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"
54 #include "stream-ssl.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto)
65 #include "sflow_api.h"
69 TABLEID_CLASSIFIER = 1
73 struct netdev *netdev;
74 struct ofp_phy_port opp; /* In host byte order. */
77 static void ofport_free(struct ofport *);
78 static void hton_ofp_phy_port(struct ofp_phy_port *);
80 static int xlate_actions(const union ofp_action *in, size_t n_in,
81 const flow_t *flow, struct ofproto *ofproto,
82 const struct ofpbuf *packet,
83 struct odp_actions *out, tag_type *tags,
84 bool *may_set_up_flow, uint16_t *nf_output_iface);
89 uint64_t flow_cookie; /* Controller-issued identifier.
90 (Kept in network-byte order.) */
91 uint16_t idle_timeout; /* In seconds from time of last use. */
92 uint16_t hard_timeout; /* In seconds from time of creation. */
93 bool send_flow_removed; /* Send a flow removed message? */
94 long long int used; /* Last-used time (0 if never used). */
95 long long int created; /* Creation time. */
96 uint64_t packet_count; /* Number of packets received. */
97 uint64_t byte_count; /* Number of bytes received. */
98 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
99 tag_type tags; /* Tags (set only by hooks). */
100 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
102 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
103 * exact-match rule (having cr.wc.wildcards of 0) generated from the
104 * wildcard rule 'super'. In this case, 'list' is an element of the
107 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
108 * a list of subrules. A super-rule with no wildcards (where
109 * cr.wc.wildcards is 0) will never have any subrules. */
115 * 'n_actions' is the number of elements in the 'actions' array. A single
116 * action may take up more more than one element's worth of space.
118 * A subrule has no actions (it uses the super-rule's actions). */
120 union ofp_action *actions;
124 * A super-rule with wildcard fields never has ODP actions (since the
125 * datapath only supports exact-match flows). */
126 bool installed; /* Installed in datapath? */
127 bool may_install; /* True ordinarily; false if actions must
128 * be reassessed for every packet. */
130 union odp_action *odp_actions;
134 rule_is_hidden(const struct rule *rule)
136 /* Subrules are merely an implementation detail, so hide them from the
138 if (rule->super != NULL) {
142 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
143 * (e.g. by in-band control) and are intentionally hidden from the
145 if (rule->cr.priority > UINT16_MAX) {
152 static struct rule *rule_create(struct ofproto *, struct rule *super,
153 const union ofp_action *, size_t n_actions,
154 uint16_t idle_timeout, uint16_t hard_timeout,
155 uint64_t flow_cookie, bool send_flow_removed);
156 static void rule_free(struct rule *);
157 static void rule_destroy(struct ofproto *, struct rule *);
158 static struct rule *rule_from_cls_rule(const struct cls_rule *);
159 static void rule_insert(struct ofproto *, struct rule *,
160 struct ofpbuf *packet, uint16_t in_port);
161 static void rule_remove(struct ofproto *, struct rule *);
162 static bool rule_make_actions(struct ofproto *, struct rule *,
163 const struct ofpbuf *packet);
164 static void rule_install(struct ofproto *, struct rule *,
165 struct rule *displaced_rule);
166 static void rule_uninstall(struct ofproto *, struct rule *);
167 static void rule_post_uninstall(struct ofproto *, struct rule *);
168 static void send_flow_removed(struct ofproto *p, struct rule *rule,
169 long long int now, uint8_t reason);
171 /* ofproto supports two kinds of OpenFlow connections:
173 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
174 * maintains persistent connections to these controllers and by default
175 * sends them asynchronous messages such as packet-ins.
177 * - "Service" connections, e.g. from ovs-ofctl. When these connections
178 * drop, it is the other side's responsibility to reconnect them if
179 * necessary. ofproto does not send them asynchronous messages by default.
181 * Currently, active (tcp, ssl, unix) connections are always "primary"
182 * connections and passive (ptcp, pssl, punix) connections are always "service"
183 * connections. There is no inherent reason for this, but it reflects the
187 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
188 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
191 /* A listener for incoming OpenFlow "service" connections. */
193 struct hmap_node node; /* In struct ofproto's "services" hmap. */
194 struct pvconn *pvconn; /* OpenFlow connection listener. */
196 /* These are not used by ofservice directly. They are settings for
197 * accepted "struct ofconn"s from the pvconn. */
198 int probe_interval; /* Max idle time before probing, in seconds. */
199 int rate_limit; /* Max packet-in rate in packets per second. */
200 int burst_limit; /* Limit on accumulating packet credits. */
203 static struct ofservice *ofservice_lookup(struct ofproto *,
205 static int ofservice_create(struct ofproto *,
206 const struct ofproto_controller *);
207 static void ofservice_reconfigure(struct ofservice *,
208 const struct ofproto_controller *);
209 static void ofservice_destroy(struct ofproto *, struct ofservice *);
211 /* An OpenFlow connection. */
213 struct ofproto *ofproto; /* The ofproto that owns this connection. */
214 struct list node; /* In struct ofproto's "all_conns" list. */
215 struct rconn *rconn; /* OpenFlow connection. */
216 enum ofconn_type type; /* Type. */
218 /* OFPT_PACKET_IN related data. */
219 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
220 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
221 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
222 int miss_send_len; /* Bytes to send of buffered packets. */
224 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
225 * requests, and the maximum number before we stop reading OpenFlow
227 #define OFCONN_REPLY_MAX 100
228 struct rconn_packet_counter *reply_counter;
230 /* type == OFCONN_PRIMARY only. */
231 enum nx_role role; /* Role. */
232 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
233 struct discovery *discovery; /* Controller discovery object, if enabled. */
234 struct status_category *ss; /* Switch status category. */
235 enum ofproto_band band; /* In-band or out-of-band? */
238 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
239 * "schedulers" array. Their values are 0 and 1, and their meanings and values
240 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
241 * case anything ever changes, check their values here. */
242 #define N_SCHEDULERS 2
243 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
244 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
245 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
246 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
248 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
250 static void ofconn_destroy(struct ofconn *);
251 static void ofconn_run(struct ofconn *, struct ofproto *);
252 static void ofconn_wait(struct ofconn *);
253 static bool ofconn_receives_async_msgs(const struct ofconn *);
254 static char *ofconn_make_name(const struct ofproto *, const char *target);
255 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
257 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
258 struct rconn_packet_counter *counter);
260 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
261 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
265 uint64_t datapath_id; /* Datapath ID. */
266 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
267 char *mfr_desc; /* Manufacturer. */
268 char *hw_desc; /* Hardware. */
269 char *sw_desc; /* Software version. */
270 char *serial_desc; /* Serial number. */
271 char *dp_desc; /* Datapath description. */
275 struct netdev_monitor *netdev_monitor;
276 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
278 struct shash port_by_name;
282 struct switch_status *switch_status;
283 struct fail_open *fail_open;
284 struct netflow *netflow;
285 struct ofproto_sflow *sflow;
287 /* In-band control. */
288 struct in_band *in_band;
289 long long int next_in_band_update;
290 struct sockaddr_in *extra_in_band_remotes;
291 size_t n_extra_remotes;
294 struct classifier cls;
295 bool need_revalidate;
296 long long int next_expiration;
297 struct tag_set revalidate_set;
298 bool tun_id_from_cookie;
300 /* OpenFlow connections. */
301 struct hmap controllers; /* Controller "struct ofconn"s. */
302 struct list all_conns; /* Contains "struct ofconn"s. */
303 enum ofproto_fail_mode fail_mode;
305 /* OpenFlow listeners. */
306 struct hmap services; /* Contains "struct ofservice"s. */
307 struct pvconn **snoops;
310 /* Hooks for ovs-vswitchd. */
311 const struct ofhooks *ofhooks;
314 /* Used by default ofhooks. */
315 struct mac_learning *ml;
318 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
320 static const struct ofhooks default_ofhooks;
322 static uint64_t pick_datapath_id(const struct ofproto *);
323 static uint64_t pick_fallback_dpid(void);
325 static void update_used(struct ofproto *);
326 static void update_stats(struct ofproto *, struct rule *,
327 const struct odp_flow_stats *);
328 static void expire_rule(struct cls_rule *, void *ofproto);
329 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
330 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
331 static void revalidate_cb(struct cls_rule *rule_, void *p_);
333 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
335 static void handle_openflow(struct ofconn *, struct ofproto *,
338 static void refresh_port_groups(struct ofproto *);
340 static void update_port(struct ofproto *, const char *devname);
341 static int init_ports(struct ofproto *);
342 static void reinit_ports(struct ofproto *);
345 ofproto_create(const char *datapath, const char *datapath_type,
346 const struct ofhooks *ofhooks, void *aux,
347 struct ofproto **ofprotop)
349 struct odp_stats stats;
356 /* Connect to datapath and start listening for messages. */
357 error = dpif_open(datapath, datapath_type, &dpif);
359 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
362 error = dpif_get_dp_stats(dpif, &stats);
364 VLOG_ERR("failed to obtain stats for datapath %s: %s",
365 datapath, strerror(error));
369 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
371 VLOG_ERR("failed to listen on datapath %s: %s",
372 datapath, strerror(error));
376 dpif_flow_flush(dpif);
377 dpif_recv_purge(dpif);
379 /* Initialize settings. */
380 p = xzalloc(sizeof *p);
381 p->fallback_dpid = pick_fallback_dpid();
382 p->datapath_id = p->fallback_dpid;
383 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
384 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
385 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
386 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
387 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
389 /* Initialize datapath. */
391 p->netdev_monitor = netdev_monitor_create();
392 port_array_init(&p->ports);
393 shash_init(&p->port_by_name);
394 p->max_ports = stats.max_ports;
396 /* Initialize submodules. */
397 p->switch_status = switch_status_create(p);
403 /* Initialize flow table. */
404 classifier_init(&p->cls);
405 p->need_revalidate = false;
406 p->next_expiration = time_msec() + 1000;
407 tag_set_init(&p->revalidate_set);
409 /* Initialize OpenFlow connections. */
410 list_init(&p->all_conns);
411 hmap_init(&p->controllers);
412 hmap_init(&p->services);
416 /* Initialize hooks. */
418 p->ofhooks = ofhooks;
422 p->ofhooks = &default_ofhooks;
424 p->ml = mac_learning_create();
427 /* Pick final datapath ID. */
428 p->datapath_id = pick_datapath_id(p);
429 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
436 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
438 uint64_t old_dpid = p->datapath_id;
439 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
440 if (p->datapath_id != old_dpid) {
441 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
443 /* Force all active connections to reconnect, since there is no way to
444 * notify a controller that the datapath ID has changed. */
445 ofproto_reconnect_controllers(p);
450 is_discovery_controller(const struct ofproto_controller *c)
452 return !strcmp(c->target, "discover");
456 is_in_band_controller(const struct ofproto_controller *c)
458 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
461 /* Creates a new controller in 'ofproto'. Some of the settings are initially
462 * drawn from 'c', but update_controller() needs to be called later to finish
463 * the new ofconn's configuration. */
465 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
467 struct discovery *discovery;
468 struct ofconn *ofconn;
470 if (is_discovery_controller(c)) {
471 int error = discovery_create(c->accept_re, c->update_resolv_conf,
472 ofproto->dpif, ofproto->switch_status,
481 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
482 ofconn->pktbuf = pktbuf_create();
483 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
485 ofconn->discovery = discovery;
487 char *name = ofconn_make_name(ofproto, c->target);
488 rconn_connect(ofconn->rconn, c->target, name);
491 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
492 hash_string(c->target, 0));
495 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
496 * target or turn discovery on or off (these are done by creating new ofconns
497 * and deleting old ones), but it can update the rest of an ofconn's
500 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
504 ofconn->band = (is_in_band_controller(c)
505 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
507 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
509 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
510 rconn_set_probe_interval(ofconn->rconn, probe_interval);
512 if (ofconn->discovery) {
513 discovery_set_update_resolv_conf(ofconn->discovery,
514 c->update_resolv_conf);
515 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
518 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
522 ofconn_get_target(const struct ofconn *ofconn)
524 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
527 static struct ofconn *
528 find_controller_by_target(struct ofproto *ofproto, const char *target)
530 struct ofconn *ofconn;
532 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
533 hash_string(target, 0), &ofproto->controllers) {
534 if (!strcmp(ofconn_get_target(ofconn), target)) {
542 update_in_band_remotes(struct ofproto *ofproto)
544 const struct ofconn *ofconn;
545 struct sockaddr_in *addrs;
546 size_t max_addrs, n_addrs;
550 /* Allocate enough memory for as many remotes as we could possibly have. */
551 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
552 addrs = xmalloc(max_addrs * sizeof *addrs);
555 /* Add all the remotes. */
557 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
558 struct sockaddr_in *sin = &addrs[n_addrs];
560 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
564 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
565 if (sin->sin_addr.s_addr) {
566 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
569 if (ofconn->discovery) {
573 for (i = 0; i < ofproto->n_extra_remotes; i++) {
574 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
577 /* Create or update or destroy in-band.
579 * Ordinarily we only enable in-band if there's at least one remote
580 * address, but discovery needs the in-band rules for DHCP to be installed
581 * even before we know any remote addresses. */
582 if (n_addrs || discovery) {
583 if (!ofproto->in_band) {
584 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
587 if (ofproto->in_band) {
588 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
590 ofproto->next_in_band_update = time_msec() + 1000;
592 in_band_destroy(ofproto->in_band);
593 ofproto->in_band = NULL;
601 update_fail_open(struct ofproto *p)
603 struct ofconn *ofconn;
605 if (!hmap_is_empty(&p->controllers)
606 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
607 struct rconn **rconns;
611 p->fail_open = fail_open_create(p, p->switch_status);
615 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
616 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
617 rconns[n++] = ofconn->rconn;
620 fail_open_set_controllers(p->fail_open, rconns, n);
621 /* p->fail_open takes ownership of 'rconns'. */
623 fail_open_destroy(p->fail_open);
629 ofproto_set_controllers(struct ofproto *p,
630 const struct ofproto_controller *controllers,
631 size_t n_controllers)
633 struct shash new_controllers;
634 struct ofconn *ofconn, *next_ofconn;
635 struct ofservice *ofservice, *next_ofservice;
639 /* Create newly configured controllers and services.
640 * Create a name to ofproto_controller mapping in 'new_controllers'. */
641 shash_init(&new_controllers);
642 for (i = 0; i < n_controllers; i++) {
643 const struct ofproto_controller *c = &controllers[i];
645 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
646 if (!find_controller_by_target(p, c->target)) {
647 add_controller(p, c);
649 } else if (!pvconn_verify_name(c->target)) {
650 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
654 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
655 dpif_name(p->dpif), c->target);
659 shash_add_once(&new_controllers, c->target, &controllers[i]);
662 /* Delete controllers that are no longer configured.
663 * Update configuration of all now-existing controllers. */
665 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node,
667 struct ofproto_controller *c;
669 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
671 ofconn_destroy(ofconn);
673 update_controller(ofconn, c);
680 /* Delete services that are no longer configured.
681 * Update configuration of all now-existing services. */
682 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
684 struct ofproto_controller *c;
686 c = shash_find_data(&new_controllers,
687 pvconn_get_name(ofservice->pvconn));
689 ofservice_destroy(p, ofservice);
691 ofservice_reconfigure(ofservice, c);
695 shash_destroy(&new_controllers);
697 update_in_band_remotes(p);
700 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
701 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
702 struct ofconn, hmap_node);
703 ofconn->ss = switch_status_register(p->switch_status, "remote",
704 rconn_status_cb, ofconn->rconn);
709 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
711 p->fail_mode = fail_mode;
715 /* Drops the connections between 'ofproto' and all of its controllers, forcing
716 * them to reconnect. */
718 ofproto_reconnect_controllers(struct ofproto *ofproto)
720 struct ofconn *ofconn;
722 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
723 rconn_reconnect(ofconn->rconn);
728 any_extras_changed(const struct ofproto *ofproto,
729 const struct sockaddr_in *extras, size_t n)
733 if (n != ofproto->n_extra_remotes) {
737 for (i = 0; i < n; i++) {
738 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
739 const struct sockaddr_in *new = &extras[i];
741 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
742 old->sin_port != new->sin_port) {
750 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
751 * in-band control should guarantee access, in the same way that in-band
752 * control guarantees access to OpenFlow controllers. */
754 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
755 const struct sockaddr_in *extras, size_t n)
757 if (!any_extras_changed(ofproto, extras, n)) {
761 free(ofproto->extra_in_band_remotes);
762 ofproto->n_extra_remotes = n;
763 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
765 update_in_band_remotes(ofproto);
769 ofproto_set_desc(struct ofproto *p,
770 const char *mfr_desc, const char *hw_desc,
771 const char *sw_desc, const char *serial_desc,
774 struct ofp_desc_stats *ods;
777 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
778 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
779 sizeof ods->mfr_desc);
782 p->mfr_desc = xstrdup(mfr_desc);
785 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
786 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
787 sizeof ods->hw_desc);
790 p->hw_desc = xstrdup(hw_desc);
793 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
794 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
795 sizeof ods->sw_desc);
798 p->sw_desc = xstrdup(sw_desc);
801 if (strlen(serial_desc) >= sizeof ods->serial_num) {
802 VLOG_WARN("truncating serial_desc, must be less than %zu "
804 sizeof ods->serial_num);
806 free(p->serial_desc);
807 p->serial_desc = xstrdup(serial_desc);
810 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
811 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
812 sizeof ods->dp_desc);
815 p->dp_desc = xstrdup(dp_desc);
820 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
821 const struct svec *svec)
823 struct pvconn **pvconns = *pvconnsp;
824 size_t n_pvconns = *n_pvconnsp;
828 for (i = 0; i < n_pvconns; i++) {
829 pvconn_close(pvconns[i]);
833 pvconns = xmalloc(svec->n * sizeof *pvconns);
835 for (i = 0; i < svec->n; i++) {
836 const char *name = svec->names[i];
837 struct pvconn *pvconn;
840 error = pvconn_open(name, &pvconn);
842 pvconns[n_pvconns++] = pvconn;
844 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
852 *n_pvconnsp = n_pvconns;
858 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
860 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
864 ofproto_set_netflow(struct ofproto *ofproto,
865 const struct netflow_options *nf_options)
867 if (nf_options && nf_options->collectors.n) {
868 if (!ofproto->netflow) {
869 ofproto->netflow = netflow_create();
871 return netflow_set_options(ofproto->netflow, nf_options);
873 netflow_destroy(ofproto->netflow);
874 ofproto->netflow = NULL;
880 ofproto_set_sflow(struct ofproto *ofproto,
881 const struct ofproto_sflow_options *oso)
883 struct ofproto_sflow *os = ofproto->sflow;
886 struct ofport *ofport;
887 unsigned int odp_port;
889 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
890 refresh_port_groups(ofproto);
891 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
892 ofproto_sflow_add_port(os, odp_port,
893 netdev_get_name(ofport->netdev));
896 ofproto_sflow_set_options(os, oso);
898 ofproto_sflow_destroy(os);
899 ofproto->sflow = NULL;
904 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
908 VLOG_WARN("STP is not yet implemented");
916 ofproto_get_datapath_id(const struct ofproto *ofproto)
918 return ofproto->datapath_id;
922 ofproto_has_primary_controller(const struct ofproto *ofproto)
924 return !hmap_is_empty(&ofproto->controllers);
927 enum ofproto_fail_mode
928 ofproto_get_fail_mode(const struct ofproto *p)
934 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
938 for (i = 0; i < ofproto->n_snoops; i++) {
939 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
944 ofproto_destroy(struct ofproto *p)
946 struct ofservice *ofservice, *next_ofservice;
947 struct ofconn *ofconn, *next_ofconn;
948 struct ofport *ofport;
949 unsigned int port_no;
956 /* Destroy fail-open and in-band early, since they touch the classifier. */
957 fail_open_destroy(p->fail_open);
960 in_band_destroy(p->in_band);
962 free(p->extra_in_band_remotes);
964 ofproto_flush_flows(p);
965 classifier_destroy(&p->cls);
967 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
969 ofconn_destroy(ofconn);
971 hmap_destroy(&p->controllers);
974 netdev_monitor_destroy(p->netdev_monitor);
975 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
978 shash_destroy(&p->port_by_name);
980 switch_status_destroy(p->switch_status);
981 netflow_destroy(p->netflow);
982 ofproto_sflow_destroy(p->sflow);
984 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node,
986 ofservice_destroy(p, ofservice);
988 hmap_destroy(&p->services);
990 for (i = 0; i < p->n_snoops; i++) {
991 pvconn_close(p->snoops[i]);
995 mac_learning_destroy(p->ml);
1000 free(p->serial_desc);
1003 port_array_destroy(&p->ports);
1009 ofproto_run(struct ofproto *p)
1011 int error = ofproto_run1(p);
1013 error = ofproto_run2(p, false);
1019 process_port_change(struct ofproto *ofproto, int error, char *devname)
1021 if (error == ENOBUFS) {
1022 reinit_ports(ofproto);
1023 } else if (!error) {
1024 update_port(ofproto, devname);
1029 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1030 * means that 'ofconn' is more interesting for monitoring than a lower return
1033 snoop_preference(const struct ofconn *ofconn)
1035 switch (ofconn->role) {
1036 case NX_ROLE_MASTER:
1043 /* Shouldn't happen. */
1048 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1049 * Connects this vconn to a controller. */
1051 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1053 struct ofconn *ofconn, *best;
1055 /* Pick a controller for monitoring. */
1057 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1058 if (ofconn->type == OFCONN_PRIMARY
1059 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1065 rconn_add_monitor(best->rconn, vconn);
1067 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1073 ofproto_run1(struct ofproto *p)
1075 struct ofconn *ofconn, *next_ofconn;
1076 struct ofservice *ofservice;
1081 if (shash_is_empty(&p->port_by_name)) {
1085 for (i = 0; i < 50; i++) {
1089 error = dpif_recv(p->dpif, &buf);
1091 if (error == ENODEV) {
1092 /* Someone destroyed the datapath behind our back. The caller
1093 * better destroy us and give up, because we're just going to
1094 * spin from here on out. */
1095 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1096 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1097 dpif_name(p->dpif));
1103 handle_odp_msg(p, buf);
1106 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1107 process_port_change(p, error, devname);
1109 while ((error = netdev_monitor_poll(p->netdev_monitor,
1110 &devname)) != EAGAIN) {
1111 process_port_change(p, error, devname);
1115 if (time_msec() >= p->next_in_band_update) {
1116 update_in_band_remotes(p);
1118 in_band_run(p->in_band);
1121 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1123 ofconn_run(ofconn, p);
1126 /* Fail-open maintenance. Do this after processing the ofconns since
1127 * fail-open checks the status of the controller rconn. */
1129 fail_open_run(p->fail_open);
1132 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1133 struct vconn *vconn;
1136 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1138 struct ofconn *ofconn;
1139 struct rconn *rconn;
1142 rconn = rconn_create(ofservice->probe_interval, 0);
1143 name = ofconn_make_name(p, vconn_get_name(vconn));
1144 rconn_connect_unreliably(rconn, vconn, name);
1147 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1148 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1149 ofservice->burst_limit);
1150 } else if (retval != EAGAIN) {
1151 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1155 for (i = 0; i < p->n_snoops; i++) {
1156 struct vconn *vconn;
1159 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1161 add_snooper(p, vconn);
1162 } else if (retval != EAGAIN) {
1163 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1167 if (time_msec() >= p->next_expiration) {
1168 COVERAGE_INC(ofproto_expiration);
1169 p->next_expiration = time_msec() + 1000;
1172 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1174 /* Let the hook know that we're at a stable point: all outstanding data
1175 * in existing flows has been accounted to the account_cb. Thus, the
1176 * hook can now reasonably do operations that depend on having accurate
1177 * flow volume accounting (currently, that's just bond rebalancing). */
1178 if (p->ofhooks->account_checkpoint_cb) {
1179 p->ofhooks->account_checkpoint_cb(p->aux);
1184 netflow_run(p->netflow);
1187 ofproto_sflow_run(p->sflow);
1193 struct revalidate_cbdata {
1194 struct ofproto *ofproto;
1195 bool revalidate_all; /* Revalidate all exact-match rules? */
1196 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1197 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1201 ofproto_run2(struct ofproto *p, bool revalidate_all)
1203 if (p->need_revalidate || revalidate_all
1204 || !tag_set_is_empty(&p->revalidate_set)) {
1205 struct revalidate_cbdata cbdata;
1207 cbdata.revalidate_all = revalidate_all;
1208 cbdata.revalidate_subrules = p->need_revalidate;
1209 cbdata.revalidate_set = p->revalidate_set;
1210 tag_set_init(&p->revalidate_set);
1211 COVERAGE_INC(ofproto_revalidate);
1212 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1213 p->need_revalidate = false;
1220 ofproto_wait(struct ofproto *p)
1222 struct ofservice *ofservice;
1223 struct ofconn *ofconn;
1226 dpif_recv_wait(p->dpif);
1227 dpif_port_poll_wait(p->dpif);
1228 netdev_monitor_poll_wait(p->netdev_monitor);
1229 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1230 ofconn_wait(ofconn);
1233 poll_timer_wait_until(p->next_in_band_update);
1234 in_band_wait(p->in_band);
1237 fail_open_wait(p->fail_open);
1240 ofproto_sflow_wait(p->sflow);
1242 if (!tag_set_is_empty(&p->revalidate_set)) {
1243 poll_immediate_wake();
1245 if (p->need_revalidate) {
1246 /* Shouldn't happen, but if it does just go around again. */
1247 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1248 poll_immediate_wake();
1249 } else if (p->next_expiration != LLONG_MAX) {
1250 poll_timer_wait_until(p->next_expiration);
1252 HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) {
1253 pvconn_wait(ofservice->pvconn);
1255 for (i = 0; i < p->n_snoops; i++) {
1256 pvconn_wait(p->snoops[i]);
1261 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1263 tag_set_add(&ofproto->revalidate_set, tag);
1267 ofproto_get_revalidate_set(struct ofproto *ofproto)
1269 return &ofproto->revalidate_set;
1273 ofproto_is_alive(const struct ofproto *p)
1275 return !hmap_is_empty(&p->controllers);
1279 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1280 const union ofp_action *actions, size_t n_actions,
1281 const struct ofpbuf *packet)
1283 struct odp_actions odp_actions;
1286 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1292 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1294 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1295 odp_actions.n_actions, packet);
1300 ofproto_add_flow(struct ofproto *p,
1301 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1302 const union ofp_action *actions, size_t n_actions,
1306 rule = rule_create(p, NULL, actions, n_actions,
1307 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1309 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1310 rule_insert(p, rule, NULL, 0);
1314 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1315 uint32_t wildcards, unsigned int priority)
1319 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1323 rule_remove(ofproto, rule);
1328 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1330 struct rule *rule = rule_from_cls_rule(rule_);
1331 struct ofproto *ofproto = ofproto_;
1333 /* Mark the flow as not installed, even though it might really be
1334 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1335 * There is no point in uninstalling it individually since we are about to
1336 * blow away all the flows with dpif_flow_flush(). */
1337 rule->installed = false;
1339 rule_remove(ofproto, rule);
1343 ofproto_flush_flows(struct ofproto *ofproto)
1345 COVERAGE_INC(ofproto_flush);
1346 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1347 dpif_flow_flush(ofproto->dpif);
1348 if (ofproto->in_band) {
1349 in_band_flushed(ofproto->in_band);
1351 if (ofproto->fail_open) {
1352 fail_open_flushed(ofproto->fail_open);
1357 reinit_ports(struct ofproto *p)
1359 struct svec devnames;
1360 struct ofport *ofport;
1361 unsigned int port_no;
1362 struct odp_port *odp_ports;
1366 svec_init(&devnames);
1367 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1368 svec_add (&devnames, (char *) ofport->opp.name);
1370 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1371 for (i = 0; i < n_odp_ports; i++) {
1372 svec_add (&devnames, odp_ports[i].devname);
1376 svec_sort_unique(&devnames);
1377 for (i = 0; i < devnames.n; i++) {
1378 update_port(p, devnames.names[i]);
1380 svec_destroy(&devnames);
1384 refresh_port_group(struct ofproto *p, unsigned int group)
1388 struct ofport *port;
1389 unsigned int port_no;
1391 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1393 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1395 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1396 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1397 ports[n_ports++] = port_no;
1400 dpif_port_group_set(p->dpif, group, ports, n_ports);
1407 refresh_port_groups(struct ofproto *p)
1409 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1410 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1412 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1416 static struct ofport *
1417 make_ofport(const struct odp_port *odp_port)
1419 struct netdev_options netdev_options;
1420 enum netdev_flags flags;
1421 struct ofport *ofport;
1422 struct netdev *netdev;
1426 memset(&netdev_options, 0, sizeof netdev_options);
1427 netdev_options.name = odp_port->devname;
1428 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1430 error = netdev_open(&netdev_options, &netdev);
1432 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1433 "cannot be opened (%s)",
1434 odp_port->devname, odp_port->port,
1435 odp_port->devname, strerror(error));
1439 ofport = xmalloc(sizeof *ofport);
1440 ofport->netdev = netdev;
1441 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1442 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1443 memcpy(ofport->opp.name, odp_port->devname,
1444 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1445 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1447 netdev_get_flags(netdev, &flags);
1448 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1450 netdev_get_carrier(netdev, &carrier);
1451 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1453 netdev_get_features(netdev,
1454 &ofport->opp.curr, &ofport->opp.advertised,
1455 &ofport->opp.supported, &ofport->opp.peer);
1460 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1462 if (port_array_get(&p->ports, odp_port->port)) {
1463 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1466 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1467 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1476 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1478 const struct ofp_phy_port *a = &a_->opp;
1479 const struct ofp_phy_port *b = &b_->opp;
1481 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1482 return (a->port_no == b->port_no
1483 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1484 && !strcmp((char *) a->name, (char *) b->name)
1485 && a->state == b->state
1486 && a->config == b->config
1487 && a->curr == b->curr
1488 && a->advertised == b->advertised
1489 && a->supported == b->supported
1490 && a->peer == b->peer);
1494 send_port_status(struct ofproto *p, const struct ofport *ofport,
1497 /* XXX Should limit the number of queued port status change messages. */
1498 struct ofconn *ofconn;
1499 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1500 struct ofp_port_status *ops;
1503 if (!ofconn_receives_async_msgs(ofconn)) {
1507 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1508 ops->reason = reason;
1509 ops->desc = ofport->opp;
1510 hton_ofp_phy_port(&ops->desc);
1511 queue_tx(b, ofconn, NULL);
1513 if (p->ofhooks->port_changed_cb) {
1514 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1519 ofport_install(struct ofproto *p, struct ofport *ofport)
1521 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1522 const char *netdev_name = (const char *) ofport->opp.name;
1524 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1525 port_array_set(&p->ports, odp_port, ofport);
1526 shash_add(&p->port_by_name, netdev_name, ofport);
1528 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1533 ofport_remove(struct ofproto *p, struct ofport *ofport)
1535 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1537 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1538 port_array_delete(&p->ports, odp_port);
1539 shash_delete(&p->port_by_name,
1540 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1542 ofproto_sflow_del_port(p->sflow, odp_port);
1547 ofport_free(struct ofport *ofport)
1550 netdev_close(ofport->netdev);
1556 update_port(struct ofproto *p, const char *devname)
1558 struct odp_port odp_port;
1559 struct ofport *old_ofport;
1560 struct ofport *new_ofport;
1563 COVERAGE_INC(ofproto_update_port);
1565 /* Query the datapath for port information. */
1566 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1568 /* Find the old ofport. */
1569 old_ofport = shash_find_data(&p->port_by_name, devname);
1572 /* There's no port named 'devname' but there might be a port with
1573 * the same port number. This could happen if a port is deleted
1574 * and then a new one added in its place very quickly, or if a port
1575 * is renamed. In the former case we want to send an OFPPR_DELETE
1576 * and an OFPPR_ADD, and in the latter case we want to send a
1577 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1578 * the old port's ifindex against the new port, or perhaps less
1579 * reliably but more portably by comparing the old port's MAC
1580 * against the new port's MAC. However, this code isn't that smart
1581 * and always sends an OFPPR_MODIFY (XXX). */
1582 old_ofport = port_array_get(&p->ports, odp_port.port);
1584 } else if (error != ENOENT && error != ENODEV) {
1585 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1586 "%s", strerror(error));
1590 /* Create a new ofport. */
1591 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1593 /* Eliminate a few pathological cases. */
1594 if (!old_ofport && !new_ofport) {
1596 } else if (old_ofport && new_ofport) {
1597 /* Most of the 'config' bits are OpenFlow soft state, but
1598 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1599 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1600 * leaves the other bits 0.) */
1601 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1603 if (ofport_equal(old_ofport, new_ofport)) {
1604 /* False alarm--no change. */
1605 ofport_free(new_ofport);
1610 /* Now deal with the normal cases. */
1612 ofport_remove(p, old_ofport);
1615 ofport_install(p, new_ofport);
1617 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1618 (!old_ofport ? OFPPR_ADD
1619 : !new_ofport ? OFPPR_DELETE
1621 ofport_free(old_ofport);
1623 /* Update port groups. */
1624 refresh_port_groups(p);
1628 init_ports(struct ofproto *p)
1630 struct odp_port *ports;
1635 error = dpif_port_list(p->dpif, &ports, &n_ports);
1640 for (i = 0; i < n_ports; i++) {
1641 const struct odp_port *odp_port = &ports[i];
1642 if (!ofport_conflicts(p, odp_port)) {
1643 struct ofport *ofport = make_ofport(odp_port);
1645 ofport_install(p, ofport);
1650 refresh_port_groups(p);
1654 static struct ofconn *
1655 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1657 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1658 ofconn->ofproto = p;
1659 list_push_back(&p->all_conns, &ofconn->node);
1660 ofconn->rconn = rconn;
1661 ofconn->type = type;
1662 ofconn->role = NX_ROLE_OTHER;
1663 ofconn->packet_in_counter = rconn_packet_counter_create ();
1664 ofconn->pktbuf = NULL;
1665 ofconn->miss_send_len = 0;
1666 ofconn->reply_counter = rconn_packet_counter_create ();
1671 ofconn_destroy(struct ofconn *ofconn)
1673 if (ofconn->type == OFCONN_PRIMARY) {
1674 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1676 discovery_destroy(ofconn->discovery);
1678 list_remove(&ofconn->node);
1679 switch_status_unregister(ofconn->ss);
1680 rconn_destroy(ofconn->rconn);
1681 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1682 rconn_packet_counter_destroy(ofconn->reply_counter);
1683 pktbuf_destroy(ofconn->pktbuf);
1688 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1693 if (ofconn->discovery) {
1694 char *controller_name;
1695 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1696 discovery_question_connectivity(ofconn->discovery);
1698 if (discovery_run(ofconn->discovery, &controller_name)) {
1699 if (controller_name) {
1700 char *ofconn_name = ofconn_make_name(p, controller_name);
1701 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1704 rconn_disconnect(ofconn->rconn);
1709 for (i = 0; i < N_SCHEDULERS; i++) {
1710 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1713 rconn_run(ofconn->rconn);
1715 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1716 /* Limit the number of iterations to prevent other tasks from
1718 for (iteration = 0; iteration < 50; iteration++) {
1719 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1724 fail_open_maybe_recover(p->fail_open);
1726 handle_openflow(ofconn, p, of_msg);
1727 ofpbuf_delete(of_msg);
1731 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1732 ofconn_destroy(ofconn);
1737 ofconn_wait(struct ofconn *ofconn)
1741 if (ofconn->discovery) {
1742 discovery_wait(ofconn->discovery);
1744 for (i = 0; i < N_SCHEDULERS; i++) {
1745 pinsched_wait(ofconn->schedulers[i]);
1747 rconn_run_wait(ofconn->rconn);
1748 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1749 rconn_recv_wait(ofconn->rconn);
1751 COVERAGE_INC(ofproto_ofconn_stuck);
1755 /* Returns true if 'ofconn' should receive asynchronous messages. */
1757 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1759 if (ofconn->type == OFCONN_PRIMARY) {
1760 /* Primary controllers always get asynchronous messages unless they
1761 * have configured themselves as "slaves". */
1762 return ofconn->role != NX_ROLE_SLAVE;
1764 /* Service connections don't get asynchronous messages unless they have
1765 * explicitly asked for them by setting a nonzero miss send length. */
1766 return ofconn->miss_send_len > 0;
1770 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1771 * and 'target', suitable for use in log messages for identifying the
1774 * The name is dynamically allocated. The caller should free it (with free())
1775 * when it is no longer needed. */
1777 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1779 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1783 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1787 for (i = 0; i < N_SCHEDULERS; i++) {
1788 struct pinsched **s = &ofconn->schedulers[i];
1792 *s = pinsched_create(rate, burst,
1793 ofconn->ofproto->switch_status);
1795 pinsched_set_limits(*s, rate, burst);
1798 pinsched_destroy(*s);
1805 ofservice_reconfigure(struct ofservice *ofservice,
1806 const struct ofproto_controller *c)
1808 ofservice->probe_interval = c->probe_interval;
1809 ofservice->rate_limit = c->rate_limit;
1810 ofservice->burst_limit = c->burst_limit;
1813 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1814 * positive errno value. */
1816 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1818 struct ofservice *ofservice;
1819 struct pvconn *pvconn;
1822 error = pvconn_open(c->target, &pvconn);
1827 ofservice = xzalloc(sizeof *ofservice);
1828 hmap_insert(&ofproto->services, &ofservice->node,
1829 hash_string(c->target, 0));
1830 ofservice->pvconn = pvconn;
1832 ofservice_reconfigure(ofservice, c);
1838 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1840 hmap_remove(&ofproto->services, &ofservice->node);
1841 pvconn_close(ofservice->pvconn);
1845 /* Finds and returns the ofservice within 'ofproto' that has the given
1846 * 'target', or a null pointer if none exists. */
1847 static struct ofservice *
1848 ofservice_lookup(struct ofproto *ofproto, const char *target)
1850 struct ofservice *ofservice;
1852 HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node,
1853 hash_string(target, 0), &ofproto->services) {
1854 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1861 /* Caller is responsible for initializing the 'cr' member of the returned
1863 static struct rule *
1864 rule_create(struct ofproto *ofproto, struct rule *super,
1865 const union ofp_action *actions, size_t n_actions,
1866 uint16_t idle_timeout, uint16_t hard_timeout,
1867 uint64_t flow_cookie, bool send_flow_removed)
1869 struct rule *rule = xzalloc(sizeof *rule);
1870 rule->idle_timeout = idle_timeout;
1871 rule->hard_timeout = hard_timeout;
1872 rule->flow_cookie = flow_cookie;
1873 rule->used = rule->created = time_msec();
1874 rule->send_flow_removed = send_flow_removed;
1875 rule->super = super;
1877 list_push_back(&super->list, &rule->list);
1879 list_init(&rule->list);
1881 rule->n_actions = n_actions;
1882 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1883 netflow_flow_clear(&rule->nf_flow);
1884 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1889 static struct rule *
1890 rule_from_cls_rule(const struct cls_rule *cls_rule)
1892 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1896 rule_free(struct rule *rule)
1898 free(rule->actions);
1899 free(rule->odp_actions);
1903 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1904 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1905 * through all of its subrules and revalidates them, destroying any that no
1906 * longer has a super-rule (which is probably all of them).
1908 * Before calling this function, the caller must make have removed 'rule' from
1909 * the classifier. If 'rule' is an exact-match rule, the caller is also
1910 * responsible for ensuring that it has been uninstalled from the datapath. */
1912 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1915 struct rule *subrule, *next;
1916 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1917 revalidate_rule(ofproto, subrule);
1920 list_remove(&rule->list);
1926 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1928 const union ofp_action *oa;
1929 struct actions_iterator i;
1931 if (out_port == htons(OFPP_NONE)) {
1934 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1935 oa = actions_next(&i)) {
1936 if (action_outputs_to_port(oa, out_port)) {
1943 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1944 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1946 * The flow that 'packet' actually contains does not need to actually match
1947 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1948 * the packet and byte counters for 'rule' will be credited for the packet sent
1949 * out whether or not the packet actually matches 'rule'.
1951 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1952 * the caller must already have accurately composed ODP actions for it given
1953 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1954 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1955 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1956 * actions and apply them to 'packet'. */
1958 rule_execute(struct ofproto *ofproto, struct rule *rule,
1959 struct ofpbuf *packet, const flow_t *flow)
1961 const union odp_action *actions;
1963 struct odp_actions a;
1965 /* Grab or compose the ODP actions.
1967 * The special case for an exact-match 'rule' where 'flow' is not the
1968 * rule's flow is important to avoid, e.g., sending a packet out its input
1969 * port simply because the ODP actions were composed for the wrong
1971 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1972 struct rule *super = rule->super ? rule->super : rule;
1973 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1974 packet, &a, NULL, 0, NULL)) {
1977 actions = a.actions;
1978 n_actions = a.n_actions;
1980 actions = rule->odp_actions;
1981 n_actions = rule->n_odp_actions;
1984 /* Execute the ODP actions. */
1985 if (!dpif_execute(ofproto->dpif, flow->in_port,
1986 actions, n_actions, packet)) {
1987 struct odp_flow_stats stats;
1988 flow_extract_stats(flow, packet, &stats);
1989 update_stats(ofproto, rule, &stats);
1990 rule->used = time_msec();
1991 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1996 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1999 struct rule *displaced_rule;
2001 /* Insert the rule in the classifier. */
2002 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2003 if (!rule->cr.wc.wildcards) {
2004 rule_make_actions(p, rule, packet);
2007 /* Send the packet and credit it to the rule. */
2010 flow_extract(packet, 0, in_port, &flow);
2011 rule_execute(p, rule, packet, &flow);
2014 /* Install the rule in the datapath only after sending the packet, to
2015 * avoid packet reordering. */
2016 if (rule->cr.wc.wildcards) {
2017 COVERAGE_INC(ofproto_add_wc_flow);
2018 p->need_revalidate = true;
2020 rule_install(p, rule, displaced_rule);
2023 /* Free the rule that was displaced, if any. */
2024 if (displaced_rule) {
2025 rule_destroy(p, displaced_rule);
2029 static struct rule *
2030 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2033 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2034 rule->idle_timeout, rule->hard_timeout,
2036 COVERAGE_INC(ofproto_subrule_create);
2037 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2038 : rule->cr.priority), &subrule->cr);
2039 classifier_insert_exact(&ofproto->cls, &subrule->cr);
2045 rule_remove(struct ofproto *ofproto, struct rule *rule)
2047 if (rule->cr.wc.wildcards) {
2048 COVERAGE_INC(ofproto_del_wc_flow);
2049 ofproto->need_revalidate = true;
2051 rule_uninstall(ofproto, rule);
2053 classifier_remove(&ofproto->cls, &rule->cr);
2054 rule_destroy(ofproto, rule);
2057 /* Returns true if the actions changed, false otherwise. */
2059 rule_make_actions(struct ofproto *p, struct rule *rule,
2060 const struct ofpbuf *packet)
2062 const struct rule *super;
2063 struct odp_actions a;
2066 assert(!rule->cr.wc.wildcards);
2068 super = rule->super ? rule->super : rule;
2070 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2071 packet, &a, &rule->tags, &rule->may_install,
2072 &rule->nf_flow.output_iface);
2074 actions_len = a.n_actions * sizeof *a.actions;
2075 if (rule->n_odp_actions != a.n_actions
2076 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2077 COVERAGE_INC(ofproto_odp_unchanged);
2078 free(rule->odp_actions);
2079 rule->n_odp_actions = a.n_actions;
2080 rule->odp_actions = xmemdup(a.actions, actions_len);
2088 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2089 struct odp_flow_put *put)
2091 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2092 put->flow.key = rule->cr.flow;
2093 put->flow.actions = rule->odp_actions;
2094 put->flow.n_actions = rule->n_odp_actions;
2095 put->flow.flags = 0;
2097 return dpif_flow_put(ofproto->dpif, put);
2101 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2103 assert(!rule->cr.wc.wildcards);
2105 if (rule->may_install) {
2106 struct odp_flow_put put;
2107 if (!do_put_flow(p, rule,
2108 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2110 rule->installed = true;
2111 if (displaced_rule) {
2112 update_stats(p, displaced_rule, &put.flow.stats);
2113 rule_post_uninstall(p, displaced_rule);
2116 } else if (displaced_rule) {
2117 rule_uninstall(p, displaced_rule);
2122 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2124 if (rule->installed) {
2125 struct odp_flow_put put;
2126 COVERAGE_INC(ofproto_dp_missed);
2127 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2129 rule_install(ofproto, rule, NULL);
2134 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2136 bool actions_changed;
2137 uint16_t new_out_iface, old_out_iface;
2139 old_out_iface = rule->nf_flow.output_iface;
2140 actions_changed = rule_make_actions(ofproto, rule, NULL);
2142 if (rule->may_install) {
2143 if (rule->installed) {
2144 if (actions_changed) {
2145 struct odp_flow_put put;
2146 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2147 | ODPPF_ZERO_STATS, &put);
2148 update_stats(ofproto, rule, &put.flow.stats);
2150 /* Temporarily set the old output iface so that NetFlow
2151 * messages have the correct output interface for the old
2153 new_out_iface = rule->nf_flow.output_iface;
2154 rule->nf_flow.output_iface = old_out_iface;
2155 rule_post_uninstall(ofproto, rule);
2156 rule->nf_flow.output_iface = new_out_iface;
2159 rule_install(ofproto, rule, NULL);
2162 rule_uninstall(ofproto, rule);
2167 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2169 uint64_t total_bytes = rule->byte_count + extra_bytes;
2171 if (ofproto->ofhooks->account_flow_cb
2172 && total_bytes > rule->accounted_bytes)
2174 ofproto->ofhooks->account_flow_cb(
2175 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2176 total_bytes - rule->accounted_bytes, ofproto->aux);
2177 rule->accounted_bytes = total_bytes;
2182 rule_uninstall(struct ofproto *p, struct rule *rule)
2184 assert(!rule->cr.wc.wildcards);
2185 if (rule->installed) {
2186 struct odp_flow odp_flow;
2188 odp_flow.key = rule->cr.flow;
2189 odp_flow.actions = NULL;
2190 odp_flow.n_actions = 0;
2192 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2193 update_stats(p, rule, &odp_flow.stats);
2195 rule->installed = false;
2197 rule_post_uninstall(p, rule);
2202 is_controller_rule(struct rule *rule)
2204 /* If the only action is send to the controller then don't report
2205 * NetFlow expiration messages since it is just part of the control
2206 * logic for the network and not real traffic. */
2210 && rule->super->n_actions == 1
2211 && action_outputs_to_port(&rule->super->actions[0],
2212 htons(OFPP_CONTROLLER)));
2216 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2218 struct rule *super = rule->super;
2220 rule_account(ofproto, rule, 0);
2222 if (ofproto->netflow && !is_controller_rule(rule)) {
2223 struct ofexpired expired;
2224 expired.flow = rule->cr.flow;
2225 expired.packet_count = rule->packet_count;
2226 expired.byte_count = rule->byte_count;
2227 expired.used = rule->used;
2228 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2231 super->packet_count += rule->packet_count;
2232 super->byte_count += rule->byte_count;
2234 /* Reset counters to prevent double counting if the rule ever gets
2236 rule->packet_count = 0;
2237 rule->byte_count = 0;
2238 rule->accounted_bytes = 0;
2240 netflow_flow_clear(&rule->nf_flow);
2245 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2246 struct rconn_packet_counter *counter)
2248 update_openflow_length(msg);
2249 if (rconn_send(ofconn->rconn, msg, counter)) {
2255 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2256 int error, const void *data, size_t len)
2259 struct ofp_error_msg *oem;
2261 if (!(error >> 16)) {
2262 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2267 COVERAGE_INC(ofproto_error);
2268 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2269 oh ? oh->xid : 0, &buf);
2270 oem->type = htons((unsigned int) error >> 16);
2271 oem->code = htons(error & 0xffff);
2272 memcpy(oem->data, data, len);
2273 queue_tx(buf, ofconn, ofconn->reply_counter);
2277 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2280 size_t oh_length = ntohs(oh->length);
2281 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2285 hton_ofp_phy_port(struct ofp_phy_port *opp)
2287 opp->port_no = htons(opp->port_no);
2288 opp->config = htonl(opp->config);
2289 opp->state = htonl(opp->state);
2290 opp->curr = htonl(opp->curr);
2291 opp->advertised = htonl(opp->advertised);
2292 opp->supported = htonl(opp->supported);
2293 opp->peer = htonl(opp->peer);
2297 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2299 struct ofp_header *rq = oh;
2300 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2305 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2306 struct ofp_header *oh)
2308 struct ofp_switch_features *osf;
2310 unsigned int port_no;
2311 struct ofport *port;
2313 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2314 osf->datapath_id = htonll(p->datapath_id);
2315 osf->n_buffers = htonl(pktbuf_capacity());
2317 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2318 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2319 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2320 (1u << OFPAT_SET_VLAN_VID) |
2321 (1u << OFPAT_SET_VLAN_PCP) |
2322 (1u << OFPAT_STRIP_VLAN) |
2323 (1u << OFPAT_SET_DL_SRC) |
2324 (1u << OFPAT_SET_DL_DST) |
2325 (1u << OFPAT_SET_NW_SRC) |
2326 (1u << OFPAT_SET_NW_DST) |
2327 (1u << OFPAT_SET_NW_TOS) |
2328 (1u << OFPAT_SET_TP_SRC) |
2329 (1u << OFPAT_SET_TP_DST) |
2330 (1u << OFPAT_ENQUEUE));
2332 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2333 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2336 queue_tx(buf, ofconn, ofconn->reply_counter);
2341 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2342 struct ofp_header *oh)
2345 struct ofp_switch_config *osc;
2349 /* Figure out flags. */
2350 dpif_get_drop_frags(p->dpif, &drop_frags);
2351 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2354 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2355 osc->flags = htons(flags);
2356 osc->miss_send_len = htons(ofconn->miss_send_len);
2357 queue_tx(buf, ofconn, ofconn->reply_counter);
2363 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2364 struct ofp_switch_config *osc)
2369 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2373 flags = ntohs(osc->flags);
2375 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2376 switch (flags & OFPC_FRAG_MASK) {
2377 case OFPC_FRAG_NORMAL:
2378 dpif_set_drop_frags(p->dpif, false);
2380 case OFPC_FRAG_DROP:
2381 dpif_set_drop_frags(p->dpif, true);
2384 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2390 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2396 add_output_group_action(struct odp_actions *actions, uint16_t group,
2397 uint16_t *nf_output_iface)
2399 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2401 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2402 *nf_output_iface = NF_OUT_FLOOD;
2407 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2409 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2410 a->controller.arg = max_len;
2413 struct action_xlate_ctx {
2415 flow_t flow; /* Flow to which these actions correspond. */
2416 int recurse; /* Recursion level, via xlate_table_action. */
2417 struct ofproto *ofproto;
2418 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2419 * null pointer if we are revalidating
2420 * without a packet to refer to. */
2423 struct odp_actions *out; /* Datapath actions. */
2424 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2425 bool may_set_up_flow; /* True ordinarily; false if the actions must
2426 * be reassessed for every packet. */
2427 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2430 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2431 struct action_xlate_ctx *ctx);
2434 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2436 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2439 if (ofport->opp.config & OFPPC_NO_FWD) {
2440 /* Forwarding disabled on port. */
2445 * We don't have an ofport record for this port, but it doesn't hurt to
2446 * allow forwarding to it anyhow. Maybe such a port will appear later
2447 * and we're pre-populating the flow table.
2451 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2452 ctx->nf_output_iface = port;
2455 static struct rule *
2456 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2459 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2461 /* The rule we found might not be valid, since we could be in need of
2462 * revalidation. If it is not valid, don't return it. */
2465 && ofproto->need_revalidate
2466 && !revalidate_rule(ofproto, rule)) {
2467 COVERAGE_INC(ofproto_invalidated);
2475 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2477 if (!ctx->recurse) {
2478 uint16_t old_in_port;
2481 /* Look up a flow with 'in_port' as the input port. Then restore the
2482 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2483 * have surprising behavior). */
2484 old_in_port = ctx->flow.in_port;
2485 ctx->flow.in_port = in_port;
2486 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2487 ctx->flow.in_port = old_in_port;
2495 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2502 xlate_output_action__(struct action_xlate_ctx *ctx,
2503 uint16_t port, uint16_t max_len)
2506 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2508 ctx->nf_output_iface = NF_OUT_DROP;
2512 add_output_action(ctx, ctx->flow.in_port);
2515 xlate_table_action(ctx, ctx->flow.in_port);
2518 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2519 ctx->out, ctx->tags,
2520 &ctx->nf_output_iface,
2521 ctx->ofproto->aux)) {
2522 COVERAGE_INC(ofproto_uninstallable);
2523 ctx->may_set_up_flow = false;
2527 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2528 &ctx->nf_output_iface);
2531 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2533 case OFPP_CONTROLLER:
2534 add_controller_action(ctx->out, max_len);
2537 add_output_action(ctx, ODPP_LOCAL);
2540 odp_port = ofp_port_to_odp_port(port);
2541 if (odp_port != ctx->flow.in_port) {
2542 add_output_action(ctx, odp_port);
2547 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2548 ctx->nf_output_iface = NF_OUT_FLOOD;
2549 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2550 ctx->nf_output_iface = prev_nf_output_iface;
2551 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2552 ctx->nf_output_iface != NF_OUT_FLOOD) {
2553 ctx->nf_output_iface = NF_OUT_MULTI;
2558 xlate_output_action(struct action_xlate_ctx *ctx,
2559 const struct ofp_action_output *oao)
2561 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2564 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2565 * optimization, because we're going to add another action that sets the
2566 * priority immediately after, or because there are no actions following the
2569 remove_pop_action(struct action_xlate_ctx *ctx)
2571 size_t n = ctx->out->n_actions;
2572 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2573 ctx->out->n_actions--;
2578 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2579 const struct ofp_action_enqueue *oae)
2581 uint16_t ofp_port, odp_port;
2585 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2588 /* Fall back to ordinary output action. */
2589 xlate_output_action__(ctx, ntohs(oae->port), 0);
2593 /* Figure out ODP output port. */
2594 ofp_port = ntohs(oae->port);
2595 if (ofp_port != OFPP_IN_PORT) {
2596 odp_port = ofp_port_to_odp_port(ofp_port);
2598 odp_port = ctx->flow.in_port;
2601 /* Add ODP actions. */
2602 remove_pop_action(ctx);
2603 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2605 add_output_action(ctx, odp_port);
2606 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2608 /* Update NetFlow output port. */
2609 if (ctx->nf_output_iface == NF_OUT_DROP) {
2610 ctx->nf_output_iface = odp_port;
2611 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2612 ctx->nf_output_iface = NF_OUT_MULTI;
2617 xlate_nicira_action(struct action_xlate_ctx *ctx,
2618 const struct nx_action_header *nah)
2620 const struct nx_action_resubmit *nar;
2621 const struct nx_action_set_tunnel *nast;
2622 union odp_action *oa;
2623 int subtype = ntohs(nah->subtype);
2625 assert(nah->vendor == htonl(NX_VENDOR_ID));
2627 case NXAST_RESUBMIT:
2628 nar = (const struct nx_action_resubmit *) nah;
2629 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2632 case NXAST_SET_TUNNEL:
2633 nast = (const struct nx_action_set_tunnel *) nah;
2634 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2635 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2638 /* If you add a new action here that modifies flow data, don't forget to
2639 * update the flow key in ctx->flow at the same time. */
2642 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2648 do_xlate_actions(const union ofp_action *in, size_t n_in,
2649 struct action_xlate_ctx *ctx)
2651 struct actions_iterator iter;
2652 const union ofp_action *ia;
2653 const struct ofport *port;
2655 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2656 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2657 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2658 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2659 /* Drop this flow. */
2663 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2664 uint16_t type = ntohs(ia->type);
2665 union odp_action *oa;
2669 xlate_output_action(ctx, &ia->output);
2672 case OFPAT_SET_VLAN_VID:
2673 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2674 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2677 case OFPAT_SET_VLAN_PCP:
2678 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2679 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2682 case OFPAT_STRIP_VLAN:
2683 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2684 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2685 ctx->flow.dl_vlan_pcp = 0;
2688 case OFPAT_SET_DL_SRC:
2689 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2690 memcpy(oa->dl_addr.dl_addr,
2691 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2692 memcpy(ctx->flow.dl_src,
2693 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2696 case OFPAT_SET_DL_DST:
2697 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2698 memcpy(oa->dl_addr.dl_addr,
2699 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2700 memcpy(ctx->flow.dl_dst,
2701 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2704 case OFPAT_SET_NW_SRC:
2705 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2706 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2709 case OFPAT_SET_NW_DST:
2710 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2711 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2714 case OFPAT_SET_NW_TOS:
2715 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2716 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2719 case OFPAT_SET_TP_SRC:
2720 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2721 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2724 case OFPAT_SET_TP_DST:
2725 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2726 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2730 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2734 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2738 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2745 xlate_actions(const union ofp_action *in, size_t n_in,
2746 const flow_t *flow, struct ofproto *ofproto,
2747 const struct ofpbuf *packet,
2748 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2749 uint16_t *nf_output_iface)
2751 tag_type no_tags = 0;
2752 struct action_xlate_ctx ctx;
2753 COVERAGE_INC(ofproto_ofp2odp);
2754 odp_actions_init(out);
2757 ctx.ofproto = ofproto;
2758 ctx.packet = packet;
2760 ctx.tags = tags ? tags : &no_tags;
2761 ctx.may_set_up_flow = true;
2762 ctx.nf_output_iface = NF_OUT_DROP;
2763 do_xlate_actions(in, n_in, &ctx);
2764 remove_pop_action(&ctx);
2766 /* Check with in-band control to see if we're allowed to set up this
2768 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2769 ctx.may_set_up_flow = false;
2772 if (may_set_up_flow) {
2773 *may_set_up_flow = ctx.may_set_up_flow;
2775 if (nf_output_iface) {
2776 *nf_output_iface = ctx.nf_output_iface;
2778 if (odp_actions_overflow(out)) {
2779 odp_actions_init(out);
2780 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2785 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2786 * error message code (composed with ofp_mkerr()) for the caller to propagate
2787 * upward. Otherwise, returns 0.
2789 * 'oh' is used to make log messages more informative. */
2791 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2793 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2794 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2797 type_name = ofp_message_type_to_string(oh->type);
2798 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2802 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2809 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2810 struct ofp_header *oh)
2812 struct ofp_packet_out *opo;
2813 struct ofpbuf payload, *buffer;
2814 struct odp_actions actions;
2820 error = reject_slave_controller(ofconn, oh);
2825 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2829 opo = (struct ofp_packet_out *) oh;
2831 COVERAGE_INC(ofproto_packet_out);
2832 if (opo->buffer_id != htonl(UINT32_MAX)) {
2833 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2835 if (error || !buffer) {
2843 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2844 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2845 &flow, p, &payload, &actions, NULL, NULL, NULL);
2850 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2852 ofpbuf_delete(buffer);
2858 update_port_config(struct ofproto *p, struct ofport *port,
2859 uint32_t config, uint32_t mask)
2861 mask &= config ^ port->opp.config;
2862 if (mask & OFPPC_PORT_DOWN) {
2863 if (config & OFPPC_PORT_DOWN) {
2864 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2866 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2869 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2870 if (mask & REVALIDATE_BITS) {
2871 COVERAGE_INC(ofproto_costly_flags);
2872 port->opp.config ^= mask & REVALIDATE_BITS;
2873 p->need_revalidate = true;
2875 #undef REVALIDATE_BITS
2876 if (mask & OFPPC_NO_FLOOD) {
2877 port->opp.config ^= OFPPC_NO_FLOOD;
2878 refresh_port_groups(p);
2880 if (mask & OFPPC_NO_PACKET_IN) {
2881 port->opp.config ^= OFPPC_NO_PACKET_IN;
2886 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2887 struct ofp_header *oh)
2889 const struct ofp_port_mod *opm;
2890 struct ofport *port;
2893 error = reject_slave_controller(ofconn, oh);
2897 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2901 opm = (struct ofp_port_mod *) oh;
2903 port = port_array_get(&p->ports,
2904 ofp_port_to_odp_port(ntohs(opm->port_no)));
2906 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2907 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2908 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2910 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2911 if (opm->advertise) {
2912 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2918 static struct ofpbuf *
2919 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2921 struct ofp_stats_reply *osr;
2924 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2925 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2927 osr->flags = htons(0);
2931 static struct ofpbuf *
2932 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2934 return make_stats_reply(request->header.xid, request->type, body_len);
2938 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2940 struct ofpbuf *msg = *msgp;
2941 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2942 if (nbytes + msg->size > UINT16_MAX) {
2943 struct ofp_stats_reply *reply = msg->data;
2944 reply->flags = htons(OFPSF_REPLY_MORE);
2945 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2946 queue_tx(msg, ofconn, ofconn->reply_counter);
2948 return ofpbuf_put_uninit(*msgp, nbytes);
2952 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2953 struct ofp_stats_request *request)
2955 struct ofp_desc_stats *ods;
2958 msg = start_stats_reply(request, sizeof *ods);
2959 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2960 memset(ods, 0, sizeof *ods);
2961 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2962 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2963 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2964 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2965 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2966 queue_tx(msg, ofconn, ofconn->reply_counter);
2972 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2974 struct rule *rule = rule_from_cls_rule(cls_rule);
2975 int *n_subrules = n_subrules_;
2983 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2984 struct ofp_stats_request *request)
2986 struct ofp_table_stats *ots;
2988 struct odp_stats dpstats;
2989 int n_exact, n_subrules, n_wild;
2991 msg = start_stats_reply(request, sizeof *ots * 2);
2993 /* Count rules of various kinds. */
2995 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2996 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2997 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3000 dpif_get_dp_stats(p->dpif, &dpstats);
3001 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3002 memset(ots, 0, sizeof *ots);
3003 ots->table_id = TABLEID_HASH;
3004 strcpy(ots->name, "hash");
3005 ots->wildcards = htonl(0);
3006 ots->max_entries = htonl(dpstats.max_capacity);
3007 ots->active_count = htonl(n_exact);
3008 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3010 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3012 /* Classifier table. */
3013 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3014 memset(ots, 0, sizeof *ots);
3015 ots->table_id = TABLEID_CLASSIFIER;
3016 strcpy(ots->name, "classifier");
3017 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3019 ots->max_entries = htonl(65536);
3020 ots->active_count = htonl(n_wild);
3021 ots->lookup_count = htonll(0); /* XXX */
3022 ots->matched_count = htonll(0); /* XXX */
3024 queue_tx(msg, ofconn, ofconn->reply_counter);
3029 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
3030 struct ofpbuf **msgp)
3032 struct netdev_stats stats;
3033 struct ofp_port_stats *ops;
3035 /* Intentionally ignore return value, since errors will set
3036 * 'stats' to all-1s, which is correct for OpenFlow, and
3037 * netdev_get_stats() will log errors. */
3038 netdev_get_stats(port->netdev, &stats);
3040 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3041 ops->port_no = htons(odp_port_to_ofp_port(port_no));
3042 memset(ops->pad, 0, sizeof ops->pad);
3043 ops->rx_packets = htonll(stats.rx_packets);
3044 ops->tx_packets = htonll(stats.tx_packets);
3045 ops->rx_bytes = htonll(stats.rx_bytes);
3046 ops->tx_bytes = htonll(stats.tx_bytes);
3047 ops->rx_dropped = htonll(stats.rx_dropped);
3048 ops->tx_dropped = htonll(stats.tx_dropped);
3049 ops->rx_errors = htonll(stats.rx_errors);
3050 ops->tx_errors = htonll(stats.tx_errors);
3051 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3052 ops->rx_over_err = htonll(stats.rx_over_errors);
3053 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3054 ops->collisions = htonll(stats.collisions);
3058 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3059 struct ofp_stats_request *osr,
3062 struct ofp_port_stats_request *psr;
3063 struct ofp_port_stats *ops;
3065 struct ofport *port;
3066 unsigned int port_no;
3068 if (arg_size != sizeof *psr) {
3069 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3071 psr = (struct ofp_port_stats_request *) osr->body;
3073 msg = start_stats_reply(osr, sizeof *ops * 16);
3074 if (psr->port_no != htons(OFPP_NONE)) {
3075 port = port_array_get(&p->ports,
3076 ofp_port_to_odp_port(ntohs(psr->port_no)));
3078 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
3081 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
3082 append_port_stat(port, port_no, ofconn, &msg);
3086 queue_tx(msg, ofconn, ofconn->reply_counter);
3090 struct flow_stats_cbdata {
3091 struct ofproto *ofproto;
3092 struct ofconn *ofconn;
3097 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3098 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3099 * returned statistic include statistics for all of 'rule''s subrules. */
3101 query_stats(struct ofproto *p, struct rule *rule,
3102 uint64_t *packet_countp, uint64_t *byte_countp)
3104 uint64_t packet_count, byte_count;
3105 struct rule *subrule;
3106 struct odp_flow *odp_flows;
3109 /* Start from historical data for 'rule' itself that are no longer tracked
3110 * by the datapath. This counts, for example, subrules that have
3112 packet_count = rule->packet_count;
3113 byte_count = rule->byte_count;
3115 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3116 * wildcarded then on all of its subrules.
3118 * Also, add any statistics that are not tracked by the datapath for each
3119 * subrule. This includes, for example, statistics for packets that were
3120 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3122 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3123 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3124 if (rule->cr.wc.wildcards) {
3126 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
3127 odp_flows[i++].key = subrule->cr.flow;
3128 packet_count += subrule->packet_count;
3129 byte_count += subrule->byte_count;
3132 odp_flows[0].key = rule->cr.flow;
3135 /* Fetch up-to-date statistics from the datapath and add them in. */
3136 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3138 for (i = 0; i < n_odp_flows; i++) {
3139 struct odp_flow *odp_flow = &odp_flows[i];
3140 packet_count += odp_flow->stats.n_packets;
3141 byte_count += odp_flow->stats.n_bytes;
3146 /* Return the stats to the caller. */
3147 *packet_countp = packet_count;
3148 *byte_countp = byte_count;
3152 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3154 struct rule *rule = rule_from_cls_rule(rule_);
3155 struct flow_stats_cbdata *cbdata = cbdata_;
3156 struct ofp_flow_stats *ofs;
3157 uint64_t packet_count, byte_count;
3158 size_t act_len, len;
3159 long long int tdiff = time_msec() - rule->created;
3160 uint32_t sec = tdiff / 1000;
3161 uint32_t msec = tdiff - (sec * 1000);
3163 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3167 act_len = sizeof *rule->actions * rule->n_actions;
3168 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3170 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3172 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3173 ofs->length = htons(len);
3174 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3176 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3177 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3178 ofs->duration_sec = htonl(sec);
3179 ofs->duration_nsec = htonl(msec * 1000000);
3180 ofs->cookie = rule->flow_cookie;
3181 ofs->priority = htons(rule->cr.priority);
3182 ofs->idle_timeout = htons(rule->idle_timeout);
3183 ofs->hard_timeout = htons(rule->hard_timeout);
3184 memset(ofs->pad2, 0, sizeof ofs->pad2);
3185 ofs->packet_count = htonll(packet_count);
3186 ofs->byte_count = htonll(byte_count);
3187 memcpy(ofs->actions, rule->actions, act_len);
3191 table_id_to_include(uint8_t table_id)
3193 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3194 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3195 : table_id == 0xff ? CLS_INC_ALL
3200 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3201 const struct ofp_stats_request *osr,
3204 struct ofp_flow_stats_request *fsr;
3205 struct flow_stats_cbdata cbdata;
3206 struct cls_rule target;
3208 if (arg_size != sizeof *fsr) {
3209 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3211 fsr = (struct ofp_flow_stats_request *) osr->body;
3213 COVERAGE_INC(ofproto_flows_req);
3215 cbdata.ofconn = ofconn;
3216 cbdata.out_port = fsr->out_port;
3217 cbdata.msg = start_stats_reply(osr, 1024);
3218 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3219 classifier_for_each_match(&p->cls, &target,
3220 table_id_to_include(fsr->table_id),
3221 flow_stats_cb, &cbdata);
3222 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3226 struct flow_stats_ds_cbdata {
3227 struct ofproto *ofproto;
3232 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3234 struct rule *rule = rule_from_cls_rule(rule_);
3235 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3236 struct ds *results = cbdata->results;
3237 struct ofp_match match;
3238 uint64_t packet_count, byte_count;
3239 size_t act_len = sizeof *rule->actions * rule->n_actions;
3241 /* Don't report on subrules. */
3242 if (rule->super != NULL) {
3246 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3247 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3248 cbdata->ofproto->tun_id_from_cookie, &match);
3250 ds_put_format(results, "duration=%llds, ",
3251 (time_msec() - rule->created) / 1000);
3252 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3253 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3254 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3255 ofp_print_match(results, &match, true);
3256 ofp_print_actions(results, &rule->actions->header, act_len);
3257 ds_put_cstr(results, "\n");
3260 /* Adds a pretty-printed description of all flows to 'results', including
3261 * those marked hidden by secchan (e.g., by in-band control). */
3263 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3265 struct ofp_match match;
3266 struct cls_rule target;
3267 struct flow_stats_ds_cbdata cbdata;
3269 memset(&match, 0, sizeof match);
3270 match.wildcards = htonl(OVSFW_ALL);
3273 cbdata.results = results;
3275 cls_rule_from_match(&match, 0, false, 0, &target);
3276 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3277 flow_stats_ds_cb, &cbdata);
3280 struct aggregate_stats_cbdata {
3281 struct ofproto *ofproto;
3283 uint64_t packet_count;
3284 uint64_t byte_count;
3289 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3291 struct rule *rule = rule_from_cls_rule(rule_);
3292 struct aggregate_stats_cbdata *cbdata = cbdata_;
3293 uint64_t packet_count, byte_count;
3295 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3299 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3301 cbdata->packet_count += packet_count;
3302 cbdata->byte_count += byte_count;
3307 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3308 const struct ofp_stats_request *osr,
3311 struct ofp_aggregate_stats_request *asr;
3312 struct ofp_aggregate_stats_reply *reply;
3313 struct aggregate_stats_cbdata cbdata;
3314 struct cls_rule target;
3317 if (arg_size != sizeof *asr) {
3318 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3320 asr = (struct ofp_aggregate_stats_request *) osr->body;
3322 COVERAGE_INC(ofproto_agg_request);
3324 cbdata.out_port = asr->out_port;
3325 cbdata.packet_count = 0;
3326 cbdata.byte_count = 0;
3328 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3329 classifier_for_each_match(&p->cls, &target,
3330 table_id_to_include(asr->table_id),
3331 aggregate_stats_cb, &cbdata);
3333 msg = start_stats_reply(osr, sizeof *reply);
3334 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3335 reply->flow_count = htonl(cbdata.n_flows);
3336 reply->packet_count = htonll(cbdata.packet_count);
3337 reply->byte_count = htonll(cbdata.byte_count);
3338 queue_tx(msg, ofconn, ofconn->reply_counter);
3342 struct queue_stats_cbdata {
3343 struct ofconn *ofconn;
3349 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3350 const struct netdev_queue_stats *stats)
3352 struct ofp_queue_stats *reply;
3354 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3355 reply->port_no = htons(cbdata->port_no);
3356 memset(reply->pad, 0, sizeof reply->pad);
3357 reply->queue_id = htonl(queue_id);
3358 reply->tx_bytes = htonll(stats->tx_bytes);
3359 reply->tx_packets = htonll(stats->tx_packets);
3360 reply->tx_errors = htonll(stats->tx_errors);
3364 handle_queue_stats_dump_cb(uint32_t queue_id,
3365 struct netdev_queue_stats *stats,
3368 struct queue_stats_cbdata *cbdata = cbdata_;
3370 put_queue_stats(cbdata, queue_id, stats);
3374 handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
3376 struct queue_stats_cbdata *cbdata)
3378 cbdata->port_no = port_no;
3379 if (queue_id == OFPQ_ALL) {
3380 netdev_dump_queue_stats(port->netdev,
3381 handle_queue_stats_dump_cb, cbdata);
3383 struct netdev_queue_stats stats;
3385 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3386 put_queue_stats(cbdata, queue_id, &stats);
3391 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3392 const struct ofp_stats_request *osr,
3395 struct ofp_queue_stats_request *qsr;
3396 struct queue_stats_cbdata cbdata;
3397 struct ofport *port;
3398 unsigned int port_no;
3401 if (arg_size != sizeof *qsr) {
3402 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3404 qsr = (struct ofp_queue_stats_request *) osr->body;
3406 COVERAGE_INC(ofproto_queue_req);
3408 cbdata.ofconn = ofconn;
3409 cbdata.msg = start_stats_reply(osr, 128);
3411 port_no = ntohs(qsr->port_no);
3412 queue_id = ntohl(qsr->queue_id);
3413 if (port_no == OFPP_ALL) {
3414 PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
3415 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3417 } else if (port_no < ofproto->max_ports) {
3418 port = port_array_get(&ofproto->ports, port_no);
3420 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3423 ofpbuf_delete(cbdata.msg);
3424 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3426 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3432 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3433 struct ofp_header *oh)
3435 struct ofp_stats_request *osr;
3439 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3444 osr = (struct ofp_stats_request *) oh;
3446 switch (ntohs(osr->type)) {
3448 return handle_desc_stats_request(p, ofconn, osr);
3451 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3453 case OFPST_AGGREGATE:
3454 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3457 return handle_table_stats_request(p, ofconn, osr);
3460 return handle_port_stats_request(p, ofconn, osr, arg_size);
3463 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3466 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3469 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3473 static long long int
3474 msec_from_nsec(uint64_t sec, uint32_t nsec)
3476 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3480 update_time(struct ofproto *ofproto, struct rule *rule,
3481 const struct odp_flow_stats *stats)
3483 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3484 if (used > rule->used) {
3486 if (rule->super && used > rule->super->used) {
3487 rule->super->used = used;
3489 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3494 update_stats(struct ofproto *ofproto, struct rule *rule,
3495 const struct odp_flow_stats *stats)
3497 if (stats->n_packets) {
3498 update_time(ofproto, rule, stats);
3499 rule->packet_count += stats->n_packets;
3500 rule->byte_count += stats->n_bytes;
3501 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3505 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3506 * in which no matching flow already exists in the flow table.
3508 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3509 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3510 * code as encoded by ofp_mkerr() on failure.
3512 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3515 add_flow(struct ofproto *p, struct ofconn *ofconn,
3516 const struct ofp_flow_mod *ofm, size_t n_actions)
3518 struct ofpbuf *packet;
3523 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3527 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3529 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3530 ntohs(ofm->priority))) {
3531 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3535 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3536 n_actions, ntohs(ofm->idle_timeout),
3537 ntohs(ofm->hard_timeout), ofm->cookie,
3538 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3539 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3540 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3543 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3544 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3548 in_port = UINT16_MAX;
3551 rule_insert(p, rule, packet, in_port);
3552 ofpbuf_delete(packet);
3556 static struct rule *
3557 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3562 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3564 return rule_from_cls_rule(classifier_find_rule_exactly(
3565 &p->cls, &flow, wildcards,
3566 ntohs(ofm->priority)));
3570 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3571 struct rule *rule, const struct ofp_flow_mod *ofm)
3573 struct ofpbuf *packet;
3578 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3582 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3588 flow_extract(packet, 0, in_port, &flow);
3589 rule_execute(ofproto, rule, packet, &flow);
3590 ofpbuf_delete(packet);
3595 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3597 struct modify_flows_cbdata {
3598 struct ofproto *ofproto;
3599 const struct ofp_flow_mod *ofm;
3604 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3605 size_t n_actions, struct rule *);
3606 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3608 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3609 * encoded by ofp_mkerr() on failure.
3611 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3614 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3615 const struct ofp_flow_mod *ofm, size_t n_actions)
3617 struct modify_flows_cbdata cbdata;
3618 struct cls_rule target;
3622 cbdata.n_actions = n_actions;
3623 cbdata.match = NULL;
3625 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3628 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3629 modify_flows_cb, &cbdata);
3631 /* This credits the packet to whichever flow happened to happened to
3632 * match last. That's weird. Maybe we should do a lookup for the
3633 * flow that actually matches the packet? Who knows. */
3634 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3637 return add_flow(p, ofconn, ofm, n_actions);
3641 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3642 * code as encoded by ofp_mkerr() on failure.
3644 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3647 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3648 struct ofp_flow_mod *ofm, size_t n_actions)
3650 struct rule *rule = find_flow_strict(p, ofm);
3651 if (rule && !rule_is_hidden(rule)) {
3652 modify_flow(p, ofm, n_actions, rule);
3653 return send_buffered_packet(p, ofconn, rule, ofm);
3655 return add_flow(p, ofconn, ofm, n_actions);
3659 /* Callback for modify_flows_loose(). */
3661 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3663 struct rule *rule = rule_from_cls_rule(rule_);
3664 struct modify_flows_cbdata *cbdata = cbdata_;
3666 if (!rule_is_hidden(rule)) {
3667 cbdata->match = rule;
3668 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3672 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3673 * been identified as a flow in 'p''s flow table to be modified, by changing
3674 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3675 * ofp_action[] structures). */
3677 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3678 size_t n_actions, struct rule *rule)
3680 size_t actions_len = n_actions * sizeof *rule->actions;
3682 rule->flow_cookie = ofm->cookie;
3684 /* If the actions are the same, do nothing. */
3685 if (n_actions == rule->n_actions
3686 && !memcmp(ofm->actions, rule->actions, actions_len))
3691 /* Replace actions. */
3692 free(rule->actions);
3693 rule->actions = xmemdup(ofm->actions, actions_len);
3694 rule->n_actions = n_actions;
3696 /* Make sure that the datapath gets updated properly. */
3697 if (rule->cr.wc.wildcards) {
3698 COVERAGE_INC(ofproto_mod_wc_flow);
3699 p->need_revalidate = true;
3701 rule_update_actions(p, rule);
3707 /* OFPFC_DELETE implementation. */
3709 struct delete_flows_cbdata {
3710 struct ofproto *ofproto;
3714 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3715 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3717 /* Implements OFPFC_DELETE. */
3719 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3721 struct delete_flows_cbdata cbdata;
3722 struct cls_rule target;
3725 cbdata.out_port = ofm->out_port;
3727 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3730 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3731 delete_flows_cb, &cbdata);
3734 /* Implements OFPFC_DELETE_STRICT. */
3736 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3738 struct rule *rule = find_flow_strict(p, ofm);
3740 delete_flow(p, rule, ofm->out_port);
3744 /* Callback for delete_flows_loose(). */
3746 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3748 struct rule *rule = rule_from_cls_rule(rule_);
3749 struct delete_flows_cbdata *cbdata = cbdata_;
3751 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3754 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3755 * been identified as a flow to delete from 'p''s flow table, by deleting the
3756 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3759 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3760 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3761 * specified 'out_port'. */
3763 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3765 if (rule_is_hidden(rule)) {
3769 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3773 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3774 rule_remove(p, rule);
3778 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3779 struct ofp_flow_mod *ofm)
3781 struct ofp_match orig_match;
3785 error = reject_slave_controller(ofconn, &ofm->header);
3789 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3790 sizeof *ofm->actions, &n_actions);
3795 /* We do not support the emergency flow cache. It will hopefully
3796 * get dropped from OpenFlow in the near future. */
3797 if (ofm->flags & htons(OFPFF_EMERG)) {
3798 /* There isn't a good fit for an error code, so just state that the
3799 * flow table is full. */
3800 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3803 /* Normalize ofp->match. If normalization actually changes anything, then
3804 * log the differences. */
3805 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3806 orig_match = ofm->match;
3807 normalize_match(&ofm->match);
3808 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3809 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3810 if (!VLOG_DROP_INFO(&normal_rl)) {
3811 char *old = ofp_match_to_literal_string(&orig_match);
3812 char *new = ofp_match_to_literal_string(&ofm->match);
3813 VLOG_INFO("%s: normalization changed ofp_match, details:",
3814 rconn_get_name(ofconn->rconn));
3815 VLOG_INFO(" pre: %s", old);
3816 VLOG_INFO("post: %s", new);
3822 if (!ofm->match.wildcards) {
3823 ofm->priority = htons(UINT16_MAX);
3826 error = validate_actions((const union ofp_action *) ofm->actions,
3827 n_actions, p->max_ports);
3832 switch (ntohs(ofm->command)) {
3834 return add_flow(p, ofconn, ofm, n_actions);
3837 return modify_flows_loose(p, ofconn, ofm, n_actions);
3839 case OFPFC_MODIFY_STRICT:
3840 return modify_flow_strict(p, ofconn, ofm, n_actions);
3843 delete_flows_loose(p, ofm);
3846 case OFPFC_DELETE_STRICT:
3847 delete_flow_strict(p, ofm);
3851 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3856 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3860 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3865 p->tun_id_from_cookie = !!msg->set;
3870 handle_role_request(struct ofproto *ofproto,
3871 struct ofconn *ofconn, struct nicira_header *msg)
3873 struct nx_role_request *nrr;
3874 struct nx_role_request *reply;
3878 if (ntohs(msg->header.length) != sizeof *nrr) {
3879 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3880 ntohs(msg->header.length), sizeof *nrr);
3881 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3883 nrr = (struct nx_role_request *) msg;
3885 if (ofconn->type != OFCONN_PRIMARY) {
3886 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3888 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3891 role = ntohl(nrr->role);
3892 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3893 && role != NX_ROLE_SLAVE) {
3894 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3896 /* There's no good error code for this. */
3897 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3900 if (role == NX_ROLE_MASTER) {
3901 struct ofconn *other;
3903 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3904 &ofproto->controllers) {
3905 if (other->role == NX_ROLE_MASTER) {
3906 other->role = NX_ROLE_SLAVE;
3910 ofconn->role = role;
3912 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3914 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3915 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3916 reply->role = htonl(role);
3917 queue_tx(buf, ofconn, ofconn->reply_counter);
3923 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3925 struct ofp_vendor_header *ovh = msg;
3926 struct nicira_header *nh;
3928 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3929 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3930 "(expected at least %zu)",
3931 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3932 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3934 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3935 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3937 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3938 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3939 "(expected at least %zu)",
3940 ntohs(ovh->header.length), sizeof(struct nicira_header));
3941 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3945 switch (ntohl(nh->subtype)) {
3946 case NXT_STATUS_REQUEST:
3947 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3950 case NXT_TUN_ID_FROM_COOKIE:
3951 return handle_tun_id_from_cookie(p, msg);
3953 case NXT_ROLE_REQUEST:
3954 return handle_role_request(p, ofconn, msg);
3957 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3961 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3963 struct ofp_header *ob;
3966 /* Currently, everything executes synchronously, so we can just
3967 * immediately send the barrier reply. */
3968 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3969 queue_tx(buf, ofconn, ofconn->reply_counter);
3974 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3975 struct ofpbuf *ofp_msg)
3977 struct ofp_header *oh = ofp_msg->data;
3980 COVERAGE_INC(ofproto_recv_openflow);
3982 case OFPT_ECHO_REQUEST:
3983 error = handle_echo_request(ofconn, oh);
3986 case OFPT_ECHO_REPLY:
3990 case OFPT_FEATURES_REQUEST:
3991 error = handle_features_request(p, ofconn, oh);
3994 case OFPT_GET_CONFIG_REQUEST:
3995 error = handle_get_config_request(p, ofconn, oh);
3998 case OFPT_SET_CONFIG:
3999 error = handle_set_config(p, ofconn, ofp_msg->data);
4002 case OFPT_PACKET_OUT:
4003 error = handle_packet_out(p, ofconn, ofp_msg->data);
4007 error = handle_port_mod(p, ofconn, oh);
4011 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4014 case OFPT_STATS_REQUEST:
4015 error = handle_stats_request(p, ofconn, oh);
4019 error = handle_vendor(p, ofconn, ofp_msg->data);
4022 case OFPT_BARRIER_REQUEST:
4023 error = handle_barrier_request(ofconn, oh);
4027 if (VLOG_IS_WARN_ENABLED()) {
4028 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4029 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4032 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4037 send_error_oh(ofconn, ofp_msg->data, error);
4042 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4044 struct odp_msg *msg = packet->data;
4046 struct ofpbuf payload;
4049 payload.data = msg + 1;
4050 payload.size = msg->length - sizeof *msg;
4051 flow_extract(&payload, msg->arg, msg->port, &flow);
4053 /* Check with in-band control to see if this packet should be sent
4054 * to the local port regardless of the flow table. */
4055 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4056 union odp_action action;
4058 memset(&action, 0, sizeof(action));
4059 action.output.type = ODPAT_OUTPUT;
4060 action.output.port = ODPP_LOCAL;
4061 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
4064 rule = lookup_valid_rule(p, &flow);
4066 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4067 struct ofport *port = port_array_get(&p->ports, msg->port);
4069 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4070 COVERAGE_INC(ofproto_no_packet_in);
4071 /* XXX install 'drop' flow entry */
4072 ofpbuf_delete(packet);
4076 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4079 COVERAGE_INC(ofproto_packet_in);
4080 send_packet_in(p, packet);
4084 if (rule->cr.wc.wildcards) {
4085 rule = rule_create_subrule(p, rule, &flow);
4086 rule_make_actions(p, rule, packet);
4088 if (!rule->may_install) {
4089 /* The rule is not installable, that is, we need to process every
4090 * packet, so process the current packet and set its actions into
4092 rule_make_actions(p, rule, packet);
4094 /* XXX revalidate rule if it needs it */
4098 rule_execute(p, rule, &payload, &flow);
4099 rule_reinstall(p, rule);
4101 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4103 * Extra-special case for fail-open mode.
4105 * We are in fail-open mode and the packet matched the fail-open rule,
4106 * but we are connected to a controller too. We should send the packet
4107 * up to the controller in the hope that it will try to set up a flow
4108 * and thereby allow us to exit fail-open.
4110 * See the top-level comment in fail-open.c for more information.
4112 send_packet_in(p, packet);
4114 ofpbuf_delete(packet);
4119 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4121 struct odp_msg *msg = packet->data;
4123 switch (msg->type) {
4124 case _ODPL_ACTION_NR:
4125 COVERAGE_INC(ofproto_ctlr_action);
4126 send_packet_in(p, packet);
4129 case _ODPL_SFLOW_NR:
4131 ofproto_sflow_received(p->sflow, msg);
4133 ofpbuf_delete(packet);
4137 handle_odp_miss_msg(p, packet);
4141 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4148 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4150 struct rule *sub = rule_from_cls_rule(sub_);
4151 struct revalidate_cbdata *cbdata = cbdata_;
4153 if (cbdata->revalidate_all
4154 || (cbdata->revalidate_subrules && sub->super)
4155 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4156 revalidate_rule(cbdata->ofproto, sub);
4161 revalidate_rule(struct ofproto *p, struct rule *rule)
4163 const flow_t *flow = &rule->cr.flow;
4165 COVERAGE_INC(ofproto_revalidate_rule);
4168 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4170 rule_remove(p, rule);
4172 } else if (super != rule->super) {
4173 COVERAGE_INC(ofproto_revalidate_moved);
4174 list_remove(&rule->list);
4175 list_push_back(&super->list, &rule->list);
4176 rule->super = super;
4177 rule->hard_timeout = super->hard_timeout;
4178 rule->idle_timeout = super->idle_timeout;
4179 rule->created = super->created;
4184 rule_update_actions(p, rule);
4188 static struct ofpbuf *
4189 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4190 long long int now, uint8_t reason)
4192 struct ofp_flow_removed *ofr;
4194 long long int tdiff = now - rule->created;
4195 uint32_t sec = tdiff / 1000;
4196 uint32_t msec = tdiff - (sec * 1000);
4198 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4199 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4201 ofr->cookie = rule->flow_cookie;
4202 ofr->priority = htons(rule->cr.priority);
4203 ofr->reason = reason;
4204 ofr->duration_sec = htonl(sec);
4205 ofr->duration_nsec = htonl(msec * 1000000);
4206 ofr->idle_timeout = htons(rule->idle_timeout);
4207 ofr->packet_count = htonll(rule->packet_count);
4208 ofr->byte_count = htonll(rule->byte_count);
4214 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
4216 assert(rule->installed);
4217 assert(!rule->cr.wc.wildcards);
4220 rule_remove(ofproto, rule);
4222 rule_uninstall(ofproto, rule);
4227 send_flow_removed(struct ofproto *p, struct rule *rule,
4228 long long int now, uint8_t reason)
4230 struct ofconn *ofconn;
4231 struct ofconn *prev;
4232 struct ofpbuf *buf = NULL;
4234 /* We limit the maximum number of queued flow expirations it by accounting
4235 * them under the counter for replies. That works because preventing
4236 * OpenFlow requests from being processed also prevents new flows from
4237 * being added (and expiring). (It also prevents processing OpenFlow
4238 * requests that would not add new flows, so it is imperfect.) */
4241 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4242 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
4243 && ofconn_receives_async_msgs(ofconn)) {
4245 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4247 buf = compose_flow_removed(p, rule, now, reason);
4253 queue_tx(buf, prev, prev->reply_counter);
4259 expire_rule(struct cls_rule *cls_rule, void *p_)
4261 struct ofproto *p = p_;
4262 struct rule *rule = rule_from_cls_rule(cls_rule);
4263 long long int hard_expire, idle_expire, expire, now;
4265 hard_expire = (rule->hard_timeout
4266 ? rule->created + rule->hard_timeout * 1000
4268 idle_expire = (rule->idle_timeout
4269 && (rule->super || list_is_empty(&rule->list))
4270 ? rule->used + rule->idle_timeout * 1000
4272 expire = MIN(hard_expire, idle_expire);
4276 if (rule->installed && now >= rule->used + 5000) {
4277 uninstall_idle_flow(p, rule);
4278 } else if (!rule->cr.wc.wildcards) {
4279 active_timeout(p, rule);
4285 COVERAGE_INC(ofproto_expired);
4287 /* Update stats. This code will be a no-op if the rule expired
4288 * due to an idle timeout. */
4289 if (rule->cr.wc.wildcards) {
4290 struct rule *subrule, *next;
4291 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
4292 rule_remove(p, subrule);
4295 rule_uninstall(p, rule);
4298 if (!rule_is_hidden(rule)) {
4299 send_flow_removed(p, rule, now,
4301 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4303 rule_remove(p, rule);
4307 active_timeout(struct ofproto *ofproto, struct rule *rule)
4309 if (ofproto->netflow && !is_controller_rule(rule) &&
4310 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4311 struct ofexpired expired;
4312 struct odp_flow odp_flow;
4314 /* Get updated flow stats. */
4315 memset(&odp_flow, 0, sizeof odp_flow);
4316 if (rule->installed) {
4317 odp_flow.key = rule->cr.flow;
4318 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4319 dpif_flow_get(ofproto->dpif, &odp_flow);
4321 if (odp_flow.stats.n_packets) {
4322 update_time(ofproto, rule, &odp_flow.stats);
4323 netflow_flow_update_flags(&rule->nf_flow,
4324 odp_flow.stats.tcp_flags);
4328 expired.flow = rule->cr.flow;
4329 expired.packet_count = rule->packet_count +
4330 odp_flow.stats.n_packets;
4331 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4332 expired.used = rule->used;
4334 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4336 /* Schedule us to send the accumulated records once we have
4337 * collected all of them. */
4338 poll_immediate_wake();
4343 update_used(struct ofproto *p)
4345 struct odp_flow *flows;
4350 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4355 for (i = 0; i < n_flows; i++) {
4356 struct odp_flow *f = &flows[i];
4359 rule = rule_from_cls_rule(
4360 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4361 if (!rule || !rule->installed) {
4362 COVERAGE_INC(ofproto_unexpected_rule);
4363 dpif_flow_del(p->dpif, f);
4367 update_time(p, rule, &f->stats);
4368 rule_account(p, rule, f->stats.n_bytes);
4373 /* pinsched callback for sending 'packet' on 'ofconn'. */
4375 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4377 struct ofconn *ofconn = ofconn_;
4379 rconn_send_with_limit(ofconn->rconn, packet,
4380 ofconn->packet_in_counter, 100);
4383 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4384 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4385 * packet scheduler for sending.
4387 * 'max_len' specifies the maximum number of bytes of the packet to send on
4388 * 'ofconn' (INT_MAX specifies no limit).
4390 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4391 * ownership is transferred to this function. */
4393 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4396 struct ofproto *ofproto = ofconn->ofproto;
4397 struct ofp_packet_in *opi = packet->data;
4398 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4399 int send_len, trim_size;
4403 if (opi->reason == OFPR_ACTION) {
4404 buffer_id = UINT32_MAX;
4405 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4406 buffer_id = pktbuf_get_null();
4407 } else if (!ofconn->pktbuf) {
4408 buffer_id = UINT32_MAX;
4410 struct ofpbuf payload;
4411 payload.data = opi->data;
4412 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4413 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4416 /* Figure out how much of the packet to send. */
4417 send_len = ntohs(opi->total_len);
4418 if (buffer_id != UINT32_MAX) {
4419 send_len = MIN(send_len, ofconn->miss_send_len);
4421 send_len = MIN(send_len, max_len);
4423 /* Adjust packet length and clone if necessary. */
4424 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4426 packet = ofpbuf_clone_data(packet->data, trim_size);
4429 packet->size = trim_size;
4432 /* Update packet headers. */
4433 opi->buffer_id = htonl(buffer_id);
4434 update_openflow_length(packet);
4436 /* Hand over to packet scheduler. It might immediately call into
4437 * do_send_packet_in() or it might buffer it for a while (until a later
4438 * call to pinsched_run()). */
4439 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4440 packet, do_send_packet_in, ofconn);
4443 /* Replace struct odp_msg header in 'packet' by equivalent struct
4444 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4445 * returned by dpif_recv()).
4447 * The conversion is not complete: the caller still needs to trim any unneeded
4448 * payload off the end of the buffer, set the length in the OpenFlow header,
4449 * and set buffer_id. Those require us to know the controller settings and so
4450 * must be done on a per-controller basis.
4452 * Returns the maximum number of bytes of the packet that should be sent to
4453 * the controller (INT_MAX if no limit). */
4455 do_convert_to_packet_in(struct ofpbuf *packet)
4457 struct odp_msg *msg = packet->data;
4458 struct ofp_packet_in *opi;
4464 /* Extract relevant header fields */
4465 if (msg->type == _ODPL_ACTION_NR) {
4466 reason = OFPR_ACTION;
4469 reason = OFPR_NO_MATCH;
4472 total_len = msg->length - sizeof *msg;
4473 in_port = odp_port_to_ofp_port(msg->port);
4475 /* Repurpose packet buffer by overwriting header. */
4476 ofpbuf_pull(packet, sizeof(struct odp_msg));
4477 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4478 opi->header.version = OFP_VERSION;
4479 opi->header.type = OFPT_PACKET_IN;
4480 opi->total_len = htons(total_len);
4481 opi->in_port = htons(in_port);
4482 opi->reason = reason;
4487 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4488 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4489 * as necessary according to their individual configurations.
4491 * 'packet' must have sufficient headroom to convert it into a struct
4492 * ofp_packet_in (e.g. as returned by dpif_recv()).
4494 * Takes ownership of 'packet'. */
4496 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4498 struct ofconn *ofconn, *prev;
4501 max_len = do_convert_to_packet_in(packet);
4504 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4505 if (ofconn_receives_async_msgs(ofconn)) {
4507 schedule_packet_in(prev, packet, max_len, true);
4513 schedule_packet_in(prev, packet, max_len, false);
4515 ofpbuf_delete(packet);
4520 pick_datapath_id(const struct ofproto *ofproto)
4522 const struct ofport *port;
4524 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4526 uint8_t ea[ETH_ADDR_LEN];
4529 error = netdev_get_etheraddr(port->netdev, ea);
4531 return eth_addr_to_uint64(ea);
4533 VLOG_WARN("could not get MAC address for %s (%s)",
4534 netdev_get_name(port->netdev), strerror(error));
4536 return ofproto->fallback_dpid;
4540 pick_fallback_dpid(void)
4542 uint8_t ea[ETH_ADDR_LEN];
4543 eth_addr_nicira_random(ea);
4544 return eth_addr_to_uint64(ea);
4548 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4549 struct odp_actions *actions, tag_type *tags,
4550 uint16_t *nf_output_iface, void *ofproto_)
4552 struct ofproto *ofproto = ofproto_;
4555 /* Drop frames for reserved multicast addresses. */
4556 if (eth_addr_is_reserved(flow->dl_dst)) {
4560 /* Learn source MAC (but don't try to learn from revalidation). */
4561 if (packet != NULL) {
4562 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4564 GRAT_ARP_LOCK_NONE);
4566 /* The log messages here could actually be useful in debugging,
4567 * so keep the rate limit relatively high. */
4568 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4569 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4570 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4571 ofproto_revalidate(ofproto, rev_tag);
4575 /* Determine output port. */
4576 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4579 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4580 } else if (out_port != flow->in_port) {
4581 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4582 *nf_output_iface = out_port;
4590 static const struct ofhooks default_ofhooks = {
4592 default_normal_ofhook_cb,