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"
36 #include "mac-learning.h"
40 #include "ofp-print.h"
42 #include "ofproto-sflow.h"
44 #include "openflow/nicira-ext.h"
45 #include "openflow/openflow.h"
46 #include "openvswitch/datapath-protocol.h"
50 #include "poll-loop.h"
54 #include "stream-ssl.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto)
65 #include "sflow_api.h"
69 TABLEID_CLASSIFIER = 1
74 struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */
75 struct netdev *netdev;
76 struct ofp_phy_port opp; /* In host byte order. */
80 static void ofport_free(struct ofport *);
81 static void hton_ofp_phy_port(struct ofp_phy_port *);
83 static int xlate_actions(const union ofp_action *in, size_t n_in,
84 const flow_t *flow, struct ofproto *ofproto,
85 const struct ofpbuf *packet,
86 struct odp_actions *out, tag_type *tags,
87 bool *may_set_up_flow, uint16_t *nf_output_iface);
92 uint64_t flow_cookie; /* Controller-issued identifier.
93 (Kept in network-byte order.) */
94 uint16_t idle_timeout; /* In seconds from time of last use. */
95 uint16_t hard_timeout; /* In seconds from time of creation. */
96 bool send_flow_removed; /* Send a flow removed message? */
97 long long int used; /* Last-used time (0 if never used). */
98 long long int created; /* Creation time. */
99 uint64_t packet_count; /* Number of packets received. */
100 uint64_t byte_count; /* Number of bytes received. */
101 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
102 tag_type tags; /* Tags (set only by hooks). */
103 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
105 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
106 * exact-match rule (having cr.wc.wildcards of 0) generated from the
107 * wildcard rule 'super'. In this case, 'list' is an element of the
110 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
111 * a list of subrules. A super-rule with no wildcards (where
112 * cr.wc.wildcards is 0) will never have any subrules. */
118 * 'n_actions' is the number of elements in the 'actions' array. A single
119 * action may take up more more than one element's worth of space.
121 * A subrule has no actions (it uses the super-rule's actions). */
123 union ofp_action *actions;
127 * A super-rule with wildcard fields never has ODP actions (since the
128 * datapath only supports exact-match flows). */
129 bool installed; /* Installed in datapath? */
130 bool may_install; /* True ordinarily; false if actions must
131 * be reassessed for every packet. */
133 union odp_action *odp_actions;
137 rule_is_hidden(const struct rule *rule)
139 /* Subrules are merely an implementation detail, so hide them from the
141 if (rule->super != NULL) {
145 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
146 * (e.g. by in-band control) and are intentionally hidden from the
148 if (rule->cr.priority > UINT16_MAX) {
155 static struct rule *rule_create(struct ofproto *, struct rule *super,
156 const union ofp_action *, size_t n_actions,
157 uint16_t idle_timeout, uint16_t hard_timeout,
158 uint64_t flow_cookie, bool send_flow_removed);
159 static void rule_free(struct rule *);
160 static void rule_destroy(struct ofproto *, struct rule *);
161 static struct rule *rule_from_cls_rule(const struct cls_rule *);
162 static void rule_insert(struct ofproto *, struct rule *,
163 struct ofpbuf *packet, uint16_t in_port);
164 static void rule_remove(struct ofproto *, struct rule *);
165 static bool rule_make_actions(struct ofproto *, struct rule *,
166 const struct ofpbuf *packet);
167 static void rule_install(struct ofproto *, struct rule *,
168 struct rule *displaced_rule);
169 static void rule_uninstall(struct ofproto *, struct rule *);
170 static void rule_post_uninstall(struct ofproto *, struct rule *);
171 static void send_flow_removed(struct ofproto *p, struct rule *rule,
172 long long int now, uint8_t reason);
174 /* ofproto supports two kinds of OpenFlow connections:
176 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
177 * maintains persistent connections to these controllers and by default
178 * sends them asynchronous messages such as packet-ins.
180 * - "Service" connections, e.g. from ovs-ofctl. When these connections
181 * drop, it is the other side's responsibility to reconnect them if
182 * necessary. ofproto does not send them asynchronous messages by default.
184 * Currently, active (tcp, ssl, unix) connections are always "primary"
185 * connections and passive (ptcp, pssl, punix) connections are always "service"
186 * connections. There is no inherent reason for this, but it reflects the
190 OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */
191 OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */
194 /* A listener for incoming OpenFlow "service" connections. */
196 struct hmap_node node; /* In struct ofproto's "services" hmap. */
197 struct pvconn *pvconn; /* OpenFlow connection listener. */
199 /* These are not used by ofservice directly. They are settings for
200 * accepted "struct ofconn"s from the pvconn. */
201 int probe_interval; /* Max idle time before probing, in seconds. */
202 int rate_limit; /* Max packet-in rate in packets per second. */
203 int burst_limit; /* Limit on accumulating packet credits. */
206 static struct ofservice *ofservice_lookup(struct ofproto *,
208 static int ofservice_create(struct ofproto *,
209 const struct ofproto_controller *);
210 static void ofservice_reconfigure(struct ofservice *,
211 const struct ofproto_controller *);
212 static void ofservice_destroy(struct ofproto *, struct ofservice *);
214 /* An OpenFlow connection. */
216 struct ofproto *ofproto; /* The ofproto that owns this connection. */
217 struct list node; /* In struct ofproto's "all_conns" list. */
218 struct rconn *rconn; /* OpenFlow connection. */
219 enum ofconn_type type; /* Type. */
221 /* OFPT_PACKET_IN related data. */
222 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
223 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
224 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
225 int miss_send_len; /* Bytes to send of buffered packets. */
227 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
228 * requests, and the maximum number before we stop reading OpenFlow
230 #define OFCONN_REPLY_MAX 100
231 struct rconn_packet_counter *reply_counter;
233 /* type == OFCONN_PRIMARY only. */
234 enum nx_role role; /* Role. */
235 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
236 struct discovery *discovery; /* Controller discovery object, if enabled. */
237 struct status_category *ss; /* Switch status category. */
238 enum ofproto_band band; /* In-band or out-of-band? */
241 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
242 * "schedulers" array. Their values are 0 and 1, and their meanings and values
243 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
244 * case anything ever changes, check their values here. */
245 #define N_SCHEDULERS 2
246 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
247 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
248 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
249 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
251 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
253 static void ofconn_destroy(struct ofconn *);
254 static void ofconn_run(struct ofconn *, struct ofproto *);
255 static void ofconn_wait(struct ofconn *);
256 static bool ofconn_receives_async_msgs(const struct ofconn *);
257 static char *ofconn_make_name(const struct ofproto *, const char *target);
258 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
260 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
261 struct rconn_packet_counter *counter);
263 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
264 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
268 uint64_t datapath_id; /* Datapath ID. */
269 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
270 char *mfr_desc; /* Manufacturer. */
271 char *hw_desc; /* Hardware. */
272 char *sw_desc; /* Software version. */
273 char *serial_desc; /* Serial number. */
274 char *dp_desc; /* Datapath description. */
278 struct netdev_monitor *netdev_monitor;
279 struct hmap ports; /* Contains "struct ofport"s. */
280 struct shash port_by_name;
284 struct switch_status *switch_status;
285 struct fail_open *fail_open;
286 struct netflow *netflow;
287 struct ofproto_sflow *sflow;
289 /* In-band control. */
290 struct in_band *in_band;
291 long long int next_in_band_update;
292 struct sockaddr_in *extra_in_band_remotes;
293 size_t n_extra_remotes;
296 struct classifier cls;
297 bool need_revalidate;
298 long long int next_expiration;
299 struct tag_set revalidate_set;
300 bool tun_id_from_cookie;
302 /* OpenFlow connections. */
303 struct hmap controllers; /* Controller "struct ofconn"s. */
304 struct list all_conns; /* Contains "struct ofconn"s. */
305 enum ofproto_fail_mode fail_mode;
307 /* OpenFlow listeners. */
308 struct hmap services; /* Contains "struct ofservice"s. */
309 struct pvconn **snoops;
312 /* Hooks for ovs-vswitchd. */
313 const struct ofhooks *ofhooks;
316 /* Used by default ofhooks. */
317 struct mac_learning *ml;
320 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
322 static const struct ofhooks default_ofhooks;
324 static uint64_t pick_datapath_id(const struct ofproto *);
325 static uint64_t pick_fallback_dpid(void);
327 static int ofproto_expire(struct ofproto *);
329 static void update_stats(struct ofproto *, struct rule *,
330 const struct odp_flow_stats *);
331 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
332 static void revalidate_cb(struct cls_rule *rule_, void *p_);
334 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
336 static void handle_openflow(struct ofconn *, struct ofproto *,
339 static void refresh_port_groups(struct ofproto *);
341 static struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
342 static void update_port(struct ofproto *, const char *devname);
343 static int init_ports(struct ofproto *);
344 static void reinit_ports(struct ofproto *);
347 ofproto_create(const char *datapath, const char *datapath_type,
348 const struct ofhooks *ofhooks, void *aux,
349 struct ofproto **ofprotop)
351 struct odp_stats stats;
358 /* Connect to datapath and start listening for messages. */
359 error = dpif_open(datapath, datapath_type, &dpif);
361 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
364 error = dpif_get_dp_stats(dpif, &stats);
366 VLOG_ERR("failed to obtain stats for datapath %s: %s",
367 datapath, strerror(error));
371 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
373 VLOG_ERR("failed to listen on datapath %s: %s",
374 datapath, strerror(error));
378 dpif_flow_flush(dpif);
379 dpif_recv_purge(dpif);
381 /* Initialize settings. */
382 p = xzalloc(sizeof *p);
383 p->fallback_dpid = pick_fallback_dpid();
384 p->datapath_id = p->fallback_dpid;
385 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
386 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
387 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
388 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
389 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
391 /* Initialize datapath. */
393 p->netdev_monitor = netdev_monitor_create();
394 hmap_init(&p->ports);
395 shash_init(&p->port_by_name);
396 p->max_ports = stats.max_ports;
398 /* Initialize submodules. */
399 p->switch_status = switch_status_create(p);
405 /* Initialize flow table. */
406 classifier_init(&p->cls);
407 p->need_revalidate = false;
408 p->next_expiration = time_msec() + 1000;
409 tag_set_init(&p->revalidate_set);
411 /* Initialize OpenFlow connections. */
412 list_init(&p->all_conns);
413 hmap_init(&p->controllers);
414 hmap_init(&p->services);
418 /* Initialize hooks. */
420 p->ofhooks = ofhooks;
424 p->ofhooks = &default_ofhooks;
426 p->ml = mac_learning_create();
429 /* Pick final datapath ID. */
430 p->datapath_id = pick_datapath_id(p);
431 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
438 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
440 uint64_t old_dpid = p->datapath_id;
441 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
442 if (p->datapath_id != old_dpid) {
443 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
445 /* Force all active connections to reconnect, since there is no way to
446 * notify a controller that the datapath ID has changed. */
447 ofproto_reconnect_controllers(p);
452 is_discovery_controller(const struct ofproto_controller *c)
454 return !strcmp(c->target, "discover");
458 is_in_band_controller(const struct ofproto_controller *c)
460 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
463 /* Creates a new controller in 'ofproto'. Some of the settings are initially
464 * drawn from 'c', but update_controller() needs to be called later to finish
465 * the new ofconn's configuration. */
467 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
469 struct discovery *discovery;
470 struct ofconn *ofconn;
472 if (is_discovery_controller(c)) {
473 int error = discovery_create(c->accept_re, c->update_resolv_conf,
474 ofproto->dpif, ofproto->switch_status,
483 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
484 ofconn->pktbuf = pktbuf_create();
485 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
487 ofconn->discovery = discovery;
489 char *name = ofconn_make_name(ofproto, c->target);
490 rconn_connect(ofconn->rconn, c->target, name);
493 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
494 hash_string(c->target, 0));
497 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
498 * target or turn discovery on or off (these are done by creating new ofconns
499 * and deleting old ones), but it can update the rest of an ofconn's
502 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
506 ofconn->band = (is_in_band_controller(c)
507 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
509 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
511 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
512 rconn_set_probe_interval(ofconn->rconn, probe_interval);
514 if (ofconn->discovery) {
515 discovery_set_update_resolv_conf(ofconn->discovery,
516 c->update_resolv_conf);
517 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
520 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
524 ofconn_get_target(const struct ofconn *ofconn)
526 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
529 static struct ofconn *
530 find_controller_by_target(struct ofproto *ofproto, const char *target)
532 struct ofconn *ofconn;
534 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
535 hash_string(target, 0), &ofproto->controllers) {
536 if (!strcmp(ofconn_get_target(ofconn), target)) {
544 update_in_band_remotes(struct ofproto *ofproto)
546 const struct ofconn *ofconn;
547 struct sockaddr_in *addrs;
548 size_t max_addrs, n_addrs;
552 /* Allocate enough memory for as many remotes as we could possibly have. */
553 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
554 addrs = xmalloc(max_addrs * sizeof *addrs);
557 /* Add all the remotes. */
559 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
560 struct sockaddr_in *sin = &addrs[n_addrs];
562 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
566 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
567 if (sin->sin_addr.s_addr) {
568 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
571 if (ofconn->discovery) {
575 for (i = 0; i < ofproto->n_extra_remotes; i++) {
576 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
579 /* Create or update or destroy in-band.
581 * Ordinarily we only enable in-band if there's at least one remote
582 * address, but discovery needs the in-band rules for DHCP to be installed
583 * even before we know any remote addresses. */
584 if (n_addrs || discovery) {
585 if (!ofproto->in_band) {
586 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
589 if (ofproto->in_band) {
590 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
592 ofproto->next_in_band_update = time_msec() + 1000;
594 in_band_destroy(ofproto->in_band);
595 ofproto->in_band = NULL;
603 update_fail_open(struct ofproto *p)
605 struct ofconn *ofconn;
607 if (!hmap_is_empty(&p->controllers)
608 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
609 struct rconn **rconns;
613 p->fail_open = fail_open_create(p, p->switch_status);
617 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
618 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
619 rconns[n++] = ofconn->rconn;
622 fail_open_set_controllers(p->fail_open, rconns, n);
623 /* p->fail_open takes ownership of 'rconns'. */
625 fail_open_destroy(p->fail_open);
631 ofproto_set_controllers(struct ofproto *p,
632 const struct ofproto_controller *controllers,
633 size_t n_controllers)
635 struct shash new_controllers;
636 struct ofconn *ofconn, *next_ofconn;
637 struct ofservice *ofservice, *next_ofservice;
641 /* Create newly configured controllers and services.
642 * Create a name to ofproto_controller mapping in 'new_controllers'. */
643 shash_init(&new_controllers);
644 for (i = 0; i < n_controllers; i++) {
645 const struct ofproto_controller *c = &controllers[i];
647 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
648 if (!find_controller_by_target(p, c->target)) {
649 add_controller(p, c);
651 } else if (!pvconn_verify_name(c->target)) {
652 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
656 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
657 dpif_name(p->dpif), c->target);
661 shash_add_once(&new_controllers, c->target, &controllers[i]);
664 /* Delete controllers that are no longer configured.
665 * Update configuration of all now-existing controllers. */
667 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
668 struct ofproto_controller *c;
670 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
672 ofconn_destroy(ofconn);
674 update_controller(ofconn, c);
681 /* Delete services that are no longer configured.
682 * Update configuration of all now-existing services. */
683 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
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, 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;
888 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
889 refresh_port_groups(ofproto);
890 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
891 ofproto_sflow_add_port(os, ofport->odp_port,
892 netdev_get_name(ofport->netdev));
895 ofproto_sflow_set_options(os, oso);
897 ofproto_sflow_destroy(os);
898 ofproto->sflow = NULL;
903 ofproto_get_datapath_id(const struct ofproto *ofproto)
905 return ofproto->datapath_id;
909 ofproto_has_primary_controller(const struct ofproto *ofproto)
911 return !hmap_is_empty(&ofproto->controllers);
914 enum ofproto_fail_mode
915 ofproto_get_fail_mode(const struct ofproto *p)
921 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
925 for (i = 0; i < ofproto->n_snoops; i++) {
926 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
931 ofproto_destroy(struct ofproto *p)
933 struct ofservice *ofservice, *next_ofservice;
934 struct ofconn *ofconn, *next_ofconn;
935 struct ofport *ofport, *next_ofport;
942 /* Destroy fail-open and in-band early, since they touch the classifier. */
943 fail_open_destroy(p->fail_open);
946 in_band_destroy(p->in_band);
948 free(p->extra_in_band_remotes);
950 ofproto_flush_flows(p);
951 classifier_destroy(&p->cls);
953 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
954 ofconn_destroy(ofconn);
956 hmap_destroy(&p->controllers);
959 netdev_monitor_destroy(p->netdev_monitor);
960 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
961 hmap_remove(&p->ports, &ofport->hmap_node);
964 shash_destroy(&p->port_by_name);
966 switch_status_destroy(p->switch_status);
967 netflow_destroy(p->netflow);
968 ofproto_sflow_destroy(p->sflow);
970 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
971 ofservice_destroy(p, ofservice);
973 hmap_destroy(&p->services);
975 for (i = 0; i < p->n_snoops; i++) {
976 pvconn_close(p->snoops[i]);
980 mac_learning_destroy(p->ml);
985 free(p->serial_desc);
988 hmap_destroy(&p->ports);
994 ofproto_run(struct ofproto *p)
996 int error = ofproto_run1(p);
998 error = ofproto_run2(p, false);
1004 process_port_change(struct ofproto *ofproto, int error, char *devname)
1006 if (error == ENOBUFS) {
1007 reinit_ports(ofproto);
1008 } else if (!error) {
1009 update_port(ofproto, devname);
1014 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1015 * means that 'ofconn' is more interesting for monitoring than a lower return
1018 snoop_preference(const struct ofconn *ofconn)
1020 switch (ofconn->role) {
1021 case NX_ROLE_MASTER:
1028 /* Shouldn't happen. */
1033 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1034 * Connects this vconn to a controller. */
1036 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1038 struct ofconn *ofconn, *best;
1040 /* Pick a controller for monitoring. */
1042 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1043 if (ofconn->type == OFCONN_PRIMARY
1044 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1050 rconn_add_monitor(best->rconn, vconn);
1052 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1058 ofproto_run1(struct ofproto *p)
1060 struct ofconn *ofconn, *next_ofconn;
1061 struct ofservice *ofservice;
1066 if (shash_is_empty(&p->port_by_name)) {
1070 for (i = 0; i < 50; i++) {
1073 error = dpif_recv(p->dpif, &buf);
1075 if (error == ENODEV) {
1076 /* Someone destroyed the datapath behind our back. The caller
1077 * better destroy us and give up, because we're just going to
1078 * spin from here on out. */
1079 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1080 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1081 dpif_name(p->dpif));
1087 handle_odp_msg(p, buf);
1090 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1091 process_port_change(p, error, devname);
1093 while ((error = netdev_monitor_poll(p->netdev_monitor,
1094 &devname)) != EAGAIN) {
1095 process_port_change(p, error, devname);
1099 if (time_msec() >= p->next_in_band_update) {
1100 update_in_band_remotes(p);
1102 in_band_run(p->in_band);
1105 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1106 ofconn_run(ofconn, p);
1109 /* Fail-open maintenance. Do this after processing the ofconns since
1110 * fail-open checks the status of the controller rconn. */
1112 fail_open_run(p->fail_open);
1115 HMAP_FOR_EACH (ofservice, node, &p->services) {
1116 struct vconn *vconn;
1119 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1121 struct rconn *rconn;
1124 rconn = rconn_create(ofservice->probe_interval, 0);
1125 name = ofconn_make_name(p, vconn_get_name(vconn));
1126 rconn_connect_unreliably(rconn, vconn, name);
1129 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1130 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1131 ofservice->burst_limit);
1132 } else if (retval != EAGAIN) {
1133 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1137 for (i = 0; i < p->n_snoops; i++) {
1138 struct vconn *vconn;
1141 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1143 add_snooper(p, vconn);
1144 } else if (retval != EAGAIN) {
1145 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1149 if (time_msec() >= p->next_expiration) {
1150 int delay = ofproto_expire(p);
1151 p->next_expiration = time_msec() + delay;
1152 COVERAGE_INC(ofproto_expiration);
1156 netflow_run(p->netflow);
1159 ofproto_sflow_run(p->sflow);
1165 struct revalidate_cbdata {
1166 struct ofproto *ofproto;
1167 bool revalidate_all; /* Revalidate all exact-match rules? */
1168 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1169 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1173 ofproto_run2(struct ofproto *p, bool revalidate_all)
1175 if (p->need_revalidate || revalidate_all
1176 || !tag_set_is_empty(&p->revalidate_set)) {
1177 struct revalidate_cbdata cbdata;
1179 cbdata.revalidate_all = revalidate_all;
1180 cbdata.revalidate_subrules = p->need_revalidate;
1181 cbdata.revalidate_set = p->revalidate_set;
1182 tag_set_init(&p->revalidate_set);
1183 COVERAGE_INC(ofproto_revalidate);
1184 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1185 p->need_revalidate = false;
1192 ofproto_wait(struct ofproto *p)
1194 struct ofservice *ofservice;
1195 struct ofconn *ofconn;
1198 dpif_recv_wait(p->dpif);
1199 dpif_port_poll_wait(p->dpif);
1200 netdev_monitor_poll_wait(p->netdev_monitor);
1201 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1202 ofconn_wait(ofconn);
1205 poll_timer_wait_until(p->next_in_band_update);
1206 in_band_wait(p->in_band);
1209 fail_open_wait(p->fail_open);
1212 ofproto_sflow_wait(p->sflow);
1214 if (!tag_set_is_empty(&p->revalidate_set)) {
1215 poll_immediate_wake();
1217 if (p->need_revalidate) {
1218 /* Shouldn't happen, but if it does just go around again. */
1219 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1220 poll_immediate_wake();
1221 } else if (p->next_expiration != LLONG_MAX) {
1222 poll_timer_wait_until(p->next_expiration);
1224 HMAP_FOR_EACH (ofservice, node, &p->services) {
1225 pvconn_wait(ofservice->pvconn);
1227 for (i = 0; i < p->n_snoops; i++) {
1228 pvconn_wait(p->snoops[i]);
1233 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1235 tag_set_add(&ofproto->revalidate_set, tag);
1239 ofproto_get_revalidate_set(struct ofproto *ofproto)
1241 return &ofproto->revalidate_set;
1245 ofproto_is_alive(const struct ofproto *p)
1247 return !hmap_is_empty(&p->controllers);
1250 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1252 * This is almost the same as calling dpif_port_del() directly on the
1253 * datapath, but it also makes 'ofproto' close its open netdev for the port
1254 * (if any). This makes it possible to create a new netdev of a different
1255 * type under the same name, which otherwise the netdev library would refuse
1256 * to do because of the conflict. (The netdev would eventually get closed on
1257 * the next trip through ofproto_run(), but this interface is more direct.)
1259 * The caller must be prepared for a callback to its port_changed_cb hook
1262 * Returns 0 if successful, otherwise a positive errno. */
1264 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1266 struct ofport *ofport = get_port(ofproto, odp_port);
1267 const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
1270 error = dpif_port_del(ofproto->dpif, odp_port);
1272 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1273 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1274 } else if (ofport) {
1275 /* 'name' is ofport->opp.name and update_port() is going to destroy
1276 * 'ofport'. Just in case update_port() refers to 'name' after it
1277 * destroys 'ofport', make a copy of it around the update_port()
1279 char *devname = xstrdup(name);
1280 update_port(ofproto, devname);
1287 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1288 const union ofp_action *actions, size_t n_actions,
1289 const struct ofpbuf *packet)
1291 struct odp_actions odp_actions;
1294 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1300 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1302 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1303 odp_actions.n_actions, packet);
1308 ofproto_add_flow(struct ofproto *p,
1309 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1310 const union ofp_action *actions, size_t n_actions,
1314 rule = rule_create(p, NULL, actions, n_actions,
1315 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1317 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1318 rule_insert(p, rule, NULL, 0);
1322 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1323 uint32_t wildcards, unsigned int priority)
1327 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1331 rule_remove(ofproto, rule);
1336 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1338 struct rule *rule = rule_from_cls_rule(rule_);
1339 struct ofproto *ofproto = ofproto_;
1341 /* Mark the flow as not installed, even though it might really be
1342 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1343 * There is no point in uninstalling it individually since we are about to
1344 * blow away all the flows with dpif_flow_flush(). */
1345 rule->installed = false;
1347 rule_remove(ofproto, rule);
1351 ofproto_flush_flows(struct ofproto *ofproto)
1353 COVERAGE_INC(ofproto_flush);
1354 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1355 dpif_flow_flush(ofproto->dpif);
1356 if (ofproto->in_band) {
1357 in_band_flushed(ofproto->in_band);
1359 if (ofproto->fail_open) {
1360 fail_open_flushed(ofproto->fail_open);
1365 reinit_ports(struct ofproto *p)
1367 struct svec devnames;
1368 struct ofport *ofport;
1369 struct odp_port *odp_ports;
1373 svec_init(&devnames);
1374 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1375 svec_add (&devnames, (char *) ofport->opp.name);
1377 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1378 for (i = 0; i < n_odp_ports; i++) {
1379 svec_add (&devnames, odp_ports[i].devname);
1383 svec_sort_unique(&devnames);
1384 for (i = 0; i < devnames.n; i++) {
1385 update_port(p, devnames.names[i]);
1387 svec_destroy(&devnames);
1391 refresh_port_group(struct ofproto *p, unsigned int group)
1395 struct ofport *port;
1397 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1399 ports = xmalloc(hmap_count(&p->ports) * sizeof *ports);
1401 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
1402 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1403 ports[n_ports++] = port->odp_port;
1406 dpif_port_group_set(p->dpif, group, ports, n_ports);
1413 refresh_port_groups(struct ofproto *p)
1415 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1416 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1418 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1422 static struct ofport *
1423 make_ofport(const struct odp_port *odp_port)
1425 struct netdev_options netdev_options;
1426 enum netdev_flags flags;
1427 struct ofport *ofport;
1428 struct netdev *netdev;
1432 memset(&netdev_options, 0, sizeof netdev_options);
1433 netdev_options.name = odp_port->devname;
1434 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1436 error = netdev_open(&netdev_options, &netdev);
1438 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1439 "cannot be opened (%s)",
1440 odp_port->devname, odp_port->port,
1441 odp_port->devname, strerror(error));
1445 ofport = xmalloc(sizeof *ofport);
1446 ofport->netdev = netdev;
1447 ofport->odp_port = odp_port->port;
1448 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1449 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1450 memcpy(ofport->opp.name, odp_port->devname,
1451 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1452 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1454 netdev_get_flags(netdev, &flags);
1455 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1457 netdev_get_carrier(netdev, &carrier);
1458 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1460 netdev_get_features(netdev,
1461 &ofport->opp.curr, &ofport->opp.advertised,
1462 &ofport->opp.supported, &ofport->opp.peer);
1467 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1469 if (get_port(p, odp_port->port)) {
1470 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1473 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1474 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1483 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1485 const struct ofp_phy_port *a = &a_->opp;
1486 const struct ofp_phy_port *b = &b_->opp;
1488 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1489 return (a->port_no == b->port_no
1490 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1491 && !strcmp((char *) a->name, (char *) b->name)
1492 && a->state == b->state
1493 && a->config == b->config
1494 && a->curr == b->curr
1495 && a->advertised == b->advertised
1496 && a->supported == b->supported
1497 && a->peer == b->peer);
1501 send_port_status(struct ofproto *p, const struct ofport *ofport,
1504 /* XXX Should limit the number of queued port status change messages. */
1505 struct ofconn *ofconn;
1506 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1507 struct ofp_port_status *ops;
1510 if (!ofconn_receives_async_msgs(ofconn)) {
1514 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1515 ops->reason = reason;
1516 ops->desc = ofport->opp;
1517 hton_ofp_phy_port(&ops->desc);
1518 queue_tx(b, ofconn, NULL);
1520 if (p->ofhooks->port_changed_cb) {
1521 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1526 ofport_install(struct ofproto *p, struct ofport *ofport)
1528 const char *netdev_name = (const char *) ofport->opp.name;
1530 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1531 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1532 shash_add(&p->port_by_name, netdev_name, ofport);
1534 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1539 ofport_remove(struct ofproto *p, struct ofport *ofport)
1541 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1542 hmap_remove(&p->ports, &ofport->hmap_node);
1543 shash_delete(&p->port_by_name,
1544 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1546 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1551 ofport_free(struct ofport *ofport)
1554 netdev_close(ofport->netdev);
1559 static struct ofport *
1560 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1562 struct ofport *port;
1564 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1565 hash_int(odp_port, 0), &ofproto->ports) {
1566 if (port->odp_port == odp_port) {
1574 update_port(struct ofproto *p, const char *devname)
1576 struct odp_port odp_port;
1577 struct ofport *old_ofport;
1578 struct ofport *new_ofport;
1581 COVERAGE_INC(ofproto_update_port);
1583 /* Query the datapath for port information. */
1584 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1586 /* Find the old ofport. */
1587 old_ofport = shash_find_data(&p->port_by_name, devname);
1590 /* There's no port named 'devname' but there might be a port with
1591 * the same port number. This could happen if a port is deleted
1592 * and then a new one added in its place very quickly, or if a port
1593 * is renamed. In the former case we want to send an OFPPR_DELETE
1594 * and an OFPPR_ADD, and in the latter case we want to send a
1595 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1596 * the old port's ifindex against the new port, or perhaps less
1597 * reliably but more portably by comparing the old port's MAC
1598 * against the new port's MAC. However, this code isn't that smart
1599 * and always sends an OFPPR_MODIFY (XXX). */
1600 old_ofport = get_port(p, odp_port.port);
1602 } else if (error != ENOENT && error != ENODEV) {
1603 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1604 "%s", strerror(error));
1608 /* Create a new ofport. */
1609 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1611 /* Eliminate a few pathological cases. */
1612 if (!old_ofport && !new_ofport) {
1614 } else if (old_ofport && new_ofport) {
1615 /* Most of the 'config' bits are OpenFlow soft state, but
1616 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1617 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1618 * leaves the other bits 0.) */
1619 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1621 if (ofport_equal(old_ofport, new_ofport)) {
1622 /* False alarm--no change. */
1623 ofport_free(new_ofport);
1628 /* Now deal with the normal cases. */
1630 ofport_remove(p, old_ofport);
1633 ofport_install(p, new_ofport);
1635 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1636 (!old_ofport ? OFPPR_ADD
1637 : !new_ofport ? OFPPR_DELETE
1639 ofport_free(old_ofport);
1641 /* Update port groups. */
1642 refresh_port_groups(p);
1646 init_ports(struct ofproto *p)
1648 struct odp_port *ports;
1653 error = dpif_port_list(p->dpif, &ports, &n_ports);
1658 for (i = 0; i < n_ports; i++) {
1659 const struct odp_port *odp_port = &ports[i];
1660 if (!ofport_conflicts(p, odp_port)) {
1661 struct ofport *ofport = make_ofport(odp_port);
1663 ofport_install(p, ofport);
1668 refresh_port_groups(p);
1672 static struct ofconn *
1673 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1675 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1676 ofconn->ofproto = p;
1677 list_push_back(&p->all_conns, &ofconn->node);
1678 ofconn->rconn = rconn;
1679 ofconn->type = type;
1680 ofconn->role = NX_ROLE_OTHER;
1681 ofconn->packet_in_counter = rconn_packet_counter_create ();
1682 ofconn->pktbuf = NULL;
1683 ofconn->miss_send_len = 0;
1684 ofconn->reply_counter = rconn_packet_counter_create ();
1689 ofconn_destroy(struct ofconn *ofconn)
1691 if (ofconn->type == OFCONN_PRIMARY) {
1692 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1694 discovery_destroy(ofconn->discovery);
1696 list_remove(&ofconn->node);
1697 switch_status_unregister(ofconn->ss);
1698 rconn_destroy(ofconn->rconn);
1699 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1700 rconn_packet_counter_destroy(ofconn->reply_counter);
1701 pktbuf_destroy(ofconn->pktbuf);
1706 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1711 if (ofconn->discovery) {
1712 char *controller_name;
1713 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1714 discovery_question_connectivity(ofconn->discovery);
1716 if (discovery_run(ofconn->discovery, &controller_name)) {
1717 if (controller_name) {
1718 char *ofconn_name = ofconn_make_name(p, controller_name);
1719 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1722 rconn_disconnect(ofconn->rconn);
1727 for (i = 0; i < N_SCHEDULERS; i++) {
1728 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1731 rconn_run(ofconn->rconn);
1733 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1734 /* Limit the number of iterations to prevent other tasks from
1736 for (iteration = 0; iteration < 50; iteration++) {
1737 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1742 fail_open_maybe_recover(p->fail_open);
1744 handle_openflow(ofconn, p, of_msg);
1745 ofpbuf_delete(of_msg);
1749 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1750 ofconn_destroy(ofconn);
1755 ofconn_wait(struct ofconn *ofconn)
1759 if (ofconn->discovery) {
1760 discovery_wait(ofconn->discovery);
1762 for (i = 0; i < N_SCHEDULERS; i++) {
1763 pinsched_wait(ofconn->schedulers[i]);
1765 rconn_run_wait(ofconn->rconn);
1766 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1767 rconn_recv_wait(ofconn->rconn);
1769 COVERAGE_INC(ofproto_ofconn_stuck);
1773 /* Returns true if 'ofconn' should receive asynchronous messages. */
1775 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1777 if (ofconn->type == OFCONN_PRIMARY) {
1778 /* Primary controllers always get asynchronous messages unless they
1779 * have configured themselves as "slaves". */
1780 return ofconn->role != NX_ROLE_SLAVE;
1782 /* Service connections don't get asynchronous messages unless they have
1783 * explicitly asked for them by setting a nonzero miss send length. */
1784 return ofconn->miss_send_len > 0;
1788 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1789 * and 'target', suitable for use in log messages for identifying the
1792 * The name is dynamically allocated. The caller should free it (with free())
1793 * when it is no longer needed. */
1795 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1797 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1801 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1805 for (i = 0; i < N_SCHEDULERS; i++) {
1806 struct pinsched **s = &ofconn->schedulers[i];
1810 *s = pinsched_create(rate, burst,
1811 ofconn->ofproto->switch_status);
1813 pinsched_set_limits(*s, rate, burst);
1816 pinsched_destroy(*s);
1823 ofservice_reconfigure(struct ofservice *ofservice,
1824 const struct ofproto_controller *c)
1826 ofservice->probe_interval = c->probe_interval;
1827 ofservice->rate_limit = c->rate_limit;
1828 ofservice->burst_limit = c->burst_limit;
1831 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1832 * positive errno value. */
1834 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1836 struct ofservice *ofservice;
1837 struct pvconn *pvconn;
1840 error = pvconn_open(c->target, &pvconn);
1845 ofservice = xzalloc(sizeof *ofservice);
1846 hmap_insert(&ofproto->services, &ofservice->node,
1847 hash_string(c->target, 0));
1848 ofservice->pvconn = pvconn;
1850 ofservice_reconfigure(ofservice, c);
1856 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1858 hmap_remove(&ofproto->services, &ofservice->node);
1859 pvconn_close(ofservice->pvconn);
1863 /* Finds and returns the ofservice within 'ofproto' that has the given
1864 * 'target', or a null pointer if none exists. */
1865 static struct ofservice *
1866 ofservice_lookup(struct ofproto *ofproto, const char *target)
1868 struct ofservice *ofservice;
1870 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1871 &ofproto->services) {
1872 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1879 /* Caller is responsible for initializing the 'cr' member of the returned
1881 static struct rule *
1882 rule_create(struct ofproto *ofproto, struct rule *super,
1883 const union ofp_action *actions, size_t n_actions,
1884 uint16_t idle_timeout, uint16_t hard_timeout,
1885 uint64_t flow_cookie, bool send_flow_removed)
1887 struct rule *rule = xzalloc(sizeof *rule);
1888 rule->idle_timeout = idle_timeout;
1889 rule->hard_timeout = hard_timeout;
1890 rule->flow_cookie = flow_cookie;
1891 rule->used = rule->created = time_msec();
1892 rule->send_flow_removed = send_flow_removed;
1893 rule->super = super;
1895 list_push_back(&super->list, &rule->list);
1897 list_init(&rule->list);
1899 if (n_actions > 0) {
1900 rule->n_actions = n_actions;
1901 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1903 netflow_flow_clear(&rule->nf_flow);
1904 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1909 static struct rule *
1910 rule_from_cls_rule(const struct cls_rule *cls_rule)
1912 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1916 rule_free(struct rule *rule)
1918 free(rule->actions);
1919 free(rule->odp_actions);
1923 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1924 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1925 * through all of its subrules and revalidates them, destroying any that no
1926 * longer has a super-rule (which is probably all of them).
1928 * Before calling this function, the caller must make have removed 'rule' from
1929 * the classifier. If 'rule' is an exact-match rule, the caller is also
1930 * responsible for ensuring that it has been uninstalled from the datapath. */
1932 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1935 struct rule *subrule, *next;
1936 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
1937 revalidate_rule(ofproto, subrule);
1940 list_remove(&rule->list);
1946 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1948 const union ofp_action *oa;
1949 struct actions_iterator i;
1951 if (out_port == htons(OFPP_NONE)) {
1954 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1955 oa = actions_next(&i)) {
1956 if (action_outputs_to_port(oa, out_port)) {
1963 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1964 * 'packet', which arrived on 'in_port'.
1966 * Takes ownership of 'packet'. */
1968 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1969 const union odp_action *actions, size_t n_actions,
1970 struct ofpbuf *packet)
1972 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1973 /* As an optimization, avoid a round-trip from userspace to kernel to
1974 * userspace. This also avoids possibly filling up kernel packet
1975 * buffers along the way. */
1976 struct odp_msg *msg;
1978 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1979 msg->type = _ODPL_ACTION_NR;
1980 msg->length = sizeof(struct odp_msg) + packet->size;
1981 msg->port = in_port;
1983 msg->arg = actions[0].controller.arg;
1985 send_packet_in(ofproto, packet);
1991 error = dpif_execute(ofproto->dpif, in_port,
1992 actions, n_actions, packet);
1993 ofpbuf_delete(packet);
1998 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1999 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
2000 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2002 * The flow that 'packet' actually contains does not need to actually match
2003 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
2004 * the packet and byte counters for 'rule' will be credited for the packet sent
2005 * out whether or not the packet actually matches 'rule'.
2007 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
2008 * the caller must already have accurately composed ODP actions for it given
2009 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
2010 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
2011 * function will compose a set of ODP actions based on 'rule''s OpenFlow
2012 * actions and apply them to 'packet'.
2014 * Takes ownership of 'packet'. */
2016 rule_execute(struct ofproto *ofproto, struct rule *rule,
2017 struct ofpbuf *packet, const flow_t *flow)
2019 const union odp_action *actions;
2020 struct odp_flow_stats stats;
2022 struct odp_actions a;
2024 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
2026 /* Grab or compose the ODP actions.
2028 * The special case for an exact-match 'rule' where 'flow' is not the
2029 * rule's flow is important to avoid, e.g., sending a packet out its input
2030 * port simply because the ODP actions were composed for the wrong
2032 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
2033 struct rule *super = rule->super ? rule->super : rule;
2034 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
2035 packet, &a, NULL, 0, NULL)) {
2036 ofpbuf_delete(packet);
2039 actions = a.actions;
2040 n_actions = a.n_actions;
2042 actions = rule->odp_actions;
2043 n_actions = rule->n_odp_actions;
2046 /* Execute the ODP actions. */
2047 flow_extract_stats(flow, packet, &stats);
2048 if (execute_odp_actions(ofproto, flow->in_port,
2049 actions, n_actions, packet)) {
2050 update_stats(ofproto, rule, &stats);
2051 rule->used = time_msec();
2052 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2056 /* Inserts 'rule' into 'p''s flow table.
2058 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2059 * actions on it and credits the statistics for sending the packet to 'rule'.
2060 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2063 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2066 struct rule *displaced_rule;
2068 /* Insert the rule in the classifier. */
2069 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2070 if (!rule->cr.wc.wildcards) {
2071 rule_make_actions(p, rule, packet);
2074 /* Send the packet and credit it to the rule. */
2077 flow_extract(packet, 0, in_port, &flow);
2078 rule_execute(p, rule, packet, &flow);
2081 /* Install the rule in the datapath only after sending the packet, to
2082 * avoid packet reordering. */
2083 if (rule->cr.wc.wildcards) {
2084 COVERAGE_INC(ofproto_add_wc_flow);
2085 p->need_revalidate = true;
2087 rule_install(p, rule, displaced_rule);
2090 /* Free the rule that was displaced, if any. */
2091 if (displaced_rule) {
2092 rule_destroy(p, displaced_rule);
2096 static struct rule *
2097 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2100 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2101 rule->idle_timeout, rule->hard_timeout,
2103 COVERAGE_INC(ofproto_subrule_create);
2104 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2105 : rule->cr.priority), &subrule->cr);
2106 classifier_insert_exact(&ofproto->cls, &subrule->cr);
2111 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2113 * - If 'rule' was installed in the datapath, uninstalls it and updates
2114 * 'rule''s statistics (or its super-rule's statistics, if it is a
2115 * subrule), via rule_uninstall().
2117 * - Removes 'rule' from the classifier.
2119 * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
2120 * uninstalls and destroys) its subrules, via rule_destroy().
2123 rule_remove(struct ofproto *ofproto, struct rule *rule)
2125 if (rule->cr.wc.wildcards) {
2126 COVERAGE_INC(ofproto_del_wc_flow);
2127 ofproto->need_revalidate = true;
2129 rule_uninstall(ofproto, rule);
2131 classifier_remove(&ofproto->cls, &rule->cr);
2132 rule_destroy(ofproto, rule);
2135 /* Returns true if the actions changed, false otherwise. */
2137 rule_make_actions(struct ofproto *p, struct rule *rule,
2138 const struct ofpbuf *packet)
2140 const struct rule *super;
2141 struct odp_actions a;
2144 assert(!rule->cr.wc.wildcards);
2146 super = rule->super ? rule->super : rule;
2148 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2149 packet, &a, &rule->tags, &rule->may_install,
2150 &rule->nf_flow.output_iface);
2152 actions_len = a.n_actions * sizeof *a.actions;
2153 if (rule->n_odp_actions != a.n_actions
2154 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2155 COVERAGE_INC(ofproto_odp_unchanged);
2156 free(rule->odp_actions);
2157 rule->n_odp_actions = a.n_actions;
2158 rule->odp_actions = xmemdup(a.actions, actions_len);
2166 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2167 struct odp_flow_put *put)
2169 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2170 put->flow.key = rule->cr.flow;
2171 put->flow.actions = rule->odp_actions;
2172 put->flow.n_actions = rule->n_odp_actions;
2173 put->flow.flags = 0;
2175 return dpif_flow_put(ofproto->dpif, put);
2179 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2181 assert(!rule->cr.wc.wildcards);
2183 if (rule->may_install) {
2184 struct odp_flow_put put;
2185 if (!do_put_flow(p, rule,
2186 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2188 rule->installed = true;
2189 if (displaced_rule) {
2190 update_stats(p, displaced_rule, &put.flow.stats);
2191 rule_post_uninstall(p, displaced_rule);
2194 } else if (displaced_rule) {
2195 rule_uninstall(p, displaced_rule);
2200 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2202 if (rule->installed) {
2203 struct odp_flow_put put;
2204 COVERAGE_INC(ofproto_dp_missed);
2205 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2207 rule_install(ofproto, rule, NULL);
2212 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2214 bool actions_changed;
2215 uint16_t new_out_iface, old_out_iface;
2217 old_out_iface = rule->nf_flow.output_iface;
2218 actions_changed = rule_make_actions(ofproto, rule, NULL);
2220 if (rule->may_install) {
2221 if (rule->installed) {
2222 if (actions_changed) {
2223 struct odp_flow_put put;
2224 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2225 | ODPPF_ZERO_STATS, &put);
2226 update_stats(ofproto, rule, &put.flow.stats);
2228 /* Temporarily set the old output iface so that NetFlow
2229 * messages have the correct output interface for the old
2231 new_out_iface = rule->nf_flow.output_iface;
2232 rule->nf_flow.output_iface = old_out_iface;
2233 rule_post_uninstall(ofproto, rule);
2234 rule->nf_flow.output_iface = new_out_iface;
2237 rule_install(ofproto, rule, NULL);
2240 rule_uninstall(ofproto, rule);
2245 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2247 uint64_t total_bytes = rule->byte_count + extra_bytes;
2249 if (ofproto->ofhooks->account_flow_cb
2250 && total_bytes > rule->accounted_bytes)
2252 ofproto->ofhooks->account_flow_cb(
2253 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2254 total_bytes - rule->accounted_bytes, ofproto->aux);
2255 rule->accounted_bytes = total_bytes;
2259 /* 'rule' must be an exact-match rule in 'p'.
2261 * If 'rule' is installed in the datapath, uninstalls it and updates's
2262 * statistics. If 'rule' is a subrule, the statistics that are updated are
2263 * actually its super-rule's statistics; otherwise 'rule''s own statistics are
2266 * If 'rule' is not installed, this function has no effect. */
2268 rule_uninstall(struct ofproto *p, struct rule *rule)
2270 assert(!rule->cr.wc.wildcards);
2271 if (rule->installed) {
2272 struct odp_flow odp_flow;
2274 odp_flow.key = rule->cr.flow;
2275 odp_flow.actions = NULL;
2276 odp_flow.n_actions = 0;
2278 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2279 update_stats(p, rule, &odp_flow.stats);
2281 rule->installed = false;
2283 rule_post_uninstall(p, rule);
2288 is_controller_rule(struct rule *rule)
2290 /* If the only action is send to the controller then don't report
2291 * NetFlow expiration messages since it is just part of the control
2292 * logic for the network and not real traffic. */
2296 && rule->super->n_actions == 1
2297 && action_outputs_to_port(&rule->super->actions[0],
2298 htons(OFPP_CONTROLLER)));
2302 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2304 struct rule *super = rule->super;
2306 rule_account(ofproto, rule, 0);
2308 if (ofproto->netflow && !is_controller_rule(rule)) {
2309 struct ofexpired expired;
2310 expired.flow = rule->cr.flow;
2311 expired.packet_count = rule->packet_count;
2312 expired.byte_count = rule->byte_count;
2313 expired.used = rule->used;
2314 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2317 super->packet_count += rule->packet_count;
2318 super->byte_count += rule->byte_count;
2320 /* Reset counters to prevent double counting if the rule ever gets
2322 rule->packet_count = 0;
2323 rule->byte_count = 0;
2324 rule->accounted_bytes = 0;
2326 netflow_flow_clear(&rule->nf_flow);
2331 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2332 struct rconn_packet_counter *counter)
2334 update_openflow_length(msg);
2335 if (rconn_send(ofconn->rconn, msg, counter)) {
2341 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2342 int error, const void *data, size_t len)
2345 struct ofp_error_msg *oem;
2347 if (!(error >> 16)) {
2348 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2353 COVERAGE_INC(ofproto_error);
2354 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2355 oh ? oh->xid : 0, &buf);
2356 oem->type = htons((unsigned int) error >> 16);
2357 oem->code = htons(error & 0xffff);
2358 memcpy(oem->data, data, len);
2359 queue_tx(buf, ofconn, ofconn->reply_counter);
2363 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2366 size_t oh_length = ntohs(oh->length);
2367 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2371 hton_ofp_phy_port(struct ofp_phy_port *opp)
2373 opp->port_no = htons(opp->port_no);
2374 opp->config = htonl(opp->config);
2375 opp->state = htonl(opp->state);
2376 opp->curr = htonl(opp->curr);
2377 opp->advertised = htonl(opp->advertised);
2378 opp->supported = htonl(opp->supported);
2379 opp->peer = htonl(opp->peer);
2383 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2385 struct ofp_header *rq = oh;
2386 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2391 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2392 struct ofp_header *oh)
2394 struct ofp_switch_features *osf;
2396 struct ofport *port;
2398 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2399 osf->datapath_id = htonll(p->datapath_id);
2400 osf->n_buffers = htonl(pktbuf_capacity());
2402 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2403 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2404 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2405 (1u << OFPAT_SET_VLAN_VID) |
2406 (1u << OFPAT_SET_VLAN_PCP) |
2407 (1u << OFPAT_STRIP_VLAN) |
2408 (1u << OFPAT_SET_DL_SRC) |
2409 (1u << OFPAT_SET_DL_DST) |
2410 (1u << OFPAT_SET_NW_SRC) |
2411 (1u << OFPAT_SET_NW_DST) |
2412 (1u << OFPAT_SET_NW_TOS) |
2413 (1u << OFPAT_SET_TP_SRC) |
2414 (1u << OFPAT_SET_TP_DST) |
2415 (1u << OFPAT_ENQUEUE));
2417 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2418 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2421 queue_tx(buf, ofconn, ofconn->reply_counter);
2426 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2427 struct ofp_header *oh)
2430 struct ofp_switch_config *osc;
2434 /* Figure out flags. */
2435 dpif_get_drop_frags(p->dpif, &drop_frags);
2436 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2439 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2440 osc->flags = htons(flags);
2441 osc->miss_send_len = htons(ofconn->miss_send_len);
2442 queue_tx(buf, ofconn, ofconn->reply_counter);
2448 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2449 struct ofp_switch_config *osc)
2454 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2458 flags = ntohs(osc->flags);
2460 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2461 switch (flags & OFPC_FRAG_MASK) {
2462 case OFPC_FRAG_NORMAL:
2463 dpif_set_drop_frags(p->dpif, false);
2465 case OFPC_FRAG_DROP:
2466 dpif_set_drop_frags(p->dpif, true);
2469 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2475 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2481 add_output_group_action(struct odp_actions *actions, uint16_t group,
2482 uint16_t *nf_output_iface)
2484 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2486 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2487 *nf_output_iface = NF_OUT_FLOOD;
2492 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2494 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2495 a->controller.arg = max_len;
2498 struct action_xlate_ctx {
2500 flow_t flow; /* Flow to which these actions correspond. */
2501 int recurse; /* Recursion level, via xlate_table_action. */
2502 struct ofproto *ofproto;
2503 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2504 * null pointer if we are revalidating
2505 * without a packet to refer to. */
2508 struct odp_actions *out; /* Datapath actions. */
2509 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2510 bool may_set_up_flow; /* True ordinarily; false if the actions must
2511 * be reassessed for every packet. */
2512 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2515 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2516 * flow translation. */
2517 #define MAX_RESUBMIT_RECURSION 8
2519 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2520 struct action_xlate_ctx *ctx);
2523 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2525 const struct ofport *ofport = get_port(ctx->ofproto, port);
2528 if (ofport->opp.config & OFPPC_NO_FWD) {
2529 /* Forwarding disabled on port. */
2534 * We don't have an ofport record for this port, but it doesn't hurt to
2535 * allow forwarding to it anyhow. Maybe such a port will appear later
2536 * and we're pre-populating the flow table.
2540 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2541 ctx->nf_output_iface = port;
2544 static struct rule *
2545 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2548 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2550 /* The rule we found might not be valid, since we could be in need of
2551 * revalidation. If it is not valid, don't return it. */
2554 && ofproto->need_revalidate
2555 && !revalidate_rule(ofproto, rule)) {
2556 COVERAGE_INC(ofproto_invalidated);
2564 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2566 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2567 uint16_t old_in_port;
2570 /* Look up a flow with 'in_port' as the input port. Then restore the
2571 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2572 * have surprising behavior). */
2573 old_in_port = ctx->flow.in_port;
2574 ctx->flow.in_port = in_port;
2575 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2576 ctx->flow.in_port = old_in_port;
2584 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2588 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2590 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2591 MAX_RESUBMIT_RECURSION);
2596 xlate_output_action__(struct action_xlate_ctx *ctx,
2597 uint16_t port, uint16_t max_len)
2600 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2602 ctx->nf_output_iface = NF_OUT_DROP;
2606 add_output_action(ctx, ctx->flow.in_port);
2609 xlate_table_action(ctx, ctx->flow.in_port);
2612 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2613 ctx->out, ctx->tags,
2614 &ctx->nf_output_iface,
2615 ctx->ofproto->aux)) {
2616 COVERAGE_INC(ofproto_uninstallable);
2617 ctx->may_set_up_flow = false;
2621 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2622 &ctx->nf_output_iface);
2625 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2627 case OFPP_CONTROLLER:
2628 add_controller_action(ctx->out, max_len);
2631 add_output_action(ctx, ODPP_LOCAL);
2634 odp_port = ofp_port_to_odp_port(port);
2635 if (odp_port != ctx->flow.in_port) {
2636 add_output_action(ctx, odp_port);
2641 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2642 ctx->nf_output_iface = NF_OUT_FLOOD;
2643 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2644 ctx->nf_output_iface = prev_nf_output_iface;
2645 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2646 ctx->nf_output_iface != NF_OUT_FLOOD) {
2647 ctx->nf_output_iface = NF_OUT_MULTI;
2652 xlate_output_action(struct action_xlate_ctx *ctx,
2653 const struct ofp_action_output *oao)
2655 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2658 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2659 * optimization, because we're going to add another action that sets the
2660 * priority immediately after, or because there are no actions following the
2663 remove_pop_action(struct action_xlate_ctx *ctx)
2665 size_t n = ctx->out->n_actions;
2666 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2667 ctx->out->n_actions--;
2672 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2673 const struct ofp_action_enqueue *oae)
2675 uint16_t ofp_port, odp_port;
2679 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2682 /* Fall back to ordinary output action. */
2683 xlate_output_action__(ctx, ntohs(oae->port), 0);
2687 /* Figure out ODP output port. */
2688 ofp_port = ntohs(oae->port);
2689 if (ofp_port != OFPP_IN_PORT) {
2690 odp_port = ofp_port_to_odp_port(ofp_port);
2692 odp_port = ctx->flow.in_port;
2695 /* Add ODP actions. */
2696 remove_pop_action(ctx);
2697 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2699 add_output_action(ctx, odp_port);
2700 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2702 /* Update NetFlow output port. */
2703 if (ctx->nf_output_iface == NF_OUT_DROP) {
2704 ctx->nf_output_iface = odp_port;
2705 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2706 ctx->nf_output_iface = NF_OUT_MULTI;
2711 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2712 const struct nx_action_set_queue *nasq)
2717 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2720 /* Couldn't translate queue to a priority, so ignore. A warning
2721 * has already been logged. */
2725 remove_pop_action(ctx);
2726 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2731 xlate_nicira_action(struct action_xlate_ctx *ctx,
2732 const struct nx_action_header *nah)
2734 const struct nx_action_resubmit *nar;
2735 const struct nx_action_set_tunnel *nast;
2736 const struct nx_action_set_queue *nasq;
2737 union odp_action *oa;
2738 int subtype = ntohs(nah->subtype);
2740 assert(nah->vendor == htonl(NX_VENDOR_ID));
2742 case NXAST_RESUBMIT:
2743 nar = (const struct nx_action_resubmit *) nah;
2744 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2747 case NXAST_SET_TUNNEL:
2748 nast = (const struct nx_action_set_tunnel *) nah;
2749 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2750 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2753 case NXAST_DROP_SPOOFED_ARP:
2754 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2755 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2759 case NXAST_SET_QUEUE:
2760 nasq = (const struct nx_action_set_queue *) nah;
2761 xlate_set_queue_action(ctx, nasq);
2764 case NXAST_POP_QUEUE:
2765 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2768 /* If you add a new action here that modifies flow data, don't forget to
2769 * update the flow key in ctx->flow at the same time. */
2772 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2778 do_xlate_actions(const union ofp_action *in, size_t n_in,
2779 struct action_xlate_ctx *ctx)
2781 struct actions_iterator iter;
2782 const union ofp_action *ia;
2783 const struct ofport *port;
2785 port = get_port(ctx->ofproto, ctx->flow.in_port);
2786 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2787 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2788 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2789 /* Drop this flow. */
2793 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2794 uint16_t type = ntohs(ia->type);
2795 union odp_action *oa;
2799 xlate_output_action(ctx, &ia->output);
2802 case OFPAT_SET_VLAN_VID:
2803 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2804 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2807 case OFPAT_SET_VLAN_PCP:
2808 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2809 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2812 case OFPAT_STRIP_VLAN:
2813 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2814 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2815 ctx->flow.dl_vlan_pcp = 0;
2818 case OFPAT_SET_DL_SRC:
2819 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2820 memcpy(oa->dl_addr.dl_addr,
2821 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2822 memcpy(ctx->flow.dl_src,
2823 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2826 case OFPAT_SET_DL_DST:
2827 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2828 memcpy(oa->dl_addr.dl_addr,
2829 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2830 memcpy(ctx->flow.dl_dst,
2831 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2834 case OFPAT_SET_NW_SRC:
2835 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2836 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2839 case OFPAT_SET_NW_DST:
2840 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2841 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2844 case OFPAT_SET_NW_TOS:
2845 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2846 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2849 case OFPAT_SET_TP_SRC:
2850 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2851 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2854 case OFPAT_SET_TP_DST:
2855 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2856 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2860 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2864 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2868 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2875 xlate_actions(const union ofp_action *in, size_t n_in,
2876 const flow_t *flow, struct ofproto *ofproto,
2877 const struct ofpbuf *packet,
2878 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2879 uint16_t *nf_output_iface)
2881 tag_type no_tags = 0;
2882 struct action_xlate_ctx ctx;
2883 COVERAGE_INC(ofproto_ofp2odp);
2884 odp_actions_init(out);
2887 ctx.ofproto = ofproto;
2888 ctx.packet = packet;
2890 ctx.tags = tags ? tags : &no_tags;
2891 ctx.may_set_up_flow = true;
2892 ctx.nf_output_iface = NF_OUT_DROP;
2893 do_xlate_actions(in, n_in, &ctx);
2894 remove_pop_action(&ctx);
2896 /* Check with in-band control to see if we're allowed to set up this
2898 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2899 ctx.may_set_up_flow = false;
2902 if (may_set_up_flow) {
2903 *may_set_up_flow = ctx.may_set_up_flow;
2905 if (nf_output_iface) {
2906 *nf_output_iface = ctx.nf_output_iface;
2908 if (odp_actions_overflow(out)) {
2909 COVERAGE_INC(odp_overflow);
2910 odp_actions_init(out);
2911 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2916 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2917 * error message code (composed with ofp_mkerr()) for the caller to propagate
2918 * upward. Otherwise, returns 0.
2920 * 'oh' is used to make log messages more informative. */
2922 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2924 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2925 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2928 type_name = ofp_message_type_to_string(oh->type);
2929 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2933 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2940 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2941 struct ofp_header *oh)
2943 struct ofp_packet_out *opo;
2944 struct ofpbuf payload, *buffer;
2945 struct odp_actions actions;
2951 error = reject_slave_controller(ofconn, oh);
2956 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2960 opo = (struct ofp_packet_out *) oh;
2962 COVERAGE_INC(ofproto_packet_out);
2963 if (opo->buffer_id != htonl(UINT32_MAX)) {
2964 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2966 if (error || !buffer) {
2974 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2975 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2976 &flow, p, &payload, &actions, NULL, NULL, NULL);
2981 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2983 ofpbuf_delete(buffer);
2989 update_port_config(struct ofproto *p, struct ofport *port,
2990 uint32_t config, uint32_t mask)
2992 mask &= config ^ port->opp.config;
2993 if (mask & OFPPC_PORT_DOWN) {
2994 if (config & OFPPC_PORT_DOWN) {
2995 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2997 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
3000 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
3001 if (mask & REVALIDATE_BITS) {
3002 COVERAGE_INC(ofproto_costly_flags);
3003 port->opp.config ^= mask & REVALIDATE_BITS;
3004 p->need_revalidate = true;
3006 #undef REVALIDATE_BITS
3007 if (mask & OFPPC_NO_FLOOD) {
3008 port->opp.config ^= OFPPC_NO_FLOOD;
3009 refresh_port_groups(p);
3011 if (mask & OFPPC_NO_PACKET_IN) {
3012 port->opp.config ^= OFPPC_NO_PACKET_IN;
3017 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
3018 struct ofp_header *oh)
3020 const struct ofp_port_mod *opm;
3021 struct ofport *port;
3024 error = reject_slave_controller(ofconn, oh);
3028 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
3032 opm = (struct ofp_port_mod *) oh;
3034 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
3036 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
3037 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
3038 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
3040 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
3041 if (opm->advertise) {
3042 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3048 static struct ofpbuf *
3049 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
3051 struct ofp_stats_reply *osr;
3054 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3055 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3057 osr->flags = htons(0);
3061 static struct ofpbuf *
3062 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3064 return make_stats_reply(request->header.xid, request->type, body_len);
3068 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
3070 struct ofpbuf *msg = *msgp;
3071 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3072 if (nbytes + msg->size > UINT16_MAX) {
3073 struct ofp_stats_reply *reply = msg->data;
3074 reply->flags = htons(OFPSF_REPLY_MORE);
3075 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
3076 queue_tx(msg, ofconn, ofconn->reply_counter);
3078 return ofpbuf_put_uninit(*msgp, nbytes);
3082 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3083 struct ofp_stats_request *request)
3085 struct ofp_desc_stats *ods;
3088 msg = start_stats_reply(request, sizeof *ods);
3089 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3090 memset(ods, 0, sizeof *ods);
3091 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3092 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3093 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3094 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3095 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3096 queue_tx(msg, ofconn, ofconn->reply_counter);
3102 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3103 struct ofp_stats_request *request)
3105 struct ofp_table_stats *ots;
3107 struct odp_stats dpstats;
3108 int n_exact, n_subrules, n_wild;
3111 msg = start_stats_reply(request, sizeof *ots * 2);
3113 /* Count rules of various kinds. */
3115 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) {
3120 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3121 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3124 dpif_get_dp_stats(p->dpif, &dpstats);
3125 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3126 memset(ots, 0, sizeof *ots);
3127 ots->table_id = TABLEID_HASH;
3128 strcpy(ots->name, "hash");
3129 ots->wildcards = htonl(0);
3130 ots->max_entries = htonl(dpstats.max_capacity);
3131 ots->active_count = htonl(n_exact);
3132 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3134 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3136 /* Classifier table. */
3137 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3138 memset(ots, 0, sizeof *ots);
3139 ots->table_id = TABLEID_CLASSIFIER;
3140 strcpy(ots->name, "classifier");
3141 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3143 ots->max_entries = htonl(65536);
3144 ots->active_count = htonl(n_wild);
3145 ots->lookup_count = htonll(0); /* XXX */
3146 ots->matched_count = htonll(0); /* XXX */
3148 queue_tx(msg, ofconn, ofconn->reply_counter);
3153 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3154 struct ofpbuf **msgp)
3156 struct netdev_stats stats;
3157 struct ofp_port_stats *ops;
3159 /* Intentionally ignore return value, since errors will set
3160 * 'stats' to all-1s, which is correct for OpenFlow, and
3161 * netdev_get_stats() will log errors. */
3162 netdev_get_stats(port->netdev, &stats);
3164 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3165 ops->port_no = htons(port->opp.port_no);
3166 memset(ops->pad, 0, sizeof ops->pad);
3167 ops->rx_packets = htonll(stats.rx_packets);
3168 ops->tx_packets = htonll(stats.tx_packets);
3169 ops->rx_bytes = htonll(stats.rx_bytes);
3170 ops->tx_bytes = htonll(stats.tx_bytes);
3171 ops->rx_dropped = htonll(stats.rx_dropped);
3172 ops->tx_dropped = htonll(stats.tx_dropped);
3173 ops->rx_errors = htonll(stats.rx_errors);
3174 ops->tx_errors = htonll(stats.tx_errors);
3175 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3176 ops->rx_over_err = htonll(stats.rx_over_errors);
3177 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3178 ops->collisions = htonll(stats.collisions);
3182 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3183 struct ofp_stats_request *osr,
3186 struct ofp_port_stats_request *psr;
3187 struct ofp_port_stats *ops;
3189 struct ofport *port;
3191 if (arg_size != sizeof *psr) {
3192 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3194 psr = (struct ofp_port_stats_request *) osr->body;
3196 msg = start_stats_reply(osr, sizeof *ops * 16);
3197 if (psr->port_no != htons(OFPP_NONE)) {
3198 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3200 append_port_stat(port, ofconn, &msg);
3203 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3204 append_port_stat(port, ofconn, &msg);
3208 queue_tx(msg, ofconn, ofconn->reply_counter);
3212 struct flow_stats_cbdata {
3213 struct ofproto *ofproto;
3214 struct ofconn *ofconn;
3219 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3220 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3221 * returned statistic include statistics for all of 'rule''s subrules. */
3223 query_stats(struct ofproto *p, struct rule *rule,
3224 uint64_t *packet_countp, uint64_t *byte_countp)
3226 uint64_t packet_count, byte_count;
3227 struct rule *subrule;
3228 struct odp_flow *odp_flows;
3231 /* Start from historical data for 'rule' itself that are no longer tracked
3232 * by the datapath. This counts, for example, subrules that have
3234 packet_count = rule->packet_count;
3235 byte_count = rule->byte_count;
3237 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3238 * wildcarded then on all of its subrules.
3240 * Also, add any statistics that are not tracked by the datapath for each
3241 * subrule. This includes, for example, statistics for packets that were
3242 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3244 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3245 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3246 if (rule->cr.wc.wildcards) {
3248 LIST_FOR_EACH (subrule, list, &rule->list) {
3249 odp_flows[i++].key = subrule->cr.flow;
3250 packet_count += subrule->packet_count;
3251 byte_count += subrule->byte_count;
3254 odp_flows[0].key = rule->cr.flow;
3257 /* Fetch up-to-date statistics from the datapath and add them in. */
3258 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3260 for (i = 0; i < n_odp_flows; i++) {
3261 struct odp_flow *odp_flow = &odp_flows[i];
3262 packet_count += odp_flow->stats.n_packets;
3263 byte_count += odp_flow->stats.n_bytes;
3268 /* Return the stats to the caller. */
3269 *packet_countp = packet_count;
3270 *byte_countp = byte_count;
3274 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3276 struct rule *rule = rule_from_cls_rule(rule_);
3277 struct flow_stats_cbdata *cbdata = cbdata_;
3278 struct ofp_flow_stats *ofs;
3279 uint64_t packet_count, byte_count;
3280 size_t act_len, len;
3281 long long int tdiff = time_msec() - rule->created;
3282 uint32_t sec = tdiff / 1000;
3283 uint32_t msec = tdiff - (sec * 1000);
3285 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3289 act_len = sizeof *rule->actions * rule->n_actions;
3290 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3292 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3294 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3295 ofs->length = htons(len);
3296 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3298 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3299 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3300 ofs->duration_sec = htonl(sec);
3301 ofs->duration_nsec = htonl(msec * 1000000);
3302 ofs->cookie = rule->flow_cookie;
3303 ofs->priority = htons(rule->cr.priority);
3304 ofs->idle_timeout = htons(rule->idle_timeout);
3305 ofs->hard_timeout = htons(rule->hard_timeout);
3306 memset(ofs->pad2, 0, sizeof ofs->pad2);
3307 ofs->packet_count = htonll(packet_count);
3308 ofs->byte_count = htonll(byte_count);
3309 if (rule->n_actions > 0) {
3310 memcpy(ofs->actions, rule->actions, act_len);
3315 table_id_to_include(uint8_t table_id)
3317 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3318 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3319 : table_id == 0xff ? CLS_INC_ALL
3324 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3325 const struct ofp_stats_request *osr,
3328 struct ofp_flow_stats_request *fsr;
3329 struct flow_stats_cbdata cbdata;
3330 struct cls_rule target;
3332 if (arg_size != sizeof *fsr) {
3333 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3335 fsr = (struct ofp_flow_stats_request *) osr->body;
3337 COVERAGE_INC(ofproto_flows_req);
3339 cbdata.ofconn = ofconn;
3340 cbdata.out_port = fsr->out_port;
3341 cbdata.msg = start_stats_reply(osr, 1024);
3342 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3343 classifier_for_each_match(&p->cls, &target,
3344 table_id_to_include(fsr->table_id),
3345 flow_stats_cb, &cbdata);
3346 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3350 struct flow_stats_ds_cbdata {
3351 struct ofproto *ofproto;
3356 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3358 struct rule *rule = rule_from_cls_rule(rule_);
3359 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3360 struct ds *results = cbdata->results;
3361 struct ofp_match match;
3362 uint64_t packet_count, byte_count;
3363 size_t act_len = sizeof *rule->actions * rule->n_actions;
3365 /* Don't report on subrules. */
3366 if (rule->super != NULL) {
3370 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3371 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3372 cbdata->ofproto->tun_id_from_cookie, &match);
3374 ds_put_format(results, "duration=%llds, ",
3375 (time_msec() - rule->created) / 1000);
3376 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3377 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3378 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3379 ofp_print_match(results, &match, true);
3381 ofp_print_actions(results, &rule->actions->header, act_len);
3383 ds_put_cstr(results, "\n");
3386 /* Adds a pretty-printed description of all flows to 'results', including
3387 * those marked hidden by secchan (e.g., by in-band control). */
3389 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3391 struct ofp_match match;
3392 struct cls_rule target;
3393 struct flow_stats_ds_cbdata cbdata;
3395 memset(&match, 0, sizeof match);
3396 match.wildcards = htonl(OVSFW_ALL);
3399 cbdata.results = results;
3401 cls_rule_from_match(&match, 0, false, 0, &target);
3402 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3403 flow_stats_ds_cb, &cbdata);
3406 struct aggregate_stats_cbdata {
3407 struct ofproto *ofproto;
3409 uint64_t packet_count;
3410 uint64_t byte_count;
3415 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3417 struct rule *rule = rule_from_cls_rule(rule_);
3418 struct aggregate_stats_cbdata *cbdata = cbdata_;
3419 uint64_t packet_count, byte_count;
3421 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3425 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3427 cbdata->packet_count += packet_count;
3428 cbdata->byte_count += byte_count;
3433 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3434 const struct ofp_stats_request *osr,
3437 struct ofp_aggregate_stats_request *asr;
3438 struct ofp_aggregate_stats_reply *reply;
3439 struct aggregate_stats_cbdata cbdata;
3440 struct cls_rule target;
3443 if (arg_size != sizeof *asr) {
3444 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3446 asr = (struct ofp_aggregate_stats_request *) osr->body;
3448 COVERAGE_INC(ofproto_agg_request);
3450 cbdata.out_port = asr->out_port;
3451 cbdata.packet_count = 0;
3452 cbdata.byte_count = 0;
3454 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3455 classifier_for_each_match(&p->cls, &target,
3456 table_id_to_include(asr->table_id),
3457 aggregate_stats_cb, &cbdata);
3459 msg = start_stats_reply(osr, sizeof *reply);
3460 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3461 reply->flow_count = htonl(cbdata.n_flows);
3462 reply->packet_count = htonll(cbdata.packet_count);
3463 reply->byte_count = htonll(cbdata.byte_count);
3464 queue_tx(msg, ofconn, ofconn->reply_counter);
3468 struct queue_stats_cbdata {
3469 struct ofconn *ofconn;
3470 struct ofport *ofport;
3475 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3476 const struct netdev_queue_stats *stats)
3478 struct ofp_queue_stats *reply;
3480 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3481 reply->port_no = htons(cbdata->ofport->opp.port_no);
3482 memset(reply->pad, 0, sizeof reply->pad);
3483 reply->queue_id = htonl(queue_id);
3484 reply->tx_bytes = htonll(stats->tx_bytes);
3485 reply->tx_packets = htonll(stats->tx_packets);
3486 reply->tx_errors = htonll(stats->tx_errors);
3490 handle_queue_stats_dump_cb(uint32_t queue_id,
3491 struct netdev_queue_stats *stats,
3494 struct queue_stats_cbdata *cbdata = cbdata_;
3496 put_queue_stats(cbdata, queue_id, stats);
3500 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3501 struct queue_stats_cbdata *cbdata)
3503 cbdata->ofport = port;
3504 if (queue_id == OFPQ_ALL) {
3505 netdev_dump_queue_stats(port->netdev,
3506 handle_queue_stats_dump_cb, cbdata);
3508 struct netdev_queue_stats stats;
3510 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3511 put_queue_stats(cbdata, queue_id, &stats);
3517 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3518 const struct ofp_stats_request *osr,
3521 struct ofp_queue_stats_request *qsr;
3522 struct queue_stats_cbdata cbdata;
3523 struct ofport *port;
3524 unsigned int port_no;
3527 if (arg_size != sizeof *qsr) {
3528 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3530 qsr = (struct ofp_queue_stats_request *) osr->body;
3532 COVERAGE_INC(ofproto_queue_req);
3534 cbdata.ofconn = ofconn;
3535 cbdata.msg = start_stats_reply(osr, 128);
3537 port_no = ntohs(qsr->port_no);
3538 queue_id = ntohl(qsr->queue_id);
3539 if (port_no == OFPP_ALL) {
3540 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3541 handle_queue_stats_for_port(port, queue_id, &cbdata);
3543 } else if (port_no < ofproto->max_ports) {
3544 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3546 handle_queue_stats_for_port(port, queue_id, &cbdata);
3549 ofpbuf_delete(cbdata.msg);
3550 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3552 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3558 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3559 struct ofp_header *oh)
3561 struct ofp_stats_request *osr;
3565 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3570 osr = (struct ofp_stats_request *) oh;
3572 switch (ntohs(osr->type)) {
3574 return handle_desc_stats_request(p, ofconn, osr);
3577 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3579 case OFPST_AGGREGATE:
3580 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3583 return handle_table_stats_request(p, ofconn, osr);
3586 return handle_port_stats_request(p, ofconn, osr, arg_size);
3589 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3592 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3595 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3599 static long long int
3600 msec_from_nsec(uint64_t sec, uint32_t nsec)
3602 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3606 update_time(struct ofproto *ofproto, struct rule *rule,
3607 const struct odp_flow_stats *stats)
3609 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3610 if (used > rule->used) {
3612 if (rule->super && used > rule->super->used) {
3613 rule->super->used = used;
3615 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3620 update_stats(struct ofproto *ofproto, struct rule *rule,
3621 const struct odp_flow_stats *stats)
3623 if (stats->n_packets) {
3624 update_time(ofproto, rule, stats);
3625 rule->packet_count += stats->n_packets;
3626 rule->byte_count += stats->n_bytes;
3627 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3631 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3632 * in which no matching flow already exists in the flow table.
3634 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3635 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3636 * code as encoded by ofp_mkerr() on failure.
3638 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3641 add_flow(struct ofproto *p, struct ofconn *ofconn,
3642 const struct ofp_flow_mod *ofm, size_t n_actions)
3644 struct ofpbuf *packet;
3649 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3653 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3655 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3656 ntohs(ofm->priority))) {
3657 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3661 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3662 n_actions, ntohs(ofm->idle_timeout),
3663 ntohs(ofm->hard_timeout), ofm->cookie,
3664 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3665 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3666 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3669 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3670 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3674 in_port = UINT16_MAX;
3677 rule_insert(p, rule, packet, in_port);
3681 static struct rule *
3682 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3687 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3689 return rule_from_cls_rule(classifier_find_rule_exactly(
3690 &p->cls, &flow, wildcards,
3691 ntohs(ofm->priority)));
3695 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3696 struct rule *rule, const struct ofp_flow_mod *ofm)
3698 struct ofpbuf *packet;
3703 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3707 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3713 flow_extract(packet, 0, in_port, &flow);
3714 rule_execute(ofproto, rule, packet, &flow);
3719 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3721 struct modify_flows_cbdata {
3722 struct ofproto *ofproto;
3723 const struct ofp_flow_mod *ofm;
3728 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3729 size_t n_actions, struct rule *);
3730 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3732 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3733 * encoded by ofp_mkerr() on failure.
3735 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3738 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3739 const struct ofp_flow_mod *ofm, size_t n_actions)
3741 struct modify_flows_cbdata cbdata;
3742 struct cls_rule target;
3746 cbdata.n_actions = n_actions;
3747 cbdata.match = NULL;
3749 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3752 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3753 modify_flows_cb, &cbdata);
3755 /* This credits the packet to whichever flow happened to happened to
3756 * match last. That's weird. Maybe we should do a lookup for the
3757 * flow that actually matches the packet? Who knows. */
3758 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3761 return add_flow(p, ofconn, ofm, n_actions);
3765 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3766 * code as encoded by ofp_mkerr() on failure.
3768 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3771 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3772 struct ofp_flow_mod *ofm, size_t n_actions)
3774 struct rule *rule = find_flow_strict(p, ofm);
3775 if (rule && !rule_is_hidden(rule)) {
3776 modify_flow(p, ofm, n_actions, rule);
3777 return send_buffered_packet(p, ofconn, rule, ofm);
3779 return add_flow(p, ofconn, ofm, n_actions);
3783 /* Callback for modify_flows_loose(). */
3785 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3787 struct rule *rule = rule_from_cls_rule(rule_);
3788 struct modify_flows_cbdata *cbdata = cbdata_;
3790 if (!rule_is_hidden(rule)) {
3791 cbdata->match = rule;
3792 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3796 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3797 * been identified as a flow in 'p''s flow table to be modified, by changing
3798 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3799 * ofp_action[] structures). */
3801 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3802 size_t n_actions, struct rule *rule)
3804 size_t actions_len = n_actions * sizeof *rule->actions;
3806 rule->flow_cookie = ofm->cookie;
3808 /* If the actions are the same, do nothing. */
3809 if (n_actions == rule->n_actions
3810 && (!n_actions || !memcmp(ofm->actions, rule->actions, actions_len)))
3815 /* Replace actions. */
3816 free(rule->actions);
3817 rule->actions = n_actions ? xmemdup(ofm->actions, actions_len) : NULL;
3818 rule->n_actions = n_actions;
3820 /* Make sure that the datapath gets updated properly. */
3821 if (rule->cr.wc.wildcards) {
3822 COVERAGE_INC(ofproto_mod_wc_flow);
3823 p->need_revalidate = true;
3825 rule_update_actions(p, rule);
3831 /* OFPFC_DELETE implementation. */
3833 struct delete_flows_cbdata {
3834 struct ofproto *ofproto;
3838 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3839 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3841 /* Implements OFPFC_DELETE. */
3843 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3845 struct delete_flows_cbdata cbdata;
3846 struct cls_rule target;
3849 cbdata.out_port = ofm->out_port;
3851 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3854 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3855 delete_flows_cb, &cbdata);
3858 /* Implements OFPFC_DELETE_STRICT. */
3860 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3862 struct rule *rule = find_flow_strict(p, ofm);
3864 delete_flow(p, rule, ofm->out_port);
3868 /* Callback for delete_flows_loose(). */
3870 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3872 struct rule *rule = rule_from_cls_rule(rule_);
3873 struct delete_flows_cbdata *cbdata = cbdata_;
3875 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3878 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3879 * been identified as a flow to delete from 'p''s flow table, by deleting the
3880 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3883 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3884 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3885 * specified 'out_port'. */
3887 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3889 if (rule_is_hidden(rule)) {
3893 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3897 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3898 rule_remove(p, rule);
3902 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3903 struct ofp_flow_mod *ofm)
3905 struct ofp_match orig_match;
3909 error = reject_slave_controller(ofconn, &ofm->header);
3913 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3914 sizeof *ofm->actions, &n_actions);
3919 /* We do not support the emergency flow cache. It will hopefully
3920 * get dropped from OpenFlow in the near future. */
3921 if (ofm->flags & htons(OFPFF_EMERG)) {
3922 /* There isn't a good fit for an error code, so just state that the
3923 * flow table is full. */
3924 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3927 /* Normalize ofp->match. If normalization actually changes anything, then
3928 * log the differences. */
3929 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3930 orig_match = ofm->match;
3931 normalize_match(&ofm->match);
3932 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3933 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3934 if (!VLOG_DROP_INFO(&normal_rl)) {
3935 char *old = ofp_match_to_literal_string(&orig_match);
3936 char *new = ofp_match_to_literal_string(&ofm->match);
3937 VLOG_INFO("%s: normalization changed ofp_match, details:",
3938 rconn_get_name(ofconn->rconn));
3939 VLOG_INFO(" pre: %s", old);
3940 VLOG_INFO("post: %s", new);
3946 if (!ofm->match.wildcards) {
3947 ofm->priority = htons(UINT16_MAX);
3950 error = validate_actions((const union ofp_action *) ofm->actions,
3951 n_actions, p->max_ports);
3956 switch (ntohs(ofm->command)) {
3958 return add_flow(p, ofconn, ofm, n_actions);
3961 return modify_flows_loose(p, ofconn, ofm, n_actions);
3963 case OFPFC_MODIFY_STRICT:
3964 return modify_flow_strict(p, ofconn, ofm, n_actions);
3967 delete_flows_loose(p, ofm);
3970 case OFPFC_DELETE_STRICT:
3971 delete_flow_strict(p, ofm);
3975 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3980 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3984 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3989 p->tun_id_from_cookie = !!msg->set;
3994 handle_role_request(struct ofproto *ofproto,
3995 struct ofconn *ofconn, struct nicira_header *msg)
3997 struct nx_role_request *nrr;
3998 struct nx_role_request *reply;
4002 if (ntohs(msg->header.length) != sizeof *nrr) {
4003 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
4004 ntohs(msg->header.length), sizeof *nrr);
4005 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4007 nrr = (struct nx_role_request *) msg;
4009 if (ofconn->type != OFCONN_PRIMARY) {
4010 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
4012 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
4015 role = ntohl(nrr->role);
4016 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
4017 && role != NX_ROLE_SLAVE) {
4018 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
4020 /* There's no good error code for this. */
4021 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
4024 if (role == NX_ROLE_MASTER) {
4025 struct ofconn *other;
4027 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
4028 if (other->role == NX_ROLE_MASTER) {
4029 other->role = NX_ROLE_SLAVE;
4033 ofconn->role = role;
4035 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
4037 reply->nxh.vendor = htonl(NX_VENDOR_ID);
4038 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
4039 reply->role = htonl(role);
4040 queue_tx(buf, ofconn, ofconn->reply_counter);
4046 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
4048 struct ofp_vendor_header *ovh = msg;
4049 struct nicira_header *nh;
4051 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
4052 VLOG_WARN_RL(&rl, "received vendor message of length %u "
4053 "(expected at least %zu)",
4054 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
4055 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4057 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
4058 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
4060 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
4061 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
4062 "(expected at least %zu)",
4063 ntohs(ovh->header.length), sizeof(struct nicira_header));
4064 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4068 switch (ntohl(nh->subtype)) {
4069 case NXT_STATUS_REQUEST:
4070 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4073 case NXT_TUN_ID_FROM_COOKIE:
4074 return handle_tun_id_from_cookie(p, msg);
4076 case NXT_ROLE_REQUEST:
4077 return handle_role_request(p, ofconn, msg);
4080 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4084 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4086 struct ofp_header *ob;
4089 /* Currently, everything executes synchronously, so we can just
4090 * immediately send the barrier reply. */
4091 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4092 queue_tx(buf, ofconn, ofconn->reply_counter);
4097 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4098 struct ofpbuf *ofp_msg)
4100 struct ofp_header *oh = ofp_msg->data;
4103 COVERAGE_INC(ofproto_recv_openflow);
4105 case OFPT_ECHO_REQUEST:
4106 error = handle_echo_request(ofconn, oh);
4109 case OFPT_ECHO_REPLY:
4113 case OFPT_FEATURES_REQUEST:
4114 error = handle_features_request(p, ofconn, oh);
4117 case OFPT_GET_CONFIG_REQUEST:
4118 error = handle_get_config_request(p, ofconn, oh);
4121 case OFPT_SET_CONFIG:
4122 error = handle_set_config(p, ofconn, ofp_msg->data);
4125 case OFPT_PACKET_OUT:
4126 error = handle_packet_out(p, ofconn, ofp_msg->data);
4130 error = handle_port_mod(p, ofconn, oh);
4134 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4137 case OFPT_STATS_REQUEST:
4138 error = handle_stats_request(p, ofconn, oh);
4142 error = handle_vendor(p, ofconn, ofp_msg->data);
4145 case OFPT_BARRIER_REQUEST:
4146 error = handle_barrier_request(ofconn, oh);
4150 if (VLOG_IS_WARN_ENABLED()) {
4151 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4152 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4155 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4160 send_error_oh(ofconn, ofp_msg->data, error);
4165 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4167 struct odp_msg *msg = packet->data;
4169 struct ofpbuf payload;
4172 payload.data = msg + 1;
4173 payload.size = msg->length - sizeof *msg;
4174 flow_extract(&payload, msg->arg, msg->port, &flow);
4176 /* Check with in-band control to see if this packet should be sent
4177 * to the local port regardless of the flow table. */
4178 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4179 union odp_action action;
4181 memset(&action, 0, sizeof(action));
4182 action.output.type = ODPAT_OUTPUT;
4183 action.output.port = ODPP_LOCAL;
4184 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
4187 rule = lookup_valid_rule(p, &flow);
4189 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4190 struct ofport *port = get_port(p, msg->port);
4192 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4193 COVERAGE_INC(ofproto_no_packet_in);
4194 /* XXX install 'drop' flow entry */
4195 ofpbuf_delete(packet);
4199 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4202 COVERAGE_INC(ofproto_packet_in);
4203 send_packet_in(p, packet);
4207 if (rule->cr.wc.wildcards) {
4208 rule = rule_create_subrule(p, rule, &flow);
4209 rule_make_actions(p, rule, packet);
4211 if (!rule->may_install) {
4212 /* The rule is not installable, that is, we need to process every
4213 * packet, so process the current packet and set its actions into
4215 rule_make_actions(p, rule, packet);
4217 /* XXX revalidate rule if it needs it */
4221 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4223 * Extra-special case for fail-open mode.
4225 * We are in fail-open mode and the packet matched the fail-open rule,
4226 * but we are connected to a controller too. We should send the packet
4227 * up to the controller in the hope that it will try to set up a flow
4228 * and thereby allow us to exit fail-open.
4230 * See the top-level comment in fail-open.c for more information.
4232 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4233 DPIF_RECV_MSG_PADDING));
4236 ofpbuf_pull(packet, sizeof *msg);
4237 rule_execute(p, rule, packet, &flow);
4238 rule_reinstall(p, rule);
4242 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4244 struct odp_msg *msg = packet->data;
4246 switch (msg->type) {
4247 case _ODPL_ACTION_NR:
4248 COVERAGE_INC(ofproto_ctlr_action);
4249 send_packet_in(p, packet);
4252 case _ODPL_SFLOW_NR:
4254 ofproto_sflow_received(p->sflow, msg);
4256 ofpbuf_delete(packet);
4260 handle_odp_miss_msg(p, packet);
4264 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4270 /* Flow expiration. */
4272 struct expire_cbdata {
4273 struct ofproto *ofproto;
4277 static int ofproto_dp_max_idle(const struct ofproto *);
4278 static void ofproto_update_used(struct ofproto *);
4279 static void rule_expire(struct cls_rule *, void *cbdata);
4281 /* This function is called periodically by ofproto_run(). Its job is to
4282 * collect updates for the flows that have been installed into the datapath,
4283 * most importantly when they last were used, and then use that information to
4284 * expire flows that have not been used recently.
4286 * Returns the number of milliseconds after which it should be called again. */
4288 ofproto_expire(struct ofproto *ofproto)
4290 struct expire_cbdata cbdata;
4292 /* Update 'used' for each flow in the datapath. */
4293 ofproto_update_used(ofproto);
4295 /* Expire idle flows.
4297 * A wildcarded flow is idle only when all of its subrules have expired due
4298 * to becoming idle, so iterate through the exact-match flows first. */
4299 cbdata.ofproto = ofproto;
4300 cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
4301 classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata);
4302 classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata);
4304 /* Let the hook know that we're at a stable point: all outstanding data
4305 * in existing flows has been accounted to the account_cb. Thus, the
4306 * hook can now reasonably do operations that depend on having accurate
4307 * flow volume accounting (currently, that's just bond rebalancing). */
4308 if (ofproto->ofhooks->account_checkpoint_cb) {
4309 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4312 return MIN(cbdata.dp_max_idle, 1000);
4315 /* Update 'used' member of each flow currently installed into the datapath. */
4317 ofproto_update_used(struct ofproto *p)
4319 struct odp_flow *flows;
4324 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4329 for (i = 0; i < n_flows; i++) {
4330 struct odp_flow *f = &flows[i];
4333 rule = rule_from_cls_rule(
4334 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4336 if (rule && rule->installed) {
4337 update_time(p, rule, &f->stats);
4338 rule_account(p, rule, f->stats.n_bytes);
4340 /* There's a flow in the datapath that we know nothing about.
4342 COVERAGE_INC(ofproto_unexpected_rule);
4343 dpif_flow_del(p->dpif, f);
4350 /* Calculates and returns the number of milliseconds of idle time after which
4351 * flows should expire from the datapath and we should fold their statistics
4352 * into their parent rules in userspace. */
4354 ofproto_dp_max_idle(const struct ofproto *ofproto)
4357 * Idle time histogram.
4359 * Most of the time a switch has a relatively small number of flows. When
4360 * this is the case we might as well keep statistics for all of them in
4361 * userspace and to cache them in the kernel datapath for performance as
4364 * As the number of flows increases, the memory required to maintain
4365 * statistics about them in userspace and in the kernel becomes
4366 * significant. However, with a large number of flows it is likely that
4367 * only a few of them are "heavy hitters" that consume a large amount of
4368 * bandwidth. At this point, only heavy hitters are worth caching in the
4369 * kernel and maintaining in userspaces; other flows we can discard.
4371 * The technique used to compute the idle time is to build a histogram with
4372 * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that
4373 * is installed in the kernel gets dropped in the appropriate bucket.
4374 * After the histogram has been built, we compute the cutoff so that only
4375 * the most-recently-used 1% of flows (but at least 1000 flows) are kept
4376 * cached. At least the most-recently-used bucket of flows is kept, so
4377 * actually an arbitrary number of flows can be kept in any given
4378 * expiration run (though the next run will delete most of those unless
4379 * they receive additional data).
4381 * This requires a second pass through the exact-match flows, in addition
4382 * to the pass made by ofproto_update_used(), because the former function
4383 * never looks at uninstallable flows.
4385 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4386 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4387 int buckets[N_BUCKETS] = { 0 };
4393 total = classifier_count_exact(&ofproto->cls);
4394 if (total <= 1000) {
4395 return N_BUCKETS * BUCKET_WIDTH;
4398 /* Build histogram. */
4400 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) {
4401 long long int idle = now - rule->used;
4402 int bucket = (idle <= 0 ? 0
4403 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4404 : (unsigned int) idle / BUCKET_WIDTH);
4408 /* Find the first bucket whose flows should be expired. */
4409 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4410 if (buckets[bucket]) {
4413 subtotal += buckets[bucket++];
4414 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4419 if (VLOG_IS_DBG_ENABLED()) {
4423 ds_put_cstr(&s, "keep");
4424 for (i = 0; i < N_BUCKETS; i++) {
4426 ds_put_cstr(&s, ", drop");
4429 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4432 VLOG_INFO("%s: %s (msec:count)",
4433 dpif_name(ofproto->dpif), ds_cstr(&s));
4437 return bucket * BUCKET_WIDTH;
4441 rule_active_timeout(struct ofproto *ofproto, struct rule *rule)
4443 if (ofproto->netflow && !is_controller_rule(rule) &&
4444 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4445 struct ofexpired expired;
4446 struct odp_flow odp_flow;
4448 /* Get updated flow stats.
4450 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4451 * updated TCP flags and (2) the dpif_flow_list_all() in
4452 * ofproto_update_used() zeroed TCP flags. */
4453 memset(&odp_flow, 0, sizeof odp_flow);
4454 if (rule->installed) {
4455 odp_flow.key = rule->cr.flow;
4456 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4457 dpif_flow_get(ofproto->dpif, &odp_flow);
4459 if (odp_flow.stats.n_packets) {
4460 update_time(ofproto, rule, &odp_flow.stats);
4461 netflow_flow_update_flags(&rule->nf_flow,
4462 odp_flow.stats.tcp_flags);
4466 expired.flow = rule->cr.flow;
4467 expired.packet_count = rule->packet_count +
4468 odp_flow.stats.n_packets;
4469 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4470 expired.used = rule->used;
4472 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4476 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4477 * rules, then delete it entirely.
4479 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4480 * the datapath and fold its statistics back into its super-rule.
4482 * (This is a callback function for classifier_for_each().) */
4484 rule_expire(struct cls_rule *cls_rule, void *cbdata_)
4486 struct expire_cbdata *cbdata = cbdata_;
4487 struct ofproto *ofproto = cbdata->ofproto;
4488 struct rule *rule = rule_from_cls_rule(cls_rule);
4489 long long int hard_expire, idle_expire, expire, now;
4491 /* Calculate OpenFlow expiration times for 'rule'. */
4492 hard_expire = (rule->hard_timeout
4493 ? rule->created + rule->hard_timeout * 1000
4495 idle_expire = (rule->idle_timeout
4496 && (rule->super || list_is_empty(&rule->list))
4497 ? rule->used + rule->idle_timeout * 1000
4499 expire = MIN(hard_expire, idle_expire);
4503 /* 'rule' has not expired according to OpenFlow rules. */
4504 if (!rule->cr.wc.wildcards) {
4505 if (now >= rule->used + cbdata->dp_max_idle) {
4506 /* This rule is idle, so drop it to free up resources. */
4508 /* It's not part of the OpenFlow flow table, so we can
4509 * delete it entirely and fold its statistics into its
4511 rule_remove(ofproto, rule);
4513 /* It is part of the OpenFlow flow table, so we have to
4514 * keep the rule but we can at least uninstall it from the
4516 rule_uninstall(ofproto, rule);
4519 /* Send NetFlow active timeout if appropriate. */
4520 rule_active_timeout(cbdata->ofproto, rule);
4524 /* 'rule' has expired according to OpenFlow rules. */
4525 COVERAGE_INC(ofproto_expired);
4527 /* Update stats. (This is a no-op if the rule expired due to an idle
4528 * timeout, because that only happens when the rule has no subrules
4530 if (rule->cr.wc.wildcards) {
4531 struct rule *subrule, *next;
4532 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
4533 rule_remove(cbdata->ofproto, subrule);
4536 rule_uninstall(cbdata->ofproto, rule);
4539 /* Get rid of the rule. */
4540 if (!rule_is_hidden(rule)) {
4541 send_flow_removed(cbdata->ofproto, rule, now,
4543 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4545 rule_remove(cbdata->ofproto, rule);
4550 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4552 struct rule *sub = rule_from_cls_rule(sub_);
4553 struct revalidate_cbdata *cbdata = cbdata_;
4555 if (cbdata->revalidate_all
4556 || (cbdata->revalidate_subrules && sub->super)
4557 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4558 revalidate_rule(cbdata->ofproto, sub);
4563 revalidate_rule(struct ofproto *p, struct rule *rule)
4565 const flow_t *flow = &rule->cr.flow;
4567 COVERAGE_INC(ofproto_revalidate_rule);
4570 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4572 rule_remove(p, rule);
4574 } else if (super != rule->super) {
4575 COVERAGE_INC(ofproto_revalidate_moved);
4576 list_remove(&rule->list);
4577 list_push_back(&super->list, &rule->list);
4578 rule->super = super;
4579 rule->hard_timeout = super->hard_timeout;
4580 rule->idle_timeout = super->idle_timeout;
4581 rule->created = super->created;
4586 rule_update_actions(p, rule);
4590 static struct ofpbuf *
4591 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4592 long long int now, uint8_t reason)
4594 struct ofp_flow_removed *ofr;
4596 long long int tdiff = now - rule->created;
4597 uint32_t sec = tdiff / 1000;
4598 uint32_t msec = tdiff - (sec * 1000);
4600 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4601 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4603 ofr->cookie = rule->flow_cookie;
4604 ofr->priority = htons(rule->cr.priority);
4605 ofr->reason = reason;
4606 ofr->duration_sec = htonl(sec);
4607 ofr->duration_nsec = htonl(msec * 1000000);
4608 ofr->idle_timeout = htons(rule->idle_timeout);
4609 ofr->packet_count = htonll(rule->packet_count);
4610 ofr->byte_count = htonll(rule->byte_count);
4616 send_flow_removed(struct ofproto *p, struct rule *rule,
4617 long long int now, uint8_t reason)
4619 struct ofconn *ofconn;
4620 struct ofconn *prev;
4621 struct ofpbuf *buf = NULL;
4623 if (!rule->send_flow_removed) {
4627 /* We limit the maximum number of queued flow expirations it by accounting
4628 * them under the counter for replies. That works because preventing
4629 * OpenFlow requests from being processed also prevents new flows from
4630 * being added (and expiring). (It also prevents processing OpenFlow
4631 * requests that would not add new flows, so it is imperfect.) */
4634 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4635 if (rconn_is_connected(ofconn->rconn)
4636 && ofconn_receives_async_msgs(ofconn)) {
4638 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4640 buf = compose_flow_removed(p, rule, now, reason);
4646 queue_tx(buf, prev, prev->reply_counter);
4650 /* pinsched callback for sending 'packet' on 'ofconn'. */
4652 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4654 struct ofconn *ofconn = ofconn_;
4656 rconn_send_with_limit(ofconn->rconn, packet,
4657 ofconn->packet_in_counter, 100);
4660 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4661 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4662 * packet scheduler for sending.
4664 * 'max_len' specifies the maximum number of bytes of the packet to send on
4665 * 'ofconn' (INT_MAX specifies no limit).
4667 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4668 * ownership is transferred to this function. */
4670 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4673 struct ofproto *ofproto = ofconn->ofproto;
4674 struct ofp_packet_in *opi = packet->data;
4675 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4676 int send_len, trim_size;
4680 if (opi->reason == OFPR_ACTION) {
4681 buffer_id = UINT32_MAX;
4682 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4683 buffer_id = pktbuf_get_null();
4684 } else if (!ofconn->pktbuf) {
4685 buffer_id = UINT32_MAX;
4687 struct ofpbuf payload;
4688 payload.data = opi->data;
4689 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4690 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4693 /* Figure out how much of the packet to send. */
4694 send_len = ntohs(opi->total_len);
4695 if (buffer_id != UINT32_MAX) {
4696 send_len = MIN(send_len, ofconn->miss_send_len);
4698 send_len = MIN(send_len, max_len);
4700 /* Adjust packet length and clone if necessary. */
4701 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4703 packet = ofpbuf_clone_data(packet->data, trim_size);
4706 packet->size = trim_size;
4709 /* Update packet headers. */
4710 opi->buffer_id = htonl(buffer_id);
4711 update_openflow_length(packet);
4713 /* Hand over to packet scheduler. It might immediately call into
4714 * do_send_packet_in() or it might buffer it for a while (until a later
4715 * call to pinsched_run()). */
4716 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4717 packet, do_send_packet_in, ofconn);
4720 /* Replace struct odp_msg header in 'packet' by equivalent struct
4721 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4722 * returned by dpif_recv()).
4724 * The conversion is not complete: the caller still needs to trim any unneeded
4725 * payload off the end of the buffer, set the length in the OpenFlow header,
4726 * and set buffer_id. Those require us to know the controller settings and so
4727 * must be done on a per-controller basis.
4729 * Returns the maximum number of bytes of the packet that should be sent to
4730 * the controller (INT_MAX if no limit). */
4732 do_convert_to_packet_in(struct ofpbuf *packet)
4734 struct odp_msg *msg = packet->data;
4735 struct ofp_packet_in *opi;
4741 /* Extract relevant header fields */
4742 if (msg->type == _ODPL_ACTION_NR) {
4743 reason = OFPR_ACTION;
4746 reason = OFPR_NO_MATCH;
4749 total_len = msg->length - sizeof *msg;
4750 in_port = odp_port_to_ofp_port(msg->port);
4752 /* Repurpose packet buffer by overwriting header. */
4753 ofpbuf_pull(packet, sizeof(struct odp_msg));
4754 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4755 opi->header.version = OFP_VERSION;
4756 opi->header.type = OFPT_PACKET_IN;
4757 opi->total_len = htons(total_len);
4758 opi->in_port = htons(in_port);
4759 opi->reason = reason;
4764 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4765 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4766 * as necessary according to their individual configurations.
4768 * 'packet' must have sufficient headroom to convert it into a struct
4769 * ofp_packet_in (e.g. as returned by dpif_recv()).
4771 * Takes ownership of 'packet'. */
4773 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4775 struct ofconn *ofconn, *prev;
4778 max_len = do_convert_to_packet_in(packet);
4781 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4782 if (ofconn_receives_async_msgs(ofconn)) {
4784 schedule_packet_in(prev, packet, max_len, true);
4790 schedule_packet_in(prev, packet, max_len, false);
4792 ofpbuf_delete(packet);
4797 pick_datapath_id(const struct ofproto *ofproto)
4799 const struct ofport *port;
4801 port = get_port(ofproto, ODPP_LOCAL);
4803 uint8_t ea[ETH_ADDR_LEN];
4806 error = netdev_get_etheraddr(port->netdev, ea);
4808 return eth_addr_to_uint64(ea);
4810 VLOG_WARN("could not get MAC address for %s (%s)",
4811 netdev_get_name(port->netdev), strerror(error));
4813 return ofproto->fallback_dpid;
4817 pick_fallback_dpid(void)
4819 uint8_t ea[ETH_ADDR_LEN];
4820 eth_addr_nicira_random(ea);
4821 return eth_addr_to_uint64(ea);
4825 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4826 struct odp_actions *actions, tag_type *tags,
4827 uint16_t *nf_output_iface, void *ofproto_)
4829 struct ofproto *ofproto = ofproto_;
4832 /* Drop frames for reserved multicast addresses. */
4833 if (eth_addr_is_reserved(flow->dl_dst)) {
4837 /* Learn source MAC (but don't try to learn from revalidation). */
4838 if (packet != NULL) {
4839 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4841 GRAT_ARP_LOCK_NONE);
4843 /* The log messages here could actually be useful in debugging,
4844 * so keep the rate limit relatively high. */
4845 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4846 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4847 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4848 ofproto_revalidate(ofproto, rev_tag);
4852 /* Determine output port. */
4853 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4856 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4857 } else if (out_port != flow->in_port) {
4858 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4859 *nf_output_iface = out_port;
4867 static const struct ofhooks default_ofhooks = {
4869 default_normal_ofhook_cb,