2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
24 #include "byte-order.h"
25 #include "classifier.h"
28 #include "dynamic-string.h"
33 #include "ofp-print.h"
36 #include "openflow/nicira-ext.h"
37 #include "openflow/openflow.h"
41 #include "poll-loop.h"
46 #include "unaligned.h"
50 VLOG_DEFINE_THIS_MODULE(ofproto);
52 COVERAGE_DEFINE(ofproto_agg_request);
53 COVERAGE_DEFINE(ofproto_error);
54 COVERAGE_DEFINE(ofproto_flows_req);
55 COVERAGE_DEFINE(ofproto_flush);
56 COVERAGE_DEFINE(ofproto_no_packet_in);
57 COVERAGE_DEFINE(ofproto_packet_out);
58 COVERAGE_DEFINE(ofproto_queue_req);
59 COVERAGE_DEFINE(ofproto_recv_openflow);
60 COVERAGE_DEFINE(ofproto_reinit_ports);
61 COVERAGE_DEFINE(ofproto_uninstallable);
62 COVERAGE_DEFINE(ofproto_update_port);
64 static void ofport_destroy__(struct ofport *);
65 static void ofport_destroy(struct ofport *);
67 static int rule_create(struct ofproto *, const struct cls_rule *,
68 const union ofp_action *, size_t n_actions,
69 uint16_t idle_timeout, uint16_t hard_timeout,
70 ovs_be64 flow_cookie, bool send_flow_removed,
73 static uint64_t pick_datapath_id(const struct ofproto *);
74 static uint64_t pick_fallback_dpid(void);
76 static void ofproto_destroy__(struct ofproto *);
77 static void ofproto_flush_flows__(struct ofproto *);
79 static void ofproto_rule_destroy__(struct rule *);
80 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
82 static void handle_openflow(struct ofconn *, struct ofpbuf *);
84 static void update_port(struct ofproto *, const char *devname);
85 static int init_ports(struct ofproto *);
86 static void reinit_ports(struct ofproto *);
88 static void ofproto_unixctl_init(void);
90 /* All registered ofproto classes, in probe order. */
91 static const struct ofproto_class **ofproto_classes;
92 static size_t n_ofproto_classes;
93 static size_t allocated_ofproto_classes;
95 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
96 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
98 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
101 ofproto_initialize(void)
107 ofproto_class_register(&ofproto_dpif_class);
111 /* 'type' should be a normalized datapath type, as returned by
112 * ofproto_normalize_type(). Returns the corresponding ofproto_class
113 * structure, or a null pointer if there is none registered for 'type'. */
114 static const struct ofproto_class *
115 ofproto_class_find__(const char *type)
119 ofproto_initialize();
120 for (i = 0; i < n_ofproto_classes; i++) {
121 const struct ofproto_class *class = ofproto_classes[i];
126 class->enumerate_types(&types);
127 found = sset_contains(&types, type);
128 sset_destroy(&types);
134 VLOG_WARN("unknown datapath type %s", type);
138 /* Registers a new ofproto class. After successful registration, new ofprotos
139 * of that type can be created using ofproto_create(). */
141 ofproto_class_register(const struct ofproto_class *new_class)
145 for (i = 0; i < n_ofproto_classes; i++) {
146 if (ofproto_classes[i] == new_class) {
151 if (n_ofproto_classes >= allocated_ofproto_classes) {
152 ofproto_classes = x2nrealloc(ofproto_classes,
153 &allocated_ofproto_classes,
154 sizeof *ofproto_classes);
156 ofproto_classes[n_ofproto_classes++] = new_class;
160 /* Unregisters a datapath provider. 'type' must have been previously
161 * registered and not currently be in use by any ofprotos. After
162 * unregistration new datapaths of that type cannot be opened using
163 * ofproto_create(). */
165 ofproto_class_unregister(const struct ofproto_class *class)
169 for (i = 0; i < n_ofproto_classes; i++) {
170 if (ofproto_classes[i] == class) {
171 for (i++; i < n_ofproto_classes; i++) {
172 ofproto_classes[i - 1] = ofproto_classes[i];
178 VLOG_WARN("attempted to unregister an ofproto class that is not "
183 /* Clears 'types' and enumerates all registered ofproto types into it. The
184 * caller must first initialize the sset. */
186 ofproto_enumerate_types(struct sset *types)
190 ofproto_initialize();
191 for (i = 0; i < n_ofproto_classes; i++) {
192 ofproto_classes[i]->enumerate_types(types);
196 /* Returns the fully spelled out name for the given ofproto 'type'.
198 * Normalized type string can be compared with strcmp(). Unnormalized type
199 * string might be the same even if they have different spellings. */
201 ofproto_normalize_type(const char *type)
203 return type && type[0] ? type : "system";
206 /* Clears 'names' and enumerates the names of all known created ofprotos with
207 * the given 'type'. The caller must first initialize the sset. Returns 0 if
208 * successful, otherwise a positive errno value.
210 * Some kinds of datapaths might not be practically enumerable. This is not
211 * considered an error. */
213 ofproto_enumerate_names(const char *type, struct sset *names)
215 const struct ofproto_class *class = ofproto_class_find__(type);
216 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
220 ofproto_create(const char *datapath_name, const char *datapath_type,
221 struct ofproto **ofprotop)
223 const struct ofproto_class *class;
224 struct ofproto *ofproto;
229 ofproto_initialize();
230 ofproto_unixctl_init();
232 datapath_type = ofproto_normalize_type(datapath_type);
233 class = ofproto_class_find__(datapath_type);
235 VLOG_WARN("could not create datapath %s of unknown type %s",
236 datapath_name, datapath_type);
240 ofproto = class->alloc();
242 VLOG_ERR("failed to allocate datapath %s of type %s",
243 datapath_name, datapath_type);
248 memset(ofproto, 0, sizeof *ofproto);
249 ofproto->ofproto_class = class;
250 ofproto->name = xstrdup(datapath_name);
251 ofproto->type = xstrdup(datapath_type);
252 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
253 hash_string(ofproto->name, 0));
254 ofproto->datapath_id = 0;
255 ofproto->fallback_dpid = pick_fallback_dpid();
256 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
257 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
258 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
259 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
260 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
261 ofproto->netdev_monitor = netdev_monitor_create();
262 hmap_init(&ofproto->ports);
263 shash_init(&ofproto->port_by_name);
264 ofproto->tables = NULL;
265 ofproto->n_tables = 0;
266 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
268 error = ofproto->ofproto_class->construct(ofproto);
270 VLOG_ERR("failed to open datapath %s: %s",
271 datapath_name, strerror(error));
272 ofproto_destroy__(ofproto);
275 assert(ofproto->n_tables > 0);
277 ofproto->datapath_id = pick_datapath_id(ofproto);
278 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
286 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
288 uint64_t old_dpid = p->datapath_id;
289 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
290 if (p->datapath_id != old_dpid) {
291 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
293 /* Force all active connections to reconnect, since there is no way to
294 * notify a controller that the datapath ID has changed. */
295 ofproto_reconnect_controllers(p);
300 ofproto_set_controllers(struct ofproto *p,
301 const struct ofproto_controller *controllers,
302 size_t n_controllers)
304 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
308 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
310 connmgr_set_fail_mode(p->connmgr, fail_mode);
313 /* Drops the connections between 'ofproto' and all of its controllers, forcing
314 * them to reconnect. */
316 ofproto_reconnect_controllers(struct ofproto *ofproto)
318 connmgr_reconnect(ofproto->connmgr);
321 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
322 * in-band control should guarantee access, in the same way that in-band
323 * control guarantees access to OpenFlow controllers. */
325 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
326 const struct sockaddr_in *extras, size_t n)
328 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
331 /* Sets the OpenFlow queue used by flows set up by in-band control on
332 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
333 * flows will use the default queue. */
335 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
337 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
341 ofproto_set_desc(struct ofproto *p,
342 const char *mfr_desc, const char *hw_desc,
343 const char *sw_desc, const char *serial_desc,
346 struct ofp_desc_stats *ods;
349 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
350 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
351 sizeof ods->mfr_desc);
354 p->mfr_desc = xstrdup(mfr_desc);
357 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
358 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
359 sizeof ods->hw_desc);
362 p->hw_desc = xstrdup(hw_desc);
365 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
366 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
367 sizeof ods->sw_desc);
370 p->sw_desc = xstrdup(sw_desc);
373 if (strlen(serial_desc) >= sizeof ods->serial_num) {
374 VLOG_WARN("truncating serial_desc, must be less than %zu "
376 sizeof ods->serial_num);
378 free(p->serial_desc);
379 p->serial_desc = xstrdup(serial_desc);
382 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
383 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
384 sizeof ods->dp_desc);
387 p->dp_desc = xstrdup(dp_desc);
392 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
394 return connmgr_set_snoops(ofproto->connmgr, snoops);
398 ofproto_set_netflow(struct ofproto *ofproto,
399 const struct netflow_options *nf_options)
401 if (nf_options && sset_is_empty(&nf_options->collectors)) {
405 if (ofproto->ofproto_class->set_netflow) {
406 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
408 return nf_options ? EOPNOTSUPP : 0;
413 ofproto_set_sflow(struct ofproto *ofproto,
414 const struct ofproto_sflow_options *oso)
416 if (oso && sset_is_empty(&oso->targets)) {
420 if (ofproto->ofproto_class->set_sflow) {
421 return ofproto->ofproto_class->set_sflow(ofproto, oso);
423 return oso ? EOPNOTSUPP : 0;
427 /* Connectivity Fault Management configuration. */
429 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
431 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
433 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
434 if (ofport && ofproto->ofproto_class->set_cfm) {
435 ofproto->ofproto_class->set_cfm(ofport, NULL, NULL, 0);
439 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
440 * basic configuration from the configuration members in 'cfm', and the set of
441 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
442 * Ignores the statistics members of 'cfm'.
444 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
446 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
447 const struct cfm *cfm,
448 const uint16_t *remote_mps, size_t n_remote_mps)
450 struct ofport *ofport;
453 ofport = ofproto_get_port(ofproto, ofp_port);
455 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
456 ofproto->name, ofp_port);
460 error = (ofproto->ofproto_class->set_cfm
461 ? ofproto->ofproto_class->set_cfm(ofport, cfm,
462 remote_mps, n_remote_mps)
465 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
466 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
471 /* Returns the connectivity fault management object associated with 'ofp_port'
472 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
473 * 'ofp_port' or if that port does not have CFM configured. The caller must
474 * not modify or destroy the returned object. */
476 ofproto_port_get_cfm(struct ofproto *ofproto, uint16_t ofp_port)
478 struct ofport *ofport;
479 const struct cfm *cfm;
481 ofport = ofproto_get_port(ofproto, ofp_port);
483 && ofproto->ofproto_class->get_cfm
484 && !ofproto->ofproto_class->get_cfm(ofport, &cfm)) ? cfm : NULL;
487 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
488 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
489 * 0 if LACP partner information is not current (generally indicating a
490 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
492 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
494 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
495 return (ofport && ofproto->ofproto_class->port_is_lacp_current
496 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
502 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
503 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
504 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
505 * configuration plus, if there is more than one slave, a bonding
508 * If 'aux' is already registered then this function updates its configuration
509 * to 's'. Otherwise, this function registers a new bundle.
511 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
514 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
515 const struct ofproto_bundle_settings *s)
517 return (ofproto->ofproto_class->bundle_set
518 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
522 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
523 * If no such bundle has been registered, this has no effect. */
525 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
527 return ofproto_bundle_register(ofproto, aux, NULL);
531 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
532 * If 'aux' is already registered then this function updates its configuration
533 * to 's'. Otherwise, this function registers a new mirror.
535 * Mirrors affect only the treatment of packets output to the OFPP_NORMAL
538 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
539 const struct ofproto_mirror_settings *s)
541 return (ofproto->ofproto_class->mirror_set
542 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
546 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
547 * If no mirror has been registered, this has no effect. */
549 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
551 return ofproto_mirror_register(ofproto, aux, NULL);
554 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
555 * which all packets are flooded, instead of using MAC learning. If
556 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
558 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
561 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
563 return (ofproto->ofproto_class->set_flood_vlans
564 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
568 /* Returns true if 'aux' is a registered bundle that is currently in use as the
569 * output for a mirror. */
571 ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux)
573 return (ofproto->ofproto_class->is_mirror_output_bundle
574 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
579 ofproto_has_snoops(const struct ofproto *ofproto)
581 return connmgr_has_snoops(ofproto->connmgr);
585 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
587 connmgr_get_snoops(ofproto->connmgr, snoops);
591 ofproto_destroy__(struct ofproto *ofproto)
595 connmgr_destroy(ofproto->connmgr);
597 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
599 free(ofproto->mfr_desc);
600 free(ofproto->hw_desc);
601 free(ofproto->sw_desc);
602 free(ofproto->serial_desc);
603 free(ofproto->dp_desc);
604 netdev_monitor_destroy(ofproto->netdev_monitor);
605 hmap_destroy(&ofproto->ports);
606 shash_destroy(&ofproto->port_by_name);
608 for (i = 0; i < ofproto->n_tables; i++) {
609 classifier_destroy(&ofproto->tables[i]);
611 free(ofproto->tables);
613 ofproto->ofproto_class->dealloc(ofproto);
617 ofproto_destroy(struct ofproto *p)
619 struct ofport *ofport, *next_ofport;
625 ofproto_flush_flows__(p);
626 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
627 ofport_destroy(ofport);
630 p->ofproto_class->destruct(p);
631 ofproto_destroy__(p);
634 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
635 * kernel datapath, for example, this destroys the datapath in the kernel, and
636 * with the netdev-based datapath, it tears down the data structures that
637 * represent the datapath.
639 * The datapath should not be currently open as an ofproto. */
641 ofproto_delete(const char *name, const char *type)
643 const struct ofproto_class *class = ofproto_class_find__(type);
644 return (!class ? EAFNOSUPPORT
645 : !class->del ? EACCES
646 : class->del(type, name));
650 process_port_change(struct ofproto *ofproto, int error, char *devname)
652 if (error == ENOBUFS) {
653 reinit_ports(ofproto);
655 update_port(ofproto, devname);
661 ofproto_run(struct ofproto *p)
666 error = p->ofproto_class->run(p);
667 if (error == ENODEV) {
668 /* Someone destroyed the datapath behind our back. The caller
669 * better destroy us and give up, because we're just going to
670 * spin from here on out. */
671 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
672 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
677 if (p->ofproto_class->port_poll) {
678 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
679 process_port_change(p, error, devname);
682 while ((error = netdev_monitor_poll(p->netdev_monitor,
683 &devname)) != EAGAIN) {
684 process_port_change(p, error, devname);
687 connmgr_run(p->connmgr, handle_openflow);
693 ofproto_wait(struct ofproto *p)
695 p->ofproto_class->wait(p);
696 if (p->ofproto_class->port_poll_wait) {
697 p->ofproto_class->port_poll_wait(p);
699 netdev_monitor_poll_wait(p->netdev_monitor);
700 connmgr_wait(p->connmgr);
704 ofproto_is_alive(const struct ofproto *p)
706 return connmgr_has_controllers(p->connmgr);
710 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
713 connmgr_get_controller_info(ofproto->connmgr, info);
717 ofproto_free_ofproto_controller_info(struct shash *info)
719 struct shash_node *node;
721 SHASH_FOR_EACH (node, info) {
722 struct ofproto_controller_info *cinfo = node->data;
723 while (cinfo->pairs.n) {
724 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
731 /* Makes a deep copy of 'old' into 'port'. */
733 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
735 port->name = xstrdup(old->name);
736 port->type = xstrdup(old->type);
737 port->ofp_port = old->ofp_port;
740 /* Frees memory allocated to members of 'ofproto_port'.
742 * Do not call this function on an ofproto_port obtained from
743 * ofproto_port_dump_next(): that function retains ownership of the data in the
746 ofproto_port_destroy(struct ofproto_port *ofproto_port)
748 free(ofproto_port->name);
749 free(ofproto_port->type);
752 /* Initializes 'dump' to begin dumping the ports in an ofproto.
754 * This function provides no status indication. An error status for the entire
755 * dump operation is provided when it is completed by calling
756 * ofproto_port_dump_done().
759 ofproto_port_dump_start(struct ofproto_port_dump *dump,
760 const struct ofproto *ofproto)
762 dump->ofproto = ofproto;
763 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
767 /* Attempts to retrieve another port from 'dump', which must have been created
768 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
769 * 'port' and returns true. On failure, returns false.
771 * Failure might indicate an actual error or merely that the last port has been
772 * dumped. An error status for the entire dump operation is provided when it
773 * is completed by calling ofproto_port_dump_done().
775 * The ofproto owns the data stored in 'port'. It will remain valid until at
776 * least the next time 'dump' is passed to ofproto_port_dump_next() or
777 * ofproto_port_dump_done(). */
779 ofproto_port_dump_next(struct ofproto_port_dump *dump,
780 struct ofproto_port *port)
782 const struct ofproto *ofproto = dump->ofproto;
788 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
791 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
797 /* Completes port table dump operation 'dump', which must have been created
798 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
799 * error-free, otherwise a positive errno value describing the problem. */
801 ofproto_port_dump_done(struct ofproto_port_dump *dump)
803 const struct ofproto *ofproto = dump->ofproto;
805 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
808 return dump->error == EOF ? 0 : dump->error;
811 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
812 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
813 * is non-null). On failure, returns a positive errno value and sets
814 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
816 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
822 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
824 update_port(ofproto, netdev_get_name(netdev));
827 *ofp_portp = error ? OFPP_NONE : ofp_port;
832 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
833 * initializes '*port' appropriately; on failure, returns a positive errno
836 * The caller owns the data in 'ofproto_port' and must free it with
837 * ofproto_port_destroy() when it is no longer needed. */
839 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
840 struct ofproto_port *port)
844 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
846 memset(port, 0, sizeof *port);
851 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
852 * Returns 0 if successful, otherwise a positive errno. */
854 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
856 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
857 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
860 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
861 if (!error && ofport) {
862 /* 'name' is the netdev's name and update_port() is going to close the
863 * netdev. Just in case update_port() refers to 'name' after it
864 * destroys 'ofport', make a copy of it around the update_port()
866 char *devname = xstrdup(name);
867 update_port(ofproto, devname);
873 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
874 * performs the 'n_actions' actions in 'actions'. The new flow will not
877 * If cls_rule->priority is in the range of priorities supported by OpenFlow
878 * (0...65535, inclusive) then the flow will be visible to OpenFlow
879 * controllers; otherwise, it will be hidden.
881 * The caller retains ownership of 'cls_rule' and 'actions'.
883 * This is a helper function for in-band control and fail-open. */
885 ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule,
886 const union ofp_action *actions, size_t n_actions)
889 rule_create(p, cls_rule, actions, n_actions, 0, 0, 0, false, &rule);
892 /* Searches for a rule with matching criteria exactly equal to 'target' in
893 * ofproto's table 0 and, if it finds one, deletes it.
895 * This is a helper function for in-band control and fail-open. */
897 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
901 rule = rule_from_cls_rule(classifier_find_rule_exactly(
902 &ofproto->tables[0], target));
903 ofproto_rule_destroy(rule);
907 ofproto_flush_flows__(struct ofproto *ofproto)
911 COVERAGE_INC(ofproto_flush);
913 if (ofproto->ofproto_class->flush) {
914 ofproto->ofproto_class->flush(ofproto);
917 for (i = 0; i < ofproto->n_tables; i++) {
918 struct rule *rule, *next_rule;
919 struct cls_cursor cursor;
921 cls_cursor_init(&cursor, &ofproto->tables[i], NULL);
922 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
923 ofproto_rule_destroy(rule);
928 /* Deletes all of the flows from all of ofproto's flow tables, then
929 * reintroduces rules required by in-band control and fail open. */
931 ofproto_flush_flows(struct ofproto *ofproto)
933 ofproto_flush_flows__(ofproto);
934 connmgr_flushed(ofproto->connmgr);
938 reinit_ports(struct ofproto *p)
940 struct ofproto_port_dump dump;
941 struct sset devnames;
942 struct ofport *ofport;
943 struct ofproto_port ofproto_port;
946 COVERAGE_INC(ofproto_reinit_ports);
948 sset_init(&devnames);
949 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
950 sset_add(&devnames, netdev_get_name(ofport->netdev));
952 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
953 sset_add(&devnames, ofproto_port.name);
956 SSET_FOR_EACH (devname, &devnames) {
957 update_port(p, devname);
959 sset_destroy(&devnames);
962 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
963 * netdev cannot be opened. On success, also fills in 'opp'. */
964 static struct netdev *
965 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
967 uint32_t curr, advertised, supported, peer;
968 struct netdev_options netdev_options;
969 enum netdev_flags flags;
970 struct netdev *netdev;
973 memset(&netdev_options, 0, sizeof netdev_options);
974 netdev_options.name = ofproto_port->name;
975 netdev_options.type = ofproto_port->type;
976 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
978 error = netdev_open(&netdev_options, &netdev);
980 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
981 "cannot be opened (%s)",
982 ofproto_port->name, ofproto_port->ofp_port,
983 ofproto_port->name, strerror(error));
987 netdev_get_flags(netdev, &flags);
988 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
990 opp->port_no = htons(ofproto_port->ofp_port);
991 netdev_get_etheraddr(netdev, opp->hw_addr);
992 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
993 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
994 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
995 opp->curr = htonl(curr);
996 opp->advertised = htonl(advertised);
997 opp->supported = htonl(supported);
998 opp->peer = htonl(peer);
1003 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1004 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1007 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1009 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1010 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1011 && a->state == b->state
1012 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1013 && a->curr == b->curr
1014 && a->advertised == b->advertised
1015 && a->supported == b->supported
1016 && a->peer == b->peer);
1019 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1020 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1021 * one with the same name or port number). */
1023 ofport_install(struct ofproto *p,
1024 struct netdev *netdev, const struct ofp_phy_port *opp)
1026 const char *netdev_name = netdev_get_name(netdev);
1027 struct ofport *ofport;
1030 /* Create ofport. */
1031 ofport = p->ofproto_class->port_alloc();
1036 ofport->ofproto = p;
1037 ofport->netdev = netdev;
1039 ofport->ofp_port = ntohs(opp->port_no);
1041 /* Add port to 'p'. */
1042 netdev_monitor_add(p->netdev_monitor, ofport->netdev);
1043 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1044 shash_add(&p->port_by_name, netdev_name, ofport);
1046 /* Let the ofproto_class initialize its private data. */
1047 error = p->ofproto_class->port_construct(ofport);
1051 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1055 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1056 p->name, netdev_name, strerror(error));
1058 ofport_destroy__(ofport);
1060 netdev_close(netdev);
1064 /* Removes 'ofport' from 'p' and destroys it. */
1066 ofport_remove(struct ofport *ofport)
1068 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1070 ofport_destroy(ofport);
1073 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1076 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1078 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1080 ofport_remove(port);
1084 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1086 * Does not handle a name or port number change. The caller must implement
1087 * such a change as a delete followed by an add. */
1089 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1091 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1092 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1093 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1094 port->opp.state = opp->state;
1095 port->opp.curr = opp->curr;
1096 port->opp.advertised = opp->advertised;
1097 port->opp.supported = opp->supported;
1098 port->opp.peer = opp->peer;
1100 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1104 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1106 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1108 if (port->ofproto->ofproto_class->set_cfm) {
1109 port->ofproto->ofproto_class->set_cfm(port, NULL, NULL, 0);
1111 if (port->ofproto->ofproto_class->bundle_remove) {
1112 port->ofproto->ofproto_class->bundle_remove(port);
1118 ofport_destroy__(struct ofport *port)
1120 struct ofproto *ofproto = port->ofproto;
1121 const char *name = netdev_get_name(port->netdev);
1123 netdev_monitor_remove(ofproto->netdev_monitor, port->netdev);
1124 hmap_remove(&ofproto->ports, &port->hmap_node);
1125 shash_delete(&ofproto->port_by_name,
1126 shash_find(&ofproto->port_by_name, name));
1128 netdev_close(port->netdev);
1129 ofproto->ofproto_class->port_dealloc(port);
1133 ofport_destroy(struct ofport *port)
1136 port->ofproto->ofproto_class->port_destruct(port);
1137 ofport_destroy__(port);
1142 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1144 struct ofport *port;
1146 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1147 hash_int(ofp_port, 0), &ofproto->ports) {
1148 if (port->ofp_port == ofp_port) {
1156 update_port(struct ofproto *ofproto, const char *name)
1158 struct ofproto_port ofproto_port;
1159 struct ofp_phy_port opp;
1160 struct netdev *netdev;
1161 struct ofport *port;
1163 COVERAGE_INC(ofproto_update_port);
1165 /* Fetch 'name''s location and properties from the datapath. */
1166 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1167 ? ofport_open(&ofproto_port, &opp)
1170 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1171 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1172 struct netdev *old_netdev = port->netdev;
1174 /* 'name' hasn't changed location. Any properties changed? */
1175 if (!ofport_equal(&port->opp, &opp)) {
1176 ofport_modified(port, &opp);
1179 /* Install the newly opened netdev in case it has changed.
1180 * Don't close the old netdev yet in case port_modified has to
1181 * remove a retained reference to it.*/
1182 netdev_monitor_remove(ofproto->netdev_monitor, port->netdev);
1183 netdev_monitor_add(ofproto->netdev_monitor, netdev);
1184 port->netdev = netdev;
1186 if (port->ofproto->ofproto_class->port_modified) {
1187 port->ofproto->ofproto_class->port_modified(port);
1190 netdev_close(old_netdev);
1192 /* If 'port' is nonnull then its name differs from 'name' and thus
1193 * we should delete it. If we think there's a port named 'name'
1194 * then its port number must be wrong now so delete it too. */
1196 ofport_remove(port);
1198 ofport_remove_with_name(ofproto, name);
1199 ofport_install(ofproto, netdev, &opp);
1202 /* Any port named 'name' is gone now. */
1203 ofport_remove_with_name(ofproto, name);
1205 ofproto_port_destroy(&ofproto_port);
1209 init_ports(struct ofproto *p)
1211 struct ofproto_port_dump dump;
1212 struct ofproto_port ofproto_port;
1214 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1215 uint16_t ofp_port = ofproto_port.ofp_port;
1216 if (ofproto_get_port(p, ofp_port)) {
1217 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1219 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1220 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1223 struct ofp_phy_port opp;
1224 struct netdev *netdev;
1226 netdev = ofport_open(&ofproto_port, &opp);
1228 ofport_install(p, netdev, &opp);
1236 /* Creates a new rule initialized as specified, inserts it into 'ofproto''s
1237 * flow table, and stores the new rule into '*rulep'. Returns 0 on success,
1238 * otherwise a positive errno value or OpenFlow error code. */
1240 rule_create(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1241 const union ofp_action *actions, size_t n_actions,
1242 uint16_t idle_timeout, uint16_t hard_timeout,
1243 ovs_be64 flow_cookie, bool send_flow_removed,
1244 struct rule **rulep)
1249 rule = ofproto->ofproto_class->rule_alloc();
1255 rule->ofproto = ofproto;
1256 rule->cr = *cls_rule;
1257 rule->flow_cookie = flow_cookie;
1258 rule->created = time_msec();
1259 rule->idle_timeout = idle_timeout;
1260 rule->hard_timeout = hard_timeout;
1261 rule->send_flow_removed = send_flow_removed;
1262 if (n_actions > 0) {
1263 rule->actions = xmemdup(actions, n_actions * sizeof *actions);
1265 rule->actions = NULL;
1267 rule->n_actions = n_actions;
1269 error = ofproto->ofproto_class->rule_construct(rule);
1271 ofproto_rule_destroy__(rule);
1279 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
1280 ofproto->name, strerror(error));
1286 ofproto_rule_destroy__(struct rule *rule)
1288 free(rule->actions);
1289 rule->ofproto->ofproto_class->rule_dealloc(rule);
1292 /* Destroys 'rule' and removes it from the flow table and the datapath. */
1294 ofproto_rule_destroy(struct rule *rule)
1297 rule->ofproto->ofproto_class->rule_destruct(rule);
1298 ofproto_rule_destroy__(rule);
1302 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1303 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1306 rule_has_out_port(const struct rule *rule, ovs_be16 out_port)
1308 const union ofp_action *oa;
1309 struct actions_iterator i;
1311 if (out_port == htons(OFPP_NONE)) {
1314 for (oa = actions_first(&i, rule->actions, rule->n_actions); oa;
1315 oa = actions_next(&i)) {
1316 if (action_outputs_to_port(oa, out_port)) {
1323 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1324 * statistics appropriately. 'packet' must have at least sizeof(struct
1325 * ofp_packet_in) bytes of headroom.
1327 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1328 * with statistics for 'packet' either way.
1330 * Takes ownership of 'packet'. */
1332 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1336 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1338 flow_extract(packet, 0, in_port, &flow);
1339 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1342 /* Returns true if 'rule' should be hidden from the controller.
1344 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1345 * (e.g. by in-band control) and are intentionally hidden from the
1348 rule_is_hidden(const struct rule *rule)
1350 return rule->cr.priority > UINT16_MAX;
1354 send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh,
1357 struct ofpbuf *buf = ofputil_encode_error_msg(error, oh);
1359 COVERAGE_INC(ofproto_error);
1360 ofconn_send_reply(ofconn, buf);
1365 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1367 ofconn_send_reply(ofconn, make_echo_reply(oh));
1372 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1374 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1375 struct ofp_switch_features *osf;
1377 struct ofport *port;
1381 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1382 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1384 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1385 osf->datapath_id = htonll(ofproto->datapath_id);
1386 osf->n_buffers = htonl(pktbuf_capacity());
1387 osf->n_tables = ofproto->n_tables;
1388 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1391 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1393 osf->actions = htonl(actions);
1395 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1396 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1399 ofconn_send_reply(ofconn, buf);
1404 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1406 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1408 struct ofp_switch_config *osc;
1412 /* Figure out flags. */
1413 drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto);
1414 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1417 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1418 osc->flags = htons(flags);
1419 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1420 ofconn_send_reply(ofconn, buf);
1426 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1428 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1429 uint16_t flags = ntohs(osc->flags);
1431 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1432 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1433 switch (flags & OFPC_FRAG_MASK) {
1434 case OFPC_FRAG_NORMAL:
1435 ofproto->ofproto_class->set_drop_frags(ofproto, false);
1437 case OFPC_FRAG_DROP:
1438 ofproto->ofproto_class->set_drop_frags(ofproto, true);
1441 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1447 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1452 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1453 * error message code (composed with ofp_mkerr()) for the caller to propagate
1454 * upward. Otherwise, returns 0.
1456 * The log message mentions 'msg_type'. */
1458 reject_slave_controller(struct ofconn *ofconn, const char *msg_type)
1460 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1461 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1462 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
1463 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
1466 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
1473 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1475 struct ofproto *p = ofconn_get_ofproto(ofconn);
1476 struct ofp_packet_out *opo;
1477 struct ofpbuf payload, *buffer;
1478 union ofp_action *ofp_actions;
1479 struct ofpbuf request;
1481 size_t n_ofp_actions;
1485 COVERAGE_INC(ofproto_packet_out);
1487 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
1492 /* Get ofp_packet_out. */
1493 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1494 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1497 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1498 &ofp_actions, &n_ofp_actions);
1504 if (opo->buffer_id != htonl(UINT32_MAX)) {
1505 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1507 if (error || !buffer) {
1516 /* Send out packet. */
1517 flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
1518 error = p->ofproto_class->packet_out(p, &payload, &flow,
1519 ofp_actions, n_ofp_actions);
1520 ofpbuf_delete(buffer);
1526 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1528 ovs_be32 old_config = port->opp.config;
1530 mask &= config ^ port->opp.config;
1531 if (mask & htonl(OFPPC_PORT_DOWN)) {
1532 if (config & htonl(OFPPC_PORT_DOWN)) {
1533 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1535 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1539 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1540 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1541 OFPPC_NO_PACKET_IN));
1542 if (port->opp.config != old_config) {
1543 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1548 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1550 struct ofproto *p = ofconn_get_ofproto(ofconn);
1551 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1552 struct ofport *port;
1555 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
1560 port = ofproto_get_port(p, ntohs(opm->port_no));
1562 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
1563 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1564 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
1566 update_port_config(port, opm->config, opm->mask);
1567 if (opm->advertise) {
1568 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1574 static struct ofpbuf *
1575 make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len)
1577 struct ofp_stats_reply *osr;
1580 msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX));
1581 osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg);
1583 osr->flags = htons(0);
1587 static struct ofpbuf *
1588 start_ofp_stats_reply(const struct ofp_header *request, size_t body_len)
1590 const struct ofp_stats_request *osr
1591 = (const struct ofp_stats_request *) request;
1592 return make_ofp_stats_reply(osr->header.xid, osr->type, body_len);
1596 append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn,
1597 struct ofpbuf **msgp)
1599 struct ofpbuf *msg = *msgp;
1600 assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply));
1601 if (nbytes + msg->size > UINT16_MAX) {
1602 struct ofp_stats_reply *reply = msg->data;
1603 reply->flags = htons(OFPSF_REPLY_MORE);
1604 *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes);
1605 ofconn_send_reply(ofconn, msg);
1607 return ofpbuf_put_uninit(*msgp, nbytes);
1610 static struct ofpbuf *
1611 make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len)
1613 struct nicira_stats_msg *nsm;
1616 msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX));
1617 nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg);
1618 nsm->type = htons(OFPST_VENDOR);
1619 nsm->flags = htons(0);
1620 nsm->vendor = htonl(NX_VENDOR_ID);
1621 nsm->subtype = subtype;
1625 static struct ofpbuf *
1626 start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len)
1628 return make_nxstats_reply(request->header.xid, request->subtype, body_len);
1632 append_nxstats_reply(size_t nbytes, struct ofconn *ofconn,
1633 struct ofpbuf **msgp)
1635 struct ofpbuf *msg = *msgp;
1636 assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg));
1637 if (nbytes + msg->size > UINT16_MAX) {
1638 struct nicira_stats_msg *reply = msg->data;
1639 reply->flags = htons(OFPSF_REPLY_MORE);
1640 *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes);
1641 ofconn_send_reply(ofconn, msg);
1643 ofpbuf_prealloc_tailroom(*msgp, nbytes);
1647 handle_desc_stats_request(struct ofconn *ofconn,
1648 const struct ofp_header *request)
1650 struct ofproto *p = ofconn_get_ofproto(ofconn);
1651 struct ofp_desc_stats *ods;
1654 msg = start_ofp_stats_reply(request, sizeof *ods);
1655 ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg);
1656 memset(ods, 0, sizeof *ods);
1657 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1658 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1659 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1660 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1661 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1662 ofconn_send_reply(ofconn, msg);
1668 handle_table_stats_request(struct ofconn *ofconn,
1669 const struct ofp_header *request)
1671 struct ofproto *p = ofconn_get_ofproto(ofconn);
1672 struct ofp_table_stats *ots;
1676 msg = start_ofp_stats_reply(request, sizeof *ots * p->n_tables);
1678 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1679 for (i = 0; i < p->n_tables; i++) {
1680 ots[i].table_id = i;
1681 sprintf(ots[i].name, "table%zu", i);
1682 ots[i].wildcards = htonl(OFPFW_ALL);
1683 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1684 ots[i].active_count = htonl(classifier_count(&p->tables[i]));
1687 p->ofproto_class->get_tables(p, ots);
1689 ofconn_send_reply(ofconn, msg);
1694 append_port_stat(struct ofport *port, struct ofconn *ofconn,
1695 struct ofpbuf **msgp)
1697 struct netdev_stats stats;
1698 struct ofp_port_stats *ops;
1700 /* Intentionally ignore return value, since errors will set
1701 * 'stats' to all-1s, which is correct for OpenFlow, and
1702 * netdev_get_stats() will log errors. */
1703 netdev_get_stats(port->netdev, &stats);
1705 ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp);
1706 ops->port_no = port->opp.port_no;
1707 memset(ops->pad, 0, sizeof ops->pad);
1708 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1709 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1710 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1711 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1712 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1713 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1714 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1715 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1716 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
1717 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
1718 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
1719 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
1723 handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
1725 struct ofproto *p = ofconn_get_ofproto(ofconn);
1726 const struct ofp_port_stats_request *psr = ofputil_stats_body(oh);
1727 struct ofp_port_stats *ops;
1729 struct ofport *port;
1731 msg = start_ofp_stats_reply(oh, sizeof *ops * 16);
1732 if (psr->port_no != htons(OFPP_NONE)) {
1733 port = ofproto_get_port(p, ntohs(psr->port_no));
1735 append_port_stat(port, ofconn, &msg);
1738 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
1739 append_port_stat(port, ofconn, &msg);
1743 ofconn_send_reply(ofconn, msg);
1748 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
1750 long long int msecs = time_msec() - start;
1751 *sec = msecs / 1000;
1752 *nsec = (msecs % 1000) * (1000 * 1000);
1756 calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be)
1760 calc_flow_duration__(start, &sec, &nsec);
1761 *sec_be = htonl(sec);
1762 *nsec_be = htonl(nsec);
1766 put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule,
1767 ovs_be16 out_port, struct ofpbuf **replyp)
1769 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1770 struct ofp_flow_stats *ofs;
1771 uint64_t packet_count, byte_count;
1772 size_t act_len, len;
1774 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
1778 act_len = sizeof *rule->actions * rule->n_actions;
1779 len = offsetof(struct ofp_flow_stats, actions) + act_len;
1781 ofproto->ofproto_class->rule_get_stats(rule, &packet_count, &byte_count);
1783 ofs = append_ofp_stats_reply(len, ofconn, replyp);
1784 ofs->length = htons(len);
1785 ofs->table_id = rule->table_id;
1787 ofputil_cls_rule_to_match(&rule->cr, &ofs->match);
1788 put_32aligned_be64(&ofs->cookie, rule->flow_cookie);
1789 calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec);
1790 ofs->priority = htons(rule->cr.priority);
1791 ofs->idle_timeout = htons(rule->idle_timeout);
1792 ofs->hard_timeout = htons(rule->hard_timeout);
1793 memset(ofs->pad2, 0, sizeof ofs->pad2);
1794 put_32aligned_be64(&ofs->packet_count, htonll(packet_count));
1795 put_32aligned_be64(&ofs->byte_count, htonll(byte_count));
1796 if (rule->n_actions > 0) {
1797 memcpy(ofs->actions, rule->actions, act_len);
1801 static struct classifier *
1802 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
1804 if (table_id == 0xff) {
1805 return &ofproto->tables[0];
1806 } else if (table_id < ofproto->n_tables) {
1807 return &ofproto->tables[table_id];
1809 /* It would probably be better to reply with an error but there doesn't
1810 * seem to be any appropriate value, so that might just be
1812 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
1818 static struct classifier *
1819 next_matching_table(struct ofproto *ofproto,
1820 struct classifier *cls, uint8_t table_id)
1822 return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
1827 /* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
1830 * - If TABLE_ID is 0xff, this iterates over every classifier table in
1833 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
1834 * only once, for that table.
1836 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning
1837 * and does not enter the loop at all.
1839 * All parameters are evaluated multiple times.
1841 #define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
1842 for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
1844 (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
1847 handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
1849 const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh);
1850 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1851 struct classifier *cls;
1852 struct cls_rule target;
1853 struct ofpbuf *reply;
1855 COVERAGE_INC(ofproto_flows_req);
1856 reply = start_ofp_stats_reply(oh, 1024);
1857 ofputil_cls_rule_from_match(&fsr->match, 0, &target);
1858 FOR_EACH_MATCHING_TABLE (cls, fsr->table_id, ofproto) {
1859 struct cls_cursor cursor;
1862 cls_cursor_init(&cursor, cls, &target);
1863 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1864 put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply);
1867 ofconn_send_reply(ofconn, reply);
1873 put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule,
1874 ovs_be16 out_port, struct ofpbuf **replyp)
1876 struct nx_flow_stats *nfs;
1877 uint64_t packet_count, byte_count;
1878 size_t act_len, start_len;
1879 struct ofpbuf *reply;
1881 if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) {
1885 rule->ofproto->ofproto_class->rule_get_stats(rule,
1886 &packet_count, &byte_count);
1888 act_len = sizeof *rule->actions * rule->n_actions;
1890 append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp);
1891 start_len = (*replyp)->size;
1894 nfs = ofpbuf_put_uninit(reply, sizeof *nfs);
1897 calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec);
1898 nfs->cookie = rule->flow_cookie;
1899 nfs->priority = htons(rule->cr.priority);
1900 nfs->idle_timeout = htons(rule->idle_timeout);
1901 nfs->hard_timeout = htons(rule->hard_timeout);
1902 nfs->match_len = htons(nx_put_match(reply, &rule->cr));
1903 memset(nfs->pad2, 0, sizeof nfs->pad2);
1904 nfs->packet_count = htonll(packet_count);
1905 nfs->byte_count = htonll(byte_count);
1906 if (rule->n_actions > 0) {
1907 ofpbuf_put(reply, rule->actions, act_len);
1909 nfs->length = htons(reply->size - start_len);
1913 handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh)
1915 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1916 struct nx_flow_stats_request *nfsr;
1917 struct classifier *cls;
1918 struct cls_rule target;
1919 struct ofpbuf *reply;
1923 ofpbuf_use_const(&b, oh, ntohs(oh->length));
1925 /* Dissect the message. */
1926 nfsr = ofpbuf_pull(&b, sizeof *nfsr);
1927 error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target);
1932 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
1935 COVERAGE_INC(ofproto_flows_req);
1936 reply = start_nxstats_reply(&nfsr->nsm, 1024);
1937 FOR_EACH_MATCHING_TABLE (cls, nfsr->table_id, ofproto) {
1938 struct cls_cursor cursor;
1941 cls_cursor_init(&cursor, cls, &target);
1942 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1943 put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply);
1946 ofconn_send_reply(ofconn, reply);
1952 flow_stats_ds(struct rule *rule, struct ds *results)
1954 uint64_t packet_count, byte_count;
1955 size_t act_len = sizeof *rule->actions * rule->n_actions;
1957 rule->ofproto->ofproto_class->rule_get_stats(rule,
1958 &packet_count, &byte_count);
1960 if (rule->table_id != 0) {
1961 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
1963 ds_put_format(results, "duration=%llds, ",
1964 (time_msec() - rule->created) / 1000);
1965 ds_put_format(results, "priority=%u, ", rule->cr.priority);
1966 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
1967 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
1968 cls_rule_format(&rule->cr, results);
1969 ds_put_char(results, ',');
1971 ofp_print_actions(results, &rule->actions->header, act_len);
1973 ds_put_cstr(results, "drop");
1975 ds_put_cstr(results, "\n");
1978 /* Adds a pretty-printed description of all flows to 'results', including
1979 * hidden flows (e.g., set up by in-band control). */
1981 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
1983 struct classifier *cls;
1985 for (cls = &p->tables[0]; cls < &p->tables[p->n_tables]; cls++) {
1986 struct cls_cursor cursor;
1989 cls_cursor_init(&cursor, cls, NULL);
1990 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1991 flow_stats_ds(rule, results);
1996 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
1997 * '*engine_type' and '*engine_id', respectively. */
1999 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2000 uint8_t *engine_type, uint8_t *engine_id)
2002 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2006 query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target,
2007 ovs_be16 out_port, uint8_t table_id,
2008 struct ofp_aggregate_stats_reply *oasr)
2010 uint64_t total_packets = 0;
2011 uint64_t total_bytes = 0;
2012 struct classifier *cls;
2015 COVERAGE_INC(ofproto_agg_request);
2017 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
2018 struct cls_cursor cursor;
2021 cls_cursor_init(&cursor, cls, target);
2022 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2023 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
2024 uint64_t packet_count;
2025 uint64_t byte_count;
2027 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2030 total_packets += packet_count;
2031 total_bytes += byte_count;
2037 oasr->flow_count = htonl(n_flows);
2038 put_32aligned_be64(&oasr->packet_count, htonll(total_packets));
2039 put_32aligned_be64(&oasr->byte_count, htonll(total_bytes));
2040 memset(oasr->pad, 0, sizeof oasr->pad);
2044 handle_aggregate_stats_request(struct ofconn *ofconn,
2045 const struct ofp_header *oh)
2047 const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh);
2048 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2049 struct ofp_aggregate_stats_reply *reply;
2050 struct cls_rule target;
2053 ofputil_cls_rule_from_match(&request->match, 0, &target);
2055 msg = start_ofp_stats_reply(oh, sizeof *reply);
2056 reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg);
2057 query_aggregate_stats(ofproto, &target, request->out_port,
2058 request->table_id, reply);
2059 ofconn_send_reply(ofconn, msg);
2064 handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh)
2066 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2067 struct nx_aggregate_stats_request *request;
2068 struct ofp_aggregate_stats_reply *reply;
2069 struct cls_rule target;
2074 ofpbuf_use_const(&b, oh, ntohs(oh->length));
2076 /* Dissect the message. */
2077 request = ofpbuf_pull(&b, sizeof *request);
2078 error = nx_pull_match(&b, ntohs(request->match_len), 0, &target);
2083 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2087 COVERAGE_INC(ofproto_flows_req);
2088 buf = start_nxstats_reply(&request->nsm, sizeof *reply);
2089 reply = ofpbuf_put_uninit(buf, sizeof *reply);
2090 query_aggregate_stats(ofproto, &target, request->out_port,
2091 request->table_id, reply);
2092 ofconn_send_reply(ofconn, buf);
2097 struct queue_stats_cbdata {
2098 struct ofconn *ofconn;
2099 struct ofport *ofport;
2104 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2105 const struct netdev_queue_stats *stats)
2107 struct ofp_queue_stats *reply;
2109 reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg);
2110 reply->port_no = cbdata->ofport->opp.port_no;
2111 memset(reply->pad, 0, sizeof reply->pad);
2112 reply->queue_id = htonl(queue_id);
2113 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2114 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2115 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2119 handle_queue_stats_dump_cb(uint32_t queue_id,
2120 struct netdev_queue_stats *stats,
2123 struct queue_stats_cbdata *cbdata = cbdata_;
2125 put_queue_stats(cbdata, queue_id, stats);
2129 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2130 struct queue_stats_cbdata *cbdata)
2132 cbdata->ofport = port;
2133 if (queue_id == OFPQ_ALL) {
2134 netdev_dump_queue_stats(port->netdev,
2135 handle_queue_stats_dump_cb, cbdata);
2137 struct netdev_queue_stats stats;
2139 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2140 put_queue_stats(cbdata, queue_id, &stats);
2146 handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh)
2148 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2149 const struct ofp_queue_stats_request *qsr;
2150 struct queue_stats_cbdata cbdata;
2151 struct ofport *port;
2152 unsigned int port_no;
2155 qsr = ofputil_stats_body(oh);
2157 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN);
2160 COVERAGE_INC(ofproto_queue_req);
2162 cbdata.ofconn = ofconn;
2163 cbdata.msg = start_ofp_stats_reply(oh, 128);
2165 port_no = ntohs(qsr->port_no);
2166 queue_id = ntohl(qsr->queue_id);
2167 if (port_no == OFPP_ALL) {
2168 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2169 handle_queue_stats_for_port(port, queue_id, &cbdata);
2171 } else if (port_no < OFPP_MAX) {
2172 port = ofproto_get_port(ofproto, port_no);
2174 handle_queue_stats_for_port(port, queue_id, &cbdata);
2177 ofpbuf_delete(cbdata.msg);
2178 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
2180 ofconn_send_reply(ofconn, cbdata.msg);
2185 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2186 * in which no matching flow already exists in the flow table.
2188 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2189 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
2190 * OpenFlow error code as encoded by ofp_mkerr() on failure.
2192 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2195 add_flow(struct ofconn *ofconn, struct flow_mod *fm)
2197 struct ofproto *p = ofconn_get_ofproto(ofconn);
2198 struct ofpbuf *packet;
2204 if (fm->flags & OFPFF_CHECK_OVERLAP) {
2205 struct classifier *cls;
2207 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, p) {
2208 if (classifier_rule_overlaps(cls, &fm->cr)) {
2209 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2214 buf_err = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id, &packet, &in_port);
2215 error = rule_create(p, &fm->cr, fm->actions, fm->n_actions,
2216 fm->idle_timeout, fm->hard_timeout, fm->cookie,
2217 fm->flags & OFPFF_SEND_FLOW_REM, &rule);
2219 ofpbuf_delete(packet);
2225 return rule_execute(rule, in_port, packet);
2230 /* Searches 'p' for an exact match for 'fm', in the table or tables indicated
2231 * by fm->table_id. Returns 0 if no match was found, 1 if exactly one match
2232 * was found, 2 if more than one match was found. If exactly one match is
2233 * found, sets '*rulep' to the match, otherwise to NULL.
2235 * This implements the rules for "strict" matching explained in the comment on
2236 * struct nxt_flow_mod_table_id in nicira-ext.h.
2238 * Ignores hidden rules. */
2240 find_flow_strict(struct ofproto *p, const struct flow_mod *fm,
2241 struct rule **rulep)
2243 struct classifier *cls;
2246 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, p) {
2249 rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, &fm->cr));
2250 if (rule && !rule_is_hidden(rule)) {
2258 return *rulep != NULL;
2262 send_buffered_packet(struct ofconn *ofconn,
2263 struct rule *rule, uint32_t buffer_id)
2265 struct ofpbuf *packet;
2269 if (buffer_id == UINT32_MAX) {
2273 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
2278 return rule_execute(rule, in_port, packet);
2281 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2283 struct modify_flows_cbdata {
2284 struct ofproto *ofproto;
2285 const struct flow_mod *fm;
2289 static int modify_flow(const struct flow_mod *, struct rule *);
2291 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2292 * encoded by ofp_mkerr() on failure.
2294 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2297 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm)
2299 struct ofproto *p = ofconn_get_ofproto(ofconn);
2300 struct rule *match = NULL;
2301 struct classifier *cls;
2305 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, p) {
2306 struct cls_cursor cursor;
2309 cls_cursor_init(&cursor, cls, &fm->cr);
2310 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2311 if (!rule_is_hidden(rule)) {
2312 int retval = modify_flow(fm, rule);
2325 /* This credits the packet to whichever flow happened to match last.
2326 * That's weird. Maybe we should do a lookup for the flow that
2327 * actually matches the packet? Who knows. */
2328 send_buffered_packet(ofconn, match, fm->buffer_id);
2331 return add_flow(ofconn, fm);
2335 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2336 * code as encoded by ofp_mkerr() on failure.
2338 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2341 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm)
2343 struct ofproto *p = ofconn_get_ofproto(ofconn);
2347 switch (find_flow_strict(p, fm, &rule)) {
2349 return add_flow(ofconn, fm);
2352 error = modify_flow(fm, rule);
2354 error = send_buffered_packet(ofconn, rule, fm->buffer_id);
2366 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
2367 * been identified as a flow to be modified, by changing the rule's actions to
2368 * match those in 'ofm' (which is followed by 'n_actions' ofp_action[]
2371 modify_flow(const struct flow_mod *fm, struct rule *rule)
2373 size_t actions_len = fm->n_actions * sizeof *rule->actions;
2376 if (fm->n_actions == rule->n_actions
2378 || !memcmp(fm->actions, rule->actions, actions_len))) {
2381 error = rule->ofproto->ofproto_class->rule_modify_actions(
2382 rule, fm->actions, fm->n_actions);
2384 free(rule->actions);
2385 rule->actions = (fm->n_actions
2386 ? xmemdup(fm->actions, actions_len)
2388 rule->n_actions = fm->n_actions;
2393 rule->flow_cookie = fm->cookie;
2399 /* OFPFC_DELETE implementation. */
2401 static void delete_flow(struct rule *, ovs_be16 out_port);
2403 /* Implements OFPFC_DELETE. */
2405 delete_flows_loose(struct ofproto *p, const struct flow_mod *fm)
2407 struct classifier *cls;
2409 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, p) {
2410 struct rule *rule, *next_rule;
2411 struct cls_cursor cursor;
2413 cls_cursor_init(&cursor, cls, &fm->cr);
2414 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
2415 delete_flow(rule, htons(fm->out_port));
2420 /* Implements OFPFC_DELETE_STRICT. */
2422 delete_flow_strict(struct ofproto *p, struct flow_mod *fm)
2425 if (find_flow_strict(p, fm, &rule) == 1) {
2426 delete_flow(rule, htons(fm->out_port));
2430 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
2431 * been identified as a flow to delete from 'p''s flow table, by deleting the
2432 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
2435 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
2436 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
2437 * specified 'out_port'. */
2439 delete_flow(struct rule *rule, ovs_be16 out_port)
2441 if (rule_is_hidden(rule)) {
2445 if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) {
2449 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2450 ofproto_rule_destroy(rule);
2454 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2456 struct ofputil_flow_removed fr;
2458 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2463 fr.cookie = rule->flow_cookie;
2465 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2466 fr.idle_timeout = rule->idle_timeout;
2467 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2470 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2473 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2474 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2477 * ofproto implementation ->run() functions should use this function to expire
2478 * OpenFlow flows. */
2480 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2482 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2483 ofproto_rule_send_removed(rule, reason);
2484 ofproto_rule_destroy(rule);
2488 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2490 struct ofproto *p = ofconn_get_ofproto(ofconn);
2494 error = reject_slave_controller(ofconn, "flow_mod");
2499 error = ofputil_decode_flow_mod(&fm, oh,
2500 ofconn_get_flow_mod_table_id(ofconn));
2505 /* We do not support the emergency flow cache. It will hopefully get
2506 * dropped from OpenFlow in the near future. */
2507 if (fm.flags & OFPFF_EMERG) {
2508 /* There isn't a good fit for an error code, so just state that the
2509 * flow table is full. */
2510 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2513 switch (fm.command) {
2515 return add_flow(ofconn, &fm);
2518 return modify_flows_loose(ofconn, &fm);
2520 case OFPFC_MODIFY_STRICT:
2521 return modify_flow_strict(ofconn, &fm);
2524 delete_flows_loose(p, &fm);
2527 case OFPFC_DELETE_STRICT:
2528 delete_flow_strict(p, &fm);
2532 if (fm.command > 0xff) {
2533 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2534 "flow_mod_table_id extension is not enabled");
2536 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2541 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2543 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2544 struct nx_role_request *reply;
2548 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2549 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
2550 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2553 role = ntohl(nrr->role);
2554 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2555 && role != NX_ROLE_SLAVE) {
2556 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
2558 /* There's no good error code for this. */
2559 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
2562 ofconn_set_role(ofconn, role);
2564 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2565 reply->role = htonl(role);
2566 ofconn_send_reply(ofconn, buf);
2572 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2573 const struct ofp_header *oh)
2575 const struct nxt_flow_mod_table_id *msg
2576 = (const struct nxt_flow_mod_table_id *) oh;
2578 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2583 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2585 const struct nxt_set_flow_format *msg
2586 = (const struct nxt_set_flow_format *) oh;
2589 format = ntohl(msg->format);
2590 if (format == NXFF_OPENFLOW10
2591 || format == NXFF_NXM) {
2592 ofconn_set_flow_format(ofconn, format);
2595 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2600 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2602 struct ofp_header *ob;
2605 /* Currently, everything executes synchronously, so we can just
2606 * immediately send the barrier reply. */
2607 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2608 ofconn_send_reply(ofconn, buf);
2613 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2615 const struct ofp_header *oh = msg->data;
2616 const struct ofputil_msg_type *type;
2619 error = ofputil_decode_msg_type(oh, &type);
2624 switch (ofputil_msg_type_code(type)) {
2625 /* OpenFlow requests. */
2626 case OFPUTIL_OFPT_ECHO_REQUEST:
2627 return handle_echo_request(ofconn, oh);
2629 case OFPUTIL_OFPT_FEATURES_REQUEST:
2630 return handle_features_request(ofconn, oh);
2632 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2633 return handle_get_config_request(ofconn, oh);
2635 case OFPUTIL_OFPT_SET_CONFIG:
2636 return handle_set_config(ofconn, msg->data);
2638 case OFPUTIL_OFPT_PACKET_OUT:
2639 return handle_packet_out(ofconn, oh);
2641 case OFPUTIL_OFPT_PORT_MOD:
2642 return handle_port_mod(ofconn, oh);
2644 case OFPUTIL_OFPT_FLOW_MOD:
2645 return handle_flow_mod(ofconn, oh);
2647 case OFPUTIL_OFPT_BARRIER_REQUEST:
2648 return handle_barrier_request(ofconn, oh);
2650 /* OpenFlow replies. */
2651 case OFPUTIL_OFPT_ECHO_REPLY:
2654 /* Nicira extension requests. */
2655 case OFPUTIL_NXT_ROLE_REQUEST:
2656 return handle_role_request(ofconn, oh);
2658 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2659 return handle_nxt_flow_mod_table_id(ofconn, oh);
2661 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2662 return handle_nxt_set_flow_format(ofconn, oh);
2664 case OFPUTIL_NXT_FLOW_MOD:
2665 return handle_flow_mod(ofconn, oh);
2667 /* OpenFlow statistics requests. */
2668 case OFPUTIL_OFPST_DESC_REQUEST:
2669 return handle_desc_stats_request(ofconn, oh);
2671 case OFPUTIL_OFPST_FLOW_REQUEST:
2672 return handle_flow_stats_request(ofconn, oh);
2674 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2675 return handle_aggregate_stats_request(ofconn, oh);
2677 case OFPUTIL_OFPST_TABLE_REQUEST:
2678 return handle_table_stats_request(ofconn, oh);
2680 case OFPUTIL_OFPST_PORT_REQUEST:
2681 return handle_port_stats_request(ofconn, oh);
2683 case OFPUTIL_OFPST_QUEUE_REQUEST:
2684 return handle_queue_stats_request(ofconn, oh);
2686 /* Nicira extension statistics requests. */
2687 case OFPUTIL_NXST_FLOW_REQUEST:
2688 return handle_nxst_flow(ofconn, oh);
2690 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2691 return handle_nxst_aggregate(ofconn, oh);
2693 case OFPUTIL_INVALID:
2694 case OFPUTIL_OFPT_HELLO:
2695 case OFPUTIL_OFPT_ERROR:
2696 case OFPUTIL_OFPT_FEATURES_REPLY:
2697 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
2698 case OFPUTIL_OFPT_PACKET_IN:
2699 case OFPUTIL_OFPT_FLOW_REMOVED:
2700 case OFPUTIL_OFPT_PORT_STATUS:
2701 case OFPUTIL_OFPT_BARRIER_REPLY:
2702 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
2703 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
2704 case OFPUTIL_OFPST_DESC_REPLY:
2705 case OFPUTIL_OFPST_FLOW_REPLY:
2706 case OFPUTIL_OFPST_QUEUE_REPLY:
2707 case OFPUTIL_OFPST_PORT_REPLY:
2708 case OFPUTIL_OFPST_TABLE_REPLY:
2709 case OFPUTIL_OFPST_AGGREGATE_REPLY:
2710 case OFPUTIL_NXT_ROLE_REPLY:
2711 case OFPUTIL_NXT_FLOW_REMOVED:
2712 case OFPUTIL_NXST_FLOW_REPLY:
2713 case OFPUTIL_NXST_AGGREGATE_REPLY:
2715 if (VLOG_IS_WARN_ENABLED()) {
2716 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2717 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2720 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
2721 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2723 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2729 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
2731 int error = handle_openflow__(ofconn, ofp_msg);
2733 send_error_oh(ofconn, ofp_msg->data, error);
2735 COVERAGE_INC(ofproto_recv_openflow);
2739 pick_datapath_id(const struct ofproto *ofproto)
2741 const struct ofport *port;
2743 port = ofproto_get_port(ofproto, OFPP_LOCAL);
2745 uint8_t ea[ETH_ADDR_LEN];
2748 error = netdev_get_etheraddr(port->netdev, ea);
2750 return eth_addr_to_uint64(ea);
2752 VLOG_WARN("could not get MAC address for %s (%s)",
2753 netdev_get_name(port->netdev), strerror(error));
2755 return ofproto->fallback_dpid;
2759 pick_fallback_dpid(void)
2761 uint8_t ea[ETH_ADDR_LEN];
2762 eth_addr_nicira_random(ea);
2763 return eth_addr_to_uint64(ea);
2766 /* unixctl commands. */
2769 ofproto_lookup(const char *name)
2771 struct ofproto *ofproto;
2773 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
2775 if (!strcmp(ofproto->name, name)) {
2783 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
2784 void *aux OVS_UNUSED)
2786 struct ofproto *ofproto;
2790 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
2791 ds_put_format(&results, "%s\n", ofproto->name);
2793 unixctl_command_reply(conn, 200, ds_cstr(&results));
2794 ds_destroy(&results);
2798 ofproto_unixctl_init(void)
2800 static bool registered;
2806 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);