2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
30 #include "fail-open.h"
32 #include "mac-learning.h"
36 #include "ofp-print.h"
37 #include "ofproto-sflow.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openvswitch/datapath-protocol.h"
45 #include "poll-loop.h"
46 #include "port-array.h"
51 #include "stream-ssl.h"
59 #define THIS_MODULE VLM_ofproto
62 #include "sflow_api.h"
66 TABLEID_CLASSIFIER = 1
70 struct netdev *netdev;
71 struct ofp_phy_port opp; /* In host byte order. */
74 static void ofport_free(struct ofport *);
75 static void hton_ofp_phy_port(struct ofp_phy_port *);
77 static int xlate_actions(const union ofp_action *in, size_t n_in,
78 const flow_t *flow, struct ofproto *ofproto,
79 const struct ofpbuf *packet,
80 struct odp_actions *out, tag_type *tags,
81 bool *may_set_up_flow, uint16_t *nf_output_iface);
86 uint64_t flow_cookie; /* Controller-issued identifier.
87 (Kept in network-byte order.) */
88 uint16_t idle_timeout; /* In seconds from time of last use. */
89 uint16_t hard_timeout; /* In seconds from time of creation. */
90 bool send_flow_removed; /* Send a flow removed message? */
91 long long int used; /* Last-used time (0 if never used). */
92 long long int created; /* Creation time. */
93 uint64_t packet_count; /* Number of packets received. */
94 uint64_t byte_count; /* Number of bytes received. */
95 uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */
96 tag_type tags; /* Tags (set only by hooks). */
97 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
99 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
100 * exact-match rule (having cr.wc.wildcards of 0) generated from the
101 * wildcard rule 'super'. In this case, 'list' is an element of the
104 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
105 * a list of subrules. A super-rule with no wildcards (where
106 * cr.wc.wildcards is 0) will never have any subrules. */
112 * 'n_actions' is the number of elements in the 'actions' array. A single
113 * action may take up more more than one element's worth of space.
115 * A subrule has no actions (it uses the super-rule's actions). */
117 union ofp_action *actions;
121 * A super-rule with wildcard fields never has ODP actions (since the
122 * datapath only supports exact-match flows). */
123 bool installed; /* Installed in datapath? */
124 bool may_install; /* True ordinarily; false if actions must
125 * be reassessed for every packet. */
127 union odp_action *odp_actions;
131 rule_is_hidden(const struct rule *rule)
133 /* Subrules are merely an implementation detail, so hide them from the
135 if (rule->super != NULL) {
139 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
140 * (e.g. by in-band control) and are intentionally hidden from the
142 if (rule->cr.priority > UINT16_MAX) {
149 static struct rule *rule_create(struct ofproto *, struct rule *super,
150 const union ofp_action *, size_t n_actions,
151 uint16_t idle_timeout, uint16_t hard_timeout,
152 uint64_t flow_cookie, bool send_flow_removed);
153 static void rule_free(struct rule *);
154 static void rule_destroy(struct ofproto *, struct rule *);
155 static struct rule *rule_from_cls_rule(const struct cls_rule *);
156 static void rule_insert(struct ofproto *, struct rule *,
157 struct ofpbuf *packet, uint16_t in_port);
158 static void rule_remove(struct ofproto *, struct rule *);
159 static bool rule_make_actions(struct ofproto *, struct rule *,
160 const struct ofpbuf *packet);
161 static void rule_install(struct ofproto *, struct rule *,
162 struct rule *displaced_rule);
163 static void rule_uninstall(struct ofproto *, struct rule *);
164 static void rule_post_uninstall(struct ofproto *, struct rule *);
165 static void send_flow_removed(struct ofproto *p, struct rule *rule,
166 long long int now, uint8_t reason);
168 /* ofproto supports two kinds of OpenFlow connections:
170 * - "Controller connections": Connections to ordinary OpenFlow controllers.
171 * ofproto maintains persistent connections to these controllers and by
172 * default sends them asynchronous messages such as packet-ins.
174 * - "Transient connections", e.g. from ovs-ofctl. When these connections
175 * drop, it is the other side's responsibility to reconnect them if
176 * necessary. ofproto does not send them asynchronous messages by default.
179 OFCONN_CONTROLLER, /* An OpenFlow controller. */
180 OFCONN_TRANSIENT /* A transient connection. */
183 /* An OpenFlow connection. */
185 struct ofproto *ofproto; /* The ofproto that owns this connection. */
186 struct list node; /* In struct ofproto's "all_conns" list. */
187 struct rconn *rconn; /* OpenFlow connection. */
188 enum ofconn_type type; /* Type. */
190 /* OFPT_PACKET_IN related data. */
191 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
192 struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */
193 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
194 int miss_send_len; /* Bytes to send of buffered packets. */
196 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
197 * requests, and the maximum number before we stop reading OpenFlow
199 #define OFCONN_REPLY_MAX 100
200 struct rconn_packet_counter *reply_counter;
202 /* type == OFCONN_CONTROLLER only. */
203 struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */
204 struct discovery *discovery; /* Controller discovery object, if enabled. */
205 struct status_category *ss; /* Switch status category. */
208 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
209 * "schedulers" array. Their values are 0 and 1, and their meanings and values
210 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
211 * case anything ever changes, check their values here. */
212 #define N_SCHEDULERS 2
213 BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0);
214 BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR);
215 BUILD_ASSERT_DECL(OFPR_ACTION == 1);
216 BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR);
218 static struct ofconn *ofconn_create(struct ofproto *, struct rconn *,
220 static void ofconn_destroy(struct ofconn *);
221 static void ofconn_run(struct ofconn *, struct ofproto *);
222 static void ofconn_wait(struct ofconn *);
223 static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
224 struct rconn_packet_counter *counter);
226 static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg);
227 static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn);
231 uint64_t datapath_id; /* Datapath ID. */
232 uint64_t fallback_dpid; /* Datapath ID if no better choice found. */
233 char *mfr_desc; /* Manufacturer. */
234 char *hw_desc; /* Hardware. */
235 char *sw_desc; /* Software version. */
236 char *serial_desc; /* Serial number. */
237 char *dp_desc; /* Datapath description. */
241 struct netdev_monitor *netdev_monitor;
242 struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is
244 struct shash port_by_name;
248 struct switch_status *switch_status;
249 struct in_band *in_band;
250 struct fail_open *fail_open;
251 struct netflow *netflow;
252 struct ofproto_sflow *sflow;
255 struct classifier cls;
256 bool need_revalidate;
257 long long int next_expiration;
258 struct tag_set revalidate_set;
259 bool tun_id_from_cookie;
261 /* OpenFlow connections. */
262 struct hmap controllers; /* Controller "struct ofconn"s. */
263 struct list all_conns; /* Contains "struct ofconn"s. */
264 struct pvconn **listeners;
266 struct pvconn **snoops;
269 /* Hooks for ovs-vswitchd. */
270 const struct ofhooks *ofhooks;
273 /* Used by default ofhooks. */
274 struct mac_learning *ml;
277 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
279 static const struct ofhooks default_ofhooks;
281 static uint64_t pick_datapath_id(const struct ofproto *);
282 static uint64_t pick_fallback_dpid(void);
284 static void update_used(struct ofproto *);
285 static void update_stats(struct ofproto *, struct rule *,
286 const struct odp_flow_stats *);
287 static void expire_rule(struct cls_rule *, void *ofproto);
288 static void active_timeout(struct ofproto *ofproto, struct rule *rule);
289 static bool revalidate_rule(struct ofproto *p, struct rule *rule);
290 static void revalidate_cb(struct cls_rule *rule_, void *p_);
292 static void handle_odp_msg(struct ofproto *, struct ofpbuf *);
294 static void handle_openflow(struct ofconn *, struct ofproto *,
297 static void refresh_port_groups(struct ofproto *);
299 static void update_port(struct ofproto *, const char *devname);
300 static int init_ports(struct ofproto *);
301 static void reinit_ports(struct ofproto *);
304 ofproto_create(const char *datapath, const char *datapath_type,
305 const struct ofhooks *ofhooks, void *aux,
306 struct ofproto **ofprotop)
308 struct odp_stats stats;
315 /* Connect to datapath and start listening for messages. */
316 error = dpif_open(datapath, datapath_type, &dpif);
318 VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error));
321 error = dpif_get_dp_stats(dpif, &stats);
323 VLOG_ERR("failed to obtain stats for datapath %s: %s",
324 datapath, strerror(error));
328 error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW);
330 VLOG_ERR("failed to listen on datapath %s: %s",
331 datapath, strerror(error));
335 dpif_flow_flush(dpif);
336 dpif_recv_purge(dpif);
338 /* Initialize settings. */
339 p = xzalloc(sizeof *p);
340 p->fallback_dpid = pick_fallback_dpid();
341 p->datapath_id = p->fallback_dpid;
342 p->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
343 p->hw_desc = xstrdup(DEFAULT_HW_DESC);
344 p->sw_desc = xstrdup(DEFAULT_SW_DESC);
345 p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
346 p->dp_desc = xstrdup(DEFAULT_DP_DESC);
348 /* Initialize datapath. */
350 p->netdev_monitor = netdev_monitor_create();
351 port_array_init(&p->ports);
352 shash_init(&p->port_by_name);
353 p->max_ports = stats.max_ports;
355 /* Initialize submodules. */
356 p->switch_status = switch_status_create(p);
362 /* Initialize flow table. */
363 classifier_init(&p->cls);
364 p->need_revalidate = false;
365 p->next_expiration = time_msec() + 1000;
366 tag_set_init(&p->revalidate_set);
368 /* Initialize OpenFlow connections. */
369 list_init(&p->all_conns);
370 hmap_init(&p->controllers);
376 /* Initialize hooks. */
378 p->ofhooks = ofhooks;
382 p->ofhooks = &default_ofhooks;
384 p->ml = mac_learning_create();
387 /* Pick final datapath ID. */
388 p->datapath_id = pick_datapath_id(p);
389 VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id);
396 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
398 uint64_t old_dpid = p->datapath_id;
399 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
400 if (p->datapath_id != old_dpid) {
401 struct ofconn *ofconn;
403 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
405 /* Force all active connections to reconnect, since there is no way to
406 * notify a controller that the datapath ID has changed. */
407 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
408 rconn_reconnect(ofconn->rconn);
414 is_discovery_controller(const struct ofproto_controller *c)
416 return !strcmp(c->target, "discover");
420 is_in_band_controller(const struct ofproto_controller *c)
422 return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND;
425 /* Creates a new controller in 'ofproto'. Some of the settings are initially
426 * drawn from 'c', but update_controller() needs to be called later to finish
427 * the new ofconn's configuration. */
429 add_controller(struct ofproto *ofproto, const struct ofproto_controller *c)
431 struct discovery *discovery;
432 struct ofconn *ofconn;
434 if (is_discovery_controller(c)) {
435 int error = discovery_create(c->accept_re, c->update_resolv_conf,
436 ofproto->dpif, ofproto->switch_status,
445 ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER);
446 ofconn->pktbuf = pktbuf_create();
447 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
449 ofconn->discovery = discovery;
451 rconn_connect(ofconn->rconn, c->target);
453 hmap_insert(&ofproto->controllers, &ofconn->hmap_node,
454 hash_string(c->target, 0));
457 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
458 * target or turn discovery on or off (these are done by creating new ofconns
459 * and deleting old ones), but it can update the rest of an ofconn's
462 update_controller(struct ofconn *ofconn, const struct ofproto_controller *c)
464 struct ofproto *ofproto = ofconn->ofproto;
468 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
470 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
471 rconn_set_probe_interval(ofconn->rconn, probe_interval);
473 if (ofconn->discovery) {
474 discovery_set_update_resolv_conf(ofconn->discovery,
475 c->update_resolv_conf);
476 discovery_set_accept_controller_re(ofconn->discovery, c->accept_re);
479 for (i = 0; i < N_SCHEDULERS; i++) {
480 struct pinsched **s = &ofconn->schedulers[i];
482 if (c->rate_limit > 0) {
484 *s = pinsched_create(c->rate_limit, c->burst_limit,
485 ofproto->switch_status);
487 pinsched_set_limits(*s, c->rate_limit, c->burst_limit);
490 pinsched_destroy(*s);
497 ofconn_get_target(const struct ofconn *ofconn)
499 return ofconn->discovery ? "discover" : rconn_get_name(ofconn->rconn);
502 static struct ofconn *
503 find_controller_by_target(struct ofproto *ofproto, const char *target)
505 struct ofconn *ofconn;
507 HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node,
508 hash_string(target, 0), &ofproto->controllers) {
509 if (!strcmp(ofconn_get_target(ofconn), target)) {
517 ofproto_set_controllers(struct ofproto *p,
518 const struct ofproto_controller *controllers,
519 size_t n_controllers)
521 struct shash new_controllers;
522 struct rconn **in_band_rconns;
523 enum ofproto_fail_mode fail_mode;
524 struct ofconn *ofconn, *next;
529 shash_init(&new_controllers);
530 for (i = 0; i < n_controllers; i++) {
531 const struct ofproto_controller *c = &controllers[i];
533 shash_add_once(&new_controllers, c->target, &controllers[i]);
534 if (!find_controller_by_target(p, c->target)) {
535 add_controller(p, c);
539 in_band_rconns = xmalloc(n_controllers * sizeof *in_band_rconns);
541 fail_mode = OFPROTO_FAIL_STANDALONE;
543 HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node,
545 struct ofproto_controller *c;
547 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
549 ofconn_destroy(ofconn);
551 update_controller(ofconn, c);
556 if (is_in_band_controller(c)) {
557 in_band_rconns[n_in_band++] = ofconn->rconn;
560 if (c->fail == OFPROTO_FAIL_SECURE) {
561 fail_mode = OFPROTO_FAIL_SECURE;
565 shash_destroy(&new_controllers);
569 in_band_create(p, p->dpif, p->switch_status, &p->in_band);
572 in_band_set_remotes(p->in_band, in_band_rconns, n_in_band);
575 in_band_destroy(p->in_band);
578 free(in_band_rconns);
580 if (!hmap_is_empty(&p->controllers)
581 && fail_mode == OFPROTO_FAIL_STANDALONE) {
582 struct rconn **rconns;
586 p->fail_open = fail_open_create(p, p->switch_status);
590 rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns);
591 HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) {
592 rconns[n++] = ofconn->rconn;
595 fail_open_set_controllers(p->fail_open, rconns, n);
596 /* p->fail_open takes ownership of 'rconns'. */
598 fail_open_destroy(p->fail_open);
602 if (!hmap_is_empty(&p->controllers) && !ss_exists) {
603 ofconn = CONTAINER_OF(hmap_first(&p->controllers),
604 struct ofconn, hmap_node);
605 ofconn->ss = switch_status_register(p->switch_status, "remote",
606 rconn_status_cb, ofconn->rconn);
611 ofproto_set_desc(struct ofproto *p,
612 const char *mfr_desc, const char *hw_desc,
613 const char *sw_desc, const char *serial_desc,
616 struct ofp_desc_stats *ods;
619 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
620 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
621 sizeof ods->mfr_desc);
624 p->mfr_desc = xstrdup(mfr_desc);
627 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
628 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
629 sizeof ods->hw_desc);
632 p->hw_desc = xstrdup(hw_desc);
635 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
636 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
637 sizeof ods->sw_desc);
640 p->sw_desc = xstrdup(sw_desc);
643 if (strlen(serial_desc) >= sizeof ods->serial_num) {
644 VLOG_WARN("truncating serial_desc, must be less than %zu "
646 sizeof ods->serial_num);
648 free(p->serial_desc);
649 p->serial_desc = xstrdup(serial_desc);
652 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
653 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
654 sizeof ods->dp_desc);
657 p->dp_desc = xstrdup(dp_desc);
662 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
663 const struct svec *svec)
665 struct pvconn **pvconns = *pvconnsp;
666 size_t n_pvconns = *n_pvconnsp;
670 for (i = 0; i < n_pvconns; i++) {
671 pvconn_close(pvconns[i]);
675 pvconns = xmalloc(svec->n * sizeof *pvconns);
677 for (i = 0; i < svec->n; i++) {
678 const char *name = svec->names[i];
679 struct pvconn *pvconn;
682 error = pvconn_open(name, &pvconn);
684 pvconns[n_pvconns++] = pvconn;
686 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
694 *n_pvconnsp = n_pvconns;
700 ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners)
702 return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners);
706 ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops)
708 return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops);
712 ofproto_set_netflow(struct ofproto *ofproto,
713 const struct netflow_options *nf_options)
715 if (nf_options && nf_options->collectors.n) {
716 if (!ofproto->netflow) {
717 ofproto->netflow = netflow_create();
719 return netflow_set_options(ofproto->netflow, nf_options);
721 netflow_destroy(ofproto->netflow);
722 ofproto->netflow = NULL;
728 ofproto_set_sflow(struct ofproto *ofproto,
729 const struct ofproto_sflow_options *oso)
731 struct ofproto_sflow *os = ofproto->sflow;
734 struct ofport *ofport;
735 unsigned int odp_port;
737 os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
738 refresh_port_groups(ofproto);
739 PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) {
740 ofproto_sflow_add_port(os, odp_port,
741 netdev_get_name(ofport->netdev));
744 ofproto_sflow_set_options(os, oso);
746 ofproto_sflow_destroy(os);
747 ofproto->sflow = NULL;
752 ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp)
756 VLOG_WARN("STP is not yet implemented");
764 ofproto_get_datapath_id(const struct ofproto *ofproto)
766 return ofproto->datapath_id;
770 ofproto_has_controller(const struct ofproto *ofproto)
772 return !hmap_is_empty(&ofproto->controllers);
776 ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners)
780 for (i = 0; i < ofproto->n_listeners; i++) {
781 svec_add(listeners, pvconn_get_name(ofproto->listeners[i]));
786 ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops)
790 for (i = 0; i < ofproto->n_snoops; i++) {
791 svec_add(snoops, pvconn_get_name(ofproto->snoops[i]));
796 ofproto_destroy(struct ofproto *p)
798 struct ofconn *ofconn, *next_ofconn;
799 struct ofport *ofport;
800 unsigned int port_no;
807 /* Destroy fail-open and in-band early, since they touch the classifier. */
808 fail_open_destroy(p->fail_open);
811 in_band_destroy(p->in_band);
814 ofproto_flush_flows(p);
815 classifier_destroy(&p->cls);
817 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
819 ofconn_destroy(ofconn);
821 hmap_destroy(&p->controllers);
824 netdev_monitor_destroy(p->netdev_monitor);
825 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
828 shash_destroy(&p->port_by_name);
830 switch_status_destroy(p->switch_status);
831 netflow_destroy(p->netflow);
832 ofproto_sflow_destroy(p->sflow);
834 for (i = 0; i < p->n_listeners; i++) {
835 pvconn_close(p->listeners[i]);
839 for (i = 0; i < p->n_snoops; i++) {
840 pvconn_close(p->snoops[i]);
844 mac_learning_destroy(p->ml);
849 free(p->serial_desc);
852 port_array_destroy(&p->ports);
858 ofproto_run(struct ofproto *p)
860 int error = ofproto_run1(p);
862 error = ofproto_run2(p, false);
868 process_port_change(struct ofproto *ofproto, int error, char *devname)
870 if (error == ENOBUFS) {
871 reinit_ports(ofproto);
873 update_port(ofproto, devname);
878 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
879 * Connects this vconn to a controller. */
881 add_snooper(struct ofproto *ofproto, struct vconn *vconn)
883 struct ofconn *ofconn;
885 /* Arbitrarily pick the first controller in the list for monitoring. We
886 * could do something smarter or more flexible later, if it ever proves
888 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
889 if (ofconn->type == OFCONN_CONTROLLER) {
890 rconn_add_monitor(ofconn->rconn, vconn);
895 VLOG_INFO_RL(&rl, "no controller connection to monitor");
900 ofproto_run1(struct ofproto *p)
902 struct ofconn *ofconn, *next_ofconn;
907 if (shash_is_empty(&p->port_by_name)) {
911 for (i = 0; i < 50; i++) {
915 error = dpif_recv(p->dpif, &buf);
917 if (error == ENODEV) {
918 /* Someone destroyed the datapath behind our back. The caller
919 * better destroy us and give up, because we're just going to
920 * spin from here on out. */
921 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
922 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
929 handle_odp_msg(p, buf);
932 while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) {
933 process_port_change(p, error, devname);
935 while ((error = netdev_monitor_poll(p->netdev_monitor,
936 &devname)) != EAGAIN) {
937 process_port_change(p, error, devname);
941 in_band_run(p->in_band);
944 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node,
946 ofconn_run(ofconn, p);
949 /* Fail-open maintenance. Do this after processing the ofconns since
950 * fail-open checks the status of the controller rconn. */
952 fail_open_run(p->fail_open);
955 for (i = 0; i < p->n_listeners; i++) {
959 retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn);
961 ofconn_create(p, rconn_new_from_vconn("passive", vconn),
963 } else if (retval != EAGAIN) {
964 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
968 for (i = 0; i < p->n_snoops; i++) {
972 retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn);
974 add_snooper(p, vconn);
975 } else if (retval != EAGAIN) {
976 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
980 if (time_msec() >= p->next_expiration) {
981 COVERAGE_INC(ofproto_expiration);
982 p->next_expiration = time_msec() + 1000;
985 classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p);
987 /* Let the hook know that we're at a stable point: all outstanding data
988 * in existing flows has been accounted to the account_cb. Thus, the
989 * hook can now reasonably do operations that depend on having accurate
990 * flow volume accounting (currently, that's just bond rebalancing). */
991 if (p->ofhooks->account_checkpoint_cb) {
992 p->ofhooks->account_checkpoint_cb(p->aux);
997 netflow_run(p->netflow);
1000 ofproto_sflow_run(p->sflow);
1006 struct revalidate_cbdata {
1007 struct ofproto *ofproto;
1008 bool revalidate_all; /* Revalidate all exact-match rules? */
1009 bool revalidate_subrules; /* Revalidate all exact-match subrules? */
1010 struct tag_set revalidate_set; /* Set of tags to revalidate. */
1014 ofproto_run2(struct ofproto *p, bool revalidate_all)
1016 if (p->need_revalidate || revalidate_all
1017 || !tag_set_is_empty(&p->revalidate_set)) {
1018 struct revalidate_cbdata cbdata;
1020 cbdata.revalidate_all = revalidate_all;
1021 cbdata.revalidate_subrules = p->need_revalidate;
1022 cbdata.revalidate_set = p->revalidate_set;
1023 tag_set_init(&p->revalidate_set);
1024 COVERAGE_INC(ofproto_revalidate);
1025 classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata);
1026 p->need_revalidate = false;
1033 ofproto_wait(struct ofproto *p)
1035 struct ofconn *ofconn;
1038 dpif_recv_wait(p->dpif);
1039 dpif_port_poll_wait(p->dpif);
1040 netdev_monitor_poll_wait(p->netdev_monitor);
1041 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1042 ofconn_wait(ofconn);
1045 in_band_wait(p->in_band);
1048 fail_open_wait(p->fail_open);
1051 ofproto_sflow_wait(p->sflow);
1053 if (!tag_set_is_empty(&p->revalidate_set)) {
1054 poll_immediate_wake();
1056 if (p->need_revalidate) {
1057 /* Shouldn't happen, but if it does just go around again. */
1058 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1059 poll_immediate_wake();
1060 } else if (p->next_expiration != LLONG_MAX) {
1061 poll_timer_wait(p->next_expiration - time_msec());
1063 for (i = 0; i < p->n_listeners; i++) {
1064 pvconn_wait(p->listeners[i]);
1066 for (i = 0; i < p->n_snoops; i++) {
1067 pvconn_wait(p->snoops[i]);
1072 ofproto_revalidate(struct ofproto *ofproto, tag_type tag)
1074 tag_set_add(&ofproto->revalidate_set, tag);
1078 ofproto_get_revalidate_set(struct ofproto *ofproto)
1080 return &ofproto->revalidate_set;
1084 ofproto_is_alive(const struct ofproto *p)
1086 return !hmap_is_empty(&p->controllers);
1090 ofproto_send_packet(struct ofproto *p, const flow_t *flow,
1091 const union ofp_action *actions, size_t n_actions,
1092 const struct ofpbuf *packet)
1094 struct odp_actions odp_actions;
1097 error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions,
1103 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1105 dpif_execute(p->dpif, flow->in_port, odp_actions.actions,
1106 odp_actions.n_actions, packet);
1111 ofproto_add_flow(struct ofproto *p,
1112 const flow_t *flow, uint32_t wildcards, unsigned int priority,
1113 const union ofp_action *actions, size_t n_actions,
1117 rule = rule_create(p, NULL, actions, n_actions,
1118 idle_timeout >= 0 ? idle_timeout : 5 /* XXX */,
1120 cls_rule_from_flow(flow, wildcards, priority, &rule->cr);
1121 rule_insert(p, rule, NULL, 0);
1125 ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow,
1126 uint32_t wildcards, unsigned int priority)
1130 rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls,
1134 rule_remove(ofproto, rule);
1139 destroy_rule(struct cls_rule *rule_, void *ofproto_)
1141 struct rule *rule = rule_from_cls_rule(rule_);
1142 struct ofproto *ofproto = ofproto_;
1144 /* Mark the flow as not installed, even though it might really be
1145 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1146 * There is no point in uninstalling it individually since we are about to
1147 * blow away all the flows with dpif_flow_flush(). */
1148 rule->installed = false;
1150 rule_remove(ofproto, rule);
1154 ofproto_flush_flows(struct ofproto *ofproto)
1156 COVERAGE_INC(ofproto_flush);
1157 classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto);
1158 dpif_flow_flush(ofproto->dpif);
1159 if (ofproto->in_band) {
1160 in_band_flushed(ofproto->in_band);
1162 if (ofproto->fail_open) {
1163 fail_open_flushed(ofproto->fail_open);
1168 reinit_ports(struct ofproto *p)
1170 struct svec devnames;
1171 struct ofport *ofport;
1172 unsigned int port_no;
1173 struct odp_port *odp_ports;
1177 svec_init(&devnames);
1178 PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) {
1179 svec_add (&devnames, (char *) ofport->opp.name);
1181 dpif_port_list(p->dpif, &odp_ports, &n_odp_ports);
1182 for (i = 0; i < n_odp_ports; i++) {
1183 svec_add (&devnames, odp_ports[i].devname);
1187 svec_sort_unique(&devnames);
1188 for (i = 0; i < devnames.n; i++) {
1189 update_port(p, devnames.names[i]);
1191 svec_destroy(&devnames);
1195 refresh_port_group(struct ofproto *p, unsigned int group)
1199 struct ofport *port;
1200 unsigned int port_no;
1202 assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD);
1204 ports = xmalloc(port_array_count(&p->ports) * sizeof *ports);
1206 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
1207 if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) {
1208 ports[n_ports++] = port_no;
1211 dpif_port_group_set(p->dpif, group, ports, n_ports);
1218 refresh_port_groups(struct ofproto *p)
1220 size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD);
1221 size_t n_all = refresh_port_group(p, DP_GROUP_ALL);
1223 ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all);
1227 static struct ofport *
1228 make_ofport(const struct odp_port *odp_port)
1230 struct netdev_options netdev_options;
1231 enum netdev_flags flags;
1232 struct ofport *ofport;
1233 struct netdev *netdev;
1237 memset(&netdev_options, 0, sizeof netdev_options);
1238 netdev_options.name = odp_port->devname;
1239 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1240 netdev_options.may_open = true;
1242 error = netdev_open(&netdev_options, &netdev);
1244 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1245 "cannot be opened (%s)",
1246 odp_port->devname, odp_port->port,
1247 odp_port->devname, strerror(error));
1251 ofport = xmalloc(sizeof *ofport);
1252 ofport->netdev = netdev;
1253 ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port);
1254 netdev_get_etheraddr(netdev, ofport->opp.hw_addr);
1255 memcpy(ofport->opp.name, odp_port->devname,
1256 MIN(sizeof ofport->opp.name, sizeof odp_port->devname));
1257 ofport->opp.name[sizeof ofport->opp.name - 1] = '\0';
1259 netdev_get_flags(netdev, &flags);
1260 ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN;
1262 netdev_get_carrier(netdev, &carrier);
1263 ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN;
1265 netdev_get_features(netdev,
1266 &ofport->opp.curr, &ofport->opp.advertised,
1267 &ofport->opp.supported, &ofport->opp.peer);
1272 ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port)
1274 if (port_array_get(&p->ports, odp_port->port)) {
1275 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1278 } else if (shash_find(&p->port_by_name, odp_port->devname)) {
1279 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1288 ofport_equal(const struct ofport *a_, const struct ofport *b_)
1290 const struct ofp_phy_port *a = &a_->opp;
1291 const struct ofp_phy_port *b = &b_->opp;
1293 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1294 return (a->port_no == b->port_no
1295 && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1296 && !strcmp((char *) a->name, (char *) b->name)
1297 && a->state == b->state
1298 && a->config == b->config
1299 && a->curr == b->curr
1300 && a->advertised == b->advertised
1301 && a->supported == b->supported
1302 && a->peer == b->peer);
1306 send_port_status(struct ofproto *p, const struct ofport *ofport,
1309 /* XXX Should limit the number of queued port status change messages. */
1310 struct ofconn *ofconn;
1311 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
1312 struct ofp_port_status *ops;
1315 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1316 ops->reason = reason;
1317 ops->desc = ofport->opp;
1318 hton_ofp_phy_port(&ops->desc);
1319 queue_tx(b, ofconn, NULL);
1321 if (p->ofhooks->port_changed_cb) {
1322 p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux);
1327 ofport_install(struct ofproto *p, struct ofport *ofport)
1329 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1330 const char *netdev_name = (const char *) ofport->opp.name;
1332 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1333 port_array_set(&p->ports, odp_port, ofport);
1334 shash_add(&p->port_by_name, netdev_name, ofport);
1336 ofproto_sflow_add_port(p->sflow, odp_port, netdev_name);
1341 ofport_remove(struct ofproto *p, struct ofport *ofport)
1343 uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no);
1345 netdev_monitor_remove(p->netdev_monitor, ofport->netdev);
1346 port_array_set(&p->ports, odp_port, NULL);
1347 shash_delete(&p->port_by_name,
1348 shash_find(&p->port_by_name, (char *) ofport->opp.name));
1350 ofproto_sflow_del_port(p->sflow, odp_port);
1355 ofport_free(struct ofport *ofport)
1358 netdev_close(ofport->netdev);
1364 update_port(struct ofproto *p, const char *devname)
1366 struct odp_port odp_port;
1367 struct ofport *old_ofport;
1368 struct ofport *new_ofport;
1371 COVERAGE_INC(ofproto_update_port);
1373 /* Query the datapath for port information. */
1374 error = dpif_port_query_by_name(p->dpif, devname, &odp_port);
1376 /* Find the old ofport. */
1377 old_ofport = shash_find_data(&p->port_by_name, devname);
1380 /* There's no port named 'devname' but there might be a port with
1381 * the same port number. This could happen if a port is deleted
1382 * and then a new one added in its place very quickly, or if a port
1383 * is renamed. In the former case we want to send an OFPPR_DELETE
1384 * and an OFPPR_ADD, and in the latter case we want to send a
1385 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1386 * the old port's ifindex against the new port, or perhaps less
1387 * reliably but more portably by comparing the old port's MAC
1388 * against the new port's MAC. However, this code isn't that smart
1389 * and always sends an OFPPR_MODIFY (XXX). */
1390 old_ofport = port_array_get(&p->ports, odp_port.port);
1392 } else if (error != ENOENT && error != ENODEV) {
1393 VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error "
1394 "%s", strerror(error));
1398 /* Create a new ofport. */
1399 new_ofport = !error ? make_ofport(&odp_port) : NULL;
1401 /* Eliminate a few pathological cases. */
1402 if (!old_ofport && !new_ofport) {
1404 } else if (old_ofport && new_ofport) {
1405 /* Most of the 'config' bits are OpenFlow soft state, but
1406 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1407 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1408 * leaves the other bits 0.) */
1409 new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN;
1411 if (ofport_equal(old_ofport, new_ofport)) {
1412 /* False alarm--no change. */
1413 ofport_free(new_ofport);
1418 /* Now deal with the normal cases. */
1420 ofport_remove(p, old_ofport);
1423 ofport_install(p, new_ofport);
1425 send_port_status(p, new_ofport ? new_ofport : old_ofport,
1426 (!old_ofport ? OFPPR_ADD
1427 : !new_ofport ? OFPPR_DELETE
1429 ofport_free(old_ofport);
1431 /* Update port groups. */
1432 refresh_port_groups(p);
1436 init_ports(struct ofproto *p)
1438 struct odp_port *ports;
1443 error = dpif_port_list(p->dpif, &ports, &n_ports);
1448 for (i = 0; i < n_ports; i++) {
1449 const struct odp_port *odp_port = &ports[i];
1450 if (!ofport_conflicts(p, odp_port)) {
1451 struct ofport *ofport = make_ofport(odp_port);
1453 ofport_install(p, ofport);
1458 refresh_port_groups(p);
1462 static struct ofconn *
1463 ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type)
1465 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
1466 ofconn->ofproto = p;
1467 list_push_back(&p->all_conns, &ofconn->node);
1468 ofconn->rconn = rconn;
1469 ofconn->type = type;
1470 ofconn->packet_in_counter = rconn_packet_counter_create ();
1471 ofconn->pktbuf = NULL;
1472 ofconn->miss_send_len = 0;
1473 ofconn->reply_counter = rconn_packet_counter_create ();
1478 ofconn_destroy(struct ofconn *ofconn)
1480 if (ofconn->type == OFCONN_CONTROLLER) {
1481 hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node);
1483 discovery_destroy(ofconn->discovery);
1485 list_remove(&ofconn->node);
1486 switch_status_unregister(ofconn->ss);
1487 rconn_destroy(ofconn->rconn);
1488 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1489 rconn_packet_counter_destroy(ofconn->reply_counter);
1490 pktbuf_destroy(ofconn->pktbuf);
1495 ofconn_run(struct ofconn *ofconn, struct ofproto *p)
1500 if (ofconn->discovery) {
1501 char *controller_name;
1502 if (rconn_is_connectivity_questionable(ofconn->rconn)) {
1503 discovery_question_connectivity(ofconn->discovery);
1505 if (discovery_run(ofconn->discovery, &controller_name)) {
1506 if (controller_name) {
1507 rconn_connect(ofconn->rconn, controller_name);
1509 rconn_disconnect(ofconn->rconn);
1514 for (i = 0; i < N_SCHEDULERS; i++) {
1515 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
1518 rconn_run(ofconn->rconn);
1520 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1521 /* Limit the number of iterations to prevent other tasks from
1523 for (iteration = 0; iteration < 50; iteration++) {
1524 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1529 fail_open_maybe_recover(p->fail_open);
1531 handle_openflow(ofconn, p, of_msg);
1532 ofpbuf_delete(of_msg);
1536 if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) {
1537 ofconn_destroy(ofconn);
1542 ofconn_wait(struct ofconn *ofconn)
1546 if (ofconn->discovery) {
1547 discovery_wait(ofconn->discovery);
1549 for (i = 0; i < N_SCHEDULERS; i++) {
1550 pinsched_wait(ofconn->schedulers[i]);
1552 rconn_run_wait(ofconn->rconn);
1553 if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) {
1554 rconn_recv_wait(ofconn->rconn);
1556 COVERAGE_INC(ofproto_ofconn_stuck);
1560 /* Caller is responsible for initializing the 'cr' member of the returned
1562 static struct rule *
1563 rule_create(struct ofproto *ofproto, struct rule *super,
1564 const union ofp_action *actions, size_t n_actions,
1565 uint16_t idle_timeout, uint16_t hard_timeout,
1566 uint64_t flow_cookie, bool send_flow_removed)
1568 struct rule *rule = xzalloc(sizeof *rule);
1569 rule->idle_timeout = idle_timeout;
1570 rule->hard_timeout = hard_timeout;
1571 rule->flow_cookie = flow_cookie;
1572 rule->used = rule->created = time_msec();
1573 rule->send_flow_removed = send_flow_removed;
1574 rule->super = super;
1576 list_push_back(&super->list, &rule->list);
1578 list_init(&rule->list);
1580 rule->n_actions = n_actions;
1581 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1582 netflow_flow_clear(&rule->nf_flow);
1583 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created);
1588 static struct rule *
1589 rule_from_cls_rule(const struct cls_rule *cls_rule)
1591 return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL;
1595 rule_free(struct rule *rule)
1597 free(rule->actions);
1598 free(rule->odp_actions);
1602 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1603 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1604 * through all of its subrules and revalidates them, destroying any that no
1605 * longer has a super-rule (which is probably all of them).
1607 * Before calling this function, the caller must make have removed 'rule' from
1608 * the classifier. If 'rule' is an exact-match rule, the caller is also
1609 * responsible for ensuring that it has been uninstalled from the datapath. */
1611 rule_destroy(struct ofproto *ofproto, struct rule *rule)
1614 struct rule *subrule, *next;
1615 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
1616 revalidate_rule(ofproto, subrule);
1619 list_remove(&rule->list);
1625 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1627 const union ofp_action *oa;
1628 struct actions_iterator i;
1630 if (out_port == htons(OFPP_NONE)) {
1633 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1634 oa = actions_next(&i)) {
1635 if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) {
1642 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1643 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1645 * The flow that 'packet' actually contains does not need to actually match
1646 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1647 * the packet and byte counters for 'rule' will be credited for the packet sent
1648 * out whether or not the packet actually matches 'rule'.
1650 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1651 * the caller must already have accurately composed ODP actions for it given
1652 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1653 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1654 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1655 * actions and apply them to 'packet'. */
1657 rule_execute(struct ofproto *ofproto, struct rule *rule,
1658 struct ofpbuf *packet, const flow_t *flow)
1660 const union odp_action *actions;
1662 struct odp_actions a;
1664 /* Grab or compose the ODP actions.
1666 * The special case for an exact-match 'rule' where 'flow' is not the
1667 * rule's flow is important to avoid, e.g., sending a packet out its input
1668 * port simply because the ODP actions were composed for the wrong
1670 if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) {
1671 struct rule *super = rule->super ? rule->super : rule;
1672 if (xlate_actions(super->actions, super->n_actions, flow, ofproto,
1673 packet, &a, NULL, 0, NULL)) {
1676 actions = a.actions;
1677 n_actions = a.n_actions;
1679 actions = rule->odp_actions;
1680 n_actions = rule->n_odp_actions;
1683 /* Execute the ODP actions. */
1684 if (!dpif_execute(ofproto->dpif, flow->in_port,
1685 actions, n_actions, packet)) {
1686 struct odp_flow_stats stats;
1687 flow_extract_stats(flow, packet, &stats);
1688 update_stats(ofproto, rule, &stats);
1689 rule->used = time_msec();
1690 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used);
1695 rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet,
1698 struct rule *displaced_rule;
1700 /* Insert the rule in the classifier. */
1701 displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr));
1702 if (!rule->cr.wc.wildcards) {
1703 rule_make_actions(p, rule, packet);
1706 /* Send the packet and credit it to the rule. */
1709 flow_extract(packet, 0, in_port, &flow);
1710 rule_execute(p, rule, packet, &flow);
1713 /* Install the rule in the datapath only after sending the packet, to
1714 * avoid packet reordering. */
1715 if (rule->cr.wc.wildcards) {
1716 COVERAGE_INC(ofproto_add_wc_flow);
1717 p->need_revalidate = true;
1719 rule_install(p, rule, displaced_rule);
1722 /* Free the rule that was displaced, if any. */
1723 if (displaced_rule) {
1724 rule_destroy(p, displaced_rule);
1728 static struct rule *
1729 rule_create_subrule(struct ofproto *ofproto, struct rule *rule,
1732 struct rule *subrule = rule_create(ofproto, rule, NULL, 0,
1733 rule->idle_timeout, rule->hard_timeout,
1735 COVERAGE_INC(ofproto_subrule_create);
1736 cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX
1737 : rule->cr.priority), &subrule->cr);
1738 classifier_insert_exact(&ofproto->cls, &subrule->cr);
1744 rule_remove(struct ofproto *ofproto, struct rule *rule)
1746 if (rule->cr.wc.wildcards) {
1747 COVERAGE_INC(ofproto_del_wc_flow);
1748 ofproto->need_revalidate = true;
1750 rule_uninstall(ofproto, rule);
1752 classifier_remove(&ofproto->cls, &rule->cr);
1753 rule_destroy(ofproto, rule);
1756 /* Returns true if the actions changed, false otherwise. */
1758 rule_make_actions(struct ofproto *p, struct rule *rule,
1759 const struct ofpbuf *packet)
1761 const struct rule *super;
1762 struct odp_actions a;
1765 assert(!rule->cr.wc.wildcards);
1767 super = rule->super ? rule->super : rule;
1769 xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p,
1770 packet, &a, &rule->tags, &rule->may_install,
1771 &rule->nf_flow.output_iface);
1773 actions_len = a.n_actions * sizeof *a.actions;
1774 if (rule->n_odp_actions != a.n_actions
1775 || memcmp(rule->odp_actions, a.actions, actions_len)) {
1776 COVERAGE_INC(ofproto_odp_unchanged);
1777 free(rule->odp_actions);
1778 rule->n_odp_actions = a.n_actions;
1779 rule->odp_actions = xmemdup(a.actions, actions_len);
1787 do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags,
1788 struct odp_flow_put *put)
1790 memset(&put->flow.stats, 0, sizeof put->flow.stats);
1791 put->flow.key = rule->cr.flow;
1792 put->flow.actions = rule->odp_actions;
1793 put->flow.n_actions = rule->n_odp_actions;
1794 put->flow.flags = 0;
1796 return dpif_flow_put(ofproto->dpif, put);
1800 rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule)
1802 assert(!rule->cr.wc.wildcards);
1804 if (rule->may_install) {
1805 struct odp_flow_put put;
1806 if (!do_put_flow(p, rule,
1807 ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS,
1809 rule->installed = true;
1810 if (displaced_rule) {
1811 update_stats(p, displaced_rule, &put.flow.stats);
1812 rule_post_uninstall(p, displaced_rule);
1815 } else if (displaced_rule) {
1816 rule_uninstall(p, displaced_rule);
1821 rule_reinstall(struct ofproto *ofproto, struct rule *rule)
1823 if (rule->installed) {
1824 struct odp_flow_put put;
1825 COVERAGE_INC(ofproto_dp_missed);
1826 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put);
1828 rule_install(ofproto, rule, NULL);
1833 rule_update_actions(struct ofproto *ofproto, struct rule *rule)
1835 bool actions_changed;
1836 uint16_t new_out_iface, old_out_iface;
1838 old_out_iface = rule->nf_flow.output_iface;
1839 actions_changed = rule_make_actions(ofproto, rule, NULL);
1841 if (rule->may_install) {
1842 if (rule->installed) {
1843 if (actions_changed) {
1844 struct odp_flow_put put;
1845 do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY
1846 | ODPPF_ZERO_STATS, &put);
1847 update_stats(ofproto, rule, &put.flow.stats);
1849 /* Temporarily set the old output iface so that NetFlow
1850 * messages have the correct output interface for the old
1852 new_out_iface = rule->nf_flow.output_iface;
1853 rule->nf_flow.output_iface = old_out_iface;
1854 rule_post_uninstall(ofproto, rule);
1855 rule->nf_flow.output_iface = new_out_iface;
1858 rule_install(ofproto, rule, NULL);
1861 rule_uninstall(ofproto, rule);
1866 rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes)
1868 uint64_t total_bytes = rule->byte_count + extra_bytes;
1870 if (ofproto->ofhooks->account_flow_cb
1871 && total_bytes > rule->accounted_bytes)
1873 ofproto->ofhooks->account_flow_cb(
1874 &rule->cr.flow, rule->odp_actions, rule->n_odp_actions,
1875 total_bytes - rule->accounted_bytes, ofproto->aux);
1876 rule->accounted_bytes = total_bytes;
1881 rule_uninstall(struct ofproto *p, struct rule *rule)
1883 assert(!rule->cr.wc.wildcards);
1884 if (rule->installed) {
1885 struct odp_flow odp_flow;
1887 odp_flow.key = rule->cr.flow;
1888 odp_flow.actions = NULL;
1889 odp_flow.n_actions = 0;
1891 if (!dpif_flow_del(p->dpif, &odp_flow)) {
1892 update_stats(p, rule, &odp_flow.stats);
1894 rule->installed = false;
1896 rule_post_uninstall(p, rule);
1901 is_controller_rule(struct rule *rule)
1903 /* If the only action is send to the controller then don't report
1904 * NetFlow expiration messages since it is just part of the control
1905 * logic for the network and not real traffic. */
1907 if (rule && rule->super) {
1908 struct rule *super = rule->super;
1910 return super->n_actions == 1 &&
1911 super->actions[0].type == htons(OFPAT_OUTPUT) &&
1912 super->actions[0].output.port == htons(OFPP_CONTROLLER);
1919 rule_post_uninstall(struct ofproto *ofproto, struct rule *rule)
1921 struct rule *super = rule->super;
1923 rule_account(ofproto, rule, 0);
1925 if (ofproto->netflow && !is_controller_rule(rule)) {
1926 struct ofexpired expired;
1927 expired.flow = rule->cr.flow;
1928 expired.packet_count = rule->packet_count;
1929 expired.byte_count = rule->byte_count;
1930 expired.used = rule->used;
1931 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
1934 super->packet_count += rule->packet_count;
1935 super->byte_count += rule->byte_count;
1937 /* Reset counters to prevent double counting if the rule ever gets
1939 rule->packet_count = 0;
1940 rule->byte_count = 0;
1941 rule->accounted_bytes = 0;
1943 netflow_flow_clear(&rule->nf_flow);
1948 queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn,
1949 struct rconn_packet_counter *counter)
1951 update_openflow_length(msg);
1952 if (rconn_send(ofconn->rconn, msg, counter)) {
1958 send_error(const struct ofconn *ofconn, const struct ofp_header *oh,
1959 int error, const void *data, size_t len)
1962 struct ofp_error_msg *oem;
1964 if (!(error >> 16)) {
1965 VLOG_WARN_RL(&rl, "not sending bad error code %d to controller",
1970 COVERAGE_INC(ofproto_error);
1971 oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR,
1972 oh ? oh->xid : 0, &buf);
1973 oem->type = htons((unsigned int) error >> 16);
1974 oem->code = htons(error & 0xffff);
1975 memcpy(oem->data, data, len);
1976 queue_tx(buf, ofconn, ofconn->reply_counter);
1980 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1983 size_t oh_length = ntohs(oh->length);
1984 send_error(ofconn, oh, error, oh, MIN(oh_length, 64));
1988 hton_ofp_phy_port(struct ofp_phy_port *opp)
1990 opp->port_no = htons(opp->port_no);
1991 opp->config = htonl(opp->config);
1992 opp->state = htonl(opp->state);
1993 opp->curr = htonl(opp->curr);
1994 opp->advertised = htonl(opp->advertised);
1995 opp->supported = htonl(opp->supported);
1996 opp->peer = htonl(opp->peer);
2000 handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh)
2002 struct ofp_header *rq = oh;
2003 queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter);
2008 handle_features_request(struct ofproto *p, struct ofconn *ofconn,
2009 struct ofp_header *oh)
2011 struct ofp_switch_features *osf;
2013 unsigned int port_no;
2014 struct ofport *port;
2016 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
2017 osf->datapath_id = htonll(p->datapath_id);
2018 osf->n_buffers = htonl(pktbuf_capacity());
2020 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
2021 OFPC_PORT_STATS | OFPC_ARP_MATCH_IP);
2022 osf->actions = htonl((1u << OFPAT_OUTPUT) |
2023 (1u << OFPAT_SET_VLAN_VID) |
2024 (1u << OFPAT_SET_VLAN_PCP) |
2025 (1u << OFPAT_STRIP_VLAN) |
2026 (1u << OFPAT_SET_DL_SRC) |
2027 (1u << OFPAT_SET_DL_DST) |
2028 (1u << OFPAT_SET_NW_SRC) |
2029 (1u << OFPAT_SET_NW_DST) |
2030 (1u << OFPAT_SET_NW_TOS) |
2031 (1u << OFPAT_SET_TP_SRC) |
2032 (1u << OFPAT_SET_TP_DST));
2034 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2035 hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp));
2038 queue_tx(buf, ofconn, ofconn->reply_counter);
2043 handle_get_config_request(struct ofproto *p, struct ofconn *ofconn,
2044 struct ofp_header *oh)
2047 struct ofp_switch_config *osc;
2051 /* Figure out flags. */
2052 dpif_get_drop_frags(p->dpif, &drop_frags);
2053 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
2056 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2057 osc->flags = htons(flags);
2058 osc->miss_send_len = htons(ofconn->miss_send_len);
2059 queue_tx(buf, ofconn, ofconn->reply_counter);
2065 handle_set_config(struct ofproto *p, struct ofconn *ofconn,
2066 struct ofp_switch_config *osc)
2071 error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc);
2075 flags = ntohs(osc->flags);
2077 if (ofconn->type == OFCONN_CONTROLLER) {
2078 switch (flags & OFPC_FRAG_MASK) {
2079 case OFPC_FRAG_NORMAL:
2080 dpif_set_drop_frags(p->dpif, false);
2082 case OFPC_FRAG_DROP:
2083 dpif_set_drop_frags(p->dpif, true);
2086 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
2092 ofconn->miss_send_len = ntohs(osc->miss_send_len);
2098 add_output_group_action(struct odp_actions *actions, uint16_t group,
2099 uint16_t *nf_output_iface)
2101 odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group;
2103 if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) {
2104 *nf_output_iface = NF_OUT_FLOOD;
2109 add_controller_action(struct odp_actions *actions,
2110 const struct ofp_action_output *oao)
2112 union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER);
2113 a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX;
2116 struct action_xlate_ctx {
2118 flow_t flow; /* Flow to which these actions correspond. */
2119 int recurse; /* Recursion level, via xlate_table_action. */
2120 struct ofproto *ofproto;
2121 const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a
2122 * null pointer if we are revalidating
2123 * without a packet to refer to. */
2126 struct odp_actions *out; /* Datapath actions. */
2127 tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */
2128 bool may_set_up_flow; /* True ordinarily; false if the actions must
2129 * be reassessed for every packet. */
2130 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
2133 static void do_xlate_actions(const union ofp_action *in, size_t n_in,
2134 struct action_xlate_ctx *ctx);
2137 add_output_action(struct action_xlate_ctx *ctx, uint16_t port)
2139 const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port);
2142 if (ofport->opp.config & OFPPC_NO_FWD) {
2143 /* Forwarding disabled on port. */
2148 * We don't have an ofport record for this port, but it doesn't hurt to
2149 * allow forwarding to it anyhow. Maybe such a port will appear later
2150 * and we're pre-populating the flow table.
2154 odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port;
2155 ctx->nf_output_iface = port;
2158 static struct rule *
2159 lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow)
2162 rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow));
2164 /* The rule we found might not be valid, since we could be in need of
2165 * revalidation. If it is not valid, don't return it. */
2168 && ofproto->need_revalidate
2169 && !revalidate_rule(ofproto, rule)) {
2170 COVERAGE_INC(ofproto_invalidated);
2178 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
2180 if (!ctx->recurse) {
2181 uint16_t old_in_port;
2184 /* Look up a flow with 'in_port' as the input port. Then restore the
2185 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2186 * have surprising behavior). */
2187 old_in_port = ctx->flow.in_port;
2188 ctx->flow.in_port = in_port;
2189 rule = lookup_valid_rule(ctx->ofproto, &ctx->flow);
2190 ctx->flow.in_port = old_in_port;
2198 do_xlate_actions(rule->actions, rule->n_actions, ctx);
2205 xlate_output_action(struct action_xlate_ctx *ctx,
2206 const struct ofp_action_output *oao)
2209 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
2211 ctx->nf_output_iface = NF_OUT_DROP;
2213 switch (ntohs(oao->port)) {
2215 add_output_action(ctx, ctx->flow.in_port);
2218 xlate_table_action(ctx, ctx->flow.in_port);
2221 if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet,
2222 ctx->out, ctx->tags,
2223 &ctx->nf_output_iface,
2224 ctx->ofproto->aux)) {
2225 COVERAGE_INC(ofproto_uninstallable);
2226 ctx->may_set_up_flow = false;
2230 add_output_group_action(ctx->out, DP_GROUP_FLOOD,
2231 &ctx->nf_output_iface);
2234 add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface);
2236 case OFPP_CONTROLLER:
2237 add_controller_action(ctx->out, oao);
2240 add_output_action(ctx, ODPP_LOCAL);
2243 odp_port = ofp_port_to_odp_port(ntohs(oao->port));
2244 if (odp_port != ctx->flow.in_port) {
2245 add_output_action(ctx, odp_port);
2250 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2251 ctx->nf_output_iface = NF_OUT_FLOOD;
2252 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
2253 ctx->nf_output_iface = prev_nf_output_iface;
2254 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2255 ctx->nf_output_iface != NF_OUT_FLOOD) {
2256 ctx->nf_output_iface = NF_OUT_MULTI;
2261 xlate_nicira_action(struct action_xlate_ctx *ctx,
2262 const struct nx_action_header *nah)
2264 const struct nx_action_resubmit *nar;
2265 const struct nx_action_set_tunnel *nast;
2266 union odp_action *oa;
2267 int subtype = ntohs(nah->subtype);
2269 assert(nah->vendor == htonl(NX_VENDOR_ID));
2271 case NXAST_RESUBMIT:
2272 nar = (const struct nx_action_resubmit *) nah;
2273 xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port)));
2276 case NXAST_SET_TUNNEL:
2277 nast = (const struct nx_action_set_tunnel *) nah;
2278 oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL);
2279 ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id;
2282 /* If you add a new action here that modifies flow data, don't forget to
2283 * update the flow key in ctx->flow in the same key. */
2286 VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype);
2292 do_xlate_actions(const union ofp_action *in, size_t n_in,
2293 struct action_xlate_ctx *ctx)
2295 struct actions_iterator iter;
2296 const union ofp_action *ia;
2297 const struct ofport *port;
2299 port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port);
2300 if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
2301 port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr)
2302 ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) {
2303 /* Drop this flow. */
2307 for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
2308 uint16_t type = ntohs(ia->type);
2309 union odp_action *oa;
2313 xlate_output_action(ctx, &ia->output);
2316 case OFPAT_SET_VLAN_VID:
2317 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID);
2318 ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid;
2321 case OFPAT_SET_VLAN_PCP:
2322 oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP);
2323 ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp;
2326 case OFPAT_STRIP_VLAN:
2327 odp_actions_add(ctx->out, ODPAT_STRIP_VLAN);
2328 ctx->flow.dl_vlan = OFP_VLAN_NONE;
2329 ctx->flow.dl_vlan_pcp = 0;
2332 case OFPAT_SET_DL_SRC:
2333 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC);
2334 memcpy(oa->dl_addr.dl_addr,
2335 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2336 memcpy(ctx->flow.dl_src,
2337 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2340 case OFPAT_SET_DL_DST:
2341 oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST);
2342 memcpy(oa->dl_addr.dl_addr,
2343 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2344 memcpy(ctx->flow.dl_dst,
2345 ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN);
2348 case OFPAT_SET_NW_SRC:
2349 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC);
2350 ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2353 case OFPAT_SET_NW_DST:
2354 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST);
2355 ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr;
2358 case OFPAT_SET_NW_TOS:
2359 oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS);
2360 ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos;
2363 case OFPAT_SET_TP_SRC:
2364 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC);
2365 ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port;
2368 case OFPAT_SET_TP_DST:
2369 oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST);
2370 ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port;
2374 xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
2378 VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type);
2385 xlate_actions(const union ofp_action *in, size_t n_in,
2386 const flow_t *flow, struct ofproto *ofproto,
2387 const struct ofpbuf *packet,
2388 struct odp_actions *out, tag_type *tags, bool *may_set_up_flow,
2389 uint16_t *nf_output_iface)
2391 tag_type no_tags = 0;
2392 struct action_xlate_ctx ctx;
2393 COVERAGE_INC(ofproto_ofp2odp);
2394 odp_actions_init(out);
2397 ctx.ofproto = ofproto;
2398 ctx.packet = packet;
2400 ctx.tags = tags ? tags : &no_tags;
2401 ctx.may_set_up_flow = true;
2402 ctx.nf_output_iface = NF_OUT_DROP;
2403 do_xlate_actions(in, n_in, &ctx);
2405 /* Check with in-band control to see if we're allowed to set up this
2407 if (!in_band_rule_check(ofproto->in_band, flow, out)) {
2408 ctx.may_set_up_flow = false;
2411 if (may_set_up_flow) {
2412 *may_set_up_flow = ctx.may_set_up_flow;
2414 if (nf_output_iface) {
2415 *nf_output_iface = ctx.nf_output_iface;
2417 if (odp_actions_overflow(out)) {
2418 odp_actions_init(out);
2419 return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY);
2425 handle_packet_out(struct ofproto *p, struct ofconn *ofconn,
2426 struct ofp_header *oh)
2428 struct ofp_packet_out *opo;
2429 struct ofpbuf payload, *buffer;
2430 struct odp_actions actions;
2436 error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports);
2440 opo = (struct ofp_packet_out *) oh;
2442 COVERAGE_INC(ofproto_packet_out);
2443 if (opo->buffer_id != htonl(UINT32_MAX)) {
2444 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id),
2446 if (error || !buffer) {
2454 flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow);
2455 error = xlate_actions((const union ofp_action *) opo->actions, n_actions,
2456 &flow, p, &payload, &actions, NULL, NULL, NULL);
2461 dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions,
2463 ofpbuf_delete(buffer);
2469 update_port_config(struct ofproto *p, struct ofport *port,
2470 uint32_t config, uint32_t mask)
2472 mask &= config ^ port->opp.config;
2473 if (mask & OFPPC_PORT_DOWN) {
2474 if (config & OFPPC_PORT_DOWN) {
2475 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2477 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2480 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2481 if (mask & REVALIDATE_BITS) {
2482 COVERAGE_INC(ofproto_costly_flags);
2483 port->opp.config ^= mask & REVALIDATE_BITS;
2484 p->need_revalidate = true;
2486 #undef REVALIDATE_BITS
2487 if (mask & OFPPC_NO_FLOOD) {
2488 port->opp.config ^= OFPPC_NO_FLOOD;
2489 refresh_port_groups(p);
2491 if (mask & OFPPC_NO_PACKET_IN) {
2492 port->opp.config ^= OFPPC_NO_PACKET_IN;
2497 handle_port_mod(struct ofproto *p, struct ofp_header *oh)
2499 const struct ofp_port_mod *opm;
2500 struct ofport *port;
2503 error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm);
2507 opm = (struct ofp_port_mod *) oh;
2509 port = port_array_get(&p->ports,
2510 ofp_port_to_odp_port(ntohs(opm->port_no)));
2512 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
2513 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
2514 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
2516 update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask));
2517 if (opm->advertise) {
2518 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
2524 static struct ofpbuf *
2525 make_stats_reply(uint32_t xid, uint16_t type, size_t body_len)
2527 struct ofp_stats_reply *osr;
2530 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
2531 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
2533 osr->flags = htons(0);
2537 static struct ofpbuf *
2538 start_stats_reply(const struct ofp_stats_request *request, size_t body_len)
2540 return make_stats_reply(request->header.xid, request->type, body_len);
2544 append_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp)
2546 struct ofpbuf *msg = *msgp;
2547 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
2548 if (nbytes + msg->size > UINT16_MAX) {
2549 struct ofp_stats_reply *reply = msg->data;
2550 reply->flags = htons(OFPSF_REPLY_MORE);
2551 *msgp = make_stats_reply(reply->header.xid, reply->type, nbytes);
2552 queue_tx(msg, ofconn, ofconn->reply_counter);
2554 return ofpbuf_put_uninit(*msgp, nbytes);
2558 handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn,
2559 struct ofp_stats_request *request)
2561 struct ofp_desc_stats *ods;
2564 msg = start_stats_reply(request, sizeof *ods);
2565 ods = append_stats_reply(sizeof *ods, ofconn, &msg);
2566 memset(ods, 0, sizeof *ods);
2567 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2568 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2569 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2570 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2571 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2572 queue_tx(msg, ofconn, ofconn->reply_counter);
2578 count_subrules(struct cls_rule *cls_rule, void *n_subrules_)
2580 struct rule *rule = rule_from_cls_rule(cls_rule);
2581 int *n_subrules = n_subrules_;
2589 handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn,
2590 struct ofp_stats_request *request)
2592 struct ofp_table_stats *ots;
2594 struct odp_stats dpstats;
2595 int n_exact, n_subrules, n_wild;
2597 msg = start_stats_reply(request, sizeof *ots * 2);
2599 /* Count rules of various kinds. */
2601 classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules);
2602 n_exact = classifier_count_exact(&p->cls) - n_subrules;
2603 n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls);
2606 dpif_get_dp_stats(p->dpif, &dpstats);
2607 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2608 memset(ots, 0, sizeof *ots);
2609 ots->table_id = TABLEID_HASH;
2610 strcpy(ots->name, "hash");
2611 ots->wildcards = htonl(0);
2612 ots->max_entries = htonl(dpstats.max_capacity);
2613 ots->active_count = htonl(n_exact);
2614 ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit +
2616 ots->matched_count = htonll(dpstats.n_hit); /* XXX */
2618 /* Classifier table. */
2619 ots = append_stats_reply(sizeof *ots, ofconn, &msg);
2620 memset(ots, 0, sizeof *ots);
2621 ots->table_id = TABLEID_CLASSIFIER;
2622 strcpy(ots->name, "classifier");
2623 ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL)
2625 ots->max_entries = htonl(65536);
2626 ots->active_count = htonl(n_wild);
2627 ots->lookup_count = htonll(0); /* XXX */
2628 ots->matched_count = htonll(0); /* XXX */
2630 queue_tx(msg, ofconn, ofconn->reply_counter);
2635 append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn,
2638 struct netdev_stats stats;
2639 struct ofp_port_stats *ops;
2641 /* Intentionally ignore return value, since errors will set
2642 * 'stats' to all-1s, which is correct for OpenFlow, and
2643 * netdev_get_stats() will log errors. */
2644 netdev_get_stats(port->netdev, &stats);
2646 ops = append_stats_reply(sizeof *ops, ofconn, &msg);
2647 ops->port_no = htons(odp_port_to_ofp_port(port_no));
2648 memset(ops->pad, 0, sizeof ops->pad);
2649 ops->rx_packets = htonll(stats.rx_packets);
2650 ops->tx_packets = htonll(stats.tx_packets);
2651 ops->rx_bytes = htonll(stats.rx_bytes);
2652 ops->tx_bytes = htonll(stats.tx_bytes);
2653 ops->rx_dropped = htonll(stats.rx_dropped);
2654 ops->tx_dropped = htonll(stats.tx_dropped);
2655 ops->rx_errors = htonll(stats.rx_errors);
2656 ops->tx_errors = htonll(stats.tx_errors);
2657 ops->rx_frame_err = htonll(stats.rx_frame_errors);
2658 ops->rx_over_err = htonll(stats.rx_over_errors);
2659 ops->rx_crc_err = htonll(stats.rx_crc_errors);
2660 ops->collisions = htonll(stats.collisions);
2664 handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn,
2665 struct ofp_stats_request *osr,
2668 struct ofp_port_stats_request *psr;
2669 struct ofp_port_stats *ops;
2671 struct ofport *port;
2672 unsigned int port_no;
2674 if (arg_size != sizeof *psr) {
2675 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2677 psr = (struct ofp_port_stats_request *) osr->body;
2679 msg = start_stats_reply(osr, sizeof *ops * 16);
2680 if (psr->port_no != htons(OFPP_NONE)) {
2681 port = port_array_get(&p->ports,
2682 ofp_port_to_odp_port(ntohs(psr->port_no)));
2684 append_port_stat(port, ntohs(psr->port_no), ofconn, msg);
2687 PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) {
2688 append_port_stat(port, port_no, ofconn, msg);
2692 queue_tx(msg, ofconn, ofconn->reply_counter);
2696 struct flow_stats_cbdata {
2697 struct ofproto *ofproto;
2698 struct ofconn *ofconn;
2703 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2704 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2705 * returned statistic include statistics for all of 'rule''s subrules. */
2707 query_stats(struct ofproto *p, struct rule *rule,
2708 uint64_t *packet_countp, uint64_t *byte_countp)
2710 uint64_t packet_count, byte_count;
2711 struct rule *subrule;
2712 struct odp_flow *odp_flows;
2715 /* Start from historical data for 'rule' itself that are no longer tracked
2716 * by the datapath. This counts, for example, subrules that have
2718 packet_count = rule->packet_count;
2719 byte_count = rule->byte_count;
2721 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2722 * wildcarded then on all of its subrules.
2724 * Also, add any statistics that are not tracked by the datapath for each
2725 * subrule. This includes, for example, statistics for packets that were
2726 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2728 n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1;
2729 odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows);
2730 if (rule->cr.wc.wildcards) {
2732 LIST_FOR_EACH (subrule, struct rule, list, &rule->list) {
2733 odp_flows[i++].key = subrule->cr.flow;
2734 packet_count += subrule->packet_count;
2735 byte_count += subrule->byte_count;
2738 odp_flows[0].key = rule->cr.flow;
2741 /* Fetch up-to-date statistics from the datapath and add them in. */
2742 if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) {
2744 for (i = 0; i < n_odp_flows; i++) {
2745 struct odp_flow *odp_flow = &odp_flows[i];
2746 packet_count += odp_flow->stats.n_packets;
2747 byte_count += odp_flow->stats.n_bytes;
2752 /* Return the stats to the caller. */
2753 *packet_countp = packet_count;
2754 *byte_countp = byte_count;
2758 flow_stats_cb(struct cls_rule *rule_, void *cbdata_)
2760 struct rule *rule = rule_from_cls_rule(rule_);
2761 struct flow_stats_cbdata *cbdata = cbdata_;
2762 struct ofp_flow_stats *ofs;
2763 uint64_t packet_count, byte_count;
2764 size_t act_len, len;
2765 long long int tdiff = time_msec() - rule->created;
2766 uint32_t sec = tdiff / 1000;
2767 uint32_t msec = tdiff - (sec * 1000);
2769 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2773 act_len = sizeof *rule->actions * rule->n_actions;
2774 len = offsetof(struct ofp_flow_stats, actions) + act_len;
2776 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2778 ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg);
2779 ofs->length = htons(len);
2780 ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH;
2782 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2783 cbdata->ofproto->tun_id_from_cookie, &ofs->match);
2784 ofs->duration_sec = htonl(sec);
2785 ofs->duration_nsec = htonl(msec * 1000000);
2786 ofs->cookie = rule->flow_cookie;
2787 ofs->priority = htons(rule->cr.priority);
2788 ofs->idle_timeout = htons(rule->idle_timeout);
2789 ofs->hard_timeout = htons(rule->hard_timeout);
2790 memset(ofs->pad2, 0, sizeof ofs->pad2);
2791 ofs->packet_count = htonll(packet_count);
2792 ofs->byte_count = htonll(byte_count);
2793 memcpy(ofs->actions, rule->actions, act_len);
2797 table_id_to_include(uint8_t table_id)
2799 return (table_id == TABLEID_HASH ? CLS_INC_EXACT
2800 : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD
2801 : table_id == 0xff ? CLS_INC_ALL
2806 handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn,
2807 const struct ofp_stats_request *osr,
2810 struct ofp_flow_stats_request *fsr;
2811 struct flow_stats_cbdata cbdata;
2812 struct cls_rule target;
2814 if (arg_size != sizeof *fsr) {
2815 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2817 fsr = (struct ofp_flow_stats_request *) osr->body;
2819 COVERAGE_INC(ofproto_flows_req);
2821 cbdata.ofconn = ofconn;
2822 cbdata.out_port = fsr->out_port;
2823 cbdata.msg = start_stats_reply(osr, 1024);
2824 cls_rule_from_match(&fsr->match, 0, false, 0, &target);
2825 classifier_for_each_match(&p->cls, &target,
2826 table_id_to_include(fsr->table_id),
2827 flow_stats_cb, &cbdata);
2828 queue_tx(cbdata.msg, ofconn, ofconn->reply_counter);
2832 struct flow_stats_ds_cbdata {
2833 struct ofproto *ofproto;
2838 flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_)
2840 struct rule *rule = rule_from_cls_rule(rule_);
2841 struct flow_stats_ds_cbdata *cbdata = cbdata_;
2842 struct ds *results = cbdata->results;
2843 struct ofp_match match;
2844 uint64_t packet_count, byte_count;
2845 size_t act_len = sizeof *rule->actions * rule->n_actions;
2847 /* Don't report on subrules. */
2848 if (rule->super != NULL) {
2852 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2853 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards,
2854 cbdata->ofproto->tun_id_from_cookie, &match);
2856 ds_put_format(results, "duration=%llds, ",
2857 (time_msec() - rule->created) / 1000);
2858 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2859 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2860 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2861 ofp_print_match(results, &match, true);
2862 ofp_print_actions(results, &rule->actions->header, act_len);
2863 ds_put_cstr(results, "\n");
2866 /* Adds a pretty-printed description of all flows to 'results', including
2867 * those marked hidden by secchan (e.g., by in-band control). */
2869 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2871 struct ofp_match match;
2872 struct cls_rule target;
2873 struct flow_stats_ds_cbdata cbdata;
2875 memset(&match, 0, sizeof match);
2876 match.wildcards = htonl(OVSFW_ALL);
2879 cbdata.results = results;
2881 cls_rule_from_match(&match, 0, false, 0, &target);
2882 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
2883 flow_stats_ds_cb, &cbdata);
2886 struct aggregate_stats_cbdata {
2887 struct ofproto *ofproto;
2889 uint64_t packet_count;
2890 uint64_t byte_count;
2895 aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_)
2897 struct rule *rule = rule_from_cls_rule(rule_);
2898 struct aggregate_stats_cbdata *cbdata = cbdata_;
2899 uint64_t packet_count, byte_count;
2901 if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) {
2905 query_stats(cbdata->ofproto, rule, &packet_count, &byte_count);
2907 cbdata->packet_count += packet_count;
2908 cbdata->byte_count += byte_count;
2913 handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn,
2914 const struct ofp_stats_request *osr,
2917 struct ofp_aggregate_stats_request *asr;
2918 struct ofp_aggregate_stats_reply *reply;
2919 struct aggregate_stats_cbdata cbdata;
2920 struct cls_rule target;
2923 if (arg_size != sizeof *asr) {
2924 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2926 asr = (struct ofp_aggregate_stats_request *) osr->body;
2928 COVERAGE_INC(ofproto_agg_request);
2930 cbdata.out_port = asr->out_port;
2931 cbdata.packet_count = 0;
2932 cbdata.byte_count = 0;
2934 cls_rule_from_match(&asr->match, 0, false, 0, &target);
2935 classifier_for_each_match(&p->cls, &target,
2936 table_id_to_include(asr->table_id),
2937 aggregate_stats_cb, &cbdata);
2939 msg = start_stats_reply(osr, sizeof *reply);
2940 reply = append_stats_reply(sizeof *reply, ofconn, &msg);
2941 reply->flow_count = htonl(cbdata.n_flows);
2942 reply->packet_count = htonll(cbdata.packet_count);
2943 reply->byte_count = htonll(cbdata.byte_count);
2944 queue_tx(msg, ofconn, ofconn->reply_counter);
2949 handle_stats_request(struct ofproto *p, struct ofconn *ofconn,
2950 struct ofp_header *oh)
2952 struct ofp_stats_request *osr;
2956 error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr,
2961 osr = (struct ofp_stats_request *) oh;
2963 switch (ntohs(osr->type)) {
2965 return handle_desc_stats_request(p, ofconn, osr);
2968 return handle_flow_stats_request(p, ofconn, osr, arg_size);
2970 case OFPST_AGGREGATE:
2971 return handle_aggregate_stats_request(p, ofconn, osr, arg_size);
2974 return handle_table_stats_request(p, ofconn, osr);
2977 return handle_port_stats_request(p, ofconn, osr, arg_size);
2980 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
2983 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2987 static long long int
2988 msec_from_nsec(uint64_t sec, uint32_t nsec)
2990 return !sec ? 0 : sec * 1000 + nsec / 1000000;
2994 update_time(struct ofproto *ofproto, struct rule *rule,
2995 const struct odp_flow_stats *stats)
2997 long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec);
2998 if (used > rule->used) {
3000 if (rule->super && used > rule->super->used) {
3001 rule->super->used = used;
3003 netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used);
3008 update_stats(struct ofproto *ofproto, struct rule *rule,
3009 const struct odp_flow_stats *stats)
3011 if (stats->n_packets) {
3012 update_time(ofproto, rule, stats);
3013 rule->packet_count += stats->n_packets;
3014 rule->byte_count += stats->n_bytes;
3015 netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos,
3020 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3021 * in which no matching flow already exists in the flow table.
3023 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3024 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3025 * code as encoded by ofp_mkerr() on failure.
3027 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3030 add_flow(struct ofproto *p, struct ofconn *ofconn,
3031 const struct ofp_flow_mod *ofm, size_t n_actions)
3033 struct ofpbuf *packet;
3038 if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) {
3042 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3044 if (classifier_rule_overlaps(&p->cls, &flow, wildcards,
3045 ntohs(ofm->priority))) {
3046 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
3050 rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions,
3051 n_actions, ntohs(ofm->idle_timeout),
3052 ntohs(ofm->hard_timeout), ofm->cookie,
3053 ofm->flags & htons(OFPFF_SEND_FLOW_REM));
3054 cls_rule_from_match(&ofm->match, ntohs(ofm->priority),
3055 p->tun_id_from_cookie, ofm->cookie, &rule->cr);
3058 if (ofm->buffer_id != htonl(UINT32_MAX)) {
3059 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3063 in_port = UINT16_MAX;
3066 rule_insert(p, rule, packet, in_port);
3067 ofpbuf_delete(packet);
3071 static struct rule *
3072 find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm)
3077 flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie,
3079 return rule_from_cls_rule(classifier_find_rule_exactly(
3080 &p->cls, &flow, wildcards,
3081 ntohs(ofm->priority)));
3085 send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn,
3086 struct rule *rule, const struct ofp_flow_mod *ofm)
3088 struct ofpbuf *packet;
3093 if (ofm->buffer_id == htonl(UINT32_MAX)) {
3097 error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id),
3103 flow_extract(packet, 0, in_port, &flow);
3104 rule_execute(ofproto, rule, packet, &flow);
3105 ofpbuf_delete(packet);
3110 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3112 struct modify_flows_cbdata {
3113 struct ofproto *ofproto;
3114 const struct ofp_flow_mod *ofm;
3119 static int modify_flow(struct ofproto *, const struct ofp_flow_mod *,
3120 size_t n_actions, struct rule *);
3121 static void modify_flows_cb(struct cls_rule *, void *cbdata_);
3123 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3124 * encoded by ofp_mkerr() on failure.
3126 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3129 modify_flows_loose(struct ofproto *p, struct ofconn *ofconn,
3130 const struct ofp_flow_mod *ofm, size_t n_actions)
3132 struct modify_flows_cbdata cbdata;
3133 struct cls_rule target;
3137 cbdata.n_actions = n_actions;
3138 cbdata.match = NULL;
3140 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3143 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3144 modify_flows_cb, &cbdata);
3146 /* This credits the packet to whichever flow happened to happened to
3147 * match last. That's weird. Maybe we should do a lookup for the
3148 * flow that actually matches the packet? Who knows. */
3149 send_buffered_packet(p, ofconn, cbdata.match, ofm);
3152 return add_flow(p, ofconn, ofm, n_actions);
3156 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3157 * code as encoded by ofp_mkerr() on failure.
3159 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3162 modify_flow_strict(struct ofproto *p, struct ofconn *ofconn,
3163 struct ofp_flow_mod *ofm, size_t n_actions)
3165 struct rule *rule = find_flow_strict(p, ofm);
3166 if (rule && !rule_is_hidden(rule)) {
3167 modify_flow(p, ofm, n_actions, rule);
3168 return send_buffered_packet(p, ofconn, rule, ofm);
3170 return add_flow(p, ofconn, ofm, n_actions);
3174 /* Callback for modify_flows_loose(). */
3176 modify_flows_cb(struct cls_rule *rule_, void *cbdata_)
3178 struct rule *rule = rule_from_cls_rule(rule_);
3179 struct modify_flows_cbdata *cbdata = cbdata_;
3181 if (!rule_is_hidden(rule)) {
3182 cbdata->match = rule;
3183 modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule);
3187 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3188 * been identified as a flow in 'p''s flow table to be modified, by changing
3189 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3190 * ofp_action[] structures). */
3192 modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm,
3193 size_t n_actions, struct rule *rule)
3195 size_t actions_len = n_actions * sizeof *rule->actions;
3197 rule->flow_cookie = ofm->cookie;
3199 /* If the actions are the same, do nothing. */
3200 if (n_actions == rule->n_actions
3201 && !memcmp(ofm->actions, rule->actions, actions_len))
3206 /* Replace actions. */
3207 free(rule->actions);
3208 rule->actions = xmemdup(ofm->actions, actions_len);
3209 rule->n_actions = n_actions;
3211 /* Make sure that the datapath gets updated properly. */
3212 if (rule->cr.wc.wildcards) {
3213 COVERAGE_INC(ofproto_mod_wc_flow);
3214 p->need_revalidate = true;
3216 rule_update_actions(p, rule);
3222 /* OFPFC_DELETE implementation. */
3224 struct delete_flows_cbdata {
3225 struct ofproto *ofproto;
3229 static void delete_flows_cb(struct cls_rule *, void *cbdata_);
3230 static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port);
3232 /* Implements OFPFC_DELETE. */
3234 delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm)
3236 struct delete_flows_cbdata cbdata;
3237 struct cls_rule target;
3240 cbdata.out_port = ofm->out_port;
3242 cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie,
3245 classifier_for_each_match(&p->cls, &target, CLS_INC_ALL,
3246 delete_flows_cb, &cbdata);
3249 /* Implements OFPFC_DELETE_STRICT. */
3251 delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm)
3253 struct rule *rule = find_flow_strict(p, ofm);
3255 delete_flow(p, rule, ofm->out_port);
3259 /* Callback for delete_flows_loose(). */
3261 delete_flows_cb(struct cls_rule *rule_, void *cbdata_)
3263 struct rule *rule = rule_from_cls_rule(rule_);
3264 struct delete_flows_cbdata *cbdata = cbdata_;
3266 delete_flow(cbdata->ofproto, rule, cbdata->out_port);
3269 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3270 * been identified as a flow to delete from 'p''s flow table, by deleting the
3271 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3274 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3275 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3276 * specified 'out_port'. */
3278 delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port)
3280 if (rule_is_hidden(rule)) {
3284 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
3288 send_flow_removed(p, rule, time_msec(), OFPRR_DELETE);
3289 rule_remove(p, rule);
3293 handle_flow_mod(struct ofproto *p, struct ofconn *ofconn,
3294 struct ofp_flow_mod *ofm)
3299 error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm,
3300 sizeof *ofm->actions, &n_actions);
3305 /* We do not support the emergency flow cache. It will hopefully
3306 * get dropped from OpenFlow in the near future. */
3307 if (ofm->flags & htons(OFPFF_EMERG)) {
3308 /* There isn't a good fit for an error code, so just state that the
3309 * flow table is full. */
3310 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
3313 normalize_match(&ofm->match);
3314 if (!ofm->match.wildcards) {
3315 ofm->priority = htons(UINT16_MAX);
3318 error = validate_actions((const union ofp_action *) ofm->actions,
3319 n_actions, p->max_ports);
3324 switch (ntohs(ofm->command)) {
3326 return add_flow(p, ofconn, ofm, n_actions);
3329 return modify_flows_loose(p, ofconn, ofm, n_actions);
3331 case OFPFC_MODIFY_STRICT:
3332 return modify_flow_strict(p, ofconn, ofm, n_actions);
3335 delete_flows_loose(p, ofm);
3338 case OFPFC_DELETE_STRICT:
3339 delete_flow_strict(p, ofm);
3343 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
3348 handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg)
3352 error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg);
3357 p->tun_id_from_cookie = !!msg->set;
3362 handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg)
3364 struct ofp_vendor_header *ovh = msg;
3365 struct nicira_header *nh;
3367 if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) {
3368 VLOG_WARN_RL(&rl, "received vendor message of length %zu "
3369 "(expected at least %zu)",
3370 ntohs(ovh->header.length), sizeof(struct ofp_vendor_header));
3371 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3373 if (ovh->vendor != htonl(NX_VENDOR_ID)) {
3374 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR);
3376 if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) {
3377 VLOG_WARN_RL(&rl, "received Nicira vendor message of length %zu "
3378 "(expected at least %zu)",
3379 ntohs(ovh->header.length), sizeof(struct nicira_header));
3380 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
3384 switch (ntohl(nh->subtype)) {
3385 case NXT_STATUS_REQUEST:
3386 return switch_status_handle_request(p->switch_status, ofconn->rconn,
3389 case NXT_TUN_ID_FROM_COOKIE:
3390 return handle_tun_id_from_cookie(p, msg);
3393 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE);
3397 handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh)
3399 struct ofp_header *ob;
3402 /* Currently, everything executes synchronously, so we can just
3403 * immediately send the barrier reply. */
3404 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
3405 queue_tx(buf, ofconn, ofconn->reply_counter);
3410 handle_openflow(struct ofconn *ofconn, struct ofproto *p,
3411 struct ofpbuf *ofp_msg)
3413 struct ofp_header *oh = ofp_msg->data;
3416 COVERAGE_INC(ofproto_recv_openflow);
3418 case OFPT_ECHO_REQUEST:
3419 error = handle_echo_request(ofconn, oh);
3422 case OFPT_ECHO_REPLY:
3426 case OFPT_FEATURES_REQUEST:
3427 error = handle_features_request(p, ofconn, oh);
3430 case OFPT_GET_CONFIG_REQUEST:
3431 error = handle_get_config_request(p, ofconn, oh);
3434 case OFPT_SET_CONFIG:
3435 error = handle_set_config(p, ofconn, ofp_msg->data);
3438 case OFPT_PACKET_OUT:
3439 error = handle_packet_out(p, ofconn, ofp_msg->data);
3443 error = handle_port_mod(p, oh);
3447 error = handle_flow_mod(p, ofconn, ofp_msg->data);
3450 case OFPT_STATS_REQUEST:
3451 error = handle_stats_request(p, ofconn, oh);
3455 error = handle_vendor(p, ofconn, ofp_msg->data);
3458 case OFPT_BARRIER_REQUEST:
3459 error = handle_barrier_request(ofconn, oh);
3463 if (VLOG_IS_WARN_ENABLED()) {
3464 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
3465 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
3468 error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
3473 send_error_oh(ofconn, ofp_msg->data, error);
3478 handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet)
3480 struct odp_msg *msg = packet->data;
3482 struct ofpbuf payload;
3485 payload.data = msg + 1;
3486 payload.size = msg->length - sizeof *msg;
3487 flow_extract(&payload, msg->arg, msg->port, &flow);
3489 /* Check with in-band control to see if this packet should be sent
3490 * to the local port regardless of the flow table. */
3491 if (in_band_msg_in_hook(p->in_band, &flow, &payload)) {
3492 union odp_action action;
3494 memset(&action, 0, sizeof(action));
3495 action.output.type = ODPAT_OUTPUT;
3496 action.output.port = ODPP_LOCAL;
3497 dpif_execute(p->dpif, flow.in_port, &action, 1, &payload);
3500 rule = lookup_valid_rule(p, &flow);
3502 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3503 struct ofport *port = port_array_get(&p->ports, msg->port);
3505 if (port->opp.config & OFPPC_NO_PACKET_IN) {
3506 COVERAGE_INC(ofproto_no_packet_in);
3507 /* XXX install 'drop' flow entry */
3508 ofpbuf_delete(packet);
3512 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port);
3515 COVERAGE_INC(ofproto_packet_in);
3516 send_packet_in(p, packet);
3520 if (rule->cr.wc.wildcards) {
3521 rule = rule_create_subrule(p, rule, &flow);
3522 rule_make_actions(p, rule, packet);
3524 if (!rule->may_install) {
3525 /* The rule is not installable, that is, we need to process every
3526 * packet, so process the current packet and set its actions into
3528 rule_make_actions(p, rule, packet);
3530 /* XXX revalidate rule if it needs it */
3534 rule_execute(p, rule, &payload, &flow);
3535 rule_reinstall(p, rule);
3537 if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) {
3539 * Extra-special case for fail-open mode.
3541 * We are in fail-open mode and the packet matched the fail-open rule,
3542 * but we are connected to a controller too. We should send the packet
3543 * up to the controller in the hope that it will try to set up a flow
3544 * and thereby allow us to exit fail-open.
3546 * See the top-level comment in fail-open.c for more information.
3548 send_packet_in(p, packet);
3550 ofpbuf_delete(packet);
3555 handle_odp_msg(struct ofproto *p, struct ofpbuf *packet)
3557 struct odp_msg *msg = packet->data;
3559 switch (msg->type) {
3560 case _ODPL_ACTION_NR:
3561 COVERAGE_INC(ofproto_ctlr_action);
3562 send_packet_in(p, packet);
3565 case _ODPL_SFLOW_NR:
3567 ofproto_sflow_received(p->sflow, msg);
3569 ofpbuf_delete(packet);
3573 handle_odp_miss_msg(p, packet);
3577 VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32,
3584 revalidate_cb(struct cls_rule *sub_, void *cbdata_)
3586 struct rule *sub = rule_from_cls_rule(sub_);
3587 struct revalidate_cbdata *cbdata = cbdata_;
3589 if (cbdata->revalidate_all
3590 || (cbdata->revalidate_subrules && sub->super)
3591 || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) {
3592 revalidate_rule(cbdata->ofproto, sub);
3597 revalidate_rule(struct ofproto *p, struct rule *rule)
3599 const flow_t *flow = &rule->cr.flow;
3601 COVERAGE_INC(ofproto_revalidate_rule);
3604 super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow));
3606 rule_remove(p, rule);
3608 } else if (super != rule->super) {
3609 COVERAGE_INC(ofproto_revalidate_moved);
3610 list_remove(&rule->list);
3611 list_push_back(&super->list, &rule->list);
3612 rule->super = super;
3613 rule->hard_timeout = super->hard_timeout;
3614 rule->idle_timeout = super->idle_timeout;
3615 rule->created = super->created;
3620 rule_update_actions(p, rule);
3624 static struct ofpbuf *
3625 compose_flow_removed(struct ofproto *p, const struct rule *rule,
3626 long long int now, uint8_t reason)
3628 struct ofp_flow_removed *ofr;
3630 long long int tdiff = now - rule->created;
3631 uint32_t sec = tdiff / 1000;
3632 uint32_t msec = tdiff - (sec * 1000);
3634 ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf);
3635 flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie,
3637 ofr->cookie = rule->flow_cookie;
3638 ofr->priority = htons(rule->cr.priority);
3639 ofr->reason = reason;
3640 ofr->duration_sec = htonl(sec);
3641 ofr->duration_nsec = htonl(msec * 1000000);
3642 ofr->idle_timeout = htons(rule->idle_timeout);
3643 ofr->packet_count = htonll(rule->packet_count);
3644 ofr->byte_count = htonll(rule->byte_count);
3650 uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule)
3652 assert(rule->installed);
3653 assert(!rule->cr.wc.wildcards);
3656 rule_remove(ofproto, rule);
3658 rule_uninstall(ofproto, rule);
3662 send_flow_removed(struct ofproto *p, struct rule *rule,
3663 long long int now, uint8_t reason)
3665 struct ofconn *ofconn;
3666 struct ofconn *prev;
3667 struct ofpbuf *buf = NULL;
3669 /* We limit the maximum number of queued flow expirations it by accounting
3670 * them under the counter for replies. That works because preventing
3671 * OpenFlow requests from being processed also prevents new flows from
3672 * being added (and expiring). (It also prevents processing OpenFlow
3673 * requests that would not add new flows, so it is imperfect.) */
3676 LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) {
3677 if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) {
3679 queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter);
3681 buf = compose_flow_removed(p, rule, now, reason);
3687 queue_tx(buf, prev, prev->reply_counter);
3693 expire_rule(struct cls_rule *cls_rule, void *p_)
3695 struct ofproto *p = p_;
3696 struct rule *rule = rule_from_cls_rule(cls_rule);
3697 long long int hard_expire, idle_expire, expire, now;
3699 hard_expire = (rule->hard_timeout
3700 ? rule->created + rule->hard_timeout * 1000
3702 idle_expire = (rule->idle_timeout
3703 && (rule->super || list_is_empty(&rule->list))
3704 ? rule->used + rule->idle_timeout * 1000
3706 expire = MIN(hard_expire, idle_expire);
3710 if (rule->installed && now >= rule->used + 5000) {
3711 uninstall_idle_flow(p, rule);
3712 } else if (!rule->cr.wc.wildcards) {
3713 active_timeout(p, rule);
3719 COVERAGE_INC(ofproto_expired);
3721 /* Update stats. This code will be a no-op if the rule expired
3722 * due to an idle timeout. */
3723 if (rule->cr.wc.wildcards) {
3724 struct rule *subrule, *next;
3725 LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) {
3726 rule_remove(p, subrule);
3729 rule_uninstall(p, rule);
3732 if (!rule_is_hidden(rule)) {
3733 send_flow_removed(p, rule, now,
3735 ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT));
3737 rule_remove(p, rule);
3741 active_timeout(struct ofproto *ofproto, struct rule *rule)
3743 if (ofproto->netflow && !is_controller_rule(rule) &&
3744 netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) {
3745 struct ofexpired expired;
3746 struct odp_flow odp_flow;
3748 /* Get updated flow stats. */
3749 memset(&odp_flow, 0, sizeof odp_flow);
3750 if (rule->installed) {
3751 odp_flow.key = rule->cr.flow;
3752 odp_flow.flags = ODPFF_ZERO_TCP_FLAGS;
3753 dpif_flow_get(ofproto->dpif, &odp_flow);
3755 if (odp_flow.stats.n_packets) {
3756 update_time(ofproto, rule, &odp_flow.stats);
3757 netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos,
3758 odp_flow.stats.tcp_flags);
3762 expired.flow = rule->cr.flow;
3763 expired.packet_count = rule->packet_count +
3764 odp_flow.stats.n_packets;
3765 expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes;
3766 expired.used = rule->used;
3768 netflow_expire(ofproto->netflow, &rule->nf_flow, &expired);
3770 /* Schedule us to send the accumulated records once we have
3771 * collected all of them. */
3772 poll_immediate_wake();
3777 update_used(struct ofproto *p)
3779 struct odp_flow *flows;
3784 error = dpif_flow_list_all(p->dpif, &flows, &n_flows);
3789 for (i = 0; i < n_flows; i++) {
3790 struct odp_flow *f = &flows[i];
3793 rule = rule_from_cls_rule(
3794 classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX));
3795 if (!rule || !rule->installed) {
3796 COVERAGE_INC(ofproto_unexpected_rule);
3797 dpif_flow_del(p->dpif, f);
3801 update_time(p, rule, &f->stats);
3802 rule_account(p, rule, f->stats.n_bytes);
3808 do_send_packet_in(struct ofpbuf *packet, void *ofconn_)
3810 struct ofconn *ofconn = ofconn_;
3811 struct ofproto *ofproto = ofconn->ofproto;
3812 struct odp_msg *msg = packet->data;
3813 struct ofpbuf payload;
3818 /* Extract packet payload from 'msg'. */
3819 payload.data = msg + 1;
3820 payload.size = msg->length - sizeof *msg;
3822 /* Construct packet-in message. */
3824 if (msg->type == _ODPL_ACTION_NR) {
3825 buffer_id = UINT32_MAX;
3827 if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) {
3828 buffer_id = pktbuf_get_null();
3830 buffer_id = pktbuf_save(ofconn->pktbuf, &payload, msg->port);
3832 if (buffer_id != UINT32_MAX) {
3833 send_len = ofconn->miss_send_len;
3836 opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port),
3837 msg->type, &payload, send_len);
3840 rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100);
3842 ofpbuf_delete(packet);
3846 send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet)
3848 struct odp_msg *msg = packet->data;
3849 struct ofconn *ofconn, *prev;
3851 assert(msg->type == _ODPL_MISS_NR || msg->type == _ODPL_ACTION_NR);
3854 LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) {
3856 pinsched_send(prev->schedulers[msg->type], msg->port,
3857 ofpbuf_clone(packet), do_send_packet_in, prev);
3862 pinsched_send(prev->schedulers[msg->type], msg->port,
3863 packet, do_send_packet_in, prev);
3865 ofpbuf_delete(packet);
3870 pick_datapath_id(const struct ofproto *ofproto)
3872 const struct ofport *port;
3874 port = port_array_get(&ofproto->ports, ODPP_LOCAL);
3876 uint8_t ea[ETH_ADDR_LEN];
3879 error = netdev_get_etheraddr(port->netdev, ea);
3881 return eth_addr_to_uint64(ea);
3883 VLOG_WARN("could not get MAC address for %s (%s)",
3884 netdev_get_name(port->netdev), strerror(error));
3886 return ofproto->fallback_dpid;
3890 pick_fallback_dpid(void)
3892 uint8_t ea[ETH_ADDR_LEN];
3893 eth_addr_nicira_random(ea);
3894 return eth_addr_to_uint64(ea);
3898 default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
3899 struct odp_actions *actions, tag_type *tags,
3900 uint16_t *nf_output_iface, void *ofproto_)
3902 struct ofproto *ofproto = ofproto_;
3905 /* Drop frames for reserved multicast addresses. */
3906 if (eth_addr_is_reserved(flow->dl_dst)) {
3910 /* Learn source MAC (but don't try to learn from revalidation). */
3911 if (packet != NULL) {
3912 tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src,
3915 /* The log messages here could actually be useful in debugging,
3916 * so keep the rate limit relatively high. */
3917 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
3918 VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16,
3919 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
3920 ofproto_revalidate(ofproto, rev_tag);
3924 /* Determine output port. */
3925 out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags);
3927 add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface);
3928 } else if (out_port != flow->in_port) {
3929 odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port;
3930 *nf_output_iface = out_port;
3938 static const struct ofhooks default_ofhooks = {
3940 default_normal_ofhook_cb,