2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "classifier.h"
29 #include "discovery.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
34 #include "mac-learning.h"
38 #include "ofp-print.h"
40 #include "ofproto-sflow.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
44 #include "openvswitch/datapath-protocol.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
54 #include "stream-ssl.h"
62 #include <linux/types.h> /* XXX */
63 #include <linux/pkt_sched.h> /* XXX */
65 #define THIS_MODULE VLM_ofproto
68 #include "sflow_api.h"
72 TABLEID_CLASSIFIER = 1
76 struct netdev *netdev;
77 struct ofp_phy_port opp; /* In host byte order. */
80 static void ofport_free(struct ofport *);
81 static void hton_ofp_phy_port(struct ofp_phy_port *);
83 static int xlate_actions(const union ofp_action *in, size_t n_in,
84 const flow_t *flow, struct ofproto *ofproto,
85 const struct ofpbuf *packet,
86 struct odp_actions *out, tag_type *tags,
87 bool *may_set_up_flow, uint16_t *nf_output_iface);
92 uint64_t flow_cookie; /* Controller-issued identifier.
93 (Kept in network-byte order.) */
94 uint16_t idle_timeout; /* In seconds from time of last use. */
95 uint16_t hard_timeout; /* In seconds from time of creation. */
96 bool send_flow_removed; /* Send a flow removed message? */
97 long long int used; /* Last-used time (0 if never used). */
98 long long int created; /* Creation time. */
99 uint64_t packet_count; /* Number of packets received. */
100 uint64_t byte_count; /* Number of bytes received. */
101 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
102 tag_type tags; /* Tags (set only by hooks). */
103 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
105 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
106 * exact-match rule (having cr.wc.wildcards of 0) generated from the
107 * wildcard rule 'super'. In this case, 'list' is an element of the
110 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
111 * a list of subrules. A super-rule with no wildcards (where
112 * cr.wc.wildcards is 0) will never have any subrules. */
118 * 'n_actions' is the number of elements in the 'actions' array. A single
119 * action may take up more more than one element's worth of space.
121 * A subrule has no actions (it uses the super-rule's actions). */
123 union ofp_action *actions;
127 * A super-rule with wildcard fields never has ODP actions (since the
128 * datapath only supports exact-match flows). */
129 bool installed; /* Installed in datapath? */
130 bool may_install; /* True ordinarily; false if actions must
131 * be reassessed for every packet. */
133 union odp_action *odp_actions;
137 rule_is_hidden(const struct rule *rule)
139 /* Subrules are merely an implementation detail, so hide them from the
141 if (rule->super != NULL) {
145 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
146 * (e.g. by in-band control) and are intentionally hidden from the
148 if (rule->cr.priority > UINT16_MAX) {
155 static struct rule *rule_create(struct ofproto *, struct rule *super,
156 const union ofp_action *, size_t n_actions,
157 uint16_t idle_timeout, uint16_t hard_timeout,
158 uint64_t flow_cookie, bool send_flow_removed);
159 static void rule_free(struct rule *);
160 static void rule_destroy(struct ofproto *, struct rule *);
161 static struct rule *rule_from_cls_rule(const struct cls_rule *);
162 static void rule_insert(struct ofproto *, struct rule *,
163 struct ofpbuf *packet, uint16_t in_port);
164 static void rule_remove(struct ofproto *, struct rule *);
165 static bool rule_make_actions(struct ofproto *, struct rule *,
166 const struct ofpbuf *packet);
167 static void rule_install(struct ofproto *, struct rule *,
168 struct rule *displaced_rule);
169 static void rule_uninstall(struct ofproto *, struct rule *);
170 static void rule_post_uninstall(struct ofproto *, struct rule *);
171 static void send_flow_removed(struct ofproto *p, struct rule *rule,
172 long long int now, uint8_t reason);
174 /* ofproto supports two kinds of OpenFlow connections:
176 * - "Controller connections": Connections to ordinary OpenFlow controllers.
177 * ofproto maintains persistent connections to these controllers and by
178 * default sends them asynchronous messages such as packet-ins.
180 * - "Transient 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.
185 OFCONN_CONTROLLER, /* An OpenFlow controller. */
186 OFCONN_TRANSIENT /* A transient connection. */
189 /* An OpenFlow connection. */
191 struct ofproto *ofproto; /* The ofproto that owns this connection. */
192 struct list node; /* In struct ofproto's "all_conns" list. */
193 struct rconn *rconn; /* OpenFlow connection. */
194 enum ofconn_type type; /* Type. */
196 /* OFPT_PACKET_IN related data. */
197 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
198 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
199 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
200 int miss_send_len; /* Bytes to send of buffered packets. */
202 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
203 * requests, and the maximum number before we stop reading OpenFlow
205 #define OFCONN_REPLY_MAX 100
206 struct rconn_packet_counter *reply_counter;
208 /* type == OFCONN_CONTROLLER only. */
209 enum nx_role role; /* Role. */
210 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
211 struct discovery *discovery; /* Controller discovery object, if enabled. */
212 struct status_category *ss; /* Switch status category. */
213 enum ofproto_band band; /* In-band or out-of-band? */
216 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
217 * "schedulers" array. Their values are 0 and 1, and their meanings and values
218 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
219 * case anything ever changes, check their values here. */
220 #define N_SCHEDULERS 2
221 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
222 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
223 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
224 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
226 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
228 static void ofconn_destroy(struct ofconn *);
229 static void ofconn_run(struct ofconn *, struct ofproto *);
230 static void ofconn_wait(struct ofconn *);
231 static bool ofconn_receives_async_msgs(const struct ofconn *);
232 static char *ofconn_make_name(const struct ofproto *, const char *target);
234 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
235 struct rconn_packet_counter *counter);
237 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
238 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
242 uint64_t datapath_id; /* Datapath ID. */
243 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
244 char *mfr_desc; /* Manufacturer. */
245 char *hw_desc; /* Hardware. */
246 char *sw_desc; /* Software version. */
247 char *serial_desc; /* Serial number. */
248 char *dp_desc; /* Datapath description. */
252 struct netdev_monitor *netdev_monitor;
253 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
255 struct shash port_by_name;
259 struct switch_status *switch_status;
260 struct fail_open *fail_open;
261 struct netflow *netflow;
262 struct ofproto_sflow *sflow;
264 /* In-band control. */
265 struct in_band *in_band;
266 long long int next_in_band_update;
267 struct sockaddr_in *extra_in_band_remotes;
268 size_t n_extra_remotes;
271 struct classifier cls;
272 bool need_revalidate;
273 long long int next_expiration;
274 struct tag_set revalidate_set;
275 bool tun_id_from_cookie;
277 /* OpenFlow connections. */
278 struct hmap controllers; /* Controller "struct ofconn"s. */
279 struct list all_conns; /* Contains "struct ofconn"s. */
280 struct pvconn **listeners;
282 struct pvconn **snoops;
285 /* Hooks for ovs-vswitchd. */
286 const struct ofhooks *ofhooks;
289 /* Used by default ofhooks. */
290 struct mac_learning *ml;
293 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
295 static const struct ofhooks default_ofhooks;
297 static uint64_t pick_datapath_id(const struct ofproto *);
298 static uint64_t pick_fallback_dpid(void);
300 static void update_used(struct ofproto *);
301 static void update_stats(struct ofproto *, struct rule *,
302 const struct odp_flow_stats *);
303 static void expire_rule(struct cls_rule *, void *ofproto);
304 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
305 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
306 static void revalidate_cb(struct cls_rule *rule_, void *p_);
308 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
310 static void handle_openflow(struct ofconn *, struct ofproto *,
313 static void refresh_port_groups(struct ofproto *);
315 static void update_port(struct ofproto *, const char *devname);
316 static int init_ports(struct ofproto *);
317 static void reinit_ports(struct ofproto *);
320 ofproto_create(const char *datapath, const char *datapath_type,
321 const struct ofhooks *ofhooks, void *aux,
322 struct ofproto **ofprotop)
324 struct odp_stats stats;
331 /* Connect to datapath and start listening for messages. */
332 error = dpif_open(datapath, datapath_type, &dpif);
334 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
337 error = dpif_get_dp_stats(dpif, &stats);
339 VLOG_ERR("failed to obtain stats for datapath %s: %s",
340 datapath, strerror(error));
344 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
346 VLOG_ERR("failed to listen on datapath %s: %s",
347 datapath, strerror(error));
351 dpif_flow_flush(dpif);
352 dpif_recv_purge(dpif);
354 /* Initialize settings. */
355 p = xzalloc(sizeof *p);
356 p->fallback_dpid = pick_fallback_dpid();
357 p->datapath_id = p->fallback_dpid;
358 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
359 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
360 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
361 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
362 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
364 /* Initialize datapath. */
366 p->netdev_monitor = netdev_monitor_create();
367 port_array_init(&p->ports);
368 shash_init(&p->port_by_name);
369 p->max_ports = stats.max_ports;
371 /* Initialize submodules. */
372 p->switch_status = switch_status_create(p);
378 /* Initialize flow table. */
379 classifier_init(&p->cls);
380 p->need_revalidate = false;
381 p->next_expiration = time_msec() + 1000;
382 tag_set_init(&p->revalidate_set);
384 /* Initialize OpenFlow connections. */
385 list_init(&p->all_conns);
386 hmap_init(&p->controllers);
392 /* Initialize hooks. */
394 p->ofhooks = ofhooks;
398 p->ofhooks = &default_ofhooks;
400 p->ml = mac_learning_create();
403 /* Pick final datapath ID. */
404 p->datapath_id = pick_datapath_id(p);
405 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
412 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
414 uint64_t old_dpid = p->datapath_id;
415 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
416 if (p->datapath_id != old_dpid) {
417 struct ofconn *ofconn;
419 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
421 /* Force all active connections to reconnect, since there is no way to
422 * notify a controller that the datapath ID has changed. */
423 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
424 rconn_reconnect(ofconn->rconn);
430 is_discovery_controller(const struct ofproto_controller *c)
432 return !strcmp(c->target, "discover");
436 is_in_band_controller(const struct ofproto_controller *c)
438 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
441 /* Creates a new controller in 'ofproto'. Some of the settings are initially
442 * drawn from 'c', but update_controller() needs to be called later to finish
443 * the new ofconn's configuration. */
445 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
447 struct discovery *discovery;
448 struct ofconn *ofconn;
450 if (is_discovery_controller(c)) {
451 int error = discovery_create(c->accept_re, c->update_resolv_conf,
452 ofproto->dpif, ofproto->switch_status,
461 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
462 ofconn->pktbuf = pktbuf_create();
463 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
465 ofconn->discovery = discovery;
467 char *name = ofconn_make_name(ofproto, c->target);
468 rconn_connect(ofconn->rconn, c->target, name);
471 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
472 hash_string(c->target, 0));
475 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
476 * target or turn discovery on or off (these are done by creating new ofconns
477 * and deleting old ones), but it can update the rest of an ofconn's
480 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
482 struct ofproto *ofproto = ofconn->ofproto;
486 ofconn->band = (is_in_band_controller(c)
487 ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND);
489 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
491 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
492 rconn_set_probe_interval(ofconn->rconn, probe_interval);
494 if (ofconn->discovery) {
495 discovery_set_update_resolv_conf(ofconn->discovery,
496 c->update_resolv_conf);
497 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
500 for (i = 0; i < N_SCHEDULERS; i++) {
501 struct pinsched **s = &ofconn->schedulers[i];
503 if (c->rate_limit > 0) {
505 *s = pinsched_create(c->rate_limit, c->burst_limit,
506 ofproto->switch_status);
508 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
511 pinsched_destroy(*s);
518 ofconn_get_target(const struct ofconn *ofconn)
520 return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn);
523 static struct ofconn *
524 find_controller_by_target(struct ofproto *ofproto, const char *target)
526 struct ofconn *ofconn;
528 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
529 hash_string(target, 0), &ofproto->controllers) {
530 if (!strcmp(ofconn_get_target(ofconn), target)) {
538 update_in_band_remotes(struct ofproto *ofproto)
540 const struct ofconn *ofconn;
541 struct sockaddr_in *addrs;
542 size_t max_addrs, n_addrs;
546 /* Allocate enough memory for as many remotes as we could possibly have. */
547 max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers);
548 addrs = xmalloc(max_addrs * sizeof *addrs);
551 /* Add all the remotes. */
553 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) {
554 struct sockaddr_in *sin = &addrs[n_addrs];
556 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
560 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
561 if (sin->sin_addr.s_addr) {
562 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
565 if (ofconn->discovery) {
569 for (i = 0; i < ofproto->n_extra_remotes; i++) {
570 addrs[n_addrs++] = ofproto->extra_in_band_remotes[i];
573 /* Create or update or destroy in-band.
575 * Ordinarily we only enable in-band if there's at least one remote
576 * address, but discovery needs the in-band rules for DHCP to be installed
577 * even before we know any remote addresses. */
578 if (n_addrs || discovery) {
579 if (!ofproto->in_band) {
580 in_band_create(ofproto, ofproto->dpif, ofproto->switch_status,
583 if (ofproto->in_band) {
584 in_band_set_remotes(ofproto->in_band, addrs, n_addrs);
586 ofproto->next_in_band_update = time_msec() + 1000;
588 in_band_destroy(ofproto->in_band);
589 ofproto->in_band = NULL;
597 ofproto_set_controllers(struct ofproto *p,
598 const struct ofproto_controller *controllers,
599 size_t n_controllers)
601 struct shash new_controllers;
602 enum ofproto_fail_mode fail_mode;
603 struct ofconn *ofconn, *next;
607 shash_init(&new_controllers);
608 for (i = 0; i < n_controllers; i++) {
609 const struct ofproto_controller *c = &controllers[i];
611 shash_add_once(&new_controllers, c->target, &controllers[i]);
612 if (!find_controller_by_target(p, c->target)) {
613 add_controller(p, c);
617 fail_mode = OFPROTO_FAIL_STANDALONE;
619 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
621 struct ofproto_controller *c;
623 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
625 ofconn_destroy(ofconn);
627 update_controller(ofconn, c);
631 if (c->fail == OFPROTO_FAIL_SECURE) {
632 fail_mode = OFPROTO_FAIL_SECURE;
636 shash_destroy(&new_controllers);
638 update_in_band_remotes(p);
640 if (!hmap_is_empty(&p->controllers)
641 && fail_mode == OFPROTO_FAIL_STANDALONE) {
642 struct rconn **rconns;
646 p->fail_open = fail_open_create(p, p->switch_status);
650 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
651 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
652 rconns[n++] = ofconn->rconn;
655 fail_open_set_controllers(p->fail_open, rconns, n);
656 /* p->fail_open takes ownership of 'rconns'. */
658 fail_open_destroy(p->fail_open);
662 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
663 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
664 struct ofconn, hmap_node);
665 ofconn->ss = switch_status_register(p->switch_status, "remote",
666 rconn_status_cb, ofconn->rconn);
671 any_extras_changed(const struct ofproto *ofproto,
672 const struct sockaddr_in *extras, size_t n)
676 if (n != ofproto->n_extra_remotes) {
680 for (i = 0; i < n; i++) {
681 const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i];
682 const struct sockaddr_in *new = &extras[i];
684 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
685 old->sin_port != new->sin_port) {
693 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
694 * in-band control should guarantee access, in the same way that in-band
695 * control guarantees access to OpenFlow controllers. */
697 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
698 const struct sockaddr_in *extras, size_t n)
700 if (!any_extras_changed(ofproto, extras, n)) {
704 free(ofproto->extra_in_band_remotes);
705 ofproto->n_extra_remotes = n;
706 ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
708 update_in_band_remotes(ofproto);
712 ofproto_set_desc(struct ofproto *p,
713 const char *mfr_desc, const char *hw_desc,
714 const char *sw_desc, const char *serial_desc,
717 struct ofp_desc_stats *ods;
720 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
721 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
722 sizeof ods->mfr_desc);
725 p->mfr_desc = xstrdup(mfr_desc);
728 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
729 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
730 sizeof ods->hw_desc);
733 p->hw_desc = xstrdup(hw_desc);
736 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
737 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
738 sizeof ods->sw_desc);
741 p->sw_desc = xstrdup(sw_desc);
744 if (strlen(serial_desc) >= sizeof ods->serial_num) {
745 VLOG_WARN("truncating serial_desc, must be less than %zu "
747 sizeof ods->serial_num);
749 free(p->serial_desc);
750 p->serial_desc = xstrdup(serial_desc);
753 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
754 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
755 sizeof ods->dp_desc);
758 p->dp_desc = xstrdup(dp_desc);
763 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
764 const struct svec *svec)
766 struct pvconn **pvconns = *pvconnsp;
767 size_t n_pvconns = *n_pvconnsp;
771 for (i = 0; i < n_pvconns; i++) {
772 pvconn_close(pvconns[i]);
776 pvconns = xmalloc(svec->n * sizeof *pvconns);
778 for (i = 0; i < svec->n; i++) {
779 const char *name = svec->names[i];
780 struct pvconn *pvconn;
783 error = pvconn_open(name, &pvconn);
785 pvconns[n_pvconns++] = pvconn;
787 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
795 *n_pvconnsp = n_pvconns;
801 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
803 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
807 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
809 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
813 ofproto_set_netflow(struct ofproto *ofproto,
814 const struct netflow_options *nf_options)
816 if (nf_options && nf_options->collectors.n) {
817 if (!ofproto->netflow) {
818 ofproto->netflow = netflow_create();
820 return netflow_set_options(ofproto->netflow, nf_options);
822 netflow_destroy(ofproto->netflow);
823 ofproto->netflow = NULL;
829 ofproto_set_sflow(struct ofproto *ofproto,
830 const struct ofproto_sflow_options *oso)
832 struct ofproto_sflow *os = ofproto->sflow;
835 struct ofport *ofport;
836 unsigned int odp_port;
838 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
839 refresh_port_groups(ofproto);
840 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
841 ofproto_sflow_add_port(os, odp_port,
842 netdev_get_name(ofport->netdev));
845 ofproto_sflow_set_options(os, oso);
847 ofproto_sflow_destroy(os);
848 ofproto->sflow = NULL;
853 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
857 VLOG_WARN("STP is not yet implemented");
865 ofproto_get_datapath_id(const struct ofproto *ofproto)
867 return ofproto->datapath_id;
871 ofproto_has_controller(const struct ofproto *ofproto)
873 return !hmap_is_empty(&ofproto->controllers);
877 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
881 for (i = 0; i < ofproto->n_listeners; i++) {
882 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
887 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
891 for (i = 0; i < ofproto->n_snoops; i++) {
892 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
897 ofproto_destroy(struct ofproto *p)
899 struct ofconn *ofconn, *next_ofconn;
900 struct ofport *ofport;
901 unsigned int port_no;
908 /* Destroy fail-open and in-band early, since they touch the classifier. */
909 fail_open_destroy(p->fail_open);
912 in_band_destroy(p->in_band);
914 free(p->extra_in_band_remotes);
916 ofproto_flush_flows(p);
917 classifier_destroy(&p->cls);
919 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
921 ofconn_destroy(ofconn);
923 hmap_destroy(&p->controllers);
926 netdev_monitor_destroy(p->netdev_monitor);
927 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
930 shash_destroy(&p->port_by_name);
932 switch_status_destroy(p->switch_status);
933 netflow_destroy(p->netflow);
934 ofproto_sflow_destroy(p->sflow);
936 for (i = 0; i < p->n_listeners; i++) {
937 pvconn_close(p->listeners[i]);
941 for (i = 0; i < p->n_snoops; i++) {
942 pvconn_close(p->snoops[i]);
946 mac_learning_destroy(p->ml);
951 free(p->serial_desc);
954 port_array_destroy(&p->ports);
960 ofproto_run(struct ofproto *p)
962 int error = ofproto_run1(p);
964 error = ofproto_run2(p, false);
970 process_port_change(struct ofproto *ofproto, int error, char *devname)
972 if (error == ENOBUFS) {
973 reinit_ports(ofproto);
975 update_port(ofproto, devname);
980 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
981 * means that 'ofconn' is more interesting for monitoring than a lower return
984 snoop_preference(const struct ofconn *ofconn)
986 switch (ofconn->role) {
994 /* Shouldn't happen. */
999 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1000 * Connects this vconn to a controller. */
1002 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
1004 struct ofconn *ofconn, *best;
1006 /* Pick a controller for monitoring. */
1008 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
1009 if (ofconn->type == OFCONN_CONTROLLER
1010 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
1016 rconn_add_monitor(best->rconn, vconn);
1018 VLOG_INFO_RL(&rl, "no controller connection to snoop");
1024 ofproto_run1(struct ofproto *p)
1026 struct ofconn *ofconn, *next_ofconn;
1031 if (shash_is_empty(&p->port_by_name)) {
1035 for (i = 0; i < 50; i++) {
1039 error = dpif_recv(p->dpif, &buf);
1041 if (error == ENODEV) {
1042 /* Someone destroyed the datapath behind our back. The caller
1043 * better destroy us and give up, because we're just going to
1044 * spin from here on out. */
1045 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
1046 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
1047 dpif_name(p->dpif));
1053 handle_odp_msg(p, buf);
1056 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
1057 process_port_change(p, error, devname);
1059 while ((error = netdev_monitor_poll(p->netdev_monitor,
1060 &devname)) != EAGAIN) {
1061 process_port_change(p, error, devname);
1065 if (time_msec() >= p->next_in_band_update) {
1066 update_in_band_remotes(p);
1068 in_band_run(p->in_band);
1071 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
1073 ofconn_run(ofconn, p);
1076 /* Fail-open maintenance. Do this after processing the ofconns since
1077 * fail-open checks the status of the controller rconn. */
1079 fail_open_run(p->fail_open);
1082 for (i = 0; i < p->n_listeners; i++) {
1083 struct vconn *vconn;
1086 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
1088 struct rconn *rconn;
1091 rconn = rconn_create(60, 0);
1092 name = ofconn_make_name(p, vconn_get_name(vconn));
1093 rconn_connect_unreliably(rconn, vconn, name);
1096 ofconn_create(p, rconn, OFCONN_TRANSIENT);
1097 } else if (retval != EAGAIN) {
1098 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1102 for (i = 0; i < p->n_snoops; i++) {
1103 struct vconn *vconn;
1106 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
1108 add_snooper(p, vconn);
1109 } else if (retval != EAGAIN) {
1110 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
1114 if (time_msec() >= p->next_expiration) {
1115 COVERAGE_INC(ofproto_expiration);
1116 p->next_expiration = time_msec() + 1000;
1119 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
1121 /* Let the hook know that we're at a stable point: all outstanding data
1122 * in existing flows has been accounted to the account_cb. Thus, the
1123 * hook can now reasonably do operations that depend on having accurate
1124 * flow volume accounting (currently, that's just bond rebalancing). */
1125 if (p->ofhooks->account_checkpoint_cb) {
1126 p->ofhooks->account_checkpoint_cb(p->aux);
1131 netflow_run(p->netflow);
1134 ofproto_sflow_run(p->sflow);
1140 struct revalidate_cbdata {
1141 struct ofproto *ofproto;
1142 bool revalidate_all; /* Revalidate all exact-match rules? */
1143 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1144 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1148 ofproto_run2(struct ofproto *p, bool revalidate_all)
1150 if (p->need_revalidate || revalidate_all
1151 || !tag_set_is_empty(&p->revalidate_set)) {
1152 struct revalidate_cbdata cbdata;
1154 cbdata.revalidate_all = revalidate_all;
1155 cbdata.revalidate_subrules = p->need_revalidate;
1156 cbdata.revalidate_set = p->revalidate_set;
1157 tag_set_init(&p->revalidate_set);
1158 COVERAGE_INC(ofproto_revalidate);
1159 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1160 p->need_revalidate = false;
1167 ofproto_wait(struct ofproto *p)
1169 struct ofconn *ofconn;
1172 dpif_recv_wait(p->dpif);
1173 dpif_port_poll_wait(p->dpif);
1174 netdev_monitor_poll_wait(p->netdev_monitor);
1175 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1176 ofconn_wait(ofconn);
1179 poll_timer_wait_until(p->next_in_band_update);
1180 in_band_wait(p->in_band);
1183 fail_open_wait(p->fail_open);
1186 ofproto_sflow_wait(p->sflow);
1188 if (!tag_set_is_empty(&p->revalidate_set)) {
1189 poll_immediate_wake();
1191 if (p->need_revalidate) {
1192 /* Shouldn't happen, but if it does just go around again. */
1193 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1194 poll_immediate_wake();
1195 } else if (p->next_expiration != LLONG_MAX) {
1196 poll_timer_wait_until(p->next_expiration);
1198 for (i = 0; i < p->n_listeners; i++) {
1199 pvconn_wait(p->listeners[i]);
1201 for (i = 0; i < p->n_snoops; i++) {
1202 pvconn_wait(p->snoops[i]);
1207 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1209 tag_set_add(&ofproto->revalidate_set, tag);
1213 ofproto_get_revalidate_set(struct ofproto *ofproto)
1215 return &ofproto->revalidate_set;
1219 ofproto_is_alive(const struct ofproto *p)
1221 return !hmap_is_empty(&p->controllers);
1225 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1226 const union ofp_action *actions, size_t n_actions,
1227 const struct ofpbuf *packet)
1229 struct odp_actions odp_actions;
1232 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1238 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1240 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1241 odp_actions.n_actions, packet);
1246 ofproto_add_flow(struct ofproto *p,
1247 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1248 const union ofp_action *actions, size_t n_actions,
1252 rule = rule_create(p, NULL, actions, n_actions,
1253 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1255 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1256 rule_insert(p, rule, NULL, 0);
1260 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1261 uint32_t wildcards, unsigned int priority)
1265 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1269 rule_remove(ofproto, rule);
1274 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1276 struct rule *rule = rule_from_cls_rule(rule_);
1277 struct ofproto *ofproto = ofproto_;
1279 /* Mark the flow as not installed, even though it might really be
1280 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1281 * There is no point in uninstalling it individually since we are about to
1282 * blow away all the flows with dpif_flow_flush(). */
1283 rule->installed = false;
1285 rule_remove(ofproto, rule);
1289 ofproto_flush_flows(struct ofproto *ofproto)
1291 COVERAGE_INC(ofproto_flush);
1292 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1293 dpif_flow_flush(ofproto->dpif);
1294 if (ofproto->in_band) {
1295 in_band_flushed(ofproto->in_band);
1297 if (ofproto->fail_open) {
1298 fail_open_flushed(ofproto->fail_open);
1303 reinit_ports(struct ofproto *p)
1305 struct svec devnames;
1306 struct ofport *ofport;
1307 unsigned int port_no;
1308 struct odp_port *odp_ports;
1312 svec_init(&devnames);
1313 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1314 svec_add (&devnames, (char *) ofport->opp.name);
1316 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1317 for (i = 0; i < n_odp_ports; i++) {
1318 svec_add (&devnames, odp_ports[i].devname);
1322 svec_sort_unique(&devnames);
1323 for (i = 0; i < devnames.n; i++) {
1324 update_port(p, devnames.names[i]);
1326 svec_destroy(&devnames);
1330 refresh_port_group(struct ofproto *p, unsigned int group)
1334 struct ofport *port;
1335 unsigned int port_no;
1337 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1339 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1341 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1342 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1343 ports[n_ports++] = port_no;
1346 dpif_port_group_set(p->dpif, group, ports, n_ports);
1353 refresh_port_groups(struct ofproto *p)
1355 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1356 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1358 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1362 static struct ofport *
1363 make_ofport(const struct odp_port *odp_port)
1365 struct netdev_options netdev_options;
1366 enum netdev_flags flags;
1367 struct ofport *ofport;
1368 struct netdev *netdev;
1372 memset(&netdev_options, 0, sizeof netdev_options);
1373 netdev_options.name = odp_port->devname;
1374 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1376 error = netdev_open(&netdev_options, &netdev);
1378 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1379 "cannot be opened (%s)",
1380 odp_port->devname, odp_port->port,
1381 odp_port->devname, strerror(error));
1385 ofport = xmalloc(sizeof *ofport);
1386 ofport->netdev = netdev;
1387 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1388 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1389 memcpy(ofport->opp.name, odp_port->devname,
1390 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1391 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1393 netdev_get_flags(netdev, &flags);
1394 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1396 netdev_get_carrier(netdev, &carrier);
1397 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1399 netdev_get_features(netdev,
1400 &ofport->opp.curr, &ofport->opp.advertised,
1401 &ofport->opp.supported, &ofport->opp.peer);
1406 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1408 if (port_array_get(&p->ports, odp_port->port)) {
1409 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1412 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1413 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1422 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1424 const struct ofp_phy_port *a = &a_->opp;
1425 const struct ofp_phy_port *b = &b_->opp;
1427 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1428 return (a->port_no == b->port_no
1429 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1430 && !strcmp((char *) a->name, (char *) b->name)
1431 && a->state == b->state
1432 && a->config == b->config
1433 && a->curr == b->curr
1434 && a->advertised == b->advertised
1435 && a->supported == b->supported
1436 && a->peer == b->peer);
1440 send_port_status(struct ofproto *p, const struct ofport *ofport,
1443 /* XXX Should limit the number of queued port status change messages. */
1444 struct ofconn *ofconn;
1445 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1446 struct ofp_port_status *ops;
1449 if (!ofconn_receives_async_msgs(ofconn)) {
1453 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1454 ops->reason = reason;
1455 ops->desc = ofport->opp;
1456 hton_ofp_phy_port(&ops->desc);
1457 queue_tx(b, ofconn, NULL);
1459 if (p->ofhooks->port_changed_cb) {
1460 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1465 ofport_install(struct ofproto *p, struct ofport *ofport)
1467 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1468 const char *netdev_name = (const char *) ofport->opp.name;
1470 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1471 port_array_set(&p->ports, odp_port, ofport);
1472 shash_add(&p->port_by_name, netdev_name, ofport);
1474 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1479 ofport_remove(struct ofproto *p, struct ofport *ofport)
1481 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1483 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1484 port_array_delete(&p->ports, odp_port);
1485 shash_delete(&p->port_by_name,
1486 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1488 ofproto_sflow_del_port(p->sflow, odp_port);
1493 ofport_free(struct ofport *ofport)
1496 netdev_close(ofport->netdev);
1502 update_port(struct ofproto *p, const char *devname)
1504 struct odp_port odp_port;
1505 struct ofport *old_ofport;
1506 struct ofport *new_ofport;
1509 COVERAGE_INC(ofproto_update_port);
1511 /* Query the datapath for port information. */
1512 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1514 /* Find the old ofport. */
1515 old_ofport = shash_find_data(&p->port_by_name, devname);
1518 /* There's no port named 'devname' but there might be a port with
1519 * the same port number. This could happen if a port is deleted
1520 * and then a new one added in its place very quickly, or if a port
1521 * is renamed. In the former case we want to send an OFPPR_DELETE
1522 * and an OFPPR_ADD, and in the latter case we want to send a
1523 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1524 * the old port's ifindex against the new port, or perhaps less
1525 * reliably but more portably by comparing the old port's MAC
1526 * against the new port's MAC. However, this code isn't that smart
1527 * and always sends an OFPPR_MODIFY (XXX). */
1528 old_ofport = port_array_get(&p->ports, odp_port.port);
1530 } else if (error != ENOENT && error != ENODEV) {
1531 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1532 "%s", strerror(error));
1536 /* Create a new ofport. */
1537 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1539 /* Eliminate a few pathological cases. */
1540 if (!old_ofport && !new_ofport) {
1542 } else if (old_ofport && new_ofport) {
1543 /* Most of the 'config' bits are OpenFlow soft state, but
1544 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1545 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1546 * leaves the other bits 0.) */
1547 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1549 if (ofport_equal(old_ofport, new_ofport)) {
1550 /* False alarm--no change. */
1551 ofport_free(new_ofport);
1556 /* Now deal with the normal cases. */
1558 ofport_remove(p, old_ofport);
1561 ofport_install(p, new_ofport);
1563 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1564 (!old_ofport ? OFPPR_ADD
1565 : !new_ofport ? OFPPR_DELETE
1567 ofport_free(old_ofport);
1569 /* Update port groups. */
1570 refresh_port_groups(p);
1574 init_ports(struct ofproto *p)
1576 struct odp_port *ports;
1581 error = dpif_port_list(p->dpif, &ports, &n_ports);
1586 for (i = 0; i < n_ports; i++) {
1587 const struct odp_port *odp_port = &ports[i];
1588 if (!ofport_conflicts(p, odp_port)) {
1589 struct ofport *ofport = make_ofport(odp_port);
1591 ofport_install(p, ofport);
1596 refresh_port_groups(p);
1600 static struct ofconn *
1601 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1603 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1604 ofconn->ofproto = p;
1605 list_push_back(&p->all_conns, &ofconn->node);
1606 ofconn->rconn = rconn;
1607 ofconn->type = type;
1608 ofconn->role = NX_ROLE_OTHER;
1609 ofconn->packet_in_counter = rconn_packet_counter_create ();
1610 ofconn->pktbuf = NULL;
1611 ofconn->miss_send_len = 0;
1612 ofconn->reply_counter = rconn_packet_counter_create ();
1617 ofconn_destroy(struct ofconn *ofconn)
1619 if (ofconn->type == OFCONN_CONTROLLER) {
1620 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1622 discovery_destroy(ofconn->discovery);
1624 list_remove(&ofconn->node);
1625 switch_status_unregister(ofconn->ss);
1626 rconn_destroy(ofconn->rconn);
1627 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1628 rconn_packet_counter_destroy(ofconn->reply_counter);
1629 pktbuf_destroy(ofconn->pktbuf);
1634 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1639 if (ofconn->discovery) {
1640 char *controller_name;
1641 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1642 discovery_question_connectivity(ofconn->discovery);
1644 if (discovery_run(ofconn->discovery, &controller_name)) {
1645 if (controller_name) {
1646 char *ofconn_name = ofconn_make_name(p, controller_name);
1647 rconn_connect(ofconn->rconn, controller_name, ofconn_name);
1650 rconn_disconnect(ofconn->rconn);
1655 for (i = 0; i < N_SCHEDULERS; i++) {
1656 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1659 rconn_run(ofconn->rconn);
1661 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1662 /* Limit the number of iterations to prevent other tasks from
1664 for (iteration = 0; iteration < 50; iteration++) {
1665 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1670 fail_open_maybe_recover(p->fail_open);
1672 handle_openflow(ofconn, p, of_msg);
1673 ofpbuf_delete(of_msg);
1677 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1678 ofconn_destroy(ofconn);
1683 ofconn_wait(struct ofconn *ofconn)
1687 if (ofconn->discovery) {
1688 discovery_wait(ofconn->discovery);
1690 for (i = 0; i < N_SCHEDULERS; i++) {
1691 pinsched_wait(ofconn->schedulers[i]);
1693 rconn_run_wait(ofconn->rconn);
1694 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1695 rconn_recv_wait(ofconn->rconn);
1697 COVERAGE_INC(ofproto_ofconn_stuck);
1701 /* Returns true if 'ofconn' should receive asynchronous messages. */
1703 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1705 if (ofconn->type == OFCONN_CONTROLLER) {
1706 /* Ordinary controllers always get asynchronous messages unless they
1707 * have configured themselves as "slaves". */
1708 return ofconn->role != NX_ROLE_SLAVE;
1710 /* Transient connections don't get asynchronous messages unless they
1711 * have explicitly asked for them by setting a nonzero miss send
1713 return ofconn->miss_send_len > 0;
1717 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1718 * and 'target', suitable for use in log messages for identifying the
1721 * The name is dynamically allocated. The caller should free it (with free())
1722 * when it is no longer needed. */
1724 ofconn_make_name(const struct ofproto *ofproto, const char *target)
1726 return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target);
1729 /* Caller is responsible for initializing the 'cr' member of the returned
1731 static struct rule *
1732 rule_create(struct ofproto *ofproto, struct rule *super,
1733 const union ofp_action *actions, size_t n_actions,
1734 uint16_t idle_timeout, uint16_t hard_timeout,
1735 uint64_t flow_cookie, bool send_flow_removed)
1737 struct rule *rule = xzalloc(sizeof *rule);
1738 rule->idle_timeout = idle_timeout;
1739 rule->hard_timeout = hard_timeout;
1740 rule->flow_cookie = flow_cookie;
1741 rule->used = rule->created = time_msec();
1742 rule->send_flow_removed = send_flow_removed;
1743 rule->super = super;
1745 list_push_back(&super->list, &rule->list);
1747 list_init(&rule->list);
1749 rule->n_actions = n_actions;
1750 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1751 netflow_flow_clear(&rule->nf_flow);
1752 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1757 static struct rule *
1758 rule_from_cls_rule(const struct cls_rule *cls_rule)
1760 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1764 rule_free(struct rule *rule)
1766 free(rule->actions);
1767 free(rule->odp_actions);
1771 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1772 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1773 * through all of its subrules and revalidates them, destroying any that no
1774 * longer has a super-rule (which is probably all of them).
1776 * Before calling this function, the caller must make have removed 'rule' from
1777 * the classifier. If 'rule' is an exact-match rule, the caller is also
1778 * responsible for ensuring that it has been uninstalled from the datapath. */
1780 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1783 struct rule *subrule, *next;
1784 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1785 revalidate_rule(ofproto, subrule);
1788 list_remove(&rule->list);
1794 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1796 const union ofp_action *oa;
1797 struct actions_iterator i;
1799 if (out_port == htons(OFPP_NONE)) {
1802 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1803 oa = actions_next(&i)) {
1804 if (action_outputs_to_port(oa, out_port)) {
1811 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1812 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1814 * The flow that 'packet' actually contains does not need to actually match
1815 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1816 * the packet and byte counters for 'rule' will be credited for the packet sent
1817 * out whether or not the packet actually matches 'rule'.
1819 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1820 * the caller must already have accurately composed ODP actions for it given
1821 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1822 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1823 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1824 * actions and apply them to 'packet'. */
1826 rule_execute(struct ofproto *ofproto, struct rule *rule,
1827 struct ofpbuf *packet, const flow_t *flow)
1829 const union odp_action *actions;
1831 struct odp_actions a;
1833 /* Grab or compose the ODP actions.
1835 * The special case for an exact-match 'rule' where 'flow' is not the
1836 * rule's flow is important to avoid, e.g., sending a packet out its input
1837 * port simply because the ODP actions were composed for the wrong
1839 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1840 struct rule *super = rule->super ? rule->super : rule;
1841 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1842 packet, &a, NULL, 0, NULL)) {
1845 actions = a.actions;
1846 n_actions = a.n_actions;
1848 actions = rule->odp_actions;
1849 n_actions = rule->n_odp_actions;
1852 /* Execute the ODP actions. */
1853 if (!dpif_execute(ofproto->dpif, flow->in_port,
1854 actions, n_actions, packet)) {
1855 struct odp_flow_stats stats;
1856 flow_extract_stats(flow, packet, &stats);
1857 update_stats(ofproto, rule, &stats);
1858 rule->used = time_msec();
1859 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1864 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1867 struct rule *displaced_rule;
1869 /* Insert the rule in the classifier. */
1870 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1871 if (!rule->cr.wc.wildcards) {
1872 rule_make_actions(p, rule, packet);
1875 /* Send the packet and credit it to the rule. */
1878 flow_extract(packet, 0, in_port, &flow);
1879 rule_execute(p, rule, packet, &flow);
1882 /* Install the rule in the datapath only after sending the packet, to
1883 * avoid packet reordering. */
1884 if (rule->cr.wc.wildcards) {
1885 COVERAGE_INC(ofproto_add_wc_flow);
1886 p->need_revalidate = true;
1888 rule_install(p, rule, displaced_rule);
1891 /* Free the rule that was displaced, if any. */
1892 if (displaced_rule) {
1893 rule_destroy(p, displaced_rule);
1897 static struct rule *
1898 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1901 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1902 rule->idle_timeout, rule->hard_timeout,
1904 COVERAGE_INC(ofproto_subrule_create);
1905 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1906 : rule->cr.priority), &subrule->cr);
1907 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1913 rule_remove(struct ofproto *ofproto, struct rule *rule)
1915 if (rule->cr.wc.wildcards) {
1916 COVERAGE_INC(ofproto_del_wc_flow);
1917 ofproto->need_revalidate = true;
1919 rule_uninstall(ofproto, rule);
1921 classifier_remove(&ofproto->cls, &rule->cr);
1922 rule_destroy(ofproto, rule);
1925 /* Returns true if the actions changed, false otherwise. */
1927 rule_make_actions(struct ofproto *p, struct rule *rule,
1928 const struct ofpbuf *packet)
1930 const struct rule *super;
1931 struct odp_actions a;
1934 assert(!rule->cr.wc.wildcards);
1936 super = rule->super ? rule->super : rule;
1938 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1939 packet, &a, &rule->tags, &rule->may_install,
1940 &rule->nf_flow.output_iface);
1942 actions_len = a.n_actions * sizeof *a.actions;
1943 if (rule->n_odp_actions != a.n_actions
1944 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1945 COVERAGE_INC(ofproto_odp_unchanged);
1946 free(rule->odp_actions);
1947 rule->n_odp_actions = a.n_actions;
1948 rule->odp_actions = xmemdup(a.actions, actions_len);
1956 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1957 struct odp_flow_put *put)
1959 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1960 put->flow.key = rule->cr.flow;
1961 put->flow.actions = rule->odp_actions;
1962 put->flow.n_actions = rule->n_odp_actions;
1963 put->flow.flags = 0;
1965 return dpif_flow_put(ofproto->dpif, put);
1969 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1971 assert(!rule->cr.wc.wildcards);
1973 if (rule->may_install) {
1974 struct odp_flow_put put;
1975 if (!do_put_flow(p, rule,
1976 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1978 rule->installed = true;
1979 if (displaced_rule) {
1980 update_stats(p, displaced_rule, &put.flow.stats);
1981 rule_post_uninstall(p, displaced_rule);
1984 } else if (displaced_rule) {
1985 rule_uninstall(p, displaced_rule);
1990 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1992 if (rule->installed) {
1993 struct odp_flow_put put;
1994 COVERAGE_INC(ofproto_dp_missed);
1995 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1997 rule_install(ofproto, rule, NULL);
2002 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
2004 bool actions_changed;
2005 uint16_t new_out_iface, old_out_iface;
2007 old_out_iface = rule->nf_flow.output_iface;
2008 actions_changed = rule_make_actions(ofproto, rule, NULL);
2010 if (rule->may_install) {
2011 if (rule->installed) {
2012 if (actions_changed) {
2013 struct odp_flow_put put;
2014 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
2015 | ODPPF_ZERO_STATS, &put);
2016 update_stats(ofproto, rule, &put.flow.stats);
2018 /* Temporarily set the old output iface so that NetFlow
2019 * messages have the correct output interface for the old
2021 new_out_iface = rule->nf_flow.output_iface;
2022 rule->nf_flow.output_iface = old_out_iface;
2023 rule_post_uninstall(ofproto, rule);
2024 rule->nf_flow.output_iface = new_out_iface;
2027 rule_install(ofproto, rule, NULL);
2030 rule_uninstall(ofproto, rule);
2035 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
2037 uint64_t total_bytes = rule->byte_count + extra_bytes;
2039 if (ofproto->ofhooks->account_flow_cb
2040 && total_bytes > rule->accounted_bytes)
2042 ofproto->ofhooks->account_flow_cb(
2043 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
2044 total_bytes - rule->accounted_bytes, ofproto->aux);
2045 rule->accounted_bytes = total_bytes;
2050 rule_uninstall(struct ofproto *p, struct rule *rule)
2052 assert(!rule->cr.wc.wildcards);
2053 if (rule->installed) {
2054 struct odp_flow odp_flow;
2056 odp_flow.key = rule->cr.flow;
2057 odp_flow.actions = NULL;
2058 odp_flow.n_actions = 0;
2060 if (!dpif_flow_del(p->dpif, &odp_flow)) {
2061 update_stats(p, rule, &odp_flow.stats);
2063 rule->installed = false;
2065 rule_post_uninstall(p, rule);
2070 is_controller_rule(struct rule *rule)
2072 /* If the only action is send to the controller then don't report
2073 * NetFlow expiration messages since it is just part of the control
2074 * logic for the network and not real traffic. */
2078 && rule->super->n_actions == 1
2079 && action_outputs_to_port(&rule->super->actions[0],
2080 htons(OFPP_CONTROLLER)));
2084 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
2086 struct rule *super = rule->super;
2088 rule_account(ofproto, rule, 0);
2090 if (ofproto->netflow && !is_controller_rule(rule)) {
2091 struct ofexpired expired;
2092 expired.flow = rule->cr.flow;
2093 expired.packet_count = rule->packet_count;
2094 expired.byte_count = rule->byte_count;
2095 expired.used = rule->used;
2096 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
2099 super->packet_count += rule->packet_count;
2100 super->byte_count += rule->byte_count;
2102 /* Reset counters to prevent double counting if the rule ever gets
2104 rule->packet_count = 0;
2105 rule->byte_count = 0;
2106 rule->accounted_bytes = 0;
2108 netflow_flow_clear(&rule->nf_flow);
2113 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
2114 struct rconn_packet_counter *counter)
2116 update_openflow_length(msg);
2117 if (rconn_send(ofconn->rconn, msg, counter)) {
2123 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
2124 int error, const void *data, size_t len)
2127 struct ofp_error_msg *oem;
2129 if (!(error >> 16)) {
2130 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
2135 COVERAGE_INC(ofproto_error);
2136 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
2137 oh ? oh->xid : 0, &buf);
2138 oem->type = htons((unsigned int) error >> 16);
2139 oem->code = htons(error & 0xffff);
2140 memcpy(oem->data, data, len);
2141 queue_tx(buf, ofconn, ofconn->reply_counter);
2145 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
2148 size_t oh_length = ntohs(oh->length);
2149 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
2153 hton_ofp_phy_port(struct ofp_phy_port *opp)
2155 opp->port_no = htons(opp->port_no);
2156 opp->config = htonl(opp->config);
2157 opp->state = htonl(opp->state);
2158 opp->curr = htonl(opp->curr);
2159 opp->advertised = htonl(opp->advertised);
2160 opp->supported = htonl(opp->supported);
2161 opp->peer = htonl(opp->peer);
2165 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2167 struct ofp_header *rq = oh;
2168 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2173 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2174 struct ofp_header *oh)
2176 struct ofp_switch_features *osf;
2178 unsigned int port_no;
2179 struct ofport *port;
2181 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2182 osf->datapath_id = htonll(p->datapath_id);
2183 osf->n_buffers = htonl(pktbuf_capacity());
2185 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2186 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2187 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2188 (1u << OFPAT_SET_VLAN_VID) |
2189 (1u << OFPAT_SET_VLAN_PCP) |
2190 (1u << OFPAT_STRIP_VLAN) |
2191 (1u << OFPAT_SET_DL_SRC) |
2192 (1u << OFPAT_SET_DL_DST) |
2193 (1u << OFPAT_SET_NW_SRC) |
2194 (1u << OFPAT_SET_NW_DST) |
2195 (1u << OFPAT_SET_NW_TOS) |
2196 (1u << OFPAT_SET_TP_SRC) |
2197 (1u << OFPAT_SET_TP_DST) |
2198 (1u << OFPAT_ENQUEUE));
2200 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2201 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2204 queue_tx(buf, ofconn, ofconn->reply_counter);
2209 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2210 struct ofp_header *oh)
2213 struct ofp_switch_config *osc;
2217 /* Figure out flags. */
2218 dpif_get_drop_frags(p->dpif, &drop_frags);
2219 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2222 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2223 osc->flags = htons(flags);
2224 osc->miss_send_len = htons(ofconn->miss_send_len);
2225 queue_tx(buf, ofconn, ofconn->reply_counter);
2231 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2232 struct ofp_switch_config *osc)
2237 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2241 flags = ntohs(osc->flags);
2243 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) {
2244 switch (flags & OFPC_FRAG_MASK) {
2245 case OFPC_FRAG_NORMAL:
2246 dpif_set_drop_frags(p->dpif, false);
2248 case OFPC_FRAG_DROP:
2249 dpif_set_drop_frags(p->dpif, true);
2252 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2258 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2264 add_output_group_action(struct odp_actions *actions, uint16_t group,
2265 uint16_t *nf_output_iface)
2267 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2269 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2270 *nf_output_iface = NF_OUT_FLOOD;
2275 add_controller_action(struct odp_actions *actions,
2276 const struct ofp_action_output *oao)
2278 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2279 a->controller.arg = ntohs(oao->max_len);
2282 struct action_xlate_ctx {
2284 flow_t flow; /* Flow to which these actions correspond. */
2285 int recurse; /* Recursion level, via xlate_table_action. */
2286 struct ofproto *ofproto;
2287 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2288 * null pointer if we are revalidating
2289 * without a packet to refer to. */
2292 struct odp_actions *out; /* Datapath actions. */
2293 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2294 bool may_set_up_flow; /* True ordinarily; false if the actions must
2295 * be reassessed for every packet. */
2296 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2299 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2300 struct action_xlate_ctx *ctx);
2303 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2305 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2308 if (ofport->opp.config & OFPPC_NO_FWD) {
2309 /* Forwarding disabled on port. */
2314 * We don't have an ofport record for this port, but it doesn't hurt to
2315 * allow forwarding to it anyhow. Maybe such a port will appear later
2316 * and we're pre-populating the flow table.
2320 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2321 ctx->nf_output_iface = port;
2324 static struct rule *
2325 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2328 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2330 /* The rule we found might not be valid, since we could be in need of
2331 * revalidation. If it is not valid, don't return it. */
2334 && ofproto->need_revalidate
2335 && !revalidate_rule(ofproto, rule)) {
2336 COVERAGE_INC(ofproto_invalidated);
2344 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2346 if (!ctx->recurse) {
2347 uint16_t old_in_port;
2350 /* Look up a flow with 'in_port' as the input port. Then restore the
2351 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2352 * have surprising behavior). */
2353 old_in_port = ctx->flow.in_port;
2354 ctx->flow.in_port = in_port;
2355 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2356 ctx->flow.in_port = old_in_port;
2364 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2371 xlate_output_action(struct action_xlate_ctx *ctx,
2372 const struct ofp_action_output *oao)
2375 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2377 ctx->nf_output_iface = NF_OUT_DROP;
2379 switch (ntohs(oao->port)) {
2381 add_output_action(ctx, ctx->flow.in_port);
2384 xlate_table_action(ctx, ctx->flow.in_port);
2387 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2388 ctx->out, ctx->tags,
2389 &ctx->nf_output_iface,
2390 ctx->ofproto->aux)) {
2391 COVERAGE_INC(ofproto_uninstallable);
2392 ctx->may_set_up_flow = false;
2396 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2397 &ctx->nf_output_iface);
2400 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2402 case OFPP_CONTROLLER:
2403 add_controller_action(ctx->out, oao);
2406 add_output_action(ctx, ODPP_LOCAL);
2409 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2410 if (odp_port != ctx->flow.in_port) {
2411 add_output_action(ctx, odp_port);
2416 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2417 ctx->nf_output_iface = NF_OUT_FLOOD;
2418 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2419 ctx->nf_output_iface = prev_nf_output_iface;
2420 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2421 ctx->nf_output_iface != NF_OUT_FLOOD) {
2422 ctx->nf_output_iface = NF_OUT_MULTI;
2426 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2427 * optimization, because we're going to add another action that sets the
2428 * priority immediately after, or because there are no actions following the
2431 remove_pop_action(struct action_xlate_ctx *ctx)
2433 size_t n = ctx->out->n_actions;
2434 if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) {
2435 ctx->out->n_actions--;
2440 xlate_enqueue_action(struct action_xlate_ctx *ctx,
2441 const struct ofp_action_enqueue *oae)
2443 uint16_t ofp_port, odp_port;
2445 /* Figure out ODP output port. */
2446 ofp_port = ntohs(oae->port);
2447 if (ofp_port != OFPP_IN_PORT) {
2448 odp_port = ofp_port_to_odp_port(ofp_port);
2450 odp_port = ctx->flow.in_port;
2453 /* Add ODP actions. */
2454 remove_pop_action(ctx);
2455 odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority
2456 = TC_H_MAKE(1, ntohl(oae->queue_id)); /* XXX */
2457 add_output_action(ctx, odp_port);
2458 odp_actions_add(ctx->out, ODPAT_POP_PRIORITY);
2460 /* Update NetFlow output port. */
2461 if (ctx->nf_output_iface == NF_OUT_DROP) {
2462 ctx->nf_output_iface = odp_port;
2463 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
2464 ctx->nf_output_iface = NF_OUT_MULTI;
2469 xlate_nicira_action(struct action_xlate_ctx *ctx,
2470 const struct nx_action_header *nah)
2472 const struct nx_action_resubmit *nar;
2473 const struct nx_action_set_tunnel *nast;
2474 union odp_action *oa;
2475 int subtype = ntohs(nah->subtype);
2477 assert(nah->vendor == htonl(NX_VENDOR_ID));
2479 case NXAST_RESUBMIT:
2480 nar = (const struct nx_action_resubmit *) nah;
2481 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2484 case NXAST_SET_TUNNEL:
2485 nast = (const struct nx_action_set_tunnel *) nah;
2486 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2487 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2490 /* If you add a new action here that modifies flow data, don't forget to
2491 * update the flow key in ctx->flow at the same time. */
2494 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2500 do_xlate_actions(const union ofp_action *in, size_t n_in,
2501 struct action_xlate_ctx *ctx)
2503 struct actions_iterator iter;
2504 const union ofp_action *ia;
2505 const struct ofport *port;
2507 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2508 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2509 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2510 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2511 /* Drop this flow. */
2515 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2516 uint16_t type = ntohs(ia->type);
2517 union odp_action *oa;
2521 xlate_output_action(ctx, &ia->output);
2524 case OFPAT_SET_VLAN_VID:
2525 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2526 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2529 case OFPAT_SET_VLAN_PCP:
2530 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2531 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2534 case OFPAT_STRIP_VLAN:
2535 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2536 ctx->flow.dl_vlan = htons(OFP_VLAN_NONE);
2537 ctx->flow.dl_vlan_pcp = 0;
2540 case OFPAT_SET_DL_SRC:
2541 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2542 memcpy(oa->dl_addr.dl_addr,
2543 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2544 memcpy(ctx->flow.dl_src,
2545 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2548 case OFPAT_SET_DL_DST:
2549 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2550 memcpy(oa->dl_addr.dl_addr,
2551 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2552 memcpy(ctx->flow.dl_dst,
2553 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2556 case OFPAT_SET_NW_SRC:
2557 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2558 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2561 case OFPAT_SET_NW_DST:
2562 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2563 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2566 case OFPAT_SET_NW_TOS:
2567 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2568 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2571 case OFPAT_SET_TP_SRC:
2572 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2573 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2576 case OFPAT_SET_TP_DST:
2577 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2578 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2582 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2586 xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
2590 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2597 xlate_actions(const union ofp_action *in, size_t n_in,
2598 const flow_t *flow, struct ofproto *ofproto,
2599 const struct ofpbuf *packet,
2600 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2601 uint16_t *nf_output_iface)
2603 tag_type no_tags = 0;
2604 struct action_xlate_ctx ctx;
2605 COVERAGE_INC(ofproto_ofp2odp);
2606 odp_actions_init(out);
2609 ctx.ofproto = ofproto;
2610 ctx.packet = packet;
2612 ctx.tags = tags ? tags : &no_tags;
2613 ctx.may_set_up_flow = true;
2614 ctx.nf_output_iface = NF_OUT_DROP;
2615 do_xlate_actions(in, n_in, &ctx);
2616 remove_pop_action(&ctx);
2618 /* Check with in-band control to see if we're allowed to set up this
2620 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2621 ctx.may_set_up_flow = false;
2624 if (may_set_up_flow) {
2625 *may_set_up_flow = ctx.may_set_up_flow;
2627 if (nf_output_iface) {
2628 *nf_output_iface = ctx.nf_output_iface;
2630 if (odp_actions_overflow(out)) {
2631 odp_actions_init(out);
2632 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2637 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2638 * error message code (composed with ofp_mkerr()) for the caller to propagate
2639 * upward. Otherwise, returns 0.
2641 * 'oh' is used to make log messages more informative. */
2643 reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh)
2645 if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) {
2646 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
2649 type_name = ofp_message_type_to_string(oh->type);
2650 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
2654 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2661 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2662 struct ofp_header *oh)
2664 struct ofp_packet_out *opo;
2665 struct ofpbuf payload, *buffer;
2666 struct odp_actions actions;
2672 error = reject_slave_controller(ofconn, oh);
2677 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2681 opo = (struct ofp_packet_out *) oh;
2683 COVERAGE_INC(ofproto_packet_out);
2684 if (opo->buffer_id != htonl(UINT32_MAX)) {
2685 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2687 if (error || !buffer) {
2695 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2696 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2697 &flow, p, &payload, &actions, NULL, NULL, NULL);
2702 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2704 ofpbuf_delete(buffer);
2710 update_port_config(struct ofproto *p, struct ofport *port,
2711 uint32_t config, uint32_t mask)
2713 mask &= config ^ port->opp.config;
2714 if (mask & OFPPC_PORT_DOWN) {
2715 if (config & OFPPC_PORT_DOWN) {
2716 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2718 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2721 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2722 if (mask & REVALIDATE_BITS) {
2723 COVERAGE_INC(ofproto_costly_flags);
2724 port->opp.config ^= mask & REVALIDATE_BITS;
2725 p->need_revalidate = true;
2727 #undef REVALIDATE_BITS
2728 if (mask & OFPPC_NO_FLOOD) {
2729 port->opp.config ^= OFPPC_NO_FLOOD;
2730 refresh_port_groups(p);
2732 if (mask & OFPPC_NO_PACKET_IN) {
2733 port->opp.config ^= OFPPC_NO_PACKET_IN;
2738 handle_port_mod(struct ofproto *p, struct ofconn *ofconn,
2739 struct ofp_header *oh)
2741 const struct ofp_port_mod *opm;
2742 struct ofport *port;
2745 error = reject_slave_controller(ofconn, oh);
2749 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2753 opm = (struct ofp_port_mod *) oh;
2755 port = port_array_get(&p->ports,
2756 ofp_port_to_odp_port(ntohs(opm->port_no)));
2758 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2759 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2760 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2762 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2763 if (opm->advertise) {
2764 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2770 static struct ofpbuf *
2771 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2773 struct ofp_stats_reply *osr;
2776 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2777 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2779 osr->flags = htons(0);
2783 static struct ofpbuf *
2784 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2786 return make_stats_reply(request->header.xid, request->type, body_len);
2790 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2792 struct ofpbuf *msg = *msgp;
2793 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2794 if (nbytes + msg->size > UINT16_MAX) {
2795 struct ofp_stats_reply *reply = msg->data;
2796 reply->flags = htons(OFPSF_REPLY_MORE);
2797 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2798 queue_tx(msg, ofconn, ofconn->reply_counter);
2800 return ofpbuf_put_uninit(*msgp, nbytes);
2804 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2805 struct ofp_stats_request *request)
2807 struct ofp_desc_stats *ods;
2810 msg = start_stats_reply(request, sizeof *ods);
2811 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2812 memset(ods, 0, sizeof *ods);
2813 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2814 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2815 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2816 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2817 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2818 queue_tx(msg, ofconn, ofconn->reply_counter);
2824 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2826 struct rule *rule = rule_from_cls_rule(cls_rule);
2827 int *n_subrules = n_subrules_;
2835 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2836 struct ofp_stats_request *request)
2838 struct ofp_table_stats *ots;
2840 struct odp_stats dpstats;
2841 int n_exact, n_subrules, n_wild;
2843 msg = start_stats_reply(request, sizeof *ots * 2);
2845 /* Count rules of various kinds. */
2847 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2848 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2849 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2852 dpif_get_dp_stats(p->dpif, &dpstats);
2853 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2854 memset(ots, 0, sizeof *ots);
2855 ots->table_id = TABLEID_HASH;
2856 strcpy(ots->name, "hash");
2857 ots->wildcards = htonl(0);
2858 ots->max_entries = htonl(dpstats.max_capacity);
2859 ots->active_count = htonl(n_exact);
2860 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2862 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2864 /* Classifier table. */
2865 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2866 memset(ots, 0, sizeof *ots);
2867 ots->table_id = TABLEID_CLASSIFIER;
2868 strcpy(ots->name, "classifier");
2869 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2871 ots->max_entries = htonl(65536);
2872 ots->active_count = htonl(n_wild);
2873 ots->lookup_count = htonll(0); /* XXX */
2874 ots->matched_count = htonll(0); /* XXX */
2876 queue_tx(msg, ofconn, ofconn->reply_counter);
2881 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2882 struct ofpbuf **msgp)
2884 struct netdev_stats stats;
2885 struct ofp_port_stats *ops;
2887 /* Intentionally ignore return value, since errors will set
2888 * 'stats' to all-1s, which is correct for OpenFlow, and
2889 * netdev_get_stats() will log errors. */
2890 netdev_get_stats(port->netdev, &stats);
2892 ops = append_stats_reply(sizeof *ops, ofconn, msgp);
2893 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2894 memset(ops->pad, 0, sizeof ops->pad);
2895 ops->rx_packets = htonll(stats.rx_packets);
2896 ops->tx_packets = htonll(stats.tx_packets);
2897 ops->rx_bytes = htonll(stats.rx_bytes);
2898 ops->tx_bytes = htonll(stats.tx_bytes);
2899 ops->rx_dropped = htonll(stats.rx_dropped);
2900 ops->tx_dropped = htonll(stats.tx_dropped);
2901 ops->rx_errors = htonll(stats.rx_errors);
2902 ops->tx_errors = htonll(stats.tx_errors);
2903 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2904 ops->rx_over_err = htonll(stats.rx_over_errors);
2905 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2906 ops->collisions = htonll(stats.collisions);
2910 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2911 struct ofp_stats_request *osr,
2914 struct ofp_port_stats_request *psr;
2915 struct ofp_port_stats *ops;
2917 struct ofport *port;
2918 unsigned int port_no;
2920 if (arg_size != sizeof *psr) {
2921 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2923 psr = (struct ofp_port_stats_request *) osr->body;
2925 msg = start_stats_reply(osr, sizeof *ops * 16);
2926 if (psr->port_no != htons(OFPP_NONE)) {
2927 port = port_array_get(&p->ports,
2928 ofp_port_to_odp_port(ntohs(psr->port_no)));
2930 append_port_stat(port, ntohs(psr->port_no), ofconn, &msg);
2933 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2934 append_port_stat(port, port_no, ofconn, &msg);
2938 queue_tx(msg, ofconn, ofconn->reply_counter);
2942 struct flow_stats_cbdata {
2943 struct ofproto *ofproto;
2944 struct ofconn *ofconn;
2949 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2950 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2951 * returned statistic include statistics for all of 'rule''s subrules. */
2953 query_stats(struct ofproto *p, struct rule *rule,
2954 uint64_t *packet_countp, uint64_t *byte_countp)
2956 uint64_t packet_count, byte_count;
2957 struct rule *subrule;
2958 struct odp_flow *odp_flows;
2961 /* Start from historical data for 'rule' itself that are no longer tracked
2962 * by the datapath. This counts, for example, subrules that have
2964 packet_count = rule->packet_count;
2965 byte_count = rule->byte_count;
2967 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2968 * wildcarded then on all of its subrules.
2970 * Also, add any statistics that are not tracked by the datapath for each
2971 * subrule. This includes, for example, statistics for packets that were
2972 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2974 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2975 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2976 if (rule->cr.wc.wildcards) {
2978 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2979 odp_flows[i++].key = subrule->cr.flow;
2980 packet_count += subrule->packet_count;
2981 byte_count += subrule->byte_count;
2984 odp_flows[0].key = rule->cr.flow;
2987 /* Fetch up-to-date statistics from the datapath and add them in. */
2988 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2990 for (i = 0; i < n_odp_flows; i++) {
2991 struct odp_flow *odp_flow = &odp_flows[i];
2992 packet_count += odp_flow->stats.n_packets;
2993 byte_count += odp_flow->stats.n_bytes;
2998 /* Return the stats to the caller. */
2999 *packet_countp = packet_count;
3000 *byte_countp = byte_count;
3004 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
3006 struct rule *rule = rule_from_cls_rule(rule_);
3007 struct flow_stats_cbdata *cbdata = cbdata_;
3008 struct ofp_flow_stats *ofs;
3009 uint64_t packet_count, byte_count;
3010 size_t act_len, len;
3011 long long int tdiff = time_msec() - rule->created;
3012 uint32_t sec = tdiff / 1000;
3013 uint32_t msec = tdiff - (sec * 1000);
3015 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3019 act_len = sizeof *rule->actions * rule->n_actions;
3020 len = offsetof(struct ofp_flow_stats, actions) + act_len;
3022 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3024 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
3025 ofs->length = htons(len);
3026 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
3028 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3029 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
3030 ofs->duration_sec = htonl(sec);
3031 ofs->duration_nsec = htonl(msec * 1000000);
3032 ofs->cookie = rule->flow_cookie;
3033 ofs->priority = htons(rule->cr.priority);
3034 ofs->idle_timeout = htons(rule->idle_timeout);
3035 ofs->hard_timeout = htons(rule->hard_timeout);
3036 memset(ofs->pad2, 0, sizeof ofs->pad2);
3037 ofs->packet_count = htonll(packet_count);
3038 ofs->byte_count = htonll(byte_count);
3039 memcpy(ofs->actions, rule->actions, act_len);
3043 table_id_to_include(uint8_t table_id)
3045 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
3046 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
3047 : table_id == 0xff ? CLS_INC_ALL
3052 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
3053 const struct ofp_stats_request *osr,
3056 struct ofp_flow_stats_request *fsr;
3057 struct flow_stats_cbdata cbdata;
3058 struct cls_rule target;
3060 if (arg_size != sizeof *fsr) {
3061 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3063 fsr = (struct ofp_flow_stats_request *) osr->body;
3065 COVERAGE_INC(ofproto_flows_req);
3067 cbdata.ofconn = ofconn;
3068 cbdata.out_port = fsr->out_port;
3069 cbdata.msg = start_stats_reply(osr, 1024);
3070 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
3071 classifier_for_each_match(&p->cls, &target,
3072 table_id_to_include(fsr->table_id),
3073 flow_stats_cb, &cbdata);
3074 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3078 struct flow_stats_ds_cbdata {
3079 struct ofproto *ofproto;
3084 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
3086 struct rule *rule = rule_from_cls_rule(rule_);
3087 struct flow_stats_ds_cbdata *cbdata = cbdata_;
3088 struct ds *results = cbdata->results;
3089 struct ofp_match match;
3090 uint64_t packet_count, byte_count;
3091 size_t act_len = sizeof *rule->actions * rule->n_actions;
3093 /* Don't report on subrules. */
3094 if (rule->super != NULL) {
3098 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3099 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
3100 cbdata->ofproto->tun_id_from_cookie, &match);
3102 ds_put_format(results, "duration=%llds, ",
3103 (time_msec() - rule->created) / 1000);
3104 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3105 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3106 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3107 ofp_print_match(results, &match, true);
3108 ofp_print_actions(results, &rule->actions->header, act_len);
3109 ds_put_cstr(results, "\n");
3112 /* Adds a pretty-printed description of all flows to 'results', including
3113 * those marked hidden by secchan (e.g., by in-band control). */
3115 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3117 struct ofp_match match;
3118 struct cls_rule target;
3119 struct flow_stats_ds_cbdata cbdata;
3121 memset(&match, 0, sizeof match);
3122 match.wildcards = htonl(OVSFW_ALL);
3125 cbdata.results = results;
3127 cls_rule_from_match(&match, 0, false, 0, &target);
3128 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3129 flow_stats_ds_cb, &cbdata);
3132 struct aggregate_stats_cbdata {
3133 struct ofproto *ofproto;
3135 uint64_t packet_count;
3136 uint64_t byte_count;
3141 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
3143 struct rule *rule = rule_from_cls_rule(rule_);
3144 struct aggregate_stats_cbdata *cbdata = cbdata_;
3145 uint64_t packet_count, byte_count;
3147 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
3151 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
3153 cbdata->packet_count += packet_count;
3154 cbdata->byte_count += byte_count;
3159 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
3160 const struct ofp_stats_request *osr,
3163 struct ofp_aggregate_stats_request *asr;
3164 struct ofp_aggregate_stats_reply *reply;
3165 struct aggregate_stats_cbdata cbdata;
3166 struct cls_rule target;
3169 if (arg_size != sizeof *asr) {
3170 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3172 asr = (struct ofp_aggregate_stats_request *) osr->body;
3174 COVERAGE_INC(ofproto_agg_request);
3176 cbdata.out_port = asr->out_port;
3177 cbdata.packet_count = 0;
3178 cbdata.byte_count = 0;
3180 cls_rule_from_match(&asr->match, 0, false, 0, &target);
3181 classifier_for_each_match(&p->cls, &target,
3182 table_id_to_include(asr->table_id),
3183 aggregate_stats_cb, &cbdata);
3185 msg = start_stats_reply(osr, sizeof *reply);
3186 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
3187 reply->flow_count = htonl(cbdata.n_flows);
3188 reply->packet_count = htonll(cbdata.packet_count);
3189 reply->byte_count = htonll(cbdata.byte_count);
3190 queue_tx(msg, ofconn, ofconn->reply_counter);
3194 struct queue_stats_cbdata {
3195 struct ofconn *ofconn;
3201 put_queue_stats(struct queue_stats_cbdata *cbdata, uint16_t queue_id,
3202 const struct netdev_queue_stats *stats)
3204 struct ofp_queue_stats *reply;
3206 reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
3207 reply->port_no = htons(cbdata->port_no);
3208 memset(reply->pad, 0, sizeof reply->pad);
3209 reply->queue_id = htonl(queue_id);
3210 reply->tx_bytes = htonll(stats->tx_bytes);
3211 reply->tx_packets = htonll(stats->tx_packets);
3212 reply->tx_errors = htonll(stats->tx_errors);
3216 handle_queue_stats_dump_cb(unsigned int queue_id,
3217 struct netdev_queue_stats *stats,
3220 struct queue_stats_cbdata *cbdata = cbdata_;
3222 put_queue_stats(cbdata, queue_id, stats);
3226 handle_queue_stats_for_port(struct ofport *port, uint16_t port_no,
3228 struct queue_stats_cbdata *cbdata)
3230 cbdata->port_no = port_no;
3231 if (queue_id == OFPQ_ALL) {
3232 netdev_dump_queue_stats(port->netdev,
3233 handle_queue_stats_dump_cb, cbdata);
3235 struct netdev_queue_stats stats;
3237 netdev_get_queue_stats(port->netdev, queue_id, &stats);
3238 put_queue_stats(cbdata, queue_id, &stats);
3243 handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn,
3244 const struct ofp_stats_request *osr,
3247 struct ofp_queue_stats_request *qsr;
3248 struct queue_stats_cbdata cbdata;
3249 struct ofport *port;
3250 unsigned int port_no;
3253 if (arg_size != sizeof *qsr) {
3254 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3256 qsr = (struct ofp_queue_stats_request *) osr->body;
3258 COVERAGE_INC(ofproto_queue_req);
3260 cbdata.ofconn = ofconn;
3261 cbdata.msg = start_stats_reply(osr, 128);
3263 port_no = ntohs(qsr->port_no);
3264 queue_id = ntohl(qsr->queue_id);
3265 if (port_no == OFPP_ALL) {
3266 PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) {
3267 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3269 } else if (port_no < ofproto->max_ports) {
3270 port = port_array_get(&ofproto->ports, port_no);
3272 handle_queue_stats_for_port(port, port_no, queue_id, &cbdata);
3275 ofpbuf_delete(cbdata.msg);
3276 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
3278 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
3284 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
3285 struct ofp_header *oh)
3287 struct ofp_stats_request *osr;
3291 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
3296 osr = (struct ofp_stats_request *) oh;
3298 switch (ntohs(osr->type)) {
3300 return handle_desc_stats_request(p, ofconn, osr);
3303 return handle_flow_stats_request(p, ofconn, osr, arg_size);
3305 case OFPST_AGGREGATE:
3306 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
3309 return handle_table_stats_request(p, ofconn, osr);
3312 return handle_port_stats_request(p, ofconn, osr, arg_size);
3315 return handle_queue_stats_request(p, ofconn, osr, arg_size);
3318 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3321 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
3325 static long long int
3326 msec_from_nsec(uint64_t sec, uint32_t nsec)
3328 return !sec ? 0 : sec * 1000 + nsec / 1000000;
3332 update_time(struct ofproto *ofproto, struct rule *rule,
3333 const struct odp_flow_stats *stats)
3335 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
3336 if (used > rule->used) {
3338 if (rule->super && used > rule->super->used) {
3339 rule->super->used = used;
3341 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3346 update_stats(struct ofproto *ofproto, struct rule *rule,
3347 const struct odp_flow_stats *stats)
3349 if (stats->n_packets) {
3350 update_time(ofproto, rule, stats);
3351 rule->packet_count += stats->n_packets;
3352 rule->byte_count += stats->n_bytes;
3353 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3358 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3359 * in which no matching flow already exists in the flow table.
3361 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3362 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3363 * code as encoded by ofp_mkerr() on failure.
3365 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3368 add_flow(struct ofproto *p, struct ofconn *ofconn,
3369 const struct ofp_flow_mod *ofm, size_t n_actions)
3371 struct ofpbuf *packet;
3376 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3380 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3382 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3383 ntohs(ofm->priority))) {
3384 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3388 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3389 n_actions, ntohs(ofm->idle_timeout),
3390 ntohs(ofm->hard_timeout), ofm->cookie,
3391 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3392 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3393 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3396 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3397 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3401 in_port = UINT16_MAX;
3404 rule_insert(p, rule, packet, in_port);
3405 ofpbuf_delete(packet);
3409 static struct rule *
3410 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3415 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3417 return rule_from_cls_rule(classifier_find_rule_exactly(
3418 &p->cls, &flow, wildcards,
3419 ntohs(ofm->priority)));
3423 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3424 struct rule *rule, const struct ofp_flow_mod *ofm)
3426 struct ofpbuf *packet;
3431 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3435 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3441 flow_extract(packet, 0, in_port, &flow);
3442 rule_execute(ofproto, rule, packet, &flow);
3443 ofpbuf_delete(packet);
3448 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3450 struct modify_flows_cbdata {
3451 struct ofproto *ofproto;
3452 const struct ofp_flow_mod *ofm;
3457 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3458 size_t n_actions, struct rule *);
3459 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3461 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3462 * encoded by ofp_mkerr() on failure.
3464 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3467 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3468 const struct ofp_flow_mod *ofm, size_t n_actions)
3470 struct modify_flows_cbdata cbdata;
3471 struct cls_rule target;
3475 cbdata.n_actions = n_actions;
3476 cbdata.match = NULL;
3478 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3481 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3482 modify_flows_cb, &cbdata);
3484 /* This credits the packet to whichever flow happened to happened to
3485 * match last. That's weird. Maybe we should do a lookup for the
3486 * flow that actually matches the packet? Who knows. */
3487 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3490 return add_flow(p, ofconn, ofm, n_actions);
3494 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3495 * code as encoded by ofp_mkerr() on failure.
3497 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3500 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3501 struct ofp_flow_mod *ofm, size_t n_actions)
3503 struct rule *rule = find_flow_strict(p, ofm);
3504 if (rule && !rule_is_hidden(rule)) {
3505 modify_flow(p, ofm, n_actions, rule);
3506 return send_buffered_packet(p, ofconn, rule, ofm);
3508 return add_flow(p, ofconn, ofm, n_actions);
3512 /* Callback for modify_flows_loose(). */
3514 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3516 struct rule *rule = rule_from_cls_rule(rule_);
3517 struct modify_flows_cbdata *cbdata = cbdata_;
3519 if (!rule_is_hidden(rule)) {
3520 cbdata->match = rule;
3521 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3525 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3526 * been identified as a flow in 'p''s flow table to be modified, by changing
3527 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3528 * ofp_action[] structures). */
3530 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3531 size_t n_actions, struct rule *rule)
3533 size_t actions_len = n_actions * sizeof *rule->actions;
3535 rule->flow_cookie = ofm->cookie;
3537 /* If the actions are the same, do nothing. */
3538 if (n_actions == rule->n_actions
3539 && !memcmp(ofm->actions, rule->actions, actions_len))
3544 /* Replace actions. */
3545 free(rule->actions);
3546 rule->actions = xmemdup(ofm->actions, actions_len);
3547 rule->n_actions = n_actions;
3549 /* Make sure that the datapath gets updated properly. */
3550 if (rule->cr.wc.wildcards) {
3551 COVERAGE_INC(ofproto_mod_wc_flow);
3552 p->need_revalidate = true;
3554 rule_update_actions(p, rule);
3560 /* OFPFC_DELETE implementation. */
3562 struct delete_flows_cbdata {
3563 struct ofproto *ofproto;
3567 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3568 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3570 /* Implements OFPFC_DELETE. */
3572 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3574 struct delete_flows_cbdata cbdata;
3575 struct cls_rule target;
3578 cbdata.out_port = ofm->out_port;
3580 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3583 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3584 delete_flows_cb, &cbdata);
3587 /* Implements OFPFC_DELETE_STRICT. */
3589 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3591 struct rule *rule = find_flow_strict(p, ofm);
3593 delete_flow(p, rule, ofm->out_port);
3597 /* Callback for delete_flows_loose(). */
3599 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3601 struct rule *rule = rule_from_cls_rule(rule_);
3602 struct delete_flows_cbdata *cbdata = cbdata_;
3604 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3607 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3608 * been identified as a flow to delete from 'p''s flow table, by deleting the
3609 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3612 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3613 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3614 * specified 'out_port'. */
3616 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3618 if (rule_is_hidden(rule)) {
3622 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3626 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3627 rule_remove(p, rule);
3631 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3632 struct ofp_flow_mod *ofm)
3637 error = reject_slave_controller(ofconn, &ofm->header);
3641 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3642 sizeof *ofm->actions, &n_actions);
3647 /* We do not support the emergency flow cache. It will hopefully
3648 * get dropped from OpenFlow in the near future. */
3649 if (ofm->flags & htons(OFPFF_EMERG)) {
3650 /* There isn't a good fit for an error code, so just state that the
3651 * flow table is full. */
3652 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3655 normalize_match(&ofm->match);
3656 if (!ofm->match.wildcards) {
3657 ofm->priority = htons(UINT16_MAX);
3660 error = validate_actions((const union ofp_action *) ofm->actions,
3661 n_actions, p->max_ports);
3666 switch (ntohs(ofm->command)) {
3668 return add_flow(p, ofconn, ofm, n_actions);
3671 return modify_flows_loose(p, ofconn, ofm, n_actions);
3673 case OFPFC_MODIFY_STRICT:
3674 return modify_flow_strict(p, ofconn, ofm, n_actions);
3677 delete_flows_loose(p, ofm);
3680 case OFPFC_DELETE_STRICT:
3681 delete_flow_strict(p, ofm);
3685 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3690 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3694 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3699 p->tun_id_from_cookie = !!msg->set;
3704 handle_role_request(struct ofproto *ofproto,
3705 struct ofconn *ofconn, struct nicira_header *msg)
3707 struct nx_role_request *nrr;
3708 struct nx_role_request *reply;
3712 if (ntohs(msg->header.length) != sizeof *nrr) {
3713 VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)",
3714 ntohs(msg->header.length), sizeof *nrr);
3715 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3717 nrr = (struct nx_role_request *) msg;
3719 if (ofconn->type != OFCONN_CONTROLLER) {
3720 VLOG_WARN_RL(&rl, "ignoring role request on non-controller "
3722 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
3725 role = ntohl(nrr->role);
3726 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3727 && role != NX_ROLE_SLAVE) {
3728 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
3730 /* There's no good error code for this. */
3731 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
3734 if (role == NX_ROLE_MASTER) {
3735 struct ofconn *other;
3737 HMAP_FOR_EACH (other, struct ofconn, hmap_node,
3738 &ofproto->controllers) {
3739 if (other->role == NX_ROLE_MASTER) {
3740 other->role = NX_ROLE_SLAVE;
3744 ofconn->role = role;
3746 reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid,
3748 reply->nxh.vendor = htonl(NX_VENDOR_ID);
3749 reply->nxh.subtype = htonl(NXT_ROLE_REPLY);
3750 reply->role = htonl(role);
3751 queue_tx(buf, ofconn, ofconn->reply_counter);
3757 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3759 struct ofp_vendor_header *ovh = msg;
3760 struct nicira_header *nh;
3762 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3763 VLOG_WARN_RL(&rl, "received vendor message of length %u "
3764 "(expected at least %zu)",
3765 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3766 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3768 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3769 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3771 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3772 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u "
3773 "(expected at least %zu)",
3774 ntohs(ovh->header.length), sizeof(struct nicira_header));
3775 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3779 switch (ntohl(nh->subtype)) {
3780 case NXT_STATUS_REQUEST:
3781 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3784 case NXT_TUN_ID_FROM_COOKIE:
3785 return handle_tun_id_from_cookie(p, msg);
3787 case NXT_ROLE_REQUEST:
3788 return handle_role_request(p, ofconn, msg);
3791 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3795 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3797 struct ofp_header *ob;
3800 /* Currently, everything executes synchronously, so we can just
3801 * immediately send the barrier reply. */
3802 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3803 queue_tx(buf, ofconn, ofconn->reply_counter);
3808 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3809 struct ofpbuf *ofp_msg)
3811 struct ofp_header *oh = ofp_msg->data;
3814 COVERAGE_INC(ofproto_recv_openflow);
3816 case OFPT_ECHO_REQUEST:
3817 error = handle_echo_request(ofconn, oh);
3820 case OFPT_ECHO_REPLY:
3824 case OFPT_FEATURES_REQUEST:
3825 error = handle_features_request(p, ofconn, oh);
3828 case OFPT_GET_CONFIG_REQUEST:
3829 error = handle_get_config_request(p, ofconn, oh);
3832 case OFPT_SET_CONFIG:
3833 error = handle_set_config(p, ofconn, ofp_msg->data);
3836 case OFPT_PACKET_OUT:
3837 error = handle_packet_out(p, ofconn, ofp_msg->data);
3841 error = handle_port_mod(p, ofconn, oh);
3845 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3848 case OFPT_STATS_REQUEST:
3849 error = handle_stats_request(p, ofconn, oh);
3853 error = handle_vendor(p, ofconn, ofp_msg->data);
3856 case OFPT_BARRIER_REQUEST:
3857 error = handle_barrier_request(ofconn, oh);
3861 if (VLOG_IS_WARN_ENABLED()) {
3862 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3863 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3866 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3871 send_error_oh(ofconn, ofp_msg->data, error);
3876 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3878 struct odp_msg *msg = packet->data;
3880 struct ofpbuf payload;
3883 payload.data = msg + 1;
3884 payload.size = msg->length - sizeof *msg;
3885 flow_extract(&payload, msg->arg, msg->port, &flow);
3887 /* Check with in-band control to see if this packet should be sent
3888 * to the local port regardless of the flow table. */
3889 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3890 union odp_action action;
3892 memset(&action, 0, sizeof(action));
3893 action.output.type = ODPAT_OUTPUT;
3894 action.output.port = ODPP_LOCAL;
3895 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3898 rule = lookup_valid_rule(p, &flow);
3900 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3901 struct ofport *port = port_array_get(&p->ports, msg->port);
3903 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3904 COVERAGE_INC(ofproto_no_packet_in);
3905 /* XXX install 'drop' flow entry */
3906 ofpbuf_delete(packet);
3910 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3913 COVERAGE_INC(ofproto_packet_in);
3914 send_packet_in(p, packet);
3918 if (rule->cr.wc.wildcards) {
3919 rule = rule_create_subrule(p, rule, &flow);
3920 rule_make_actions(p, rule, packet);
3922 if (!rule->may_install) {
3923 /* The rule is not installable, that is, we need to process every
3924 * packet, so process the current packet and set its actions into
3926 rule_make_actions(p, rule, packet);
3928 /* XXX revalidate rule if it needs it */
3932 rule_execute(p, rule, &payload, &flow);
3933 rule_reinstall(p, rule);
3935 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3937 * Extra-special case for fail-open mode.
3939 * We are in fail-open mode and the packet matched the fail-open rule,
3940 * but we are connected to a controller too. We should send the packet
3941 * up to the controller in the hope that it will try to set up a flow
3942 * and thereby allow us to exit fail-open.
3944 * See the top-level comment in fail-open.c for more information.
3946 send_packet_in(p, packet);
3948 ofpbuf_delete(packet);
3953 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3955 struct odp_msg *msg = packet->data;
3957 switch (msg->type) {
3958 case _ODPL_ACTION_NR:
3959 COVERAGE_INC(ofproto_ctlr_action);
3960 send_packet_in(p, packet);
3963 case _ODPL_SFLOW_NR:
3965 ofproto_sflow_received(p->sflow, msg);
3967 ofpbuf_delete(packet);
3971 handle_odp_miss_msg(p, packet);
3975 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3982 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3984 struct rule *sub = rule_from_cls_rule(sub_);
3985 struct revalidate_cbdata *cbdata = cbdata_;
3987 if (cbdata->revalidate_all
3988 || (cbdata->revalidate_subrules && sub->super)
3989 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3990 revalidate_rule(cbdata->ofproto, sub);
3995 revalidate_rule(struct ofproto *p, struct rule *rule)
3997 const flow_t *flow = &rule->cr.flow;
3999 COVERAGE_INC(ofproto_revalidate_rule);
4002 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
4004 rule_remove(p, rule);
4006 } else if (super != rule->super) {
4007 COVERAGE_INC(ofproto_revalidate_moved);
4008 list_remove(&rule->list);
4009 list_push_back(&super->list, &rule->list);
4010 rule->super = super;
4011 rule->hard_timeout = super->hard_timeout;
4012 rule->idle_timeout = super->idle_timeout;
4013 rule->created = super->created;
4018 rule_update_actions(p, rule);
4022 static struct ofpbuf *
4023 compose_flow_removed(struct ofproto *p, const struct rule *rule,
4024 long long int now, uint8_t reason)
4026 struct ofp_flow_removed *ofr;
4028 long long int tdiff = now - rule->created;
4029 uint32_t sec = tdiff / 1000;
4030 uint32_t msec = tdiff - (sec * 1000);
4032 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
4033 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
4035 ofr->cookie = rule->flow_cookie;
4036 ofr->priority = htons(rule->cr.priority);
4037 ofr->reason = reason;
4038 ofr->duration_sec = htonl(sec);
4039 ofr->duration_nsec = htonl(msec * 1000000);
4040 ofr->idle_timeout = htons(rule->idle_timeout);
4041 ofr->packet_count = htonll(rule->packet_count);
4042 ofr->byte_count = htonll(rule->byte_count);
4048 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
4050 assert(rule->installed);
4051 assert(!rule->cr.wc.wildcards);
4054 rule_remove(ofproto, rule);
4056 rule_uninstall(ofproto, rule);
4061 send_flow_removed(struct ofproto *p, struct rule *rule,
4062 long long int now, uint8_t reason)
4064 struct ofconn *ofconn;
4065 struct ofconn *prev;
4066 struct ofpbuf *buf = NULL;
4068 /* We limit the maximum number of queued flow expirations it by accounting
4069 * them under the counter for replies. That works because preventing
4070 * OpenFlow requests from being processed also prevents new flows from
4071 * being added (and expiring). (It also prevents processing OpenFlow
4072 * requests that would not add new flows, so it is imperfect.) */
4075 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
4076 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)
4077 && ofconn_receives_async_msgs(ofconn)) {
4079 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
4081 buf = compose_flow_removed(p, rule, now, reason);
4087 queue_tx(buf, prev, prev->reply_counter);
4093 expire_rule(struct cls_rule *cls_rule, void *p_)
4095 struct ofproto *p = p_;
4096 struct rule *rule = rule_from_cls_rule(cls_rule);
4097 long long int hard_expire, idle_expire, expire, now;
4099 hard_expire = (rule->hard_timeout
4100 ? rule->created + rule->hard_timeout * 1000
4102 idle_expire = (rule->idle_timeout
4103 && (rule->super || list_is_empty(&rule->list))
4104 ? rule->used + rule->idle_timeout * 1000
4106 expire = MIN(hard_expire, idle_expire);
4110 if (rule->installed && now >= rule->used + 5000) {
4111 uninstall_idle_flow(p, rule);
4112 } else if (!rule->cr.wc.wildcards) {
4113 active_timeout(p, rule);
4119 COVERAGE_INC(ofproto_expired);
4121 /* Update stats. This code will be a no-op if the rule expired
4122 * due to an idle timeout. */
4123 if (rule->cr.wc.wildcards) {
4124 struct rule *subrule, *next;
4125 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
4126 rule_remove(p, subrule);
4129 rule_uninstall(p, rule);
4132 if (!rule_is_hidden(rule)) {
4133 send_flow_removed(p, rule, now,
4135 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
4137 rule_remove(p, rule);
4141 active_timeout(struct ofproto *ofproto, struct rule *rule)
4143 if (ofproto->netflow && !is_controller_rule(rule) &&
4144 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
4145 struct ofexpired expired;
4146 struct odp_flow odp_flow;
4148 /* Get updated flow stats. */
4149 memset(&odp_flow, 0, sizeof odp_flow);
4150 if (rule->installed) {
4151 odp_flow.key = rule->cr.flow;
4152 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
4153 dpif_flow_get(ofproto->dpif, &odp_flow);
4155 if (odp_flow.stats.n_packets) {
4156 update_time(ofproto, rule, &odp_flow.stats);
4157 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
4158 odp_flow.stats.tcp_flags);
4162 expired.flow = rule->cr.flow;
4163 expired.packet_count = rule->packet_count +
4164 odp_flow.stats.n_packets;
4165 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
4166 expired.used = rule->used;
4168 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
4170 /* Schedule us to send the accumulated records once we have
4171 * collected all of them. */
4172 poll_immediate_wake();
4177 update_used(struct ofproto *p)
4179 struct odp_flow *flows;
4184 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
4189 for (i = 0; i < n_flows; i++) {
4190 struct odp_flow *f = &flows[i];
4193 rule = rule_from_cls_rule(
4194 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
4195 if (!rule || !rule->installed) {
4196 COVERAGE_INC(ofproto_unexpected_rule);
4197 dpif_flow_del(p->dpif, f);
4201 update_time(p, rule, &f->stats);
4202 rule_account(p, rule, f->stats.n_bytes);
4207 /* pinsched callback for sending 'packet' on 'ofconn'. */
4209 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
4211 struct ofconn *ofconn = ofconn_;
4213 rconn_send_with_limit(ofconn->rconn, packet,
4214 ofconn->packet_in_counter, 100);
4217 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4218 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4219 * packet scheduler for sending.
4221 * 'max_len' specifies the maximum number of bytes of the packet to send on
4222 * 'ofconn' (INT_MAX specifies no limit).
4224 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4225 * ownership is transferred to this function. */
4227 schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len,
4230 struct ofproto *ofproto = ofconn->ofproto;
4231 struct ofp_packet_in *opi = packet->data;
4232 uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port));
4233 int send_len, trim_size;
4237 if (opi->reason == OFPR_ACTION) {
4238 buffer_id = UINT32_MAX;
4239 } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
4240 buffer_id = pktbuf_get_null();
4241 } else if (!ofconn->pktbuf) {
4242 buffer_id = UINT32_MAX;
4244 struct ofpbuf payload;
4245 payload.data = opi->data;
4246 payload.size = packet->size - offsetof(struct ofp_packet_in, data);
4247 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port);
4250 /* Figure out how much of the packet to send. */
4251 send_len = ntohs(opi->total_len);
4252 if (buffer_id != UINT32_MAX) {
4253 send_len = MIN(send_len, ofconn->miss_send_len);
4255 send_len = MIN(send_len, max_len);
4257 /* Adjust packet length and clone if necessary. */
4258 trim_size = offsetof(struct ofp_packet_in, data) + send_len;
4260 packet = ofpbuf_clone_data(packet->data, trim_size);
4263 packet->size = trim_size;
4266 /* Update packet headers. */
4267 opi->buffer_id = htonl(buffer_id);
4268 update_openflow_length(packet);
4270 /* Hand over to packet scheduler. It might immediately call into
4271 * do_send_packet_in() or it might buffer it for a while (until a later
4272 * call to pinsched_run()). */
4273 pinsched_send(ofconn->schedulers[opi->reason], in_port,
4274 packet, do_send_packet_in, ofconn);
4277 /* Replace struct odp_msg header in 'packet' by equivalent struct
4278 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4279 * returned by dpif_recv()).
4281 * The conversion is not complete: the caller still needs to trim any unneeded
4282 * payload off the end of the buffer, set the length in the OpenFlow header,
4283 * and set buffer_id. Those require us to know the controller settings and so
4284 * must be done on a per-controller basis.
4286 * Returns the maximum number of bytes of the packet that should be sent to
4287 * the controller (INT_MAX if no limit). */
4289 do_convert_to_packet_in(struct ofpbuf *packet)
4291 struct odp_msg *msg = packet->data;
4292 struct ofp_packet_in *opi;
4298 /* Extract relevant header fields */
4299 if (msg->type == _ODPL_ACTION_NR) {
4300 reason = OFPR_ACTION;
4303 reason = OFPR_NO_MATCH;
4306 total_len = msg->length - sizeof *msg;
4307 in_port = odp_port_to_ofp_port(msg->port);
4309 /* Repurpose packet buffer by overwriting header. */
4310 ofpbuf_pull(packet, sizeof(struct odp_msg));
4311 opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data));
4312 opi->header.version = OFP_VERSION;
4313 opi->header.type = OFPT_PACKET_IN;
4314 opi->total_len = htons(total_len);
4315 opi->in_port = htons(in_port);
4316 opi->reason = reason;
4321 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4322 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4323 * as necessary according to their individual configurations.
4325 * 'packet' must have sufficient headroom to convert it into a struct
4326 * ofp_packet_in (e.g. as returned by dpif_recv()).
4328 * Takes ownership of 'packet'. */
4330 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
4332 struct ofconn *ofconn, *prev;
4335 max_len = do_convert_to_packet_in(packet);
4338 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
4339 if (ofconn_receives_async_msgs(ofconn)) {
4341 schedule_packet_in(prev, packet, max_len, true);
4347 schedule_packet_in(prev, packet, max_len, false);
4349 ofpbuf_delete(packet);
4354 pick_datapath_id(const struct ofproto *ofproto)
4356 const struct ofport *port;
4358 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
4360 uint8_t ea[ETH_ADDR_LEN];
4363 error = netdev_get_etheraddr(port->netdev, ea);
4365 return eth_addr_to_uint64(ea);
4367 VLOG_WARN("could not get MAC address for %s (%s)",
4368 netdev_get_name(port->netdev), strerror(error));
4370 return ofproto->fallback_dpid;
4374 pick_fallback_dpid(void)
4376 uint8_t ea[ETH_ADDR_LEN];
4377 eth_addr_nicira_random(ea);
4378 return eth_addr_to_uint64(ea);
4382 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
4383 struct odp_actions *actions, tag_type *tags,
4384 uint16_t *nf_output_iface, void *ofproto_)
4386 struct ofproto *ofproto = ofproto_;
4389 /* Drop frames for reserved multicast addresses. */
4390 if (eth_addr_is_reserved(flow->dl_dst)) {
4394 /* Learn source MAC (but don't try to learn from revalidation). */
4395 if (packet != NULL) {
4396 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
4398 GRAT_ARP_LOCK_NONE);
4400 /* The log messages here could actually be useful in debugging,
4401 * so keep the rate limit relatively high. */
4402 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
4403 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
4404 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
4405 ofproto_revalidate(ofproto, rev_tag);
4409 /* Determine output port. */
4410 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags,
4413 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
4414 } else if (out_port != flow->in_port) {
4415 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
4416 *nf_output_iface = out_port;
4424 static const struct ofhooks default_ofhooks = {
4426 default_normal_ofhook_cb,