserve(copy, DPIF_RECV_MSG_PADDING);
1945 msg = ofpbuf_put_uninit(copy, sizeof *msg);
1946 msg->type = _ODPL_ACTION_NR;
1947 msg->length = sizeof(struct odp_msg) + packet->size;
1948 msg->port = in_port;
1950 msg->arg = actions[0].controller.arg;
1951 ofpbuf_put(copy, packet->data, packet->size);
1953 send_packet_in(ofproto, copy);
1959 return !n_actions || !dpif_execute(ofproto->dpif, in_port,
1960 actions, n_actions, 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'.
1966 * The flow that 'packet' actually contains does not need to actually match
1967 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1968 * the packet and byte counters for 'rule' will be credited for the packet sent
1969 * out whether or not the packet actually matches 'rule'.
1971 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1972 * the caller must already have accurately composed ODP actions for it given
1973 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1974 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1975 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1976 * actions and apply them to 'packet'. */
1978 rule_execute(struct ofproto *ofproto, struct rule *rule,
1979 struct ofpbuf *packet, const flow_t *flow)
1981 const union odp_action *actions;
1983 struct odp_actions a;
1985 /* Grab or compose the ODP actions.
1987 * The special case for an exact-match 'rule' where 'flow' is not the
1988 * rule's flow is important to avoid, e.g., sending a packet out its input
1989 * port simply because the ODP actions were composed for the wrong
1991 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1992 struct rule *super = rule->super ? rule->super : rule;
1993 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1994 packet, &a, NULL, 0, NULL)) {
1997 actions = a.actions;
1998 n_actions = a.n_actions;
2000 actions = rule->odp_actions;
2001 n_actions = rule->n_odp_actions;
2004 /* Execute the ODP actions. */
2005 if (execute_odp_actions(ofproto, flow->in_port,
2006 actions, n_actions, packet)) {
2007 struct odp_flow_stats stats;
2008 flow_extract_stats(flow, packet, &stats);
2009 update_stats(ofproto, rule, &stats);
2010 rule->used = time_msec();
2011 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2016 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2019 struct rule *displaced_rule;
2021 /* Insert the rule in the classifier. */
2022 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2023 if (!rule->cr.wc.wildcards) {
2024 rule_make_actions(p, rule, packet);
2027 /* Send the packet and credit it to the rule. */
2030 flow_extract(packet, 0, in_port, &flow);
2031 rule_execute(p, rule, packet, &flow);
2034 /* Install the rule in the datapath only after sending the packet, to
2035 * avoid packet reordering. */
2036 if (rule->cr.wc.wildcards) {
2037 COVERAGE_INC(ofproto_add_wc_flow);
2038 p->need_revalidate = true;
2040 rule_install(p, rule, displaced_rule);
2043 /* Free the rule that was displaced, if any. */
2044 if (displaced_rule) {
2045 rule_destroy(p, displaced_rule);
2049 static struct rule *
2050 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2053 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2054 rule->idle_timeout, rule->hard_timeout,
2056 COVERAGE_INC(ofproto_subrule_create);
2057 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2058 : rule->cr.priority), &subrule->cr);
2059 classifier_insert_exact(&ofproto->cls, &subrule->cr);
2065 rule_remove(struct ofproto *ofproto, struct rule *rule)
2067 if (rule->cr.wc.wildcards) {
2068 COVERAGE_INC(ofproto_del_wc_flow);
2069 ofproto->need_revalidate = true;
2071 rule_uninstall(ofproto, rule);
2073 classifier_remove(&ofproto->cls, &rule->cr);
2074 rule_destroy(ofproto, rule);
2077 /* Returns true if the actions changed, false otherwise. */
2079 rule_make_actions(struct ofproto *p, struct rule *rule,
2080 const struct ofpbuf *packet)
2082 const struct rule *super;
2083 struct odp_actions a;
2086 assert(!rule->cr.wc.wildcards);
2088 super = rule->super ? rule->super : rule;
2090 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2091 packet, &a, &rule->tags, &rule->may_install,
2092 &rule->nf_flow.output_iface);
2094 actions_len = a.n_actions * sizeof *a.actions;
2095 if (rule->n_odp_actions != a.n_actions
2096 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2097 COVERAGE_INC(ofproto_odp_unchanged);
2098 free(rule->odp_actions);
2099 rule->n_odp_actions = a.n_actions;
2100 rule->odp_actions = xmemdup(a.actions, actions_len);
2108 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2109 struct odp_flow_put *put)
2111 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2112 put->flow.key = rule->cr.flow;
2113 put->flow.actions = rule->odp_actions;
2114 put->flow.n_actions = rule->n_odp_actions;
2115 put->flow.flags = 0;
2117 return dpif_flow_put(ofproto->dpif, put);
2121 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2123 assert(!rule->cr.wc.wildcards);
2125 if (rule->may_install) {
2126 struct odp_flow_put put;
2127 if (!do_put_flow(p, rule,
2128 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2130 rule->installed = true;
2131 if (displaced_rule) {
2132 update_stats(p, displaced_rule, &put.flow.stats);
2133 rule_post_uninstall(p, displaced_rule);
2136 } else if (displaced_rule) {
2137 rule_uninstall(p, displaced_rule);
2142 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2144 if (rule->installed) {
2145 struct odp_flow_put put;
2146 COVERAGE_INC(ofproto_dp_missed);
2147 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2149 rule_install(ofproto, rule, NULL);
2154 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2156 bool actions_changed;
2157 uint16_t new_out_iface, old_out_iface;
2159 old_out_iface = rule->nf_flow.output_iface;
2160 actions_changed = rule_make_actions(ofproto, rule, NULL);
2162 if (rule->may_install) {
2163 if (rule->installed) {
2164 if (actions_changed) {
2165 struct odp_flow_put put;
2166 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2167 | ODPPF_ZERO_STATS, &put);
2168 update_stats(ofproto, rule, &put.flow.stats);
2170 /* Temporarily set the old output iface so that NetFlow
2171 * messages have the correct output interface for the old
2173 new_out_iface = rule->nf_flow.output_iface;
2174 rule->nf_flow.output_iface = old_out_iface;
2175 rule_post_uninstall(ofproto, rule);
2176 rule->nf_flow.output_iface = new_out_iface;
2179 rule_install(ofproto, rule, NULL);
2182 rule_uninstall(ofproto, rule);
2187 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2189 uint64_t total_bytes = rule->byte_count + extra_bytes;
2191 if (ofproto->ofhooks->account_flow_cb
2192 && total_bytes > rule->accounted_bytes)
2194 ofproto->ofhooks->account_flow_cb(
2195 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2196 total_bytes - rule->accounted_bytes, ofproto->aux);
2197 rule->accounted_bytes = total_bytes;
2202 rule_uninstall(struct ofproto *p, struct rule *rule)
2204 assert(!rule->cr.wc.wildcards);
2205 if (rule->installed) {
2206 struct odp_flow odp_flow;
2208 odp_flow.key = rule->cr.flow;
2209 odp_flow.actions = NULL;
2210 odp_flow.n_actions = 0;
2212 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2213 update_stats(p, rule, &odp_flow.stats);
2215 rule->installed = false;
2217 rule_post_uninstall(p, rule);
2222 is_controller_rule(struct rule *rule)
2224 /* If the only action is send to the controller then don't report
2225 * NetFlow expiration messages since it is just part of the control
2226 * logic for the network and not real traffic. */
2230 && rule->super->n_actions == 1
2231 && action_outputs_to_port(&rule->super->actions[0],
2232 htons(OFPP_CONTROLLER)));
2236 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2238 struct rule *super = rule->super;
2240 rule_account(ofproto, rule, 0);
2242 if (ofproto->netflow && !is_controller_rule(rule)) {
2243 struct ofexpired expired;
2244 expired.flow = rule->cr.flow;
2245 expired.packet_count = rule->packet_count;
2246 expired.byte_count = rule->byte_count;
2247 expired.used = rule->used;
2248 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2251 super->packet_count += rule->packet_count;
2252 super->byte_count += rule->byte_count;
2254 /* Reset counters to prevent double counting if the rule ever gets
2256 rule->packet_count = 0;
2257 rule->byte_count = 0;
2258 rule->accounted_bytes = 0;
2260 netflow_flow_clear(&rule->nf_flow);
2265 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2266 struct rconn_packet_counter *counter)
2268 update_openflow_length(msg);
2269 if (rconn_send(ofconn->rconn, msg, counter)) {
2275 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2276 int error, const void *data, size_t len)
2279 struct ofp_error_msg *oem;
2281 if (!(error >> 16)) {
2282 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2287 COVERAGE_INC(ofproto_error);
2288 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2289 oh ? oh->xid : 0, &buf);
2290 oem->type = htons((unsigned int) error >> 16);
2291 oem->code = htons(error & 0xffff);
2292 memcpy(oem->data, data, len);
2293 queue_tx(buf, ofconn, ofconn->reply_counter);
2297 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2300 size_t oh_length = ntohs(oh->length);
2301 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2305 hton_ofp_phy_port(struct ofp_phy_port *opp)
2307 opp->port_no = htons(opp->port_no);
2308 opp->config = htonl(opp->config);
2309 opp->state = htonl(opp->state);
2310 opp->curr = htonl(opp->curr);
2311 opp->advertised = htonl(opp->advertised);
2312 opp->supported = htonl(opp->supported);
2313 opp->peer = htonl(opp->peer);
2317 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2319 struct ofp_header *rq = oh;
2320 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2325 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2326 struct ofp_header *oh)
2328 struct ofp_switch_features *osf;
2330 unsigned int port_no;
2331 struct ofport *port;
2333 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2334 osf->datapath_id = htonll(p->datapath_id);
2335 osf->n_buffers = htonl(pktbuf_capacity());
2337 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2338 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2339 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2340 (1u << OFPAT_SET_VLAN_VID) |
2341 (1u << OFPAT_SET_VLAN_PCP) |
2342 (1u << OFPAT_STRIP_VLAN) |
2343 (1u << OFPAT_SET_DL_SRC) |
2344 (1u << OFPAT_SET_DL_DST) |
2345 (1u << OFPAT_SET_NW_SRC) |
2346 (1u << OFPAT_SET_NW_DST) |
2347 (1u << OFPAT_SET_NW_TOS) |
2348 (1u << OFPAT_SET_TP_SRC) |
2349 (1u << OFPAT_SET_TP_DST) |
2350 (1u << OFPAT_ENQUEUE));
2352 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2353 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2356 queue_tx(buf, ofconn, ofconn->reply_counter);
2361 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2362 struct ofp_header *oh)
2365 struct ofp_switch_config *osc;
2369 /* Figure out flags. */
2370 dpif_get_drop_frags(p->dpif, &drop_frags);
2371 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2374 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2375 osc->flags = htons(flags);
2376 osc->miss_send_len = htons(ofconn->miss_send_len);
2377 queue_tx(buf, ofconn, ofconn->reply_counter);
2383 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2384 struct ofp_switch_config *osc)
2389 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2393 flags = ntohs(osc->flags);
2395 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2396 switch (flags & OFPC_FRAG_MASK) {
2397 case OFPC_FRAG_NORMAL:
2398 dpif_set_drop_frags(p->dpif, false);
2400 case OFPC_FRAG_DROP:
2401 dpif_set_drop_frags(p->dpif, true);
2404 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2410 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2416 add_output_group_action(struct odp_actions *actions, uint16_t group,
2417 uint16_t *nf_output_iface)
2419 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2421 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2422 *nf_output_iface = NF_OUT_FLOOD;
2427 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2429 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2430 a->controller.arg = max_len;
2433 struct action_xlate_ctx {
2435 flow_t flow; /* Flow to which these actions correspond. */
2436 int recurse; /* Recursion level, via xlate_table_action. */
2437 struct ofproto *ofproto;
2438 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2439 * null pointer if we are revalidating
2440 * without a packet to refer to. */
2443 struct odp_actions *out; /* Datapath actions. */
2444 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2445 bool may_set_up_flow; /* True ordinarily; false if the actions must
2446 * be reassessed for every packet. */
2447 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2450 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2451 * flow translation. */
2452 #define MAX_RESUBMIT_RECURSION 8
2454 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2455 struct action_xlate_ctx *ctx);
2458 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2460 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2463 if (ofport->opp.config & OFPPC_NO_FWD) {
2464 /* Forwarding disabled on port. */
2469 * We don't have an ofport record for this port, but it doesn't hurt to
2470 * allow forwarding to it anyhow. Maybe such a port will appear later
2471 * and we're pre-populating the flow table.
2475 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2476 ctx->nf_output_iface = port;
2479 static struct rule *
2480 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2483 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2485 /* The rule we found might not be valid, since we could be in need of
2486 * revalidation. If it is not valid, don't return it. */
2489 && ofproto->need_revalidate
2490 && !revalidate_rule(ofproto, rule)) {
2491 COVERAGE_INC(ofproto_invalidated);
2499 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2501 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2502 uint16_t old_in_port;
2505 /* Look up a flow with 'in_port' as the input port. Then restore the
2506 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2507 * have surprising behavior). */
2508 old_in_port = ctx->flow.in_port;
2509 ctx->flow.in_port = in_port;
2510 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2511 ctx->flow.in_port = old_in_port;
2519 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2523 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2525 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2526 MAX_RESUBMIT_RECURSION);
2531 xlate_output_action__(struct action_xlate_ctx *ctx,
2532 uint16_t port, uint16_t max_len)
2535 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2537 ctx->nf_output_iface = NF_OUT_DROP;
2541 add_output_action(ctx, ctx->flow.in_port);
2544 xlate_table_action(ctx, ctx->flow.in_port);
2547 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2548 ctx->out, ctx->tags,
2549 &ctx->nf_output_iface,
2550 ctx->ofproto->aux)) {
2551 COVERAGE_INC(ofproto_uninstallable);
2552 ctx->may_set_up_flow = false;
2556 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2557 &ctx->nf_output_iface);
2560 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2562 case OFPP_CONTROLLER:
2563 add_controller_action(ctx->out, max_len);
2566 add_output_action(ctx, ODPP_LOCAL);
2569 odp_port = ofp_port_to_odp_port(port);
2570 if (odp_port != ctx->flow.in_port) {
2571 add_output_action(ctx, odp_port);
2576 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2577 ctx->nf_output_iface = NF_OUT_FLOOD;
2578 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2579 ctx->nf_output_iface = prev_nf_output_iface;
2580 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2581 ctx->nf_output_iface != NF_OUT_FLOOD) {
2582 ctx->nf_output_iface = NF_OUT_MULTI;
2587 xlate_output_action(struct action_xlate_ctx *ctx,
2588 const struct ofp_action_output *oao)
2590 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2593 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2594 * optimization, because we're going to add another action that sets the
2595 * priority immediately after, or because there are no actions following the
2598 remove_pop_action(struct action_xlate_ctx *ctx)
2600 size_t n = ctx->out->n_actions;
2601 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2602 ctx->out->n_actions--;
2607 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2608 const struct ofp_action_enqueue *oae)
2610 uint16_t ofp_port, odp_port;
2614 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2617 /* Fall back to ordinary output action. */
2618 xlate_output_action__(ctx, ntohs(oae->port), 0);
2622 /* Figure out ODP output port. */
2623 ofp_port = ntohs(oae->port);
2624 if (ofp_port != OFPP_IN_PORT) {
2625 odp_port = ofp_port_to_odp_port(ofp_port);
2627 odp_port = ctx->flow.in_port;
2630 /* Add ODP actions. */
2631 remove_pop_action(ctx);
2632 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2634 add_output_action(ctx, odp_port);
2635 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2637 /* Update NetFlow output port. */
2638 if (ctx->nf_output_iface == NF_OUT_DROP) {
2639 ctx->nf_output_iface = odp_port;
2640 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2641 ctx->nf_output_iface = NF_OUT_MULTI;
2646 xlate_nicira_action(struct action_xlate_ctx *ctx,
2647 const struct nx_action_header *nah)
2649 const struct nx_action_resubmit *nar;
2650 const struct nx_action_set_tunnel *nast;
2651 union odp_action *oa;
2652 int subtype = ntohs(nah->subtype);
2654 assert(nah->vendor == htonl(NX_VENDOR_ID));
2656 case NXAST_RESUBMIT:
2657 nar = (const struct nx_action_resubmit *) nah;
2658 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2661 case NXAST_SET_TUNNEL:
2662 nast = (const struct nx_action_set_tunnel *) nah;
2663 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2664 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2667 case NXAST_DROP_SPOOFED_ARP:
2668 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2669 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2673 /* If you add a new action here that modifies flow data, don't forget to
2674 * update the flow key in ctx->flow at the same time. */
2677 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2683 do_xlate_actions(const union ofp_action *in, size_t n_in,
2684 struct action_xlate_ctx *ctx)
2686 struct actions_iterator iter;
2687 const union ofp_action *ia;
2688 const struct ofport *port;
2690 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2691 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2692 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2693 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2694 /* Drop this flow. */
2698 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2699 uint16_t type = ntohs(ia->type);
2700 union odp_action *oa;
2704 xlate_output_action(ctx, &ia->output);
2707 case OFPAT_SET_VLAN_VID:
2708 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2709 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2712 case OFPAT_SET_VLAN_PCP:
2713 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2714 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2717 case OFPAT_STRIP_VLAN:
2718 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2719 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2720 ctx->flow.dl_vlan_pcp = 0;
2723 case OFPAT_SET_DL_SRC:
2724 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2725 memcpy(oa->dl_addr.dl_addr,
2726 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2727 memcpy(ctx->flow.dl_src,
2728 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2731 case OFPAT_SET_DL_DST:
2732 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2733 memcpy(oa->dl_addr.dl_addr,
2734 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2735 memcpy(ctx->flow.dl_dst,
2736 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2739 case OFPAT_SET_NW_SRC:
2740 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2741 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2744 case OFPAT_SET_NW_DST:
2745 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2746 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2749 case OFPAT_SET_NW_TOS:
2750 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2751 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2754 case OFPAT_SET_TP_SRC:
2755 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2756 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2759 case OFPAT_SET_TP_DST:
2760 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2761 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2765 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2769 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2773 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2780 xlate_actions(const union ofp_action *in, size_t n_in,
2781 const flow_t *flow, struct ofproto *ofproto,
2782 const struct ofpbuf *packet,
2783 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2784 uint16_t *nf_output_iface)
2786 tag_type no_tags = 0;
2787 struct action_xlate_ctx ctx;
2788 COVERAGE_INC(ofproto_ofp2odp);
2789 odp_actions_init(out);
2792 ctx.ofproto = ofproto;
2793 ctx.packet = packet;
2795 ctx.tags = tags ? tags : &no_tags;
2796 ctx.may_set_up_flow = true;
2797 ctx.nf_output_iface = NF_OUT_DROP;
2798 do_xlate_actions(in, n_in, &ctx);
2799 remove_pop_action(&ctx);
2801 /* Check with in-band control to see if we're allowed to set up this
2803 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2804 ctx.may_set_up_flow = false;
2807 if (may_set_up_flow) {
2808 *may_set_up_flow = ctx.may_set_up_flow;
2810 if (nf_output_iface) {
2811 *nf_output_iface = ctx.nf_output_iface;
2813 if (odp_actions_overflow(out)) {
2814 COVERAGE_INC(odp_overflow);
2815 odp_actions_init(out);
2816 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2821 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2822 * error message code (composed with ofp_mkerr()) for the caller to propagate
2823 * upward. Otherwise, returns 0.
2825 * 'oh' is used to make log messages more informative. */
2827 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2829 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2830 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2833 type_name = ofp_message_type_to_string(oh->type);
2834 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2838 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2845 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2846 struct ofp_header *oh)
2848 struct ofp_packet_out *opo;
2849 struct ofpbuf payload, *buffer;
2850 struct odp_actions actions;
2856 error = reject_slave_controller(ofconn, oh);
2861 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2865 opo = (struct ofp_packet_out *) oh;
2867 COVERAGE_INC(ofproto_packet_out);
2868 if (opo->buffer_id != htonl(UINT32_MAX)) {
2869 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2871 if (error || !buffer) {
2879 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2880 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2881 &flow, p, &payload, &actions, NULL, NULL, NULL);
2886 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2888 ofpbuf_delete(buffer);
2894 update_port_config(struct ofproto *p, struct ofport *port,
2895 uint32_t config, uint32_t mask)
2897 mask &= config ^ port->opp.config;
2898 if (mask & OFPPC_PORT_DOWN) {
2899 if (config & OFPPC_PORT_DOWN) {
2900 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2902 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2905 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2906 if (mask & REVALIDATE_BITS) {
2907 COVERAGE_INC(ofproto_costly_flags);
2908 port->opp.config ^= mask & REVALIDATE_BITS;
2909 p->need_revalidate = true;
2911 #undef REVALIDATE_BITS
2912 if (mask & OFPPC_NO_FLOOD) {
2913 port->opp.config ^= OFPPC_NO_FLOOD;
2914 refresh_port_groups(p);
2916 if (mask & OFPPC_NO_PACKET_IN) {
2917 port->opp.config ^= OFPPC_NO_PACKET_IN;
2922 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2923 struct ofp_header *oh)
2925 const struct ofp_port_mod *opm;
2926 struct ofport *port;
2929 error = reject_slave_controller(ofconn, oh);
2933 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2937 opm = (struct ofp_port_mod *) oh;
2939 port = port_array_get(&p->ports,
2940 ofp_port_to_odp_port(ntohs(opm->port_no)));
2942 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2943 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2944 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2946 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2947 if (opm->advertise) {
2948 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2954 static struct ofpbuf *
2955 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2957 struct ofp_stats_reply *osr;
2960 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2961 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2963 osr->flags = htons(0);
2967 static struct ofpbuf *
2968 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2970 return make_stats_reply(request->header.xid, request->type, body_len);
2974 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2976 struct ofpbuf *msg = *msgp;
2977 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2978 if (nbytes + msg->size > UINT16_MAX) {
2979 struct ofp_stats_reply *reply = msg->data;
2980 reply->flags = htons(OFPSF_REPLY_MORE);
2981 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2982 queue_tx(msg, ofconn, ofconn->reply_counter);
2984 return ofpbuf_put_uninit(*msgp, nbytes);
2988 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2989 struct ofp_stats_request *request)
2991 struct ofp_desc_stats *ods;
2994 msg = start_stats_reply(request, sizeof *ods);
2995 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2996 memset(ods, 0, sizeof *ods);
2997 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2998 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2999 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3000 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3001 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3002 queue_tx(msg, ofconn, ofconn->reply_counter);
3008 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
3010 struct rule *rule = rule_from_cls_rule(cls_rule);
3011 int *n_subrules = n_subrules_;
3019 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3020 struct ofp_stats_request *request)
3022 struct ofp_table_stats *ots;
3024 struct odp_stats dpstats;
3025 int n_exact, n_subrules, n_wild;
3027 msg = start_stats_reply(request, sizeof *ots * 2);
3029 /* Count rules of various kinds. */
3031 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
3032 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3033 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3036 dpif_get_dp_stats(p->dpif, &dpstats);
3037 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3038 memset(ots, 0, sizeof *ots);
3039 ots->table_id = TABLEID_HASH;
3040 strcpy(ots->name, "hash");
3041 ots->wildcards = htonl(0);
3042 ots->max_entries = htonl(dpstats.max_capacity);
3043 ots->active_count = htonl(n_exact);
3044 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3046 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3048 /* Classifier table. */
3049 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3050 memset(ots, 0, sizeof *ots);
3051 ots->table_id = TABLEID_CLASSIFIER;
3052 strcpy(ots->name, "classifier");
3053 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3055 ots->max_entries = htonl(65536);
3056 ots->active_count = htonl(n_wild);
3057 ots->lookup_count = htonll(0); /* XXX */
3058 ots->matched_count = htonll(0); /* XXX */
3060 queue_tx(msg, ofconn, ofconn->reply_counter);
3065 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
3066 struct ofpbuf **msgp)
3068 struct netdev_stats stats;
3069 struct ofp_port_stats *ops;
3071 /* Intentionally ignore return value, since errors will set
3072 * 'stats' to all-1s, which is correct for OpenFlow, and
3073 * netdev_get_stats() will log errors. */
3074 netdev_get_stats(port->netdev, &stats);
3076 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3077 ops->port_no = htons(odp_port_to_ofp_port(port_no));
3078 memset(ops->pad, 0, sizeof ops->pad);
3079 ops->rx_packets = htonll(stats.rx_packets);
3080 ops->tx_packets = htonll(stats.tx_packets);
3081 ops->rx_bytes = htonll(stats.rx_bytes);
3082 ops->tx_bytes = htonll(stats.tx_bytes);
3083 ops->rx_dropped = htonll(stats.rx_dropped);
3084 ops->tx_dropped = htonll(stats.tx_dropped);
3085 ops->rx_errors = htonll(stats.rx_errors);
3086 ops->tx_errors = htonll(stats.tx_errors);
3087 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3088 ops->rx_over_err = htonll(stats.rx_over_errors);
3089 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3090 ops->collisions = htonll(stats.collisions);
3094 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3095 struct ofp_stats_request *osr,
3098 struct ofp_port_stats_request *psr;
3099 struct ofp_port_stats *ops;
3101 struct ofport *port;
3102 unsigned int port_no;
3104 if (arg_size != sizeof *psr) {
3105 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3107 psr = (struct ofp_port_stats_request *) osr->body;
3109 msg = start_stats_reply(osr, sizeof *ops * 16);
3110 if (psr->port_no != htons(OFPP_NONE)) {
3111 port = port_array_get(&p->ports,
3112 ofp_port_to_odp_port(ntohs(psr->port_no)));
3114 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
3117 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
3118 append_port_stat(port, port_no, ofconn, &msg);
3122 queue_tx(msg, ofconn, ofconn->reply_counter);
3126 struct flow_stats_cbdata {
3127 struct ofproto *ofproto;
3128 struct ofconn *ofconn;
3133 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3134 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3135 * returned statistic include statistics for all of 'rule''s subrules. */
3137 query_stats(struct ofproto *p, struct rule *rule,
3138 uint64_t *packet_countp, uint64_t *byte_countp)
3140 uint64_t packet_count, byte_count;
3141 struct rule *subrule;
3142 struct odp_flow *odp_flows;
3145 /* Start from historical data for 'rule' itself that are no longer tracked
3146 * by the datapath. This counts, for example, subrules that have
3148 packet_count = rule->packet_count;
3149 byte_count = rule->byte_count;
3151 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3152 * wildcarded then on all of its subrules.
3154 * Also, add any statistics that are not tracked by the datapath for each
3155 * subrule. This includes, for example, statistics for packets that were
3156 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3158 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3159 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3160 if (rule->cr.wc.wildcards) {
3162 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
3163 odp_flows[i++].key = subrule->cr.flow;
3164 packet_count += subrule->packet_count;
3165 byte_count += subrule->byte_count;
3168 odp_flows[0].key = rule->cr.flow;
3171 /* Fetch up-to-date statistics from the datapath and add them in. */
3172 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3174 for (i = 0; i < n_odp_flows; i++) {
3175 struct odp_flow *odp_flow = &odp_flows[i];
3176 packet_count += odp_flow->stats.n_packets;
3177 byte_count += odp_flow->stats.n_bytes;
3182 /* Return the stats to the caller. */
3183 *packet_countp = packet_count;
3184 *byte_countp = byte_count;
3188 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3190 struct rule *rule = rule_from_cls_rule(rule_);
3191 struct flow_stats_cbdata *cbdata = cbdata_;
3192 struct ofp_flow_stats *ofs;
3193 uint64_t packet_count, byte_count;
3194 size_t act_len, len;
3195 long long int tdiff = time_msec() - rule->created;
3196 uint32_t sec = tdiff / 1000;
3197 uint32_t msec = tdiff - (sec * 1000);
3199 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3203 act_len = sizeof *rule->actions * rule->n_actions;
3204 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3206 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3208 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3209 ofs->length = htons(len);
3210 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3212 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3213 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3214 ofs->duration_sec = htonl(sec);
3215 ofs->duration_nsec = htonl(msec * 1000000);
3216 ofs->cookie = rule->flow_cookie;
3217 ofs->priority = htons(rule->cr.priority);
3218 ofs->idle_timeout = htons(rule->idle_timeout);
3219 ofs->hard_timeout = htons(rule->hard_timeout);
3220 memset(ofs->pad2, 0, sizeof ofs->pad2);
3221 ofs->packet_count = htonll(packet_count);
3222 ofs->byte_count = htonll(byte_count);
3223 memcpy(ofs->actions, rule->actions, act_len);
3227 table_id_to_include(uint8_t table_id)
3229 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3230 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3231 : table_id == 0xff ? CLS_INC_ALL
3236 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3237 const struct ofp_stats_request *osr,
3240 struct ofp_flow_stats_request *fsr;
3241 struct flow_stats_cbdata cbdata;
3242 struct cls_rule target;
3244 if (arg_size != sizeof *fsr) {
3245 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3247 fsr = (struct ofp_flow_stats_request *) osr->body;
3249 COVERAGE_INC(ofproto_flows_req);
3251 cbdata.ofconn = ofconn;
3252 cbdata.out_port = fsr->out_port;
3253 cbdata.msg = start_stats_reply(osr, 1024);
3254 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3255 classifier_for_each_match(&p->cls, &target,
3256 table_id_to_include(fsr->table_id),
3257 flow_stats_cb, &cbdata);
3258 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3262 struct flow_stats_ds_cbdata {
3263 struct ofproto *ofproto;
3268 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3270 struct rule *rule = rule_from_cls_rule(rule_);
3271 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3272 struct ds *results = cbdata->results;
3273 struct ofp_match match;
3274 uint64_t packet_count, byte_count;
3275 size_t act_len = sizeof *rule->actions * rule->n_actions;
3277 /* Don't report on subrules. */
3278 if (rule->super != NULL) {
3282 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3283 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3284 cbdata->ofproto->tun_id_from_cookie, &match);
3286 ds_put_format(results, "duration=%llds, ",
3287 (time_msec() - rule->created) / 1000);
3288 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3289 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3290 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3291 ofp_print_match(results, &match, true);
3292 ofp_print_actions(results, &rule->actions->header, act_len);
3293 ds_put_cstr(results, "\n");
3296 /* Adds a pretty-printed description of all flows to 'results', including
3297 * those marked hidden by secchan (e.g., by in-band control). */
3299 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3301 struct ofp_match match;
3302 struct cls_rule target;
3303 struct flow_stats_ds_cbdata cbdata;
3305 memset(&match, 0, sizeof match);
3306 match.wildcards = htonl(OVSFW_ALL);
3309 cbdata.results = results;
3311 cls_rule_from_match(&match, 0, false, 0, &target);
3312 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3313 flow_stats_ds_cb, &cbdata);
3316 struct aggregate_stats_cbdata {
3317 struct ofproto *ofproto;
3319 uint64_t packet_count;
3320 uint64_t byte_count;
3325 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3327 struct rule *rule = rule_from_cls_rule(rule_);
3328 struct aggregate_stats_cbdata *cbdata = cbdata_;
3329 uint64_t packet_count, byte_count;
3331 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3335 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3337 cbdata->packet_count += packet_count;
3338 cbdata->byte_count += byte_count;
3343 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3344 const struct ofp_stats_request *osr,
3347 struct ofp_aggregate_stats_request *asr;
3348 struct ofp_aggregate_stats_reply *reply;
3349 struct aggregate_stats_cbdata cbdata;
3350 struct cls_rule target;
3353 if (arg_size != sizeof *asr) {
3354 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3356 asr = (struct ofp_aggregate_stats_request *) osr->body;
3358 COVERAGE_INC(ofproto_agg_request);
3360 cbdata.out_port = asr->out_port;
3361 cbdata.packet_count = 0;
3362 cbdata.byte_count = 0;
3364 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3365 classifier_for_each_match(&p->cls, &target,
3366 table_id_to_include(asr->table_id),
3367 aggregate_stats_cb, &cbdata);
3369 msg = start_stats_reply(osr, sizeof *reply);
3370 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3371 reply->flow_count = htonl(cbdata.n_flows);
3372 reply->packet_count = htonll(cbdata.packet_count);
3373 reply->byte_count = htonll(cbdata.byte_count);
3374 queue_tx(msg, ofconn, ofconn->reply_counter);
3378 struct queue_stats_cbdata {
3379 struct ofconn *ofconn;
3385 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3386 const struct netdev_queue_stats *stats)
3388 struct ofp_queue_stats *reply;
3390 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3391 reply->port_no = htons(cbdata->port_no);
3392 memset(reply->pad, 0, sizeof reply->pad);
3393 reply->queue_id = htonl(queue_id);
3394 reply->tx_bytes = htonll(stats->tx_bytes);
3395 reply->tx_packets = htonll(stats->tx_packets);
3396 reply->tx_errors = htonll(stats->tx_errors);
3400 handle_queue_stats_dump_cb(uint32_t queue_id,
3401 struct netdev_queue_stats *stats,
3404 struct queue_stats_cbdata *cbdata = cbdata_;
3406 put_queue_stats(cbdata, queue_id, stats);
3410 handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
3412 struct queue_stats_cbdata *cbdata)
3414 cbdata->port_no = port_no;
3415 if (queue_id == OFPQ_ALL) {
3416 netdev_dump_queue_stats(port->netdev,
3417 handle_queue_stats_dump_cb, cbdata);
3419 struct netdev_queue_stats stats;
3421 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3422 put_queue_stats(cbdata, queue_id, &stats);
3427 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3428 const struct ofp_stats_request *osr,
3431 struct ofp_queue_stats_request *qsr;
3432 struct queue_stats_cbdata cbdata;
3433 struct ofport *port;
3434 unsigned int port_no;
3437 if (arg_size != sizeof *qsr) {
3438 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3440 qsr = (struct ofp_queue_stats_request *) osr->body;
3442 COVERAGE_INC(ofproto_queue_req);
3444 cbdata.ofconn = ofconn;
3445 cbdata.msg = start_stats_reply(osr, 128);
3447 port_no = ntohs(qsr->port_no);
3448 queue_id = ntohl(qsr->queue_id);
3449 if (port_no == OFPP_ALL) {
3450 PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
3451 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3453 } else if (port_no < ofproto->max_ports) {
3454 port = port_array_get(&ofproto->ports, port_no);
3456 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3459 ofpbuf_delete(cbdata.msg);
3460 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3462 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3468 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3469 struct ofp_header *oh)
3471 struct ofp_stats_request *osr;
3475 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3480 osr = (struct ofp_stats_request *) oh;
3482 switch (ntohs(osr->type)) {
3484 return handle_desc_stats_request(p, ofconn, osr);
3487 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3489 case OFPST_AGGREGATE:
3490 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3493 return handle_table_stats_request(p, ofconn, osr);
3496 return handle_port_stats_request(p, ofconn, osr, arg_size);
3499 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3502 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3505 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3509 static long long int
3510 msec_from_nsec(uint64_t sec, uint32_t nsec)
3512 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3516 update_time(struct ofproto *ofproto, struct rule *rule,
3517 const struct odp_flow_stats *stats)
3519 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3520 if (used > rule->used) {
3522 if (rule->super && used > rule->super->used) {
3523 rule->super->used = used;
3525 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3530 update_stats(struct ofproto *ofproto, struct rule *rule,
3531 const struct odp_flow_stats *stats)
3533 if (stats->n_packets) {
3534 update_time(ofproto, rule, stats);
3535 rule->packet_count += stats->n_packets;
3536 rule->byte_count += stats->n_bytes;
3537 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3541 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3542 * in which no matching flow already exists in the flow table.
3544 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3545 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3546 * code as encoded by ofp_mkerr() on failure.
3548 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3551 add_flow(struct ofproto *p, struct ofconn *ofconn,
3552 const struct ofp_flow_mod *ofm, size_t n_actions)
3554 struct ofpbuf *packet;
3559 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3563 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3565 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3566 ntohs(ofm->priority))) {
3567 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3571 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3572 n_actions, ntohs(ofm->idle_timeout),
3573 ntohs(ofm->hard_timeout), ofm->cookie,
3574 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3575 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3576 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3579 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3580 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3584 in_port = UINT16_MAX;
3587 rule_insert(p, rule, packet, in_port);
3588 ofpbuf_delete(packet);
3592 static struct rule *
3593 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3598 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3600 return rule_from_cls_rule(classifier_find_rule_exactly(
3601 &p->cls, &flow, wildcards,
3602 ntohs(ofm->priority)));
3606 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3607 struct rule *rule, const struct ofp_flow_mod *ofm)
3609 struct ofpbuf *packet;
3614 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3618 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3624 flow_extract(packet, 0, in_port, &flow);
3625 rule_execute(ofproto, rule, packet, &flow);
3626 ofpbuf_delete(packet);
3631 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3633 struct modify_flows_cbdata {
3634 struct ofproto *ofproto;
3635 const struct ofp_flow_mod *ofm;
3640 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3641 size_t n_actions, struct rule *);
3642 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3644 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3645 * encoded by ofp_mkerr() on failure.
3647 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3650 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3651 const struct ofp_flow_mod *ofm, size_t n_actions)
3653 struct modify_flows_cbdata cbdata;
3654 struct cls_rule target;
3658 cbdata.n_actions = n_actions;
3659 cbdata.match = NULL;
3661 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3664 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3665 modify_flows_cb, &cbdata);
3667 /* This credits the packet to whichever flow happened to happened to
3668 * match last. That's weird. Maybe we should do a lookup for the
3669 * flow that actually matches the packet? Who knows. */
3670 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3673 return add_flow(p, ofconn, ofm, n_actions);
3677 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3678 * code as encoded by ofp_mkerr() on failure.
3680 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3683 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3684 struct ofp_flow_mod *ofm, size_t n_actions)
3686 struct rule *rule = find_flow_strict(p, ofm);
3687 if (rule && !rule_is_hidden(rule)) {
3688 modify_flow(p, ofm, n_actions, rule);
3689 return send_buffered_packet(p, ofconn, rule, ofm);
3691 return add_flow(p, ofconn, ofm, n_actions);
3695 /* Callback for modify_flows_loose(). */
3697 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3699 struct rule *rule = rule_from_cls_rule(rule_);
3700 struct modify_flows_cbdata *cbdata = cbdata_;
3702 if (!rule_is_hidden(rule)) {
3703 cbdata->match = rule;
3704 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3708 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3709 * been identified as a flow in 'p''s flow table to be modified, by changing
3710 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3711 * ofp_action[] structures). */
3713 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3714 size_t n_actions, struct rule *rule)
3716 size_t actions_len = n_actions * sizeof *rule->actions;
3718 rule->flow_cookie = ofm->cookie;
3720 /* If the actions are the same, do nothing. */
3721 if (n_actions == rule->n_actions
3722 && !memcmp(ofm->actions, rule->actions, actions_len))
3727 /* Replace actions. */
3728 free(rule->actions);
3729 rule->actions = xmemdup(ofm->actions, actions_len);
3730 rule->n_actions = n_actions;
3732 /* Make sure that the datapath gets updated properly. */
3733 if (rule->cr.wc.wildcards) {
3734 COVERAGE_INC(ofproto_mod_wc_flow);
3735 p->need_revalidate = true;
3737 rule_update_actions(p, rule);
3743 /* OFPFC_DELETE implementation. */
3745 struct delete_flows_cbdata {
3746 struct ofproto *ofproto;
3750 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3751 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3753 /* Implements OFPFC_DELETE. */
3755 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3757 struct delete_flows_cbdata cbdata;
3758 struct cls_rule target;
3761 cbdata.out_port = ofm->out_port;
3763 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3766 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3767 delete_flows_cb, &cbdata);
3770 /* Implements OFPFC_DELETE_STRICT. */
3772 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3774 struct rule *rule = find_flow_strict(p, ofm);
3776 delete_flow(p, rule, ofm->out_port);
3780 /* Callback for delete_flows_loose(). */
3782 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3784 struct rule *rule = rule_from_cls_rule(rule_);
3785 struct delete_flows_cbdata *cbdata = cbdata_;
3787 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3790 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3791 * been identified as a flow to delete from 'p''s flow table, by deleting the
3792 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3795 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3796 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3797 * specified 'out_port'. */
3799 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3801 if (rule_is_hidden(rule)) {
3805 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3809 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3810 rule_remove(p, rule);
3814 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3815 struct ofp_flow_mod *ofm)
3817 struct ofp_match orig_match;
3821 error = reject_slave_controller(ofconn, &ofm->header);
3825 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3826 sizeof *ofm->actions, &n_actions);
3831 /* We do not support the emergency flow cache. It will hopefully
3832 * get dropped from OpenFlow in the near future. */
3833 if (ofm->flags & htons(OFPFF_EMERG)) {
3834 /* There isn't a good fit for an error code, so just state that the
3835 * flow table is full. */
3836 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3839 /* Normalize ofp->match. If normalization actually changes anything, then
3840 * log the differences. */
3841 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3842 orig_match = ofm->match;
3843 normalize_match(&ofm->match);
3844 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3845 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3846 if (!VLOG_DROP_INFO(&normal_rl)) {
3847 char *old = ofp_match_to_literal_string(&orig_match);
3848 char *new = ofp_match_to_literal_string(&ofm->match);
3849 VLOG_INFO("%s: normalization changed ofp_match, details:",
3850 rconn_get_name(ofconn->rconn));
3851 VLOG_INFO(" pre: %s", old);
3852 VLOG_INFO("post: %s", new);
3858 if (!ofm->match.wildcards) {
3859 ofm->priority = htons(UINT16_MAX);
3862 error = validate_actions((const union ofp_action *) ofm->actions,
3863 n_actions, p->max_ports);
3868 switch (ntohs(ofm->command)) {
3870 return add_flow(p, ofconn, ofm, n_actions);
3873 return modify_flows_loose(p, ofconn, ofm, n_actions);
3875 case OFPFC_MODIFY_STRICT:
3876 return modify_flow_strict(p, ofconn, ofm, n_actions);
3879 delete_flows_loose(p, ofm);
3882 case OFPFC_DELETE_STRICT:
3883 delete_flow_strict(p, ofm);
3887 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3892 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3896 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3901 p->tun_id_from_cookie = !!msg->set;
3906 handle_role_request(struct ofproto *ofproto,
3907 struct ofconn *ofconn, struct nicira_header *msg)
3909 struct nx_role_request *nrr;
3910 struct nx_role_request *reply;
3914 if (ntohs(msg->header.length) != sizeof *nrr) {
3915 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3916 ntohs(msg->header.length), sizeof *nrr);
3917 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3919 nrr = (struct nx_role_request *) msg;
3921 if (ofconn->type != OFCONN_PRIMARY) {
3922 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3924 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3927 role = ntohl(nrr->role);
3928 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3929 && role != NX_ROLE_SLAVE) {
3930 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3932 /* There's no good error code for this. */
3933 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3936 if (role == NX_ROLE_MASTER) {
3937 struct ofconn *other;
3939 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3940 &ofproto->controllers) {
3941 if (other->role == NX_ROLE_MASTER) {
3942 other->role = NX_ROLE_SLAVE;
3946 ofconn->role = role;
3948 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3950 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3951 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3952 reply->role = htonl(role);
3953 queue_tx(buf, ofconn, ofconn->reply_counter);
3959 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3961 struct ofp_vendor_header *ovh = msg;
3962 struct nicira_header *nh;
3964 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3965 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3966 "(expected at least %zu)",
3967 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3968 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3970 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3971 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3973 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3974 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3975 "(expected at least %zu)",
3976 ntohs(ovh->header.length), sizeof(struct nicira_header));
3977 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3981 switch (ntohl(nh->subtype)) {
3982 case NXT_STATUS_REQUEST:
3983 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3986 case NXT_TUN_ID_FROM_COOKIE:
3987 return handle_tun_id_from_cookie(p, msg);
3989 case NXT_ROLE_REQUEST:
3990 return handle_role_request(p, ofconn, msg);
3993 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3997 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3999 struct ofp_header *ob;
4002 /* Currently, everything executes synchronously, so we can just
4003 * immediately send the barrier reply. */
4004 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4005 queue_tx(buf, ofconn, ofconn->reply_counter);
4010 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4011 struct ofpbuf *ofp_msg)
4013 struct ofp_header *oh = ofp_msg->data;
4016 COVERAGE_INC(ofproto_recv_openflow);
4018 case OFPT_ECHO_REQUEST:
4019 error = handle_echo_request(ofconn, oh);
4022 case OFPT_ECHO_REPLY:
4026 case OFPT_FEATURES_REQUEST:
4027 error = handle_features_request(p, ofconn, oh);
4030 case OFPT_GET_CONFIG_REQUEST:
4031 error = handle_get_config_request(p, ofconn, oh);
4034 case OFPT_SET_CONFIG:
4035 error = handle_set_config(p, ofconn, ofp_msg->data);
4038 case OFPT_PACKET_OUT:
4039 error = handle_packet_out(p, ofconn, ofp_msg->data);
4043 error = handle_port_mod(p, ofconn, oh);
4047 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4050 case OFPT_STATS_REQUEST:
4051 error = handle_stats_request(p, ofconn, oh);
4055 error = handle_vendor(p, ofconn, ofp_msg->data);
4058 case OFPT_BARRIER_REQUEST:
4059 error = handle_barrier_request(ofconn, oh);
4063 if (VLOG_IS_WARN_ENABLED()) {
4064 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4065 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4068 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4073 send_error_oh(ofconn, ofp_msg->data, error);
4078 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4080 struct odp_msg *msg = packet->data;
4082 struct ofpbuf payload;
4085 payload.data = msg + 1;
4086 payload.size = msg->length - sizeof *msg;
4087 flow_extract(&payload, msg->arg, msg->port, &flow);
4089 /* Check with in-band control to see if this packet should be sent
4090 * to the local port regardless of the flow table. */
4091 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4092 union odp_action action;
4094 memset(&action, 0, sizeof(action));
4095 action.output.type = ODPAT_OUTPUT;
4096 action.output.port = ODPP_LOCAL;
4097 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
4100 rule = lookup_valid_rule(p, &flow);
4102 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4103 struct ofport *port = port_array_get(&p->ports, msg->port);
4105 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4106 COVERAGE_INC(ofproto_no_packet_in);
4107 /* XXX install 'drop' flow entry */
4108 ofpbuf_delete(packet);
4112 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4115 COVERAGE_INC(ofproto_packet_in);
4116 send_packet_in(p, packet);
4120 if (rule->cr.wc.wildcards) {
4121 rule = rule_create_subrule(p, rule, &flow);
4122 rule_make_actions(p, rule, packet);
4124 if (!rule->may_install) {
4125 /* The rule is not installable, that is, we need to process every
4126 * packet, so process the current packet and set its actions into
4128 rule_make_actions(p, rule, packet);
4130 /* XXX revalidate rule if it needs it */
4134 rule_execute(p, rule, &payload, &flow);
4135 rule_reinstall(p, rule);
4137 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4139 * Extra-special case for fail-open mode.
4141 * We are in fail-open mode and the packet matched the fail-open rule,
4142 * but we are connected to a controller too. We should send the packet
4143 * up to the controller in the hope that it will try to set up a flow
4144 * and thereby allow us to exit fail-open.
4146 * See the top-level comment in fail-open.c for more information.
4148 send_packet_in(p, packet);
4150 ofpbuf_delete(packet);
4155 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4157 struct odp_msg *msg = packet->data;
4159 switch (msg->type) {
4160 case _ODPL_ACTION_NR:
4161 COVERAGE_INC(ofproto_ctlr_action);
4162 send_packet_in(p, packet);
4165 case _ODPL_SFLOW_NR:
4167 ofproto_sflow_received(p->sflow, msg);
4169 ofpbuf_delete(packet);
4173 handle_odp_miss_msg(p, packet);
4177 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4184 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4186 struct rule *sub = rule_from_cls_rule(sub_);
4187 struct revalidate_cbdata *cbdata = cbdata_;
4189 if (cbdata->revalidate_all
4190 || (cbdata->revalidate_subrules && sub->super)
4191 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4192 revalidate_rule(cbdata->ofproto, sub);
4197 revalidate_rule(struct ofproto *p, struct rule *rule)
4199 const flow_t *flow = &rule->cr.flow;
4201 COVERAGE_INC(ofproto_revalidate_rule);
4204 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4206 rule_remove(p, rule);
4208 } else if (super != rule->super) {
4209 COVERAGE_INC(ofproto_revalidate_moved);
4210 list_remove(&rule->list);
4211 list_push_back(&super->list, &rule->list);
4212 rule->super = super;
4213 rule->hard_timeout = super->hard_timeout;
4214 rule->idle_timeout = super->idle_timeout;
4215 rule->created = super->created;
4220 rule_update_actions(p, rule);
4224 static struct ofpbuf *
4225 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4226 long long int now, uint8_t reason)
4228 struct ofp_flow_removed *ofr;
4230 long long int tdiff = now - rule->created;
4231 uint32_t sec = tdiff / 1000;
4232 uint32_t msec = tdiff - (sec * 1000);
4234 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4235 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4237 ofr->cookie = rule->flow_cookie;
4238 ofr->priority = htons(rule->cr.priority);
4239 ofr->reason = reason;
4240 ofr->duration_sec = htonl(sec);
4241 ofr->duration_nsec = htonl(msec * 1000000);
4242 ofr->idle_timeout = htons(rule->idle_timeout);
4243 ofr->packet_count = htonll(rule->packet_count);
4244 ofr->byte_count = htonll(rule->byte_count);
4250 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
4252 assert(rule->installed);
4253 assert(!rule->cr.wc.wildcards);
4256 rule_remove(ofproto, rule);
4258 rule_uninstall(ofproto, rule);
4263 send_flow_removed(struct ofproto *p, struct rule *rule,
4264 long long int now, uint8_t reason)
4266 struct ofconn *ofconn;
4267 struct ofconn *prev;
4268 struct ofpbuf *buf = NULL;
4270 /* We limit the maximum number of queued flow expirations it by accounting
4271 * them under the counter for replies. That works because preventing
4272 * OpenFlow requests from being processed also prevents new flows from
4273 * being added (and expiring). (It also prevents processing OpenFlow
4274 * requests that would not add new flows, so it is imperfect.) */
4277 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4278 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
4279 && ofconn_receives_async_msgs(ofconn)) {
4281 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4283 buf = compose_flow_removed(p, rule, now, reason);
4289 queue_tx(buf, prev, prev->reply_counter);
4295 expire_rule(struct cls_rule *cls_rule, void *p_)
4297 struct ofproto *p = p_;
4298 struct rule *rule = rule_from_cls_rule(cls_rule);
4299 long long int hard_expire, idle_expire, expire, now;
4301 hard_expire = (rule->hard_timeout
4302 ? rule->created + rule->hard_timeout * 1000
4304 idle_expire = (rule->idle_timeout
4305 && (rule->super || list_is_empty(&rule->list))
4306 ? rule->used + rule->idle_timeout * 1000
4308 expire = MIN(hard_expire, idle_expire);
4312 if (rule->installed && now >= rule->used + 5000) {
4313 uninstall_idle_flow(p, rule);
4314 } else if (!rule->cr.wc.wildcards) {
4315 active_timeout(p, rule);
4321 COVERAGE_INC(ofproto_expired);
4323 /* Update stats. This code will be a no-op if the rule expired
4324 * due to an idle timeout. */
4325 if (rule->cr.wc.wildcards) {
4326 struct rule *subrule, *next;
4327 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
4328 rule_remove(p, subrule);
4331 rule_uninstall(p, rule);
4334 if (!rule_is_hidden(rule)) {
4335 send_flow_removed(p, rule, now,
4337 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4339 rule_remove(p, rule);
4343 active_timeout(struct ofproto *ofproto, struct rule *rule)
4345 if (ofproto->netflow && !is_controller_rule(rule) &&
4346 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4347 struct ofexpired expired;
4348 struct odp_flow odp_flow;
4350 /* Get updated flow stats. */
4351 memset(&odp_flow, 0, sizeof odp_flow);
4352 if (rule->installed) {
4353 odp_flow.key = rule->cr.flow;
4354 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4355 dpif_flow_get(ofproto->dpif, &odp_flow);
4357 if (odp_flow.stats.n_packets) {
4358 update_time(ofproto, rule, &odp_flow.stats);
4359 netflow_flow_update_flags(&rule->nf_flow,
4360 odp_flow.stats.tcp_flags);
4364 expired.flow = rule->cr.flow;
4365 expired.packet_count = rule->packet_count +
4366 odp_flow.stats.n_packets;
4367 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4368 expired.used = rule->used;
4370 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4372 /* Schedule us to send the accumulated records once we have
4373 * collected all of them. */
4374 poll_immediate_wake();
4379 update_used(struct ofproto *p)
4381 struct odp_flow *flows;
4386 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4391 for (i = 0; i < n_flows; i++) {
4392 struct odp_flow *f = &flows[i];
4395 rule = rule_from_cls_rule(
4396 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4397 if (!rule || !rule->installed) {
4398 COVERAGE_INC(ofproto_unexpected_rule);
4399 dpif_flow_del(p->dpif, f);
4403 update_time(p, rule, &f->stats);
4404 rule_account(p, rule, f->stats.n_bytes);
4409 /* pinsched callback for sending 'packet' on 'ofconn'. */
4411 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4413 struct ofconn *ofconn = ofconn_;
4415 rconn_send_with_limit(ofconn->rconn, packet,
4416 ofconn->packet_in_counter, 100);
4419 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4420 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4421 * packet scheduler for sending.
4423 * 'max_len' specifies the maximum number of bytes of the packet to send on
4424 * 'ofconn' (INT_MAX specifies no limit).
4426 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4427 * ownership is transferred to this function. */
4429 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4432 struct ofproto *ofproto = ofconn->ofproto;
4433 struct ofp_packet_in *opi = packet->data;
4434 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4435 int send_len, trim_size;
4439 if (opi->reason == OFPR_ACTION) {
4440 buffer_id = UINT32_MAX;
4441 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4442 buffer_id = pktbuf_get_null();
4443 } else if (!ofconn->pktbuf) {
4444 buffer_id = UINT32_MAX;
4446 struct ofpbuf payload;
4447 payload.data = opi->data;
4448 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4449 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4452 /* Figure out how much of the packet to send. */
4453 send_len = ntohs(opi->total_len);
4454 if (buffer_id != UINT32_MAX) {
4455 send_len = MIN(send_len, ofconn->miss_send_len);
4457 send_len = MIN(send_len, max_len);
4459 /* Adjust packet length and clone if necessary. */
4460 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4462 packet = ofpbuf_clone_data(packet->data, trim_size);
4465 packet->size = trim_size;
4468 /* Update packet headers. */
4469 opi->buffer_id = htonl(buffer_id);
4470 update_openflow_length(packet);
4472 /* Hand over to packet scheduler. It might immediately call into
4473 * do_send_packet_in() or it might buffer it for a while (until a later
4474 * call to pinsched_run()). */
4475 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4476 packet, do_send_packet_in, ofconn);
4479 /* Replace struct odp_msg header in 'packet' by equivalent struct
4480 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4481 * returned by dpif_recv()).
4483 * The conversion is not complete: the caller still needs to trim any unneeded
4484 * payload off the end of the buffer, set the length in the OpenFlow header,
4485 * and set buffer_id. Those require us to know the controller settings and so
4486 * must be done on a per-controller basis.
4488 * Returns the maximum number of bytes of the packet that should be sent to
4489 * the controller (INT_MAX if no limit). */
4491 do_convert_to_packet_in(struct ofpbuf *packet)
4493 struct odp_msg *msg = packet->data;
4494 struct ofp_packet_in *opi;
4500 /* Extract relevant header fields */
4501 if (msg->type == _ODPL_ACTION_NR) {
4502 reason = OFPR_ACTION;
4505 reason = OFPR_NO_MATCH;
4508 total_len = msg->length - sizeof *msg;
4509 in_port = odp_port_to_ofp_port(msg->port);
4511 /* Repurpose packet buffer by overwriting header. */
4512 ofpbuf_pull(packet, sizeof(struct odp_msg));
4513 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4514 opi->header.version = OFP_VERSION;
4515 opi->header.type = OFPT_PACKET_IN;
4516 opi->total_len = htons(total_len);
4517 opi->in_port = htons(in_port);
4518 opi->reason = reason;
4523 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4524 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4525 * as necessary according to their individual configurations.
4527 * 'packet' must have sufficient headroom to convert it into a struct
4528 * ofp_packet_in (e.g. as returned by dpif_recv()).
4530 * Takes ownership of 'packet'. */
4532 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4534 struct ofconn *ofconn, *prev;
4537 max_len = do_convert_to_packet_in(packet);
4540 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4541 if (ofconn_receives_async_msgs(ofconn)) {
4543 schedule_packet_in(prev, packet, max_len, true);
4549 schedule_packet_in(prev, packet, max_len, false);
4551 ofpbuf_delete(packet);
4556 pick_datapath_id(const struct ofproto *ofproto)
4558 const struct ofport *port;
4560 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4562 uint8_t ea[ETH_ADDR_LEN];
4565 error = netdev_get_etheraddr(port->netdev, ea);
4567 return eth_addr_to_uint64(ea);
4569 VLOG_WARN("could not get MAC address for %s (%s)",
4570 netdev_get_name(port->netdev), strerror(error));
4572 return ofproto->fallback_dpid;
4576 pick_fallback_dpid(void)
4578 uint8_t ea[ETH_ADDR_LEN];
4579 eth_addr_nicira_random(ea);
4580 return eth_addr_to_uint64(ea);
4584 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4585 struct odp_actions *actions, tag_type *tags,
4586 uint16_t *nf_output_iface, void *ofproto_)
4588 struct ofproto *ofproto = ofproto_;
4591 /* Drop frames for reserved multicast addresses. */
4592 if (eth_addr_is_reserved(flow->dl_dst)) {
4596 /* Learn source MAC (but don't try to learn from revalidation). */
4597 if (packet != NULL) {
4598 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4600 GRAT_ARP_LOCK_NONE);
4602 /* The log messages here could actually be useful in debugging,
4603 * so keep the rate limit relatively high. */
4604 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4605 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4606 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4607 ofproto_revalidate(ofproto, rev_tag);
4611 /* Determine output port. */
4612 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4615 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4616 } else if (out_port != flow->in_port) {
4617 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4618 *nf_output_iface = out_port;
4626 static const struct ofhooks default_ofhooks = {
4628 default_normal_ofhook_cb,