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 struct flow *, struct 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 struct ofport *get_port(const struct ofproto *, uint16_t odp_port);
340 static void update_port(struct ofproto *, const char *devname);
341 static int init_ports(struct ofproto *);
342 static void reinit_ports(struct ofproto *);
345 ofproto_create(const char *datapath, const char *datapath_type,
346 const struct ofhooks *ofhooks, void *aux,
347 struct ofproto **ofprotop)
349 struct odp_stats stats;
356 /* Connect to datapath and start listening for messages. */
357 error = dpif_open(datapath, datapath_type, &dpif);
359 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
362 error = dpif_get_dp_stats(dpif, &stats);
364 VLOG_ERR("failed to obtain stats for datapath %s: %s",
365 datapath, strerror(error));
369 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
371 VLOG_ERR("failed to listen on datapath %s: %s",
372 datapath, strerror(error));
376 dpif_flow_flush(dpif);
377 dpif_recv_purge(dpif);
379 /* Initialize settings. */
380 p = xzalloc(sizeof *p);
381 p->fallback_dpid = pick_fallback_dpid();
382 p->datapath_id = p->fallback_dpid;
383 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
384 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
385 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
386 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
387 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
389 /* Initialize datapath. */
391 p->netdev_monitor = netdev_monitor_create();
392 hmap_init(&p->ports);
393 shash_init(&p->port_by_name);
394 p->max_ports = stats.max_ports;
396 /* Initialize submodules. */
397 p->switch_status = switch_status_create(p);
403 /* Initialize flow table. */
404 classifier_init(&p->cls);
405 p->need_revalidate = false;
406 p->next_expiration = time_msec() + 1000;
407 tag_set_init(&p->revalidate_set);
409 /* Initialize OpenFlow connections. */
410 list_init(&p->all_conns);
411 hmap_init(&p->controllers);
412 hmap_init(&p->services);
416 /* Initialize hooks. */
418 p->ofhooks = ofhooks;
422 p->ofhooks = &default_ofhooks;
424 p->ml = mac_learning_create();
427 /* Pick final datapath ID. */
428 p->datapath_id = pick_datapath_id(p);
429 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
436 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
438 uint64_t old_dpid = p->datapath_id;
439 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
440 if (p->datapath_id != old_dpid) {
441 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
443 /* Force all active connections to reconnect, since there is no way to
444 * notify a controller that the datapath ID has changed. */
445 ofproto_reconnect_controllers(p);
450 is_discovery_controller(const struct ofproto_controller *c)
452 return !strcmp(c->target, "discover");
456 is_in_band_controller(const struct ofproto_controller *c)
458 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
461 /* Creates a new controller in 'ofproto'. Some of the settings are initially
462 * drawn from 'c', but update_controller() needs to be called later to finish
463 * the new ofconn's configuration. */
465 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
467 struct discovery *discovery;
468 struct ofconn *ofconn;
470 if (is_discovery_controller(c)) {
471 int error = discovery_create(c->accept_re, c->update_resolv_conf,
472 ofproto->dpif, ofproto->switch_status,
481 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY);
482 ofconn->pktbuf = pktbuf_create();
483 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
485 ofconn->discovery = discovery;
487 char *name = ofconn_make_name(ofproto, c->target);
488 rconn_connect(ofconn->rconn, c->target, name);
491 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
492 hash_string(c->target, 0));
495 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
496 * target or turn discovery on or off (these are done by creating new ofconns
497 * and deleting old ones), but it can update the rest of an ofconn's
500 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
504 ofconn->band = (is_in_band_controller(c)
505 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
507 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
509 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
510 rconn_set_probe_interval(ofconn->rconn, probe_interval);
512 if (ofconn->discovery) {
513 discovery_set_update_resolv_conf(ofconn->discovery,
514 c->update_resolv_conf);
515 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
518 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
522 ofconn_get_target(const struct ofconn *ofconn)
524 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
527 static struct ofconn *
528 find_controller_by_target(struct ofproto *ofproto, const char *target)
530 struct ofconn *ofconn;
532 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
533 hash_string(target, 0), &ofproto->controllers) {
534 if (!strcmp(ofconn_get_target(ofconn), target)) {
542 update_in_band_remotes(struct ofproto *ofproto)
544 const struct ofconn *ofconn;
545 struct sockaddr_in *addrs;
546 size_t max_addrs, n_addrs;
550 /* Allocate enough memory for as many remotes as we could possibly have. */
551 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
552 addrs = xmalloc(max_addrs * sizeof *addrs);
555 /* Add all the remotes. */
557 HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) {
558 struct sockaddr_in *sin = &addrs[n_addrs];
560 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
564 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
565 if (sin->sin_addr.s_addr) {
566 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
569 if (ofconn->discovery) {
573 for (i = 0; i < ofproto->n_extra_remotes; i++) {
574 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
577 /* Create or update or destroy in-band.
579 * Ordinarily we only enable in-band if there's at least one remote
580 * address, but discovery needs the in-band rules for DHCP to be installed
581 * even before we know any remote addresses. */
582 if (n_addrs || discovery) {
583 if (!ofproto->in_band) {
584 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
587 if (ofproto->in_band) {
588 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
590 ofproto->next_in_band_update = time_msec() + 1000;
592 in_band_destroy(ofproto->in_band);
593 ofproto->in_band = NULL;
601 update_fail_open(struct ofproto *p)
603 struct ofconn *ofconn;
605 if (!hmap_is_empty(&p->controllers)
606 && p->fail_mode == OFPROTO_FAIL_STANDALONE) {
607 struct rconn **rconns;
611 p->fail_open = fail_open_create(p, p->switch_status);
615 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
616 HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) {
617 rconns[n++] = ofconn->rconn;
620 fail_open_set_controllers(p->fail_open, rconns, n);
621 /* p->fail_open takes ownership of 'rconns'. */
623 fail_open_destroy(p->fail_open);
629 ofproto_set_controllers(struct ofproto *p,
630 const struct ofproto_controller *controllers,
631 size_t n_controllers)
633 struct shash new_controllers;
634 struct ofconn *ofconn, *next_ofconn;
635 struct ofservice *ofservice, *next_ofservice;
639 /* Create newly configured controllers and services.
640 * Create a name to ofproto_controller mapping in 'new_controllers'. */
641 shash_init(&new_controllers);
642 for (i = 0; i < n_controllers; i++) {
643 const struct ofproto_controller *c = &controllers[i];
645 if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) {
646 if (!find_controller_by_target(p, c->target)) {
647 add_controller(p, c);
649 } else if (!pvconn_verify_name(c->target)) {
650 if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) {
654 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
655 dpif_name(p->dpif), c->target);
659 shash_add_once(&new_controllers, c->target, &controllers[i]);
662 /* Delete controllers that are no longer configured.
663 * Update configuration of all now-existing controllers. */
665 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) {
666 struct ofproto_controller *c;
668 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
670 ofconn_destroy(ofconn);
672 update_controller(ofconn, c);
679 /* Delete services that are no longer configured.
680 * Update configuration of all now-existing services. */
681 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
682 struct ofproto_controller *c;
684 c = shash_find_data(&new_controllers,
685 pvconn_get_name(ofservice->pvconn));
687 ofservice_destroy(p, ofservice);
689 ofservice_reconfigure(ofservice, c);
693 shash_destroy(&new_controllers);
695 update_in_band_remotes(p);
698 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
699 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
700 struct ofconn, hmap_node);
701 ofconn->ss = switch_status_register(p->switch_status, "remote",
702 rconn_status_cb, ofconn->rconn);
707 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
709 p->fail_mode = fail_mode;
713 /* Drops the connections between 'ofproto' and all of its controllers, forcing
714 * them to reconnect. */
716 ofproto_reconnect_controllers(struct ofproto *ofproto)
718 struct ofconn *ofconn;
720 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
721 rconn_reconnect(ofconn->rconn);
726 any_extras_changed(const struct ofproto *ofproto,
727 const struct sockaddr_in *extras, size_t n)
731 if (n != ofproto->n_extra_remotes) {
735 for (i = 0; i < n; i++) {
736 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
737 const struct sockaddr_in *new = &extras[i];
739 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
740 old->sin_port != new->sin_port) {
748 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
749 * in-band control should guarantee access, in the same way that in-band
750 * control guarantees access to OpenFlow controllers. */
752 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
753 const struct sockaddr_in *extras, size_t n)
755 if (!any_extras_changed(ofproto, extras, n)) {
759 free(ofproto->extra_in_band_remotes);
760 ofproto->n_extra_remotes = n;
761 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
763 update_in_band_remotes(ofproto);
767 ofproto_set_desc(struct ofproto *p,
768 const char *mfr_desc, const char *hw_desc,
769 const char *sw_desc, const char *serial_desc,
772 struct ofp_desc_stats *ods;
775 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
776 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
777 sizeof ods->mfr_desc);
780 p->mfr_desc = xstrdup(mfr_desc);
783 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
784 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
785 sizeof ods->hw_desc);
788 p->hw_desc = xstrdup(hw_desc);
791 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
792 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
793 sizeof ods->sw_desc);
796 p->sw_desc = xstrdup(sw_desc);
799 if (strlen(serial_desc) >= sizeof ods->serial_num) {
800 VLOG_WARN("truncating serial_desc, must be less than %zu "
802 sizeof ods->serial_num);
804 free(p->serial_desc);
805 p->serial_desc = xstrdup(serial_desc);
808 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
809 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
810 sizeof ods->dp_desc);
813 p->dp_desc = xstrdup(dp_desc);
818 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
819 const struct svec *svec)
821 struct pvconn **pvconns = *pvconnsp;
822 size_t n_pvconns = *n_pvconnsp;
826 for (i = 0; i < n_pvconns; i++) {
827 pvconn_close(pvconns[i]);
831 pvconns = xmalloc(svec->n * sizeof *pvconns);
833 for (i = 0; i < svec->n; i++) {
834 const char *name = svec->names[i];
835 struct pvconn *pvconn;
838 error = pvconn_open(name, &pvconn);
840 pvconns[n_pvconns++] = pvconn;
842 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
850 *n_pvconnsp = n_pvconns;
856 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
858 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
862 ofproto_set_netflow(struct ofproto *ofproto,
863 const struct netflow_options *nf_options)
865 if (nf_options && nf_options->collectors.n) {
866 if (!ofproto->netflow) {
867 ofproto->netflow = netflow_create();
869 return netflow_set_options(ofproto->netflow, nf_options);
871 netflow_destroy(ofproto->netflow);
872 ofproto->netflow = NULL;
878 ofproto_set_sflow(struct ofproto *ofproto,
879 const struct ofproto_sflow_options *oso)
881 struct ofproto_sflow *os = ofproto->sflow;
884 struct ofport *ofport;
886 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
887 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
888 ofproto_sflow_add_port(os, ofport->odp_port,
889 netdev_get_name(ofport->netdev));
892 ofproto_sflow_set_options(os, oso);
894 ofproto_sflow_destroy(os);
895 ofproto->sflow = NULL;
900 ofproto_get_datapath_id(const struct ofproto *ofproto)
902 return ofproto->datapath_id;
906 ofproto_has_primary_controller(const struct ofproto *ofproto)
908 return !hmap_is_empty(&ofproto->controllers);
911 enum ofproto_fail_mode
912 ofproto_get_fail_mode(const struct ofproto *p)
918 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
922 for (i = 0; i < ofproto->n_snoops; i++) {
923 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
928 ofproto_destroy(struct ofproto *p)
930 struct ofservice *ofservice, *next_ofservice;
931 struct ofconn *ofconn, *next_ofconn;
932 struct ofport *ofport, *next_ofport;
939 /* Destroy fail-open and in-band early, since they touch the classifier. */
940 fail_open_destroy(p->fail_open);
943 in_band_destroy(p->in_band);
945 free(p->extra_in_band_remotes);
947 ofproto_flush_flows(p);
948 classifier_destroy(&p->cls);
950 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
951 ofconn_destroy(ofconn);
953 hmap_destroy(&p->controllers);
956 netdev_monitor_destroy(p->netdev_monitor);
957 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
958 hmap_remove(&p->ports, &ofport->hmap_node);
961 shash_destroy(&p->port_by_name);
963 switch_status_destroy(p->switch_status);
964 netflow_destroy(p->netflow);
965 ofproto_sflow_destroy(p->sflow);
967 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) {
968 ofservice_destroy(p, ofservice);
970 hmap_destroy(&p->services);
972 for (i = 0; i < p->n_snoops; i++) {
973 pvconn_close(p->snoops[i]);
977 mac_learning_destroy(p->ml);
982 free(p->serial_desc);
985 hmap_destroy(&p->ports);
991 ofproto_run(struct ofproto *p)
993 int error = ofproto_run1(p);
995 error = ofproto_run2(p, false);
1001 process_port_change(struct ofproto *ofproto, int error, char *devname)
1003 if (error == ENOBUFS) {
1004 reinit_ports(ofproto);
1005 } else if (!error) {
1006 update_port(ofproto, devname);
1011 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1012 * means that 'ofconn' is more interesting for monitoring than a lower return
1015 snoop_preference(const struct ofconn *ofconn)
1017 switch (ofconn->role) {
1018 case NX_ROLE_MASTER:
1025 /* Shouldn't happen. */
1030 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1031 * Connects this vconn to a controller. */
1033 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1035 struct ofconn *ofconn, *best;
1037 /* Pick a controller for monitoring. */
1039 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
1040 if (ofconn->type == OFCONN_PRIMARY
1041 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1047 rconn_add_monitor(best->rconn, vconn);
1049 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1055 ofproto_run1(struct ofproto *p)
1057 struct ofconn *ofconn, *next_ofconn;
1058 struct ofservice *ofservice;
1063 if (shash_is_empty(&p->port_by_name)) {
1067 for (i = 0; i < 50; i++) {
1070 error = dpif_recv(p->dpif, &buf);
1072 if (error == ENODEV) {
1073 /* Someone destroyed the datapath behind our back. The caller
1074 * better destroy us and give up, because we're just going to
1075 * spin from here on out. */
1076 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1077 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1078 dpif_name(p->dpif));
1084 handle_odp_msg(p, buf);
1087 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1088 process_port_change(p, error, devname);
1090 while ((error = netdev_monitor_poll(p->netdev_monitor,
1091 &devname)) != EAGAIN) {
1092 process_port_change(p, error, devname);
1096 if (time_msec() >= p->next_in_band_update) {
1097 update_in_band_remotes(p);
1099 in_band_run(p->in_band);
1102 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) {
1103 ofconn_run(ofconn, p);
1106 /* Fail-open maintenance. Do this after processing the ofconns since
1107 * fail-open checks the status of the controller rconn. */
1109 fail_open_run(p->fail_open);
1112 HMAP_FOR_EACH (ofservice, node, &p->services) {
1113 struct vconn *vconn;
1116 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
1118 struct rconn *rconn;
1121 rconn = rconn_create(ofservice->probe_interval, 0);
1122 name = ofconn_make_name(p, vconn_get_name(vconn));
1123 rconn_connect_unreliably(rconn, vconn, name);
1126 ofconn = ofconn_create(p, rconn, OFCONN_SERVICE);
1127 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
1128 ofservice->burst_limit);
1129 } else if (retval != EAGAIN) {
1130 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1134 for (i = 0; i < p->n_snoops; i++) {
1135 struct vconn *vconn;
1138 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1140 add_snooper(p, vconn);
1141 } else if (retval != EAGAIN) {
1142 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1146 if (time_msec() >= p->next_expiration) {
1147 int delay = ofproto_expire(p);
1148 p->next_expiration = time_msec() + delay;
1149 COVERAGE_INC(ofproto_expiration);
1153 netflow_run(p->netflow);
1156 ofproto_sflow_run(p->sflow);
1162 struct revalidate_cbdata {
1163 struct ofproto *ofproto;
1164 bool revalidate_all; /* Revalidate all exact-match rules? */
1165 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1166 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1170 ofproto_run2(struct ofproto *p, bool revalidate_all)
1172 if (p->need_revalidate || revalidate_all
1173 || !tag_set_is_empty(&p->revalidate_set)) {
1174 struct revalidate_cbdata cbdata;
1176 cbdata.revalidate_all = revalidate_all;
1177 cbdata.revalidate_subrules = p->need_revalidate;
1178 cbdata.revalidate_set = p->revalidate_set;
1179 tag_set_init(&p->revalidate_set);
1180 COVERAGE_INC(ofproto_revalidate);
1181 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1182 p->need_revalidate = false;
1189 ofproto_wait(struct ofproto *p)
1191 struct ofservice *ofservice;
1192 struct ofconn *ofconn;
1195 dpif_recv_wait(p->dpif);
1196 dpif_port_poll_wait(p->dpif);
1197 netdev_monitor_poll_wait(p->netdev_monitor);
1198 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1199 ofconn_wait(ofconn);
1202 poll_timer_wait_until(p->next_in_band_update);
1203 in_band_wait(p->in_band);
1206 fail_open_wait(p->fail_open);
1209 ofproto_sflow_wait(p->sflow);
1211 if (!tag_set_is_empty(&p->revalidate_set)) {
1212 poll_immediate_wake();
1214 if (p->need_revalidate) {
1215 /* Shouldn't happen, but if it does just go around again. */
1216 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1217 poll_immediate_wake();
1218 } else if (p->next_expiration != LLONG_MAX) {
1219 poll_timer_wait_until(p->next_expiration);
1221 HMAP_FOR_EACH (ofservice, node, &p->services) {
1222 pvconn_wait(ofservice->pvconn);
1224 for (i = 0; i < p->n_snoops; i++) {
1225 pvconn_wait(p->snoops[i]);
1230 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1232 tag_set_add(&ofproto->revalidate_set, tag);
1236 ofproto_get_revalidate_set(struct ofproto *ofproto)
1238 return &ofproto->revalidate_set;
1242 ofproto_is_alive(const struct ofproto *p)
1244 return !hmap_is_empty(&p->controllers);
1247 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1249 * This is almost the same as calling dpif_port_del() directly on the
1250 * datapath, but it also makes 'ofproto' close its open netdev for the port
1251 * (if any). This makes it possible to create a new netdev of a different
1252 * type under the same name, which otherwise the netdev library would refuse
1253 * to do because of the conflict. (The netdev would eventually get closed on
1254 * the next trip through ofproto_run(), but this interface is more direct.)
1256 * Returns 0 if successful, otherwise a positive errno. */
1258 ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port)
1260 struct ofport *ofport = get_port(ofproto, odp_port);
1261 const char *name = ofport ? (char *) ofport->opp.name : "<unknown>";
1264 error = dpif_port_del(ofproto->dpif, odp_port);
1266 VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)",
1267 dpif_name(ofproto->dpif), odp_port, name, strerror(error));
1268 } else if (ofport) {
1269 /* 'name' is ofport->opp.name and update_port() is going to destroy
1270 * 'ofport'. Just in case update_port() refers to 'name' after it
1271 * destroys 'ofport', make a copy of it around the update_port()
1273 char *devname = xstrdup(name);
1274 update_port(ofproto, devname);
1281 ofproto_send_packet(struct ofproto *p, const struct flow *flow,
1282 const union ofp_action *actions, size_t n_actions,
1283 const struct ofpbuf *packet)
1285 struct odp_actions odp_actions;
1288 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1294 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1296 dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet);
1301 ofproto_add_flow(struct ofproto *p, const struct flow *flow,
1302 uint32_t wildcards, unsigned int priority,
1303 const union ofp_action *actions, size_t n_actions,
1307 rule = rule_create(p, NULL, actions, n_actions,
1308 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1310 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1311 rule_insert(p, rule, NULL, 0);
1315 ofproto_delete_flow(struct ofproto *ofproto, const struct flow *flow,
1316 uint32_t wildcards, unsigned int priority)
1320 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1324 rule_remove(ofproto, rule);
1329 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1331 struct rule *rule = rule_from_cls_rule(rule_);
1332 struct ofproto *ofproto = ofproto_;
1334 /* Mark the flow as not installed, even though it might really be
1335 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1336 * There is no point in uninstalling it individually since we are about to
1337 * blow away all the flows with dpif_flow_flush(). */
1338 rule->installed = false;
1340 rule_remove(ofproto, rule);
1344 ofproto_flush_flows(struct ofproto *ofproto)
1346 COVERAGE_INC(ofproto_flush);
1347 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1348 dpif_flow_flush(ofproto->dpif);
1349 if (ofproto->in_band) {
1350 in_band_flushed(ofproto->in_band);
1352 if (ofproto->fail_open) {
1353 fail_open_flushed(ofproto->fail_open);
1358 reinit_ports(struct ofproto *p)
1360 struct svec devnames;
1361 struct ofport *ofport;
1362 struct odp_port *odp_ports;
1366 svec_init(&devnames);
1367 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1368 svec_add (&devnames, (char *) ofport->opp.name);
1370 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1371 for (i = 0; i < n_odp_ports; i++) {
1372 svec_add (&devnames, odp_ports[i].devname);
1376 svec_sort_unique(&devnames);
1377 for (i = 0; i < devnames.n; i++) {
1378 update_port(p, devnames.names[i]);
1380 svec_destroy(&devnames);
1383 static struct ofport *
1384 make_ofport(const struct odp_port *odp_port)
1386 struct netdev_options netdev_options;
1387 enum netdev_flags flags;
1388 struct ofport *ofport;
1389 struct netdev *netdev;
1393 memset(&netdev_options, 0, sizeof netdev_options);
1394 netdev_options.name = odp_port->devname;
1395 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1397 error = netdev_open(&netdev_options, &netdev);
1399 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1400 "cannot be opened (%s)",
1401 odp_port->devname, odp_port->port,
1402 odp_port->devname, strerror(error));
1406 ofport = xmalloc(sizeof *ofport);
1407 ofport->netdev = netdev;
1408 ofport->odp_port = odp_port->port;
1409 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1410 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1411 memcpy(ofport->opp.name, odp_port->devname,
1412 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1413 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1415 netdev_get_flags(netdev, &flags);
1416 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1418 netdev_get_carrier(netdev, &carrier);
1419 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1421 netdev_get_features(netdev,
1422 &ofport->opp.curr, &ofport->opp.advertised,
1423 &ofport->opp.supported, &ofport->opp.peer);
1428 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1430 if (get_port(p, odp_port->port)) {
1431 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1434 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1435 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1444 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1446 const struct ofp_phy_port *a = &a_->opp;
1447 const struct ofp_phy_port *b = &b_->opp;
1449 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1450 return (a->port_no == b->port_no
1451 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1452 && !strcmp((char *) a->name, (char *) b->name)
1453 && a->state == b->state
1454 && a->config == b->config
1455 && a->curr == b->curr
1456 && a->advertised == b->advertised
1457 && a->supported == b->supported
1458 && a->peer == b->peer);
1462 send_port_status(struct ofproto *p, const struct ofport *ofport,
1465 /* XXX Should limit the number of queued port status change messages. */
1466 struct ofconn *ofconn;
1467 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
1468 struct ofp_port_status *ops;
1471 if (!ofconn_receives_async_msgs(ofconn)) {
1475 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1476 ops->reason = reason;
1477 ops->desc = ofport->opp;
1478 hton_ofp_phy_port(&ops->desc);
1479 queue_tx(b, ofconn, NULL);
1484 ofport_install(struct ofproto *p, struct ofport *ofport)
1486 const char *netdev_name = (const char *) ofport->opp.name;
1488 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1489 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0));
1490 shash_add(&p->port_by_name, netdev_name, ofport);
1492 ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name);
1497 ofport_remove(struct ofproto *p, struct ofport *ofport)
1499 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1500 hmap_remove(&p->ports, &ofport->hmap_node);
1501 shash_delete(&p->port_by_name,
1502 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1504 ofproto_sflow_del_port(p->sflow, ofport->odp_port);
1509 ofport_free(struct ofport *ofport)
1512 netdev_close(ofport->netdev);
1517 static struct ofport *
1518 get_port(const struct ofproto *ofproto, uint16_t odp_port)
1520 struct ofport *port;
1522 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1523 hash_int(odp_port, 0), &ofproto->ports) {
1524 if (port->odp_port == odp_port) {
1532 update_port(struct ofproto *p, const char *devname)
1534 struct odp_port odp_port;
1535 struct ofport *old_ofport;
1536 struct ofport *new_ofport;
1539 COVERAGE_INC(ofproto_update_port);
1541 /* Query the datapath for port information. */
1542 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1544 /* Find the old ofport. */
1545 old_ofport = shash_find_data(&p->port_by_name, devname);
1548 /* There's no port named 'devname' but there might be a port with
1549 * the same port number. This could happen if a port is deleted
1550 * and then a new one added in its place very quickly, or if a port
1551 * is renamed. In the former case we want to send an OFPPR_DELETE
1552 * and an OFPPR_ADD, and in the latter case we want to send a
1553 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1554 * the old port's ifindex against the new port, or perhaps less
1555 * reliably but more portably by comparing the old port's MAC
1556 * against the new port's MAC. However, this code isn't that smart
1557 * and always sends an OFPPR_MODIFY (XXX). */
1558 old_ofport = get_port(p, odp_port.port);
1560 } else if (error != ENOENT && error != ENODEV) {
1561 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1562 "%s", strerror(error));
1566 /* Create a new ofport. */
1567 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1569 /* Eliminate a few pathological cases. */
1570 if (!old_ofport && !new_ofport) {
1572 } else if (old_ofport && new_ofport) {
1573 /* Most of the 'config' bits are OpenFlow soft state, but
1574 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1575 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1576 * leaves the other bits 0.) */
1577 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1579 if (ofport_equal(old_ofport, new_ofport)) {
1580 /* False alarm--no change. */
1581 ofport_free(new_ofport);
1586 /* Now deal with the normal cases. */
1588 ofport_remove(p, old_ofport);
1591 ofport_install(p, new_ofport);
1593 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1594 (!old_ofport ? OFPPR_ADD
1595 : !new_ofport ? OFPPR_DELETE
1597 ofport_free(old_ofport);
1601 init_ports(struct ofproto *p)
1603 struct odp_port *ports;
1608 error = dpif_port_list(p->dpif, &ports, &n_ports);
1613 for (i = 0; i < n_ports; i++) {
1614 const struct odp_port *odp_port = &ports[i];
1615 if (!ofport_conflicts(p, odp_port)) {
1616 struct ofport *ofport = make_ofport(odp_port);
1618 ofport_install(p, ofport);
1626 static struct ofconn *
1627 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1629 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1630 ofconn->ofproto = p;
1631 list_push_back(&p->all_conns, &ofconn->node);
1632 ofconn->rconn = rconn;
1633 ofconn->type = type;
1634 ofconn->role = NX_ROLE_OTHER;
1635 ofconn->packet_in_counter = rconn_packet_counter_create ();
1636 ofconn->pktbuf = NULL;
1637 ofconn->miss_send_len = 0;
1638 ofconn->reply_counter = rconn_packet_counter_create ();
1643 ofconn_destroy(struct ofconn *ofconn)
1645 if (ofconn->type == OFCONN_PRIMARY) {
1646 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1648 discovery_destroy(ofconn->discovery);
1650 list_remove(&ofconn->node);
1651 switch_status_unregister(ofconn->ss);
1652 rconn_destroy(ofconn->rconn);
1653 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1654 rconn_packet_counter_destroy(ofconn->reply_counter);
1655 pktbuf_destroy(ofconn->pktbuf);
1660 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1665 if (ofconn->discovery) {
1666 char *controller_name;
1667 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1668 discovery_question_connectivity(ofconn->discovery);
1670 if (discovery_run(ofconn->discovery, &controller_name)) {
1671 if (controller_name) {
1672 char *ofconn_name = ofconn_make_name(p, controller_name);
1673 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1676 rconn_disconnect(ofconn->rconn);
1681 for (i = 0; i < N_SCHEDULERS; i++) {
1682 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1685 rconn_run(ofconn->rconn);
1687 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1688 /* Limit the number of iterations to prevent other tasks from
1690 for (iteration = 0; iteration < 50; iteration++) {
1691 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1696 fail_open_maybe_recover(p->fail_open);
1698 handle_openflow(ofconn, p, of_msg);
1699 ofpbuf_delete(of_msg);
1703 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1704 ofconn_destroy(ofconn);
1709 ofconn_wait(struct ofconn *ofconn)
1713 if (ofconn->discovery) {
1714 discovery_wait(ofconn->discovery);
1716 for (i = 0; i < N_SCHEDULERS; i++) {
1717 pinsched_wait(ofconn->schedulers[i]);
1719 rconn_run_wait(ofconn->rconn);
1720 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1721 rconn_recv_wait(ofconn->rconn);
1723 COVERAGE_INC(ofproto_ofconn_stuck);
1727 /* Returns true if 'ofconn' should receive asynchronous messages. */
1729 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1731 if (ofconn->type == OFCONN_PRIMARY) {
1732 /* Primary controllers always get asynchronous messages unless they
1733 * have configured themselves as "slaves". */
1734 return ofconn->role != NX_ROLE_SLAVE;
1736 /* Service connections don't get asynchronous messages unless they have
1737 * explicitly asked for them by setting a nonzero miss send length. */
1738 return ofconn->miss_send_len > 0;
1742 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1743 * and 'target', suitable for use in log messages for identifying the
1746 * The name is dynamically allocated. The caller should free it (with free())
1747 * when it is no longer needed. */
1749 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1751 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1755 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1759 for (i = 0; i < N_SCHEDULERS; i++) {
1760 struct pinsched **s = &ofconn->schedulers[i];
1764 *s = pinsched_create(rate, burst,
1765 ofconn->ofproto->switch_status);
1767 pinsched_set_limits(*s, rate, burst);
1770 pinsched_destroy(*s);
1777 ofservice_reconfigure(struct ofservice *ofservice,
1778 const struct ofproto_controller *c)
1780 ofservice->probe_interval = c->probe_interval;
1781 ofservice->rate_limit = c->rate_limit;
1782 ofservice->burst_limit = c->burst_limit;
1785 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1786 * positive errno value. */
1788 ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c)
1790 struct ofservice *ofservice;
1791 struct pvconn *pvconn;
1794 error = pvconn_open(c->target, &pvconn);
1799 ofservice = xzalloc(sizeof *ofservice);
1800 hmap_insert(&ofproto->services, &ofservice->node,
1801 hash_string(c->target, 0));
1802 ofservice->pvconn = pvconn;
1804 ofservice_reconfigure(ofservice, c);
1810 ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice)
1812 hmap_remove(&ofproto->services, &ofservice->node);
1813 pvconn_close(ofservice->pvconn);
1817 /* Finds and returns the ofservice within 'ofproto' that has the given
1818 * 'target', or a null pointer if none exists. */
1819 static struct ofservice *
1820 ofservice_lookup(struct ofproto *ofproto, const char *target)
1822 struct ofservice *ofservice;
1824 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1825 &ofproto->services) {
1826 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1833 /* Caller is responsible for initializing the 'cr' member of the returned
1835 static struct rule *
1836 rule_create(struct ofproto *ofproto, struct rule *super,
1837 const union ofp_action *actions, size_t n_actions,
1838 uint16_t idle_timeout, uint16_t hard_timeout,
1839 uint64_t flow_cookie, bool send_flow_removed)
1841 struct rule *rule = xzalloc(sizeof *rule);
1842 rule->idle_timeout = idle_timeout;
1843 rule->hard_timeout = hard_timeout;
1844 rule->flow_cookie = flow_cookie;
1845 rule->used = rule->created = time_msec();
1846 rule->send_flow_removed = send_flow_removed;
1847 rule->super = super;
1849 list_push_back(&super->list, &rule->list);
1851 list_init(&rule->list);
1853 if (n_actions > 0) {
1854 rule->n_actions = n_actions;
1855 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1857 netflow_flow_clear(&rule->nf_flow);
1858 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1863 static struct rule *
1864 rule_from_cls_rule(const struct cls_rule *cls_rule)
1866 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1870 rule_free(struct rule *rule)
1872 free(rule->actions);
1873 free(rule->odp_actions);
1877 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1878 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1879 * through all of its subrules and revalidates them, destroying any that no
1880 * longer has a super-rule (which is probably all of them).
1882 * Before calling this function, the caller must make have removed 'rule' from
1883 * the classifier. If 'rule' is an exact-match rule, the caller is also
1884 * responsible for ensuring that it has been uninstalled from the datapath. */
1886 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1889 struct rule *subrule, *next;
1890 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
1891 revalidate_rule(ofproto, subrule);
1894 list_remove(&rule->list);
1900 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1902 const union ofp_action *oa;
1903 struct actions_iterator i;
1905 if (out_port == htons(OFPP_NONE)) {
1908 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1909 oa = actions_next(&i)) {
1910 if (action_outputs_to_port(oa, out_port)) {
1917 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1918 * 'packet', which arrived on 'in_port'.
1920 * Takes ownership of 'packet'. */
1922 execute_odp_actions(struct ofproto *ofproto, uint16_t in_port,
1923 const union odp_action *actions, size_t n_actions,
1924 struct ofpbuf *packet)
1926 if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) {
1927 /* As an optimization, avoid a round-trip from userspace to kernel to
1928 * userspace. This also avoids possibly filling up kernel packet
1929 * buffers along the way. */
1930 struct odp_msg *msg;
1932 msg = ofpbuf_push_uninit(packet, sizeof *msg);
1933 msg->type = _ODPL_ACTION_NR;
1934 msg->length = sizeof(struct odp_msg) + packet->size;
1935 msg->port = in_port;
1937 msg->arg = actions[0].controller.arg;
1939 send_packet_in(ofproto, packet);
1945 error = dpif_execute(ofproto->dpif, actions, n_actions, packet);
1946 ofpbuf_delete(packet);
1951 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1952 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1953 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1955 * The flow that 'packet' actually contains does not need to actually match
1956 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1957 * the packet and byte counters for 'rule' will be credited for the packet sent
1958 * out whether or not the packet actually matches 'rule'.
1960 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1961 * the caller must already have accurately composed ODP actions for it given
1962 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1963 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1964 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1965 * actions and apply them to 'packet'.
1967 * Takes ownership of 'packet'. */
1969 rule_execute(struct ofproto *ofproto, struct rule *rule,
1970 struct ofpbuf *packet, const struct flow *flow)
1972 const union odp_action *actions;
1973 struct odp_flow_stats stats;
1975 struct odp_actions a;
1977 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1979 /* Grab or compose the ODP actions.
1981 * The special case for an exact-match 'rule' where 'flow' is not the
1982 * rule's flow is important to avoid, e.g., sending a packet out its input
1983 * port simply because the ODP actions were composed for the wrong
1985 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1986 struct rule *super = rule->super ? rule->super : rule;
1987 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1988 packet, &a, NULL, 0, NULL)) {
1989 ofpbuf_delete(packet);
1992 actions = a.actions;
1993 n_actions = a.n_actions;
1995 actions = rule->odp_actions;
1996 n_actions = rule->n_odp_actions;
1999 /* Execute the ODP actions. */
2000 flow_extract_stats(flow, packet, &stats);
2001 if (execute_odp_actions(ofproto, flow->in_port,
2002 actions, n_actions, packet)) {
2003 update_stats(ofproto, rule, &stats);
2004 rule->used = time_msec();
2005 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
2009 /* Inserts 'rule' into 'p''s flow table.
2011 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2012 * actions on it and credits the statistics for sending the packet to 'rule'.
2013 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2016 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
2019 struct rule *displaced_rule;
2021 /* Insert the rule in the classifier. */
2022 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
2023 if (!rule->cr.wc.wildcards) {
2024 rule_make_actions(p, rule, packet);
2027 /* Send the packet and credit it to the rule. */
2030 flow_extract(packet, 0, in_port, &flow);
2031 rule_execute(p, rule, packet, &flow);
2034 /* Install the rule in the datapath only after sending the packet, to
2035 * avoid packet reordering. */
2036 if (rule->cr.wc.wildcards) {
2037 COVERAGE_INC(ofproto_add_wc_flow);
2038 p->need_revalidate = true;
2040 rule_install(p, rule, displaced_rule);
2043 /* Free the rule that was displaced, if any. */
2044 if (displaced_rule) {
2045 rule_destroy(p, displaced_rule);
2049 static struct rule *
2050 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
2051 const struct flow *flow)
2053 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
2054 rule->idle_timeout, rule->hard_timeout,
2056 COVERAGE_INC(ofproto_subrule_create);
2057 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
2058 : rule->cr.priority), &subrule->cr);
2059 classifier_insert_exact(&ofproto->cls, &subrule->cr);
2064 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2066 * - If 'rule' was installed in the datapath, uninstalls it and updates
2067 * 'rule''s statistics (or its super-rule's statistics, if it is a
2068 * subrule), via rule_uninstall().
2070 * - Removes 'rule' from the classifier.
2072 * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
2073 * uninstalls and destroys) its subrules, via rule_destroy().
2076 rule_remove(struct ofproto *ofproto, struct rule *rule)
2078 if (rule->cr.wc.wildcards) {
2079 COVERAGE_INC(ofproto_del_wc_flow);
2080 ofproto->need_revalidate = true;
2082 rule_uninstall(ofproto, rule);
2084 classifier_remove(&ofproto->cls, &rule->cr);
2085 rule_destroy(ofproto, rule);
2088 /* Returns true if the actions changed, false otherwise. */
2090 rule_make_actions(struct ofproto *p, struct rule *rule,
2091 const struct ofpbuf *packet)
2093 const struct rule *super;
2094 struct odp_actions a;
2097 assert(!rule->cr.wc.wildcards);
2099 super = rule->super ? rule->super : rule;
2101 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
2102 packet, &a, &rule->tags, &rule->may_install,
2103 &rule->nf_flow.output_iface);
2105 actions_len = a.n_actions * sizeof *a.actions;
2106 if (rule->n_odp_actions != a.n_actions
2107 || memcmp(rule->odp_actions, a.actions, actions_len)) {
2108 COVERAGE_INC(ofproto_odp_unchanged);
2109 free(rule->odp_actions);
2110 rule->n_odp_actions = a.n_actions;
2111 rule->odp_actions = xmemdup(a.actions, actions_len);
2119 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
2120 struct odp_flow_put *put)
2122 memset(&put->flow.stats, 0, sizeof put->flow.stats);
2123 odp_flow_key_from_flow(&put->flow.key, &rule->cr.flow);
2124 put->flow.actions = rule->odp_actions;
2125 put->flow.n_actions = rule->n_odp_actions;
2126 put->flow.flags = 0;
2128 return dpif_flow_put(ofproto->dpif, put);
2132 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
2134 assert(!rule->cr.wc.wildcards);
2136 if (rule->may_install) {
2137 struct odp_flow_put put;
2138 if (!do_put_flow(p, rule,
2139 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
2141 rule->installed = true;
2142 if (displaced_rule) {
2143 update_stats(p, displaced_rule, &put.flow.stats);
2144 rule_post_uninstall(p, displaced_rule);
2147 } else if (displaced_rule) {
2148 rule_uninstall(p, displaced_rule);
2153 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
2155 if (rule->installed) {
2156 struct odp_flow_put put;
2157 COVERAGE_INC(ofproto_dp_missed);
2158 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
2160 rule_install(ofproto, rule, NULL);
2165 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2167 bool actions_changed;
2168 uint16_t new_out_iface, old_out_iface;
2170 old_out_iface = rule->nf_flow.output_iface;
2171 actions_changed = rule_make_actions(ofproto, rule, NULL);
2173 if (rule->may_install) {
2174 if (rule->installed) {
2175 if (actions_changed) {
2176 struct odp_flow_put put;
2177 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2178 | ODPPF_ZERO_STATS, &put);
2179 update_stats(ofproto, rule, &put.flow.stats);
2181 /* Temporarily set the old output iface so that NetFlow
2182 * messages have the correct output interface for the old
2184 new_out_iface = rule->nf_flow.output_iface;
2185 rule->nf_flow.output_iface = old_out_iface;
2186 rule_post_uninstall(ofproto, rule);
2187 rule->nf_flow.output_iface = new_out_iface;
2190 rule_install(ofproto, rule, NULL);
2193 rule_uninstall(ofproto, rule);
2198 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2200 uint64_t total_bytes = rule->byte_count + extra_bytes;
2202 if (ofproto->ofhooks->account_flow_cb
2203 && total_bytes > rule->accounted_bytes)
2205 ofproto->ofhooks->account_flow_cb(
2206 &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions,
2207 total_bytes - rule->accounted_bytes, ofproto->aux);
2208 rule->accounted_bytes = total_bytes;
2212 /* 'rule' must be an exact-match rule in 'p'.
2214 * If 'rule' is installed in the datapath, uninstalls it and updates's
2215 * statistics. If 'rule' is a subrule, the statistics that are updated are
2216 * actually its super-rule's statistics; otherwise 'rule''s own statistics are
2219 * If 'rule' is not installed, this function has no effect. */
2221 rule_uninstall(struct ofproto *p, struct rule *rule)
2223 assert(!rule->cr.wc.wildcards);
2224 if (rule->installed) {
2225 struct odp_flow odp_flow;
2227 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
2228 odp_flow.actions = NULL;
2229 odp_flow.n_actions = 0;
2231 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2232 update_stats(p, rule, &odp_flow.stats);
2234 rule->installed = false;
2236 rule_post_uninstall(p, rule);
2241 is_controller_rule(struct rule *rule)
2243 /* If the only action is send to the controller then don't report
2244 * NetFlow expiration messages since it is just part of the control
2245 * logic for the network and not real traffic. */
2249 && rule->super->n_actions == 1
2250 && action_outputs_to_port(&rule->super->actions[0],
2251 htons(OFPP_CONTROLLER)));
2255 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2257 struct rule *super = rule->super;
2259 rule_account(ofproto, rule, 0);
2261 if (ofproto->netflow && !is_controller_rule(rule)) {
2262 struct ofexpired expired;
2263 expired.flow = rule->cr.flow;
2264 expired.packet_count = rule->packet_count;
2265 expired.byte_count = rule->byte_count;
2266 expired.used = rule->used;
2267 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2270 super->packet_count += rule->packet_count;
2271 super->byte_count += rule->byte_count;
2273 /* Reset counters to prevent double counting if the rule ever gets
2275 rule->packet_count = 0;
2276 rule->byte_count = 0;
2277 rule->accounted_bytes = 0;
2279 netflow_flow_clear(&rule->nf_flow);
2284 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2285 struct rconn_packet_counter *counter)
2287 update_openflow_length(msg);
2288 if (rconn_send(ofconn->rconn, msg, counter)) {
2294 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2295 int error, const void *data, size_t len)
2298 struct ofp_error_msg *oem;
2300 if (!(error >> 16)) {
2301 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2306 COVERAGE_INC(ofproto_error);
2307 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2308 oh ? oh->xid : 0, &buf);
2309 oem->type = htons((unsigned int) error >> 16);
2310 oem->code = htons(error & 0xffff);
2311 memcpy(oem->data, data, len);
2312 queue_tx(buf, ofconn, ofconn->reply_counter);
2316 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2319 size_t oh_length = ntohs(oh->length);
2320 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2324 hton_ofp_phy_port(struct ofp_phy_port *opp)
2326 opp->port_no = htons(opp->port_no);
2327 opp->config = htonl(opp->config);
2328 opp->state = htonl(opp->state);
2329 opp->curr = htonl(opp->curr);
2330 opp->advertised = htonl(opp->advertised);
2331 opp->supported = htonl(opp->supported);
2332 opp->peer = htonl(opp->peer);
2336 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2338 struct ofp_header *rq = oh;
2339 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2344 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2345 struct ofp_header *oh)
2347 struct ofp_switch_features *osf;
2349 struct ofport *port;
2351 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2352 osf->datapath_id = htonll(p->datapath_id);
2353 osf->n_buffers = htonl(pktbuf_capacity());
2355 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2356 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2357 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2358 (1u << OFPAT_SET_VLAN_VID) |
2359 (1u << OFPAT_SET_VLAN_PCP) |
2360 (1u << OFPAT_STRIP_VLAN) |
2361 (1u << OFPAT_SET_DL_SRC) |
2362 (1u << OFPAT_SET_DL_DST) |
2363 (1u << OFPAT_SET_NW_SRC) |
2364 (1u << OFPAT_SET_NW_DST) |
2365 (1u << OFPAT_SET_NW_TOS) |
2366 (1u << OFPAT_SET_TP_SRC) |
2367 (1u << OFPAT_SET_TP_DST) |
2368 (1u << OFPAT_ENQUEUE));
2370 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2371 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2374 queue_tx(buf, ofconn, ofconn->reply_counter);
2379 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2380 struct ofp_header *oh)
2383 struct ofp_switch_config *osc;
2387 /* Figure out flags. */
2388 dpif_get_drop_frags(p->dpif, &drop_frags);
2389 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2392 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2393 osc->flags = htons(flags);
2394 osc->miss_send_len = htons(ofconn->miss_send_len);
2395 queue_tx(buf, ofconn, ofconn->reply_counter);
2401 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2402 struct ofp_switch_config *osc)
2407 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2411 flags = ntohs(osc->flags);
2413 if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) {
2414 switch (flags & OFPC_FRAG_MASK) {
2415 case OFPC_FRAG_NORMAL:
2416 dpif_set_drop_frags(p->dpif, false);
2418 case OFPC_FRAG_DROP:
2419 dpif_set_drop_frags(p->dpif, true);
2422 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2428 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2434 add_controller_action(struct odp_actions *actions, uint16_t max_len)
2436 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2437 a->controller.arg = max_len;
2440 struct action_xlate_ctx {
2442 struct flow flow; /* Flow to which these actions correspond. */
2443 int recurse; /* Recursion level, via xlate_table_action. */
2444 struct ofproto *ofproto;
2445 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2446 * null pointer if we are revalidating
2447 * without a packet to refer to. */
2450 struct odp_actions *out; /* Datapath actions. */
2451 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2452 bool may_set_up_flow; /* True ordinarily; false if the actions must
2453 * be reassessed for every packet. */
2454 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2457 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2458 * flow translation. */
2459 #define MAX_RESUBMIT_RECURSION 8
2461 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2462 struct action_xlate_ctx *ctx);
2465 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2467 const struct ofport *ofport = get_port(ctx->ofproto, port);
2470 if (ofport->opp.config & OFPPC_NO_FWD) {
2471 /* Forwarding disabled on port. */
2476 * We don't have an ofport record for this port, but it doesn't hurt to
2477 * allow forwarding to it anyhow. Maybe such a port will appear later
2478 * and we're pre-populating the flow table.
2482 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2483 ctx->nf_output_iface = port;
2486 static struct rule *
2487 lookup_valid_rule(struct ofproto *ofproto, const struct flow *flow)
2490 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2492 /* The rule we found might not be valid, since we could be in need of
2493 * revalidation. If it is not valid, don't return it. */
2496 && ofproto->need_revalidate
2497 && !revalidate_rule(ofproto, rule)) {
2498 COVERAGE_INC(ofproto_invalidated);
2506 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2508 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
2509 uint16_t old_in_port;
2512 /* Look up a flow with 'in_port' as the input port. Then restore the
2513 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2514 * have surprising behavior). */
2515 old_in_port = ctx->flow.in_port;
2516 ctx->flow.in_port = in_port;
2517 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2518 ctx->flow.in_port = old_in_port;
2526 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2530 struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2532 VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
2533 MAX_RESUBMIT_RECURSION);
2538 flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask,
2539 uint16_t *nf_output_iface, struct odp_actions *actions)
2541 struct ofport *ofport;
2543 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) {
2544 uint16_t odp_port = ofport->odp_port;
2545 if (odp_port != odp_in_port && !(ofport->opp.config & mask)) {
2546 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port;
2549 *nf_output_iface = NF_OUT_FLOOD;
2553 xlate_output_action__(struct action_xlate_ctx *ctx,
2554 uint16_t port, uint16_t max_len)
2557 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2559 ctx->nf_output_iface = NF_OUT_DROP;
2563 add_output_action(ctx, ctx->flow.in_port);
2566 xlate_table_action(ctx, ctx->flow.in_port);
2569 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2570 ctx->out, ctx->tags,
2571 &ctx->nf_output_iface,
2572 ctx->ofproto->aux)) {
2573 COVERAGE_INC(ofproto_uninstallable);
2574 ctx->may_set_up_flow = false;
2578 flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD,
2579 &ctx->nf_output_iface, ctx->out);
2581 flood_packets(ctx->ofproto, ctx->flow.in_port, 0,
2582 &ctx->nf_output_iface, ctx->out);
2584 case OFPP_CONTROLLER:
2585 add_controller_action(ctx->out, max_len);
2588 add_output_action(ctx, ODPP_LOCAL);
2591 odp_port = ofp_port_to_odp_port(port);
2592 if (odp_port != ctx->flow.in_port) {
2593 add_output_action(ctx, odp_port);
2598 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2599 ctx->nf_output_iface = NF_OUT_FLOOD;
2600 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2601 ctx->nf_output_iface = prev_nf_output_iface;
2602 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2603 ctx->nf_output_iface != NF_OUT_FLOOD) {
2604 ctx->nf_output_iface = NF_OUT_MULTI;
2609 xlate_output_action(struct action_xlate_ctx *ctx,
2610 const struct ofp_action_output *oao)
2612 xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
2615 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2616 * optimization, because we're going to add another action that sets the
2617 * priority immediately after, or because there are no actions following the
2620 remove_pop_action(struct action_xlate_ctx *ctx)
2622 size_t n = ctx->out->n_actions;
2623 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2624 ctx->out->n_actions--;
2629 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2630 const struct ofp_action_enqueue *oae)
2632 uint16_t ofp_port, odp_port;
2636 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
2639 /* Fall back to ordinary output action. */
2640 xlate_output_action__(ctx, ntohs(oae->port), 0);
2644 /* Figure out ODP output port. */
2645 ofp_port = ntohs(oae->port);
2646 if (ofp_port != OFPP_IN_PORT) {
2647 odp_port = ofp_port_to_odp_port(ofp_port);
2649 odp_port = ctx->flow.in_port;
2652 /* Add ODP actions. */
2653 remove_pop_action(ctx);
2654 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2656 add_output_action(ctx, odp_port);
2657 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2659 /* Update NetFlow output port. */
2660 if (ctx->nf_output_iface == NF_OUT_DROP) {
2661 ctx->nf_output_iface = odp_port;
2662 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2663 ctx->nf_output_iface = NF_OUT_MULTI;
2668 xlate_set_queue_action(struct action_xlate_ctx *ctx,
2669 const struct nx_action_set_queue *nasq)
2674 error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
2677 /* Couldn't translate queue to a priority, so ignore. A warning
2678 * has already been logged. */
2682 remove_pop_action(ctx);
2683 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2688 xlate_nicira_action(struct action_xlate_ctx *ctx,
2689 const struct nx_action_header *nah)
2691 const struct nx_action_resubmit *nar;
2692 const struct nx_action_set_tunnel *nast;
2693 const struct nx_action_set_queue *nasq;
2694 union odp_action *oa;
2695 int subtype = ntohs(nah->subtype);
2697 assert(nah->vendor == htonl(NX_VENDOR_ID));
2699 case NXAST_RESUBMIT:
2700 nar = (const struct nx_action_resubmit *) nah;
2701 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2704 case NXAST_SET_TUNNEL:
2705 nast = (const struct nx_action_set_tunnel *) nah;
2706 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2707 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2710 case NXAST_DROP_SPOOFED_ARP:
2711 if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
2712 odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP);
2716 case NXAST_SET_QUEUE:
2717 nasq = (const struct nx_action_set_queue *) nah;
2718 xlate_set_queue_action(ctx, nasq);
2721 case NXAST_POP_QUEUE:
2722 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2725 /* If you add a new action here that modifies flow data, don't forget to
2726 * update the flow key in ctx->flow at the same time. */
2729 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2735 do_xlate_actions(const union ofp_action *in, size_t n_in,
2736 struct action_xlate_ctx *ctx)
2738 struct actions_iterator iter;
2739 const union ofp_action *ia;
2740 const struct ofport *port;
2742 port = get_port(ctx->ofproto, ctx->flow.in_port);
2743 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2744 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
2745 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2746 /* Drop this flow. */
2750 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2751 uint16_t type = ntohs(ia->type);
2752 union odp_action *oa;
2756 xlate_output_action(ctx, &ia->output);
2759 case OFPAT_SET_VLAN_VID:
2760 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2761 oa->dl_tci.tci = ia->vlan_vid.vlan_vid & htons(VLAN_VID_MASK);
2762 oa->dl_tci.mask = htons(VLAN_VID_MASK);
2765 case OFPAT_SET_VLAN_PCP:
2766 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI);
2767 oa->dl_tci.tci = htons((ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT)
2769 oa->dl_tci.mask = htons(VLAN_PCP_MASK);
2772 case OFPAT_STRIP_VLAN:
2773 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2774 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2775 ctx->flow.dl_vlan_pcp = 0;
2778 case OFPAT_SET_DL_SRC:
2779 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2780 memcpy(oa->dl_addr.dl_addr,
2781 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2782 memcpy(ctx->flow.dl_src,
2783 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2786 case OFPAT_SET_DL_DST:
2787 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2788 memcpy(oa->dl_addr.dl_addr,
2789 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2790 memcpy(ctx->flow.dl_dst,
2791 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2794 case OFPAT_SET_NW_SRC:
2795 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2796 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2799 case OFPAT_SET_NW_DST:
2800 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2801 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2804 case OFPAT_SET_NW_TOS:
2805 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2806 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2809 case OFPAT_SET_TP_SRC:
2810 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2811 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2814 case OFPAT_SET_TP_DST:
2815 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2816 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2820 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2824 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2828 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2835 xlate_actions(const union ofp_action *in, size_t n_in,
2836 const struct flow *flow, struct ofproto *ofproto,
2837 const struct ofpbuf *packet,
2838 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2839 uint16_t *nf_output_iface)
2841 tag_type no_tags = 0;
2842 struct action_xlate_ctx ctx;
2843 COVERAGE_INC(ofproto_ofp2odp);
2844 odp_actions_init(out);
2847 ctx.ofproto = ofproto;
2848 ctx.packet = packet;
2850 ctx.tags = tags ? tags : &no_tags;
2851 ctx.may_set_up_flow = true;
2852 ctx.nf_output_iface = NF_OUT_DROP;
2853 do_xlate_actions(in, n_in, &ctx);
2854 remove_pop_action(&ctx);
2856 /* Check with in-band control to see if we're allowed to set up this
2858 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2859 ctx.may_set_up_flow = false;
2862 if (may_set_up_flow) {
2863 *may_set_up_flow = ctx.may_set_up_flow;
2865 if (nf_output_iface) {
2866 *nf_output_iface = ctx.nf_output_iface;
2868 if (odp_actions_overflow(out)) {
2869 COVERAGE_INC(odp_overflow);
2870 odp_actions_init(out);
2871 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2876 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2877 * error message code (composed with ofp_mkerr()) for the caller to propagate
2878 * upward. Otherwise, returns 0.
2880 * 'oh' is used to make log messages more informative. */
2882 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2884 if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) {
2885 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2888 type_name = ofp_message_type_to_string(oh->type);
2889 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2893 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2900 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2901 struct ofp_header *oh)
2903 struct ofp_packet_out *opo;
2904 struct ofpbuf payload, *buffer;
2905 struct odp_actions actions;
2911 error = reject_slave_controller(ofconn, oh);
2916 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2920 opo = (struct ofp_packet_out *) oh;
2922 COVERAGE_INC(ofproto_packet_out);
2923 if (opo->buffer_id != htonl(UINT32_MAX)) {
2924 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2926 if (error || !buffer) {
2934 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2935 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2936 &flow, p, &payload, &actions, NULL, NULL, NULL);
2941 dpif_execute(p->dpif, actions.actions, actions.n_actions, &payload);
2942 ofpbuf_delete(buffer);
2948 update_port_config(struct ofproto *p, struct ofport *port,
2949 uint32_t config, uint32_t mask)
2951 mask &= config ^ port->opp.config;
2952 if (mask & OFPPC_PORT_DOWN) {
2953 if (config & OFPPC_PORT_DOWN) {
2954 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2956 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2959 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2960 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2961 if (mask & REVALIDATE_BITS) {
2962 COVERAGE_INC(ofproto_costly_flags);
2963 port->opp.config ^= mask & REVALIDATE_BITS;
2964 p->need_revalidate = true;
2966 #undef REVALIDATE_BITS
2967 if (mask & OFPPC_NO_PACKET_IN) {
2968 port->opp.config ^= OFPPC_NO_PACKET_IN;
2973 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2974 struct ofp_header *oh)
2976 const struct ofp_port_mod *opm;
2977 struct ofport *port;
2980 error = reject_slave_controller(ofconn, oh);
2984 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2988 opm = (struct ofp_port_mod *) oh;
2990 port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no)));
2992 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2993 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2994 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2996 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2997 if (opm->advertise) {
2998 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
3004 static struct ofpbuf *
3005 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
3007 struct ofp_stats_reply *osr;
3010 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
3011 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
3013 osr->flags = htons(0);
3017 static struct ofpbuf *
3018 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
3020 return make_stats_reply(request->header.xid, request->type, body_len);
3024 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
3026 struct ofpbuf *msg = *msgp;
3027 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
3028 if (nbytes + msg->size > UINT16_MAX) {
3029 struct ofp_stats_reply *reply = msg->data;
3030 reply->flags = htons(OFPSF_REPLY_MORE);
3031 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
3032 queue_tx(msg, ofconn, ofconn->reply_counter);
3034 return ofpbuf_put_uninit(*msgp, nbytes);
3038 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
3039 struct ofp_stats_request *request)
3041 struct ofp_desc_stats *ods;
3044 msg = start_stats_reply(request, sizeof *ods);
3045 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
3046 memset(ods, 0, sizeof *ods);
3047 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
3048 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
3049 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
3050 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
3051 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
3052 queue_tx(msg, ofconn, ofconn->reply_counter);
3058 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
3059 struct ofp_stats_request *request)
3061 struct ofp_table_stats *ots;
3063 struct odp_stats dpstats;
3064 int n_exact, n_subrules, n_wild;
3067 msg = start_stats_reply(request, sizeof *ots * 2);
3069 /* Count rules of various kinds. */
3071 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) {
3076 n_exact = classifier_count_exact(&p->cls) - n_subrules;
3077 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
3080 dpif_get_dp_stats(p->dpif, &dpstats);
3081 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3082 memset(ots, 0, sizeof *ots);
3083 ots->table_id = TABLEID_HASH;
3084 strcpy(ots->name, "hash");
3085 ots->wildcards = htonl(0);
3086 ots->max_entries = htonl(dpstats.max_capacity);
3087 ots->active_count = htonl(n_exact);
3088 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
3090 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
3092 /* Classifier table. */
3093 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
3094 memset(ots, 0, sizeof *ots);
3095 ots->table_id = TABLEID_CLASSIFIER;
3096 strcpy(ots->name, "classifier");
3097 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
3099 ots->max_entries = htonl(65536);
3100 ots->active_count = htonl(n_wild);
3101 ots->lookup_count = htonll(0); /* XXX */
3102 ots->matched_count = htonll(0); /* XXX */
3104 queue_tx(msg, ofconn, ofconn->reply_counter);
3109 append_port_stat(struct ofport *port, struct ofconn *ofconn,
3110 struct ofpbuf **msgp)
3112 struct netdev_stats stats;
3113 struct ofp_port_stats *ops;
3115 /* Intentionally ignore return value, since errors will set
3116 * 'stats' to all-1s, which is correct for OpenFlow, and
3117 * netdev_get_stats() will log errors. */
3118 netdev_get_stats(port->netdev, &stats);
3120 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
3121 ops->port_no = htons(port->opp.port_no);
3122 memset(ops->pad, 0, sizeof ops->pad);
3123 ops->rx_packets = htonll(stats.rx_packets);
3124 ops->tx_packets = htonll(stats.tx_packets);
3125 ops->rx_bytes = htonll(stats.rx_bytes);
3126 ops->tx_bytes = htonll(stats.tx_bytes);
3127 ops->rx_dropped = htonll(stats.rx_dropped);
3128 ops->tx_dropped = htonll(stats.tx_dropped);
3129 ops->rx_errors = htonll(stats.rx_errors);
3130 ops->tx_errors = htonll(stats.tx_errors);
3131 ops->rx_frame_err = htonll(stats.rx_frame_errors);
3132 ops->rx_over_err = htonll(stats.rx_over_errors);
3133 ops->rx_crc_err = htonll(stats.rx_crc_errors);
3134 ops->collisions = htonll(stats.collisions);
3138 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
3139 struct ofp_stats_request *osr,
3142 struct ofp_port_stats_request *psr;
3143 struct ofp_port_stats *ops;
3145 struct ofport *port;
3147 if (arg_size != sizeof *psr) {
3148 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3150 psr = (struct ofp_port_stats_request *) osr->body;
3152 msg = start_stats_reply(osr, sizeof *ops * 16);
3153 if (psr->port_no != htons(OFPP_NONE)) {
3154 port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no)));
3156 append_port_stat(port, ofconn, &msg);
3159 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3160 append_port_stat(port, ofconn, &msg);
3164 queue_tx(msg, ofconn, ofconn->reply_counter);
3168 struct flow_stats_cbdata {
3169 struct ofproto *ofproto;
3170 struct ofconn *ofconn;
3175 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3176 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3177 * returned statistic include statistics for all of 'rule''s subrules. */
3179 query_stats(struct ofproto *p, struct rule *rule,
3180 uint64_t *packet_countp, uint64_t *byte_countp)
3182 uint64_t packet_count, byte_count;
3183 struct rule *subrule;
3184 struct odp_flow *odp_flows;
3187 /* Start from historical data for 'rule' itself that are no longer tracked
3188 * by the datapath. This counts, for example, subrules that have
3190 packet_count = rule->packet_count;
3191 byte_count = rule->byte_count;
3193 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3194 * wildcarded then on all of its subrules.
3196 * Also, add any statistics that are not tracked by the datapath for each
3197 * subrule. This includes, for example, statistics for packets that were
3198 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3200 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
3201 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
3202 if (rule->cr.wc.wildcards) {
3204 LIST_FOR_EACH (subrule, list, &rule->list) {
3205 odp_flow_key_from_flow(&odp_flows[i++].key, &subrule->cr.flow);
3206 packet_count += subrule->packet_count;
3207 byte_count += subrule->byte_count;
3210 odp_flow_key_from_flow(&odp_flows[0].key, &rule->cr.flow);
3213 /* Fetch up-to-date statistics from the datapath and add them in. */
3214 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
3216 for (i = 0; i < n_odp_flows; i++) {
3217 struct odp_flow *odp_flow = &odp_flows[i];
3218 packet_count += odp_flow->stats.n_packets;
3219 byte_count += odp_flow->stats.n_bytes;
3224 /* Return the stats to the caller. */
3225 *packet_countp = packet_count;
3226 *byte_countp = byte_count;
3230 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3232 struct rule *rule = rule_from_cls_rule(rule_);
3233 struct flow_stats_cbdata *cbdata = cbdata_;
3234 struct ofp_flow_stats *ofs;
3235 uint64_t packet_count, byte_count;
3236 size_t act_len, len;
3237 long long int tdiff = time_msec() - rule->created;
3238 uint32_t sec = tdiff / 1000;
3239 uint32_t msec = tdiff - (sec * 1000);
3241 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3245 act_len = sizeof *rule->actions * rule->n_actions;
3246 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3248 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3250 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3251 ofs->length = htons(len);
3252 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3254 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3255 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3256 ofs->duration_sec = htonl(sec);
3257 ofs->duration_nsec = htonl(msec * 1000000);
3258 ofs->cookie = rule->flow_cookie;
3259 ofs->priority = htons(rule->cr.priority);
3260 ofs->idle_timeout = htons(rule->idle_timeout);
3261 ofs->hard_timeout = htons(rule->hard_timeout);
3262 memset(ofs->pad2, 0, sizeof ofs->pad2);
3263 ofs->packet_count = htonll(packet_count);
3264 ofs->byte_count = htonll(byte_count);
3265 if (rule->n_actions > 0) {
3266 memcpy(ofs->actions, rule->actions, act_len);
3271 table_id_to_include(uint8_t table_id)
3273 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3274 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3275 : table_id == 0xff ? CLS_INC_ALL
3280 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3281 const struct ofp_stats_request *osr,
3284 struct ofp_flow_stats_request *fsr;
3285 struct flow_stats_cbdata cbdata;
3286 struct cls_rule target;
3288 if (arg_size != sizeof *fsr) {
3289 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3291 fsr = (struct ofp_flow_stats_request *) osr->body;
3293 COVERAGE_INC(ofproto_flows_req);
3295 cbdata.ofconn = ofconn;
3296 cbdata.out_port = fsr->out_port;
3297 cbdata.msg = start_stats_reply(osr, 1024);
3298 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3299 classifier_for_each_match(&p->cls, &target,
3300 table_id_to_include(fsr->table_id),
3301 flow_stats_cb, &cbdata);
3302 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3306 struct flow_stats_ds_cbdata {
3307 struct ofproto *ofproto;
3312 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3314 struct rule *rule = rule_from_cls_rule(rule_);
3315 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3316 struct ds *results = cbdata->results;
3317 struct ofp_match match;
3318 uint64_t packet_count, byte_count;
3319 size_t act_len = sizeof *rule->actions * rule->n_actions;
3321 /* Don't report on subrules. */
3322 if (rule->super != NULL) {
3326 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3327 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3328 cbdata->ofproto->tun_id_from_cookie, &match);
3330 ds_put_format(results, "duration=%llds, ",
3331 (time_msec() - rule->created) / 1000);
3332 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3333 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3334 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3335 ofp_print_match(results, &match, true);
3337 ofp_print_actions(results, &rule->actions->header, act_len);
3339 ds_put_cstr(results, "\n");
3342 /* Adds a pretty-printed description of all flows to 'results', including
3343 * those marked hidden by secchan (e.g., by in-band control). */
3345 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3347 struct ofp_match match;
3348 struct cls_rule target;
3349 struct flow_stats_ds_cbdata cbdata;
3351 memset(&match, 0, sizeof match);
3352 match.wildcards = htonl(OVSFW_ALL);
3355 cbdata.results = results;
3357 cls_rule_from_match(&match, 0, false, 0, &target);
3358 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3359 flow_stats_ds_cb, &cbdata);
3362 struct aggregate_stats_cbdata {
3363 struct ofproto *ofproto;
3365 uint64_t packet_count;
3366 uint64_t byte_count;
3371 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3373 struct rule *rule = rule_from_cls_rule(rule_);
3374 struct aggregate_stats_cbdata *cbdata = cbdata_;
3375 uint64_t packet_count, byte_count;
3377 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3381 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3383 cbdata->packet_count += packet_count;
3384 cbdata->byte_count += byte_count;
3389 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3390 const struct ofp_stats_request *osr,
3393 struct ofp_aggregate_stats_request *asr;
3394 struct ofp_aggregate_stats_reply *reply;
3395 struct aggregate_stats_cbdata cbdata;
3396 struct cls_rule target;
3399 if (arg_size != sizeof *asr) {
3400 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3402 asr = (struct ofp_aggregate_stats_request *) osr->body;
3404 COVERAGE_INC(ofproto_agg_request);
3406 cbdata.out_port = asr->out_port;
3407 cbdata.packet_count = 0;
3408 cbdata.byte_count = 0;
3410 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3411 classifier_for_each_match(&p->cls, &target,
3412 table_id_to_include(asr->table_id),
3413 aggregate_stats_cb, &cbdata);
3415 msg = start_stats_reply(osr, sizeof *reply);
3416 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3417 reply->flow_count = htonl(cbdata.n_flows);
3418 reply->packet_count = htonll(cbdata.packet_count);
3419 reply->byte_count = htonll(cbdata.byte_count);
3420 queue_tx(msg, ofconn, ofconn->reply_counter);
3424 struct queue_stats_cbdata {
3425 struct ofconn *ofconn;
3426 struct ofport *ofport;
3431 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3432 const struct netdev_queue_stats *stats)
3434 struct ofp_queue_stats *reply;
3436 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3437 reply->port_no = htons(cbdata->ofport->opp.port_no);
3438 memset(reply->pad, 0, sizeof reply->pad);
3439 reply->queue_id = htonl(queue_id);
3440 reply->tx_bytes = htonll(stats->tx_bytes);
3441 reply->tx_packets = htonll(stats->tx_packets);
3442 reply->tx_errors = htonll(stats->tx_errors);
3446 handle_queue_stats_dump_cb(uint32_t queue_id,
3447 struct netdev_queue_stats *stats,
3450 struct queue_stats_cbdata *cbdata = cbdata_;
3452 put_queue_stats(cbdata, queue_id, stats);
3456 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3457 struct queue_stats_cbdata *cbdata)
3459 cbdata->ofport = port;
3460 if (queue_id == OFPQ_ALL) {
3461 netdev_dump_queue_stats(port->netdev,
3462 handle_queue_stats_dump_cb, cbdata);
3464 struct netdev_queue_stats stats;
3466 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3467 put_queue_stats(cbdata, queue_id, &stats);
3473 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3474 const struct ofp_stats_request *osr,
3477 struct ofp_queue_stats_request *qsr;
3478 struct queue_stats_cbdata cbdata;
3479 struct ofport *port;
3480 unsigned int port_no;
3483 if (arg_size != sizeof *qsr) {
3484 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3486 qsr = (struct ofp_queue_stats_request *) osr->body;
3488 COVERAGE_INC(ofproto_queue_req);
3490 cbdata.ofconn = ofconn;
3491 cbdata.msg = start_stats_reply(osr, 128);
3493 port_no = ntohs(qsr->port_no);
3494 queue_id = ntohl(qsr->queue_id);
3495 if (port_no == OFPP_ALL) {
3496 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3497 handle_queue_stats_for_port(port, queue_id, &cbdata);
3499 } else if (port_no < ofproto->max_ports) {
3500 port = get_port(ofproto, ofp_port_to_odp_port(port_no));
3502 handle_queue_stats_for_port(port, queue_id, &cbdata);
3505 ofpbuf_delete(cbdata.msg);
3506 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3508 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3514 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3515 struct ofp_header *oh)
3517 struct ofp_stats_request *osr;
3521 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3526 osr = (struct ofp_stats_request *) oh;
3528 switch (ntohs(osr->type)) {
3530 return handle_desc_stats_request(p, ofconn, osr);
3533 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3535 case OFPST_AGGREGATE:
3536 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3539 return handle_table_stats_request(p, ofconn, osr);
3542 return handle_port_stats_request(p, ofconn, osr, arg_size);
3545 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3548 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3551 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3555 static long long int
3556 msec_from_nsec(uint64_t sec, uint32_t nsec)
3558 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3562 update_time(struct ofproto *ofproto, struct rule *rule,
3563 const struct odp_flow_stats *stats)
3565 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3566 if (used > rule->used) {
3568 if (rule->super && used > rule->super->used) {
3569 rule->super->used = used;
3571 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3576 update_stats(struct ofproto *ofproto, struct rule *rule,
3577 const struct odp_flow_stats *stats)
3579 if (stats->n_packets) {
3580 update_time(ofproto, rule, stats);
3581 rule->packet_count += stats->n_packets;
3582 rule->byte_count += stats->n_bytes;
3583 netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags);
3587 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3588 * in which no matching flow already exists in the flow table.
3590 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3591 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3592 * code as encoded by ofp_mkerr() on failure.
3594 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3597 add_flow(struct ofproto *p, struct ofconn *ofconn,
3598 const struct ofp_flow_mod *ofm, size_t n_actions)
3600 struct ofpbuf *packet;
3605 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3609 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3611 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3612 ntohs(ofm->priority))) {
3613 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3617 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3618 n_actions, ntohs(ofm->idle_timeout),
3619 ntohs(ofm->hard_timeout), ofm->cookie,
3620 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3621 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3622 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3625 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3626 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3630 in_port = UINT16_MAX;
3633 rule_insert(p, rule, packet, in_port);
3637 static struct rule *
3638 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3643 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3645 return rule_from_cls_rule(classifier_find_rule_exactly(
3646 &p->cls, &flow, wildcards,
3647 ntohs(ofm->priority)));
3651 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3652 struct rule *rule, const struct ofp_flow_mod *ofm)
3654 struct ofpbuf *packet;
3659 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3663 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3669 flow_extract(packet, 0, in_port, &flow);
3670 rule_execute(ofproto, rule, packet, &flow);
3675 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3677 struct modify_flows_cbdata {
3678 struct ofproto *ofproto;
3679 const struct ofp_flow_mod *ofm;
3684 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3685 size_t n_actions, struct rule *);
3686 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3688 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3689 * encoded by ofp_mkerr() on failure.
3691 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3694 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3695 const struct ofp_flow_mod *ofm, size_t n_actions)
3697 struct modify_flows_cbdata cbdata;
3698 struct cls_rule target;
3702 cbdata.n_actions = n_actions;
3703 cbdata.match = NULL;
3705 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3708 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3709 modify_flows_cb, &cbdata);
3711 /* This credits the packet to whichever flow happened to happened to
3712 * match last. That's weird. Maybe we should do a lookup for the
3713 * flow that actually matches the packet? Who knows. */
3714 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3717 return add_flow(p, ofconn, ofm, n_actions);
3721 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3722 * code as encoded by ofp_mkerr() on failure.
3724 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3727 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3728 struct ofp_flow_mod *ofm, size_t n_actions)
3730 struct rule *rule = find_flow_strict(p, ofm);
3731 if (rule && !rule_is_hidden(rule)) {
3732 modify_flow(p, ofm, n_actions, rule);
3733 return send_buffered_packet(p, ofconn, rule, ofm);
3735 return add_flow(p, ofconn, ofm, n_actions);
3739 /* Callback for modify_flows_loose(). */
3741 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3743 struct rule *rule = rule_from_cls_rule(rule_);
3744 struct modify_flows_cbdata *cbdata = cbdata_;
3746 if (!rule_is_hidden(rule)) {
3747 cbdata->match = rule;
3748 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3752 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3753 * been identified as a flow in 'p''s flow table to be modified, by changing
3754 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3755 * ofp_action[] structures). */
3757 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3758 size_t n_actions, struct rule *rule)
3760 size_t actions_len = n_actions * sizeof *rule->actions;
3762 rule->flow_cookie = ofm->cookie;
3764 /* If the actions are the same, do nothing. */
3765 if (n_actions == rule->n_actions
3766 && (!n_actions || !memcmp(ofm->actions, rule->actions, actions_len)))
3771 /* Replace actions. */
3772 free(rule->actions);
3773 rule->actions = n_actions ? xmemdup(ofm->actions, actions_len) : NULL;
3774 rule->n_actions = n_actions;
3776 /* Make sure that the datapath gets updated properly. */
3777 if (rule->cr.wc.wildcards) {
3778 COVERAGE_INC(ofproto_mod_wc_flow);
3779 p->need_revalidate = true;
3781 rule_update_actions(p, rule);
3787 /* OFPFC_DELETE implementation. */
3789 struct delete_flows_cbdata {
3790 struct ofproto *ofproto;
3794 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3795 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3797 /* Implements OFPFC_DELETE. */
3799 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3801 struct delete_flows_cbdata cbdata;
3802 struct cls_rule target;
3805 cbdata.out_port = ofm->out_port;
3807 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3810 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3811 delete_flows_cb, &cbdata);
3814 /* Implements OFPFC_DELETE_STRICT. */
3816 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3818 struct rule *rule = find_flow_strict(p, ofm);
3820 delete_flow(p, rule, ofm->out_port);
3824 /* Callback for delete_flows_loose(). */
3826 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3828 struct rule *rule = rule_from_cls_rule(rule_);
3829 struct delete_flows_cbdata *cbdata = cbdata_;
3831 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3834 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3835 * been identified as a flow to delete from 'p''s flow table, by deleting the
3836 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3839 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3840 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3841 * specified 'out_port'. */
3843 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3845 if (rule_is_hidden(rule)) {
3849 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3853 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3854 rule_remove(p, rule);
3858 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3859 struct ofp_flow_mod *ofm)
3861 struct ofp_match orig_match;
3865 error = reject_slave_controller(ofconn, &ofm->header);
3869 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3870 sizeof *ofm->actions, &n_actions);
3875 /* We do not support the emergency flow cache. It will hopefully
3876 * get dropped from OpenFlow in the near future. */
3877 if (ofm->flags & htons(OFPFF_EMERG)) {
3878 /* There isn't a good fit for an error code, so just state that the
3879 * flow table is full. */
3880 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3883 /* Normalize ofp->match. If normalization actually changes anything, then
3884 * log the differences. */
3885 ofm->match.pad1[0] = ofm->match.pad2[0] = 0;
3886 orig_match = ofm->match;
3887 normalize_match(&ofm->match);
3888 if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) {
3889 static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1);
3890 if (!VLOG_DROP_INFO(&normal_rl)) {
3891 char *old = ofp_match_to_literal_string(&orig_match);
3892 char *new = ofp_match_to_literal_string(&ofm->match);
3893 VLOG_INFO("%s: normalization changed ofp_match, details:",
3894 rconn_get_name(ofconn->rconn));
3895 VLOG_INFO(" pre: %s", old);
3896 VLOG_INFO("post: %s", new);
3902 if (!ofm->match.wildcards) {
3903 ofm->priority = htons(UINT16_MAX);
3906 error = validate_actions((const union ofp_action *) ofm->actions,
3907 n_actions, p->max_ports);
3912 switch (ntohs(ofm->command)) {
3914 return add_flow(p, ofconn, ofm, n_actions);
3917 return modify_flows_loose(p, ofconn, ofm, n_actions);
3919 case OFPFC_MODIFY_STRICT:
3920 return modify_flow_strict(p, ofconn, ofm, n_actions);
3923 delete_flows_loose(p, ofm);
3926 case OFPFC_DELETE_STRICT:
3927 delete_flow_strict(p, ofm);
3931 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3936 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3940 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3945 p->tun_id_from_cookie = !!msg->set;
3950 handle_role_request(struct ofproto *ofproto,
3951 struct ofconn *ofconn, struct nicira_header *msg)
3953 struct nx_role_request *nrr;
3954 struct nx_role_request *reply;
3958 if (ntohs(msg->header.length) != sizeof *nrr) {
3959 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3960 ntohs(msg->header.length), sizeof *nrr);
3961 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3963 nrr = (struct nx_role_request *) msg;
3965 if (ofconn->type != OFCONN_PRIMARY) {
3966 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3968 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3971 role = ntohl(nrr->role);
3972 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3973 && role != NX_ROLE_SLAVE) {
3974 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3976 /* There's no good error code for this. */
3977 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3980 if (role == NX_ROLE_MASTER) {
3981 struct ofconn *other;
3983 HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) {
3984 if (other->role == NX_ROLE_MASTER) {
3985 other->role = NX_ROLE_SLAVE;
3989 ofconn->role = role;
3991 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3993 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3994 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3995 reply->role = htonl(role);
3996 queue_tx(buf, ofconn, ofconn->reply_counter);
4002 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
4004 struct ofp_vendor_header *ovh = msg;
4005 struct nicira_header *nh;
4007 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
4008 VLOG_WARN_RL(&rl, "received vendor message of length %u "
4009 "(expected at least %zu)",
4010 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
4011 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4013 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
4014 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
4016 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
4017 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
4018 "(expected at least %zu)",
4019 ntohs(ovh->header.length), sizeof(struct nicira_header));
4020 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
4024 switch (ntohl(nh->subtype)) {
4025 case NXT_STATUS_REQUEST:
4026 return switch_status_handle_request(p->switch_status, ofconn->rconn,
4029 case NXT_TUN_ID_FROM_COOKIE:
4030 return handle_tun_id_from_cookie(p, msg);
4032 case NXT_ROLE_REQUEST:
4033 return handle_role_request(p, ofconn, msg);
4036 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
4040 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
4042 struct ofp_header *ob;
4045 /* Currently, everything executes synchronously, so we can just
4046 * immediately send the barrier reply. */
4047 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
4048 queue_tx(buf, ofconn, ofconn->reply_counter);
4053 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
4054 struct ofpbuf *ofp_msg)
4056 struct ofp_header *oh = ofp_msg->data;
4059 COVERAGE_INC(ofproto_recv_openflow);
4061 case OFPT_ECHO_REQUEST:
4062 error = handle_echo_request(ofconn, oh);
4065 case OFPT_ECHO_REPLY:
4069 case OFPT_FEATURES_REQUEST:
4070 error = handle_features_request(p, ofconn, oh);
4073 case OFPT_GET_CONFIG_REQUEST:
4074 error = handle_get_config_request(p, ofconn, oh);
4077 case OFPT_SET_CONFIG:
4078 error = handle_set_config(p, ofconn, ofp_msg->data);
4081 case OFPT_PACKET_OUT:
4082 error = handle_packet_out(p, ofconn, ofp_msg->data);
4086 error = handle_port_mod(p, ofconn, oh);
4090 error = handle_flow_mod(p, ofconn, ofp_msg->data);
4093 case OFPT_STATS_REQUEST:
4094 error = handle_stats_request(p, ofconn, oh);
4098 error = handle_vendor(p, ofconn, ofp_msg->data);
4101 case OFPT_BARRIER_REQUEST:
4102 error = handle_barrier_request(ofconn, oh);
4106 if (VLOG_IS_WARN_ENABLED()) {
4107 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
4108 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
4111 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
4116 send_error_oh(ofconn, ofp_msg->data, error);
4121 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
4123 struct odp_msg *msg = packet->data;
4125 struct ofpbuf payload;
4128 payload.data = msg + 1;
4129 payload.size = msg->length - sizeof *msg;
4130 flow_extract(&payload, msg->arg, msg->port, &flow);
4132 /* Check with in-band control to see if this packet should be sent
4133 * to the local port regardless of the flow table. */
4134 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
4135 union odp_action action;
4137 memset(&action, 0, sizeof(action));
4138 action.output.type = ODPAT_OUTPUT;
4139 action.output.port = ODPP_LOCAL;
4140 dpif_execute(p->dpif, &action, 1, &payload);
4143 rule = lookup_valid_rule(p, &flow);
4145 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4146 struct ofport *port = get_port(p, msg->port);
4148 if (port->opp.config & OFPPC_NO_PACKET_IN) {
4149 COVERAGE_INC(ofproto_no_packet_in);
4150 /* XXX install 'drop' flow entry */
4151 ofpbuf_delete(packet);
4155 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
4158 COVERAGE_INC(ofproto_packet_in);
4159 send_packet_in(p, packet);
4163 if (rule->cr.wc.wildcards) {
4164 rule = rule_create_subrule(p, rule, &flow);
4165 rule_make_actions(p, rule, packet);
4167 if (!rule->may_install) {
4168 /* The rule is not installable, that is, we need to process every
4169 * packet, so process the current packet and set its actions into
4171 rule_make_actions(p, rule, packet);
4173 /* XXX revalidate rule if it needs it */
4177 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
4179 * Extra-special case for fail-open mode.
4181 * We are in fail-open mode and the packet matched the fail-open rule,
4182 * but we are connected to a controller too. We should send the packet
4183 * up to the controller in the hope that it will try to set up a flow
4184 * and thereby allow us to exit fail-open.
4186 * See the top-level comment in fail-open.c for more information.
4188 send_packet_in(p, ofpbuf_clone_with_headroom(packet,
4189 DPIF_RECV_MSG_PADDING));
4192 ofpbuf_pull(packet, sizeof *msg);
4193 rule_execute(p, rule, packet, &flow);
4194 rule_reinstall(p, rule);
4198 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
4200 struct odp_msg *msg = packet->data;
4202 switch (msg->type) {
4203 case _ODPL_ACTION_NR:
4204 COVERAGE_INC(ofproto_ctlr_action);
4205 send_packet_in(p, packet);
4208 case _ODPL_SFLOW_NR:
4210 ofproto_sflow_received(p->sflow, msg);
4212 ofpbuf_delete(packet);
4216 handle_odp_miss_msg(p, packet);
4220 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
4226 /* Flow expiration. */
4228 struct expire_cbdata {
4229 struct ofproto *ofproto;
4233 static int ofproto_dp_max_idle(const struct ofproto *);
4234 static void ofproto_update_used(struct ofproto *);
4235 static void rule_expire(struct cls_rule *, void *cbdata);
4237 /* This function is called periodically by ofproto_run(). Its job is to
4238 * collect updates for the flows that have been installed into the datapath,
4239 * most importantly when they last were used, and then use that information to
4240 * expire flows that have not been used recently.
4242 * Returns the number of milliseconds after which it should be called again. */
4244 ofproto_expire(struct ofproto *ofproto)
4246 struct expire_cbdata cbdata;
4248 /* Update 'used' for each flow in the datapath. */
4249 ofproto_update_used(ofproto);
4251 /* Expire idle flows.
4253 * A wildcarded flow is idle only when all of its subrules have expired due
4254 * to becoming idle, so iterate through the exact-match flows first. */
4255 cbdata.ofproto = ofproto;
4256 cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto);
4257 classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata);
4258 classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata);
4260 /* Let the hook know that we're at a stable point: all outstanding data
4261 * in existing flows has been accounted to the account_cb. Thus, the
4262 * hook can now reasonably do operations that depend on having accurate
4263 * flow volume accounting (currently, that's just bond rebalancing). */
4264 if (ofproto->ofhooks->account_checkpoint_cb) {
4265 ofproto->ofhooks->account_checkpoint_cb(ofproto->aux);
4268 return MIN(cbdata.dp_max_idle, 1000);
4271 /* Update 'used' member of each flow currently installed into the datapath. */
4273 ofproto_update_used(struct ofproto *p)
4275 struct odp_flow *flows;
4280 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4285 for (i = 0; i < n_flows; i++) {
4286 struct odp_flow *f = &flows[i];
4290 odp_flow_key_to_flow(&f->key, &flow);
4292 rule = rule_from_cls_rule(
4293 classifier_find_rule_exactly(&p->cls, &flow, 0, UINT16_MAX));
4295 if (rule && rule->installed) {
4296 update_time(p, rule, &f->stats);
4297 rule_account(p, rule, f->stats.n_bytes);
4299 /* There's a flow in the datapath that we know nothing about.
4301 COVERAGE_INC(ofproto_unexpected_rule);
4302 dpif_flow_del(p->dpif, f);
4309 /* Calculates and returns the number of milliseconds of idle time after which
4310 * flows should expire from the datapath and we should fold their statistics
4311 * into their parent rules in userspace. */
4313 ofproto_dp_max_idle(const struct ofproto *ofproto)
4316 * Idle time histogram.
4318 * Most of the time a switch has a relatively small number of flows. When
4319 * this is the case we might as well keep statistics for all of them in
4320 * userspace and to cache them in the kernel datapath for performance as
4323 * As the number of flows increases, the memory required to maintain
4324 * statistics about them in userspace and in the kernel becomes
4325 * significant. However, with a large number of flows it is likely that
4326 * only a few of them are "heavy hitters" that consume a large amount of
4327 * bandwidth. At this point, only heavy hitters are worth caching in the
4328 * kernel and maintaining in userspaces; other flows we can discard.
4330 * The technique used to compute the idle time is to build a histogram with
4331 * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that
4332 * is installed in the kernel gets dropped in the appropriate bucket.
4333 * After the histogram has been built, we compute the cutoff so that only
4334 * the most-recently-used 1% of flows (but at least 1000 flows) are kept
4335 * cached. At least the most-recently-used bucket of flows is kept, so
4336 * actually an arbitrary number of flows can be kept in any given
4337 * expiration run (though the next run will delete most of those unless
4338 * they receive additional data).
4340 * This requires a second pass through the exact-match flows, in addition
4341 * to the pass made by ofproto_update_used(), because the former function
4342 * never looks at uninstallable flows.
4344 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4345 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4346 int buckets[N_BUCKETS] = { 0 };
4352 total = classifier_count_exact(&ofproto->cls);
4353 if (total <= 1000) {
4354 return N_BUCKETS * BUCKET_WIDTH;
4357 /* Build histogram. */
4359 CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) {
4360 long long int idle = now - rule->used;
4361 int bucket = (idle <= 0 ? 0
4362 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4363 : (unsigned int) idle / BUCKET_WIDTH);
4367 /* Find the first bucket whose flows should be expired. */
4368 for (bucket = 0; bucket < N_BUCKETS; bucket++) {
4369 if (buckets[bucket]) {
4372 subtotal += buckets[bucket++];
4373 } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
4378 if (VLOG_IS_DBG_ENABLED()) {
4382 ds_put_cstr(&s, "keep");
4383 for (i = 0; i < N_BUCKETS; i++) {
4385 ds_put_cstr(&s, ", drop");
4388 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4391 VLOG_INFO("%s: %s (msec:count)",
4392 dpif_name(ofproto->dpif), ds_cstr(&s));
4396 return bucket * BUCKET_WIDTH;
4400 rule_active_timeout(struct ofproto *ofproto, struct rule *rule)
4402 if (ofproto->netflow && !is_controller_rule(rule) &&
4403 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4404 struct ofexpired expired;
4405 struct odp_flow odp_flow;
4407 /* Get updated flow stats.
4409 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4410 * updated TCP flags and (2) the dpif_flow_list_all() in
4411 * ofproto_update_used() zeroed TCP flags. */
4412 memset(&odp_flow, 0, sizeof odp_flow);
4413 if (rule->installed) {
4414 odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow);
4415 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4416 dpif_flow_get(ofproto->dpif, &odp_flow);
4418 if (odp_flow.stats.n_packets) {
4419 update_time(ofproto, rule, &odp_flow.stats);
4420 netflow_flow_update_flags(&rule->nf_flow,
4421 odp_flow.stats.tcp_flags);
4425 expired.flow = rule->cr.flow;
4426 expired.packet_count = rule->packet_count +
4427 odp_flow.stats.n_packets;
4428 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4429 expired.used = rule->used;
4431 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4435 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4436 * rules, then delete it entirely.
4438 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4439 * the datapath and fold its statistics back into its super-rule.
4441 * (This is a callback function for classifier_for_each().) */
4443 rule_expire(struct cls_rule *cls_rule, void *cbdata_)
4445 struct expire_cbdata *cbdata = cbdata_;
4446 struct ofproto *ofproto = cbdata->ofproto;
4447 struct rule *rule = rule_from_cls_rule(cls_rule);
4448 long long int hard_expire, idle_expire, expire, now;
4450 /* Calculate OpenFlow expiration times for 'rule'. */
4451 hard_expire = (rule->hard_timeout
4452 ? rule->created + rule->hard_timeout * 1000
4454 idle_expire = (rule->idle_timeout
4455 && (rule->super || list_is_empty(&rule->list))
4456 ? rule->used + rule->idle_timeout * 1000
4458 expire = MIN(hard_expire, idle_expire);
4462 /* 'rule' has not expired according to OpenFlow rules. */
4463 if (!rule->cr.wc.wildcards) {
4464 if (now >= rule->used + cbdata->dp_max_idle) {
4465 /* This rule is idle, so drop it to free up resources. */
4467 /* It's not part of the OpenFlow flow table, so we can
4468 * delete it entirely and fold its statistics into its
4470 rule_remove(ofproto, rule);
4472 /* It is part of the OpenFlow flow table, so we have to
4473 * keep the rule but we can at least uninstall it from the
4475 rule_uninstall(ofproto, rule);
4478 /* Send NetFlow active timeout if appropriate. */
4479 rule_active_timeout(cbdata->ofproto, rule);
4483 /* 'rule' has expired according to OpenFlow rules. */
4484 COVERAGE_INC(ofproto_expired);
4486 /* Update stats. (This is a no-op if the rule expired due to an idle
4487 * timeout, because that only happens when the rule has no subrules
4489 if (rule->cr.wc.wildcards) {
4490 struct rule *subrule, *next;
4491 LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) {
4492 rule_remove(cbdata->ofproto, subrule);
4495 rule_uninstall(cbdata->ofproto, rule);
4498 /* Get rid of the rule. */
4499 if (!rule_is_hidden(rule)) {
4500 send_flow_removed(cbdata->ofproto, rule, now,
4502 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4504 rule_remove(cbdata->ofproto, rule);
4509 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
4511 struct rule *sub = rule_from_cls_rule(sub_);
4512 struct revalidate_cbdata *cbdata = cbdata_;
4514 if (cbdata->revalidate_all
4515 || (cbdata->revalidate_subrules && sub->super)
4516 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
4517 revalidate_rule(cbdata->ofproto, sub);
4522 revalidate_rule(struct ofproto *p, struct rule *rule)
4524 const struct flow *flow = &rule->cr.flow;
4526 COVERAGE_INC(ofproto_revalidate_rule);
4529 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4531 rule_remove(p, rule);
4533 } else if (super != rule->super) {
4534 COVERAGE_INC(ofproto_revalidate_moved);
4535 list_remove(&rule->list);
4536 list_push_back(&super->list, &rule->list);
4537 rule->super = super;
4538 rule->hard_timeout = super->hard_timeout;
4539 rule->idle_timeout = super->idle_timeout;
4540 rule->created = super->created;
4545 rule_update_actions(p, rule);
4549 static struct ofpbuf *
4550 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4551 long long int now, uint8_t reason)
4553 struct ofp_flow_removed *ofr;
4555 long long int tdiff = now - rule->created;
4556 uint32_t sec = tdiff / 1000;
4557 uint32_t msec = tdiff - (sec * 1000);
4559 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4560 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4562 ofr->cookie = rule->flow_cookie;
4563 ofr->priority = htons(rule->cr.priority);
4564 ofr->reason = reason;
4565 ofr->duration_sec = htonl(sec);
4566 ofr->duration_nsec = htonl(msec * 1000000);
4567 ofr->idle_timeout = htons(rule->idle_timeout);
4568 ofr->packet_count = htonll(rule->packet_count);
4569 ofr->byte_count = htonll(rule->byte_count);
4575 send_flow_removed(struct ofproto *p, struct rule *rule,
4576 long long int now, uint8_t reason)
4578 struct ofconn *ofconn;
4579 struct ofconn *prev;
4580 struct ofpbuf *buf = NULL;
4582 if (!rule->send_flow_removed) {
4586 /* We limit the maximum number of queued flow expirations it by accounting
4587 * them under the counter for replies. That works because preventing
4588 * OpenFlow requests from being processed also prevents new flows from
4589 * being added (and expiring). (It also prevents processing OpenFlow
4590 * requests that would not add new flows, so it is imperfect.) */
4593 LIST_FOR_EACH (ofconn, node, &p->all_conns) {
4594 if (rconn_is_connected(ofconn->rconn)
4595 && ofconn_receives_async_msgs(ofconn)) {
4597 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4599 buf = compose_flow_removed(p, rule, now, reason);
4605 queue_tx(buf, prev, prev->reply_counter);
4609 /* pinsched callback for sending 'packet' on 'ofconn'. */
4611 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4613 struct ofconn *ofconn = ofconn_;
4615 rconn_send_with_limit(ofconn->rconn, packet,
4616 ofconn->packet_in_counter, 100);
4619 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4620 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4621 * packet scheduler for sending.
4623 * 'max_len' specifies the maximum number of bytes of the packet to send on
4624 * 'ofconn' (INT_MAX specifies no limit).
4626 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4627 * ownership is transferred to this function. */
4629 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4632 struct ofproto *ofproto = ofconn->ofproto;
4633 struct ofp_packet_in *opi = packet->data;
4634 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4635 int send_len, trim_size;
4639 if (opi->reason == OFPR_ACTION) {
4640 buffer_id = UINT32_MAX;
4641 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4642 buffer_id = pktbuf_get_null();
4643 } else if (!ofconn->pktbuf) {
4644 buffer_id = UINT32_MAX;
4646 struct ofpbuf payload;
4647 payload.data = opi->data;
4648 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4649 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4652 /* Figure out how much of the packet to send. */
4653 send_len = ntohs(opi->total_len);
4654 if (buffer_id != UINT32_MAX) {
4655 send_len = MIN(send_len, ofconn->miss_send_len);
4657 send_len = MIN(send_len, max_len);
4659 /* Adjust packet length and clone if necessary. */
4660 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4662 packet = ofpbuf_clone_data(packet->data, trim_size);
4665 packet->size = trim_size;
4668 /* Update packet headers. */
4669 opi->buffer_id = htonl(buffer_id);
4670 update_openflow_length(packet);
4672 /* Hand over to packet scheduler. It might immediately call into
4673 * do_send_packet_in() or it might buffer it for a while (until a later
4674 * call to pinsched_run()). */
4675 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4676 packet, do_send_packet_in, ofconn);
4679 /* Replace struct odp_msg header in 'packet' by equivalent struct
4680 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4681 * returned by dpif_recv()).
4683 * The conversion is not complete: the caller still needs to trim any unneeded
4684 * payload off the end of the buffer, set the length in the OpenFlow header,
4685 * and set buffer_id. Those require us to know the controller settings and so
4686 * must be done on a per-controller basis.
4688 * Returns the maximum number of bytes of the packet that should be sent to
4689 * the controller (INT_MAX if no limit). */
4691 do_convert_to_packet_in(struct ofpbuf *packet)
4693 struct odp_msg *msg = packet->data;
4694 struct ofp_packet_in *opi;
4700 /* Extract relevant header fields */
4701 if (msg->type == _ODPL_ACTION_NR) {
4702 reason = OFPR_ACTION;
4705 reason = OFPR_NO_MATCH;
4708 total_len = msg->length - sizeof *msg;
4709 in_port = odp_port_to_ofp_port(msg->port);
4711 /* Repurpose packet buffer by overwriting header. */
4712 ofpbuf_pull(packet, sizeof(struct odp_msg));
4713 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4714 opi->header.version = OFP_VERSION;
4715 opi->header.type = OFPT_PACKET_IN;
4716 opi->total_len = htons(total_len);
4717 opi->in_port = htons(in_port);
4718 opi->reason = reason;
4723 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4724 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4725 * as necessary according to their individual configurations.
4727 * 'packet' must have sufficient headroom to convert it into a struct
4728 * ofp_packet_in (e.g. as returned by dpif_recv()).
4730 * Takes ownership of 'packet'. */
4732 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4734 struct ofconn *ofconn, *prev;
4737 max_len = do_convert_to_packet_in(packet);
4740 LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) {
4741 if (ofconn_receives_async_msgs(ofconn)) {
4743 schedule_packet_in(prev, packet, max_len, true);
4749 schedule_packet_in(prev, packet, max_len, false);
4751 ofpbuf_delete(packet);
4756 pick_datapath_id(const struct ofproto *ofproto)
4758 const struct ofport *port;
4760 port = get_port(ofproto, ODPP_LOCAL);
4762 uint8_t ea[ETH_ADDR_LEN];
4765 error = netdev_get_etheraddr(port->netdev, ea);
4767 return eth_addr_to_uint64(ea);
4769 VLOG_WARN("could not get MAC address for %s (%s)",
4770 netdev_get_name(port->netdev), strerror(error));
4772 return ofproto->fallback_dpid;
4776 pick_fallback_dpid(void)
4778 uint8_t ea[ETH_ADDR_LEN];
4779 eth_addr_nicira_random(ea);
4780 return eth_addr_to_uint64(ea);
4784 default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet,
4785 struct odp_actions *actions, tag_type *tags,
4786 uint16_t *nf_output_iface, void *ofproto_)
4788 struct ofproto *ofproto = ofproto_;
4791 /* Drop frames for reserved multicast addresses. */
4792 if (eth_addr_is_reserved(flow->dl_dst)) {
4796 /* Learn source MAC (but don't try to learn from revalidation). */
4797 if (packet != NULL) {
4798 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4800 GRAT_ARP_LOCK_NONE);
4802 /* The log messages here could actually be useful in debugging,
4803 * so keep the rate limit relatively high. */
4804 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4805 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4806 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4807 ofproto_revalidate(ofproto, rev_tag);
4811 /* Determine output port. */
4812 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4815 flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD,
4816 nf_output_iface, actions);
4817 } else if (out_port != flow->in_port) {
4818 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4819 *nf_output_iface = out_port;
4827 static const struct ofhooks default_ofhooks = {
4828 default_normal_ofhook_cb,