2 * Copyright (c) 2009, 2010, 2011, 2012 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.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
34 #include "ofp-errors.h"
35 #include "ofp-print.h"
38 #include "ofproto-provider.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
44 #include "poll-loop.h"
48 #include "unaligned.h"
52 VLOG_DEFINE_THIS_MODULE(ofproto);
54 COVERAGE_DEFINE(ofproto_error);
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);
65 S_OPENFLOW, /* Processing OpenFlow commands. */
66 S_FLUSH, /* Deleting all flow table rules. */
69 enum ofoperation_type {
75 /* A single OpenFlow request can execute any number of operations. The
76 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
77 * ofconn to which an error reply should be sent if necessary.
79 * ofproto initiates some operations internally. These operations are still
80 * assigned to groups but will not have an associated ofconn. */
82 struct ofproto *ofproto; /* Owning ofproto. */
83 struct list ofproto_node; /* In ofproto's "pending" list. */
84 struct list ops; /* List of "struct ofoperation"s. */
86 /* Data needed to send OpenFlow reply on failure or to send a buffered
89 * If list_is_empty(ofconn_node) then this ofopgroup never had an
90 * associated ofconn or its ofconn's connection dropped after it initiated
91 * the operation. In the latter case 'ofconn' is a wild pointer that
92 * refers to freed memory, so the 'ofconn' member must be used only if
93 * !list_is_empty(ofconn_node).
95 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
96 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
97 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
98 uint32_t buffer_id; /* Buffer id from original request. */
99 int error; /* 0 if no error yet, otherwise error code. */
102 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
103 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
104 const struct ofp_header *,
106 static void ofopgroup_submit(struct ofopgroup *);
107 static void ofopgroup_destroy(struct ofopgroup *);
109 /* A single flow table operation. */
111 struct ofopgroup *group; /* Owning group. */
112 struct list group_node; /* In ofopgroup's "ops" list. */
113 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
114 struct rule *rule; /* Rule being operated upon. */
115 enum ofoperation_type type; /* Type of operation. */
116 int status; /* -1 if pending, otherwise 0 or error code. */
117 struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
118 union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */
119 int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */
120 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
123 static void ofoperation_create(struct ofopgroup *, struct rule *,
124 enum ofoperation_type);
125 static void ofoperation_destroy(struct ofoperation *);
128 static void oftable_init(struct oftable *);
129 static void oftable_destroy(struct oftable *);
131 static void oftable_remove_rule(struct rule *);
132 static struct rule *oftable_replace_rule(struct rule *);
133 static void oftable_substitute_rule(struct rule *old, struct rule *new);
136 static void ofproto_rule_destroy__(struct rule *);
137 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
140 static void ofport_destroy__(struct ofport *);
141 static void ofport_destroy(struct ofport *);
143 static void update_port(struct ofproto *, const char *devname);
144 static int init_ports(struct ofproto *);
145 static void reinit_ports(struct ofproto *);
148 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
149 const struct ofputil_flow_mod *,
150 const struct ofp_header *);
152 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
153 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
154 const struct ofputil_flow_mod *,
155 const struct ofp_header *);
158 static uint64_t pick_datapath_id(const struct ofproto *);
159 static uint64_t pick_fallback_dpid(void);
160 static void ofproto_destroy__(struct ofproto *);
161 static void set_internal_devs_mtu(struct ofproto *);
164 static void ofproto_unixctl_init(void);
166 /* All registered ofproto classes, in probe order. */
167 static const struct ofproto_class **ofproto_classes;
168 static size_t n_ofproto_classes;
169 static size_t allocated_ofproto_classes;
171 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
172 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
174 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
177 ofproto_initialize(void)
183 ofproto_class_register(&ofproto_dpif_class);
187 /* 'type' should be a normalized datapath type, as returned by
188 * ofproto_normalize_type(). Returns the corresponding ofproto_class
189 * structure, or a null pointer if there is none registered for 'type'. */
190 static const struct ofproto_class *
191 ofproto_class_find__(const char *type)
195 ofproto_initialize();
196 for (i = 0; i < n_ofproto_classes; i++) {
197 const struct ofproto_class *class = ofproto_classes[i];
202 class->enumerate_types(&types);
203 found = sset_contains(&types, type);
204 sset_destroy(&types);
210 VLOG_WARN("unknown datapath type %s", type);
214 /* Registers a new ofproto class. After successful registration, new ofprotos
215 * of that type can be created using ofproto_create(). */
217 ofproto_class_register(const struct ofproto_class *new_class)
221 for (i = 0; i < n_ofproto_classes; i++) {
222 if (ofproto_classes[i] == new_class) {
227 if (n_ofproto_classes >= allocated_ofproto_classes) {
228 ofproto_classes = x2nrealloc(ofproto_classes,
229 &allocated_ofproto_classes,
230 sizeof *ofproto_classes);
232 ofproto_classes[n_ofproto_classes++] = new_class;
236 /* Unregisters a datapath provider. 'type' must have been previously
237 * registered and not currently be in use by any ofprotos. After
238 * unregistration new datapaths of that type cannot be opened using
239 * ofproto_create(). */
241 ofproto_class_unregister(const struct ofproto_class *class)
245 for (i = 0; i < n_ofproto_classes; i++) {
246 if (ofproto_classes[i] == class) {
247 for (i++; i < n_ofproto_classes; i++) {
248 ofproto_classes[i - 1] = ofproto_classes[i];
254 VLOG_WARN("attempted to unregister an ofproto class that is not "
259 /* Clears 'types' and enumerates all registered ofproto types into it. The
260 * caller must first initialize the sset. */
262 ofproto_enumerate_types(struct sset *types)
266 ofproto_initialize();
267 for (i = 0; i < n_ofproto_classes; i++) {
268 ofproto_classes[i]->enumerate_types(types);
272 /* Returns the fully spelled out name for the given ofproto 'type'.
274 * Normalized type string can be compared with strcmp(). Unnormalized type
275 * string might be the same even if they have different spellings. */
277 ofproto_normalize_type(const char *type)
279 return type && type[0] ? type : "system";
282 /* Clears 'names' and enumerates the names of all known created ofprotos with
283 * the given 'type'. The caller must first initialize the sset. Returns 0 if
284 * successful, otherwise a positive errno value.
286 * Some kinds of datapaths might not be practically enumerable. This is not
287 * considered an error. */
289 ofproto_enumerate_names(const char *type, struct sset *names)
291 const struct ofproto_class *class = ofproto_class_find__(type);
292 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
296 ofproto_create(const char *datapath_name, const char *datapath_type,
297 struct ofproto **ofprotop)
299 const struct ofproto_class *class;
300 struct ofproto *ofproto;
301 struct oftable *table;
307 ofproto_initialize();
308 ofproto_unixctl_init();
310 datapath_type = ofproto_normalize_type(datapath_type);
311 class = ofproto_class_find__(datapath_type);
313 VLOG_WARN("could not create datapath %s of unknown type %s",
314 datapath_name, datapath_type);
318 ofproto = class->alloc();
320 VLOG_ERR("failed to allocate datapath %s of type %s",
321 datapath_name, datapath_type);
326 memset(ofproto, 0, sizeof *ofproto);
327 ofproto->ofproto_class = class;
328 ofproto->name = xstrdup(datapath_name);
329 ofproto->type = xstrdup(datapath_type);
330 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
331 hash_string(ofproto->name, 0));
332 ofproto->datapath_id = 0;
333 ofproto_set_flow_eviction_threshold(ofproto,
334 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
335 ofproto->forward_bpdu = false;
336 ofproto->fallback_dpid = pick_fallback_dpid();
337 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
338 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
339 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
340 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
341 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
342 ofproto->frag_handling = OFPC_FRAG_NORMAL;
343 hmap_init(&ofproto->ports);
344 shash_init(&ofproto->port_by_name);
345 ofproto->tables = NULL;
346 ofproto->n_tables = 0;
347 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
348 ofproto->state = S_OPENFLOW;
349 list_init(&ofproto->pending);
350 ofproto->n_pending = 0;
351 hmap_init(&ofproto->deletions);
352 ofproto->vlan_bitmap = NULL;
353 ofproto->vlans_changed = false;
355 error = ofproto->ofproto_class->construct(ofproto, &n_tables);
357 VLOG_ERR("failed to open datapath %s: %s",
358 datapath_name, strerror(error));
359 ofproto_destroy__(ofproto);
363 assert(n_tables >= 1 && n_tables <= 255);
364 ofproto->n_tables = n_tables;
365 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
366 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
370 ofproto->datapath_id = pick_datapath_id(ofproto);
371 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
379 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
381 uint64_t old_dpid = p->datapath_id;
382 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
383 if (p->datapath_id != old_dpid) {
384 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
386 /* Force all active connections to reconnect, since there is no way to
387 * notify a controller that the datapath ID has changed. */
388 ofproto_reconnect_controllers(p);
393 ofproto_set_controllers(struct ofproto *p,
394 const struct ofproto_controller *controllers,
395 size_t n_controllers)
397 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
401 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
403 connmgr_set_fail_mode(p->connmgr, fail_mode);
406 /* Drops the connections between 'ofproto' and all of its controllers, forcing
407 * them to reconnect. */
409 ofproto_reconnect_controllers(struct ofproto *ofproto)
411 connmgr_reconnect(ofproto->connmgr);
414 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
415 * in-band control should guarantee access, in the same way that in-band
416 * control guarantees access to OpenFlow controllers. */
418 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
419 const struct sockaddr_in *extras, size_t n)
421 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
424 /* Sets the OpenFlow queue used by flows set up by in-band control on
425 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
426 * flows will use the default queue. */
428 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
430 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
433 /* Sets the number of flows at which eviction from the kernel flow table
436 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
438 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
439 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
441 ofproto->flow_eviction_threshold = threshold;
445 /* If forward_bpdu is true, the NORMAL action will forward frames with
446 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
447 * the NORMAL action will drop these frames. */
449 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
451 bool old_val = ofproto->forward_bpdu;
452 ofproto->forward_bpdu = forward_bpdu;
453 if (old_val != ofproto->forward_bpdu) {
454 if (ofproto->ofproto_class->forward_bpdu_changed) {
455 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
461 ofproto_set_desc(struct ofproto *p,
462 const char *mfr_desc, const char *hw_desc,
463 const char *sw_desc, const char *serial_desc,
466 struct ofp_desc_stats *ods;
469 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
470 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
471 sizeof ods->mfr_desc);
474 p->mfr_desc = xstrdup(mfr_desc);
477 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
478 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
479 sizeof ods->hw_desc);
482 p->hw_desc = xstrdup(hw_desc);
485 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
486 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
487 sizeof ods->sw_desc);
490 p->sw_desc = xstrdup(sw_desc);
493 if (strlen(serial_desc) >= sizeof ods->serial_num) {
494 VLOG_WARN("truncating serial_desc, must be less than %zu "
496 sizeof ods->serial_num);
498 free(p->serial_desc);
499 p->serial_desc = xstrdup(serial_desc);
502 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
503 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
504 sizeof ods->dp_desc);
507 p->dp_desc = xstrdup(dp_desc);
512 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
514 return connmgr_set_snoops(ofproto->connmgr, snoops);
518 ofproto_set_netflow(struct ofproto *ofproto,
519 const struct netflow_options *nf_options)
521 if (nf_options && sset_is_empty(&nf_options->collectors)) {
525 if (ofproto->ofproto_class->set_netflow) {
526 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
528 return nf_options ? EOPNOTSUPP : 0;
533 ofproto_set_sflow(struct ofproto *ofproto,
534 const struct ofproto_sflow_options *oso)
536 if (oso && sset_is_empty(&oso->targets)) {
540 if (ofproto->ofproto_class->set_sflow) {
541 return ofproto->ofproto_class->set_sflow(ofproto, oso);
543 return oso ? EOPNOTSUPP : 0;
547 /* Spanning Tree Protocol (STP) configuration. */
549 /* Configures STP on 'ofproto' using the settings defined in 's'. If
550 * 's' is NULL, disables STP.
552 * Returns 0 if successful, otherwise a positive errno value. */
554 ofproto_set_stp(struct ofproto *ofproto,
555 const struct ofproto_stp_settings *s)
557 return (ofproto->ofproto_class->set_stp
558 ? ofproto->ofproto_class->set_stp(ofproto, s)
562 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
563 * 'enabled' member of 's' is false, then the other members are not
566 * Returns 0 if successful, otherwise a positive errno value. */
568 ofproto_get_stp_status(struct ofproto *ofproto,
569 struct ofproto_stp_status *s)
571 return (ofproto->ofproto_class->get_stp_status
572 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
576 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
577 * in 's'. The caller is responsible for assigning STP port numbers
578 * (using the 'port_num' member in the range of 1 through 255, inclusive)
579 * and ensuring there are no duplicates. If the 's' is NULL, then STP
580 * is disabled on the port.
582 * Returns 0 if successful, otherwise a positive errno value.*/
584 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
585 const struct ofproto_port_stp_settings *s)
587 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
589 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
590 ofproto->name, ofp_port);
594 return (ofproto->ofproto_class->set_stp_port
595 ? ofproto->ofproto_class->set_stp_port(ofport, s)
599 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
600 * 's'. If the 'enabled' member in 's' is false, then the other members
601 * are not meaningful.
603 * Returns 0 if successful, otherwise a positive errno value.*/
605 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
606 struct ofproto_port_stp_status *s)
608 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
610 VLOG_WARN("%s: cannot get STP status on nonexistent port %"PRIu16,
611 ofproto->name, ofp_port);
615 return (ofproto->ofproto_class->get_stp_port_status
616 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
620 /* Queue DSCP configuration. */
622 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
623 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
624 * to implement QoS. Instead, it is used to implement features which require
625 * knowledge of what queues exist on a port, and some basic information about
628 * Returns 0 if successful, otherwise a positive errno value. */
630 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
631 const struct ofproto_port_queue *queues,
634 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
637 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
638 ofproto->name, ofp_port);
642 return (ofproto->ofproto_class->set_queues
643 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
647 /* Connectivity Fault Management configuration. */
649 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
651 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
653 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
654 if (ofport && ofproto->ofproto_class->set_cfm) {
655 ofproto->ofproto_class->set_cfm(ofport, NULL);
659 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
660 * basic configuration from the configuration members in 'cfm', and the remote
661 * maintenance point ID from remote_mpid. Ignores the statistics members of
664 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
666 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
667 const struct cfm_settings *s)
669 struct ofport *ofport;
672 ofport = ofproto_get_port(ofproto, ofp_port);
674 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
675 ofproto->name, ofp_port);
679 /* XXX: For configuration simplicity, we only support one remote_mpid
680 * outside of the CFM module. It's not clear if this is the correct long
681 * term solution or not. */
682 error = (ofproto->ofproto_class->set_cfm
683 ? ofproto->ofproto_class->set_cfm(ofport, s)
686 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
687 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
692 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
693 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
694 * 0 if LACP partner information is not current (generally indicating a
695 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
697 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
699 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
700 return (ofport && ofproto->ofproto_class->port_is_lacp_current
701 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
707 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
708 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
709 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
710 * configuration plus, if there is more than one slave, a bonding
713 * If 'aux' is already registered then this function updates its configuration
714 * to 's'. Otherwise, this function registers a new bundle.
716 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
719 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
720 const struct ofproto_bundle_settings *s)
722 return (ofproto->ofproto_class->bundle_set
723 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
727 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
728 * If no such bundle has been registered, this has no effect. */
730 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
732 return ofproto_bundle_register(ofproto, aux, NULL);
736 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
737 * If 'aux' is already registered then this function updates its configuration
738 * to 's'. Otherwise, this function registers a new mirror. */
740 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
741 const struct ofproto_mirror_settings *s)
743 return (ofproto->ofproto_class->mirror_set
744 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
748 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
749 * If no mirror has been registered, this has no effect. */
751 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
753 return ofproto_mirror_register(ofproto, aux, NULL);
756 /* Retrieves statistics from mirror associated with client data pointer
757 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
758 * 'bytes', respectively. If a particular counters is not supported,
759 * the appropriate argument is set to UINT64_MAX. */
761 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
762 uint64_t *packets, uint64_t *bytes)
764 if (!ofproto->ofproto_class->mirror_get_stats) {
765 *packets = *bytes = UINT64_MAX;
769 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
773 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
774 * which all packets are flooded, instead of using MAC learning. If
775 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
777 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
780 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
782 return (ofproto->ofproto_class->set_flood_vlans
783 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
787 /* Returns true if 'aux' is a registered bundle that is currently in use as the
788 * output for a mirror. */
790 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
792 return (ofproto->ofproto_class->is_mirror_output_bundle
793 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
798 ofproto_has_snoops(const struct ofproto *ofproto)
800 return connmgr_has_snoops(ofproto->connmgr);
804 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
806 connmgr_get_snoops(ofproto->connmgr, snoops);
810 ofproto_flush__(struct ofproto *ofproto)
812 struct ofopgroup *group;
813 struct oftable *table;
815 if (ofproto->ofproto_class->flush) {
816 ofproto->ofproto_class->flush(ofproto);
819 group = ofopgroup_create_unattached(ofproto);
820 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
821 struct rule *rule, *next_rule;
822 struct cls_cursor cursor;
824 cls_cursor_init(&cursor, &table->cls, NULL);
825 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
826 if (!rule->pending) {
827 ofoperation_create(group, rule, OFOPERATION_DELETE);
828 oftable_remove_rule(rule);
829 ofproto->ofproto_class->rule_destruct(rule);
833 ofopgroup_submit(group);
837 ofproto_destroy__(struct ofproto *ofproto)
839 struct oftable *table;
841 assert(list_is_empty(&ofproto->pending));
842 assert(!ofproto->n_pending);
844 connmgr_destroy(ofproto->connmgr);
846 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
849 free(ofproto->mfr_desc);
850 free(ofproto->hw_desc);
851 free(ofproto->sw_desc);
852 free(ofproto->serial_desc);
853 free(ofproto->dp_desc);
854 hmap_destroy(&ofproto->ports);
855 shash_destroy(&ofproto->port_by_name);
857 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
858 oftable_destroy(table);
860 free(ofproto->tables);
862 hmap_destroy(&ofproto->deletions);
864 free(ofproto->vlan_bitmap);
866 ofproto->ofproto_class->dealloc(ofproto);
870 ofproto_destroy(struct ofproto *p)
872 struct ofport *ofport, *next_ofport;
879 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
880 ofport_destroy(ofport);
883 p->ofproto_class->destruct(p);
884 ofproto_destroy__(p);
887 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
888 * kernel datapath, for example, this destroys the datapath in the kernel, and
889 * with the netdev-based datapath, it tears down the data structures that
890 * represent the datapath.
892 * The datapath should not be currently open as an ofproto. */
894 ofproto_delete(const char *name, const char *type)
896 const struct ofproto_class *class = ofproto_class_find__(type);
897 return (!class ? EAFNOSUPPORT
898 : !class->del ? EACCES
899 : class->del(type, name));
903 process_port_change(struct ofproto *ofproto, int error, char *devname)
905 if (error == ENOBUFS) {
906 reinit_ports(ofproto);
908 update_port(ofproto, devname);
914 ofproto_run(struct ofproto *p)
916 struct ofport *ofport;
920 error = p->ofproto_class->run(p);
921 if (error && error != EAGAIN) {
922 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
925 if (p->ofproto_class->port_poll) {
926 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
927 process_port_change(p, error, devname);
931 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
932 unsigned int change_seq = netdev_change_seq(ofport->netdev);
933 if (ofport->change_seq != change_seq) {
934 ofport->change_seq = change_seq;
935 update_port(p, netdev_get_name(ofport->netdev));
942 connmgr_run(p->connmgr, handle_openflow);
946 connmgr_run(p->connmgr, NULL);
948 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
949 connmgr_flushed(p->connmgr);
950 p->state = S_OPENFLOW;
961 /* Performs periodic activity required by 'ofproto' that needs to be done
962 * with the least possible latency.
964 * It makes sense to call this function a couple of times per poll loop, to
965 * provide a significant performance boost on some benchmarks with the
966 * ofproto-dpif implementation. */
968 ofproto_run_fast(struct ofproto *p)
972 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
973 if (error && error != EAGAIN) {
974 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
975 p->name, strerror(error));
981 ofproto_wait(struct ofproto *p)
983 struct ofport *ofport;
985 p->ofproto_class->wait(p);
986 if (p->ofproto_class->port_poll_wait) {
987 p->ofproto_class->port_poll_wait(p);
990 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
991 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
992 poll_immediate_wake();
998 connmgr_wait(p->connmgr, true);
1002 connmgr_wait(p->connmgr, false);
1003 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1004 poll_immediate_wake();
1011 ofproto_is_alive(const struct ofproto *p)
1013 return connmgr_has_controllers(p->connmgr);
1017 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1020 connmgr_get_controller_info(ofproto->connmgr, info);
1024 ofproto_free_ofproto_controller_info(struct shash *info)
1026 connmgr_free_controller_info(info);
1029 /* Makes a deep copy of 'old' into 'port'. */
1031 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1033 port->name = xstrdup(old->name);
1034 port->type = xstrdup(old->type);
1035 port->ofp_port = old->ofp_port;
1038 /* Frees memory allocated to members of 'ofproto_port'.
1040 * Do not call this function on an ofproto_port obtained from
1041 * ofproto_port_dump_next(): that function retains ownership of the data in the
1044 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1046 free(ofproto_port->name);
1047 free(ofproto_port->type);
1050 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1052 * This function provides no status indication. An error status for the entire
1053 * dump operation is provided when it is completed by calling
1054 * ofproto_port_dump_done().
1057 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1058 const struct ofproto *ofproto)
1060 dump->ofproto = ofproto;
1061 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1065 /* Attempts to retrieve another port from 'dump', which must have been created
1066 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1067 * 'port' and returns true. On failure, returns false.
1069 * Failure might indicate an actual error or merely that the last port has been
1070 * dumped. An error status for the entire dump operation is provided when it
1071 * is completed by calling ofproto_port_dump_done().
1073 * The ofproto owns the data stored in 'port'. It will remain valid until at
1074 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1075 * ofproto_port_dump_done(). */
1077 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1078 struct ofproto_port *port)
1080 const struct ofproto *ofproto = dump->ofproto;
1086 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1089 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1095 /* Completes port table dump operation 'dump', which must have been created
1096 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1097 * error-free, otherwise a positive errno value describing the problem. */
1099 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1101 const struct ofproto *ofproto = dump->ofproto;
1103 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1106 return dump->error == EOF ? 0 : dump->error;
1109 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1110 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1111 * is non-null). On failure, returns a positive errno value and sets
1112 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1114 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1115 uint16_t *ofp_portp)
1120 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1122 update_port(ofproto, netdev_get_name(netdev));
1125 *ofp_portp = error ? OFPP_NONE : ofp_port;
1130 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1131 * initializes '*port' appropriately; on failure, returns a positive errno
1134 * The caller owns the data in 'ofproto_port' and must free it with
1135 * ofproto_port_destroy() when it is no longer needed. */
1137 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1138 struct ofproto_port *port)
1142 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1144 memset(port, 0, sizeof *port);
1149 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1150 * Returns 0 if successful, otherwise a positive errno. */
1152 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1154 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1155 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1158 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1159 if (!error && ofport) {
1160 /* 'name' is the netdev's name and update_port() is going to close the
1161 * netdev. Just in case update_port() refers to 'name' after it
1162 * destroys 'ofport', make a copy of it around the update_port()
1164 char *devname = xstrdup(name);
1165 update_port(ofproto, devname);
1171 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1172 * performs the 'n_actions' actions in 'actions'. The new flow will not
1175 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1176 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1177 * controllers; otherwise, it will be hidden.
1179 * The caller retains ownership of 'cls_rule' and 'actions'.
1181 * This is a helper function for in-band control and fail-open. */
1183 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1184 const union ofp_action *actions, size_t n_actions)
1186 const struct rule *rule;
1188 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1189 &ofproto->tables[0].cls, cls_rule));
1190 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1191 actions, n_actions)) {
1192 struct ofputil_flow_mod fm;
1194 memset(&fm, 0, sizeof fm);
1196 fm.buffer_id = UINT32_MAX;
1197 fm.actions = (union ofp_action *) actions;
1198 fm.n_actions = n_actions;
1199 add_flow(ofproto, NULL, &fm, NULL);
1203 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1204 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1205 * operation cannot be initiated now but may be retried later.
1207 * This is a helper function for in-band control and fail-open. */
1209 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1211 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1214 /* Searches for a rule with matching criteria exactly equal to 'target' in
1215 * ofproto's table 0 and, if it finds one, deletes it.
1217 * This is a helper function for in-band control and fail-open. */
1219 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1223 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1224 &ofproto->tables[0].cls, target));
1226 /* No such rule -> success. */
1228 } else if (rule->pending) {
1229 /* An operation on the rule is already pending -> failure.
1230 * Caller must retry later if it's important. */
1233 /* Initiate deletion -> success. */
1234 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1235 ofoperation_create(group, rule, OFOPERATION_DELETE);
1236 oftable_remove_rule(rule);
1237 rule->ofproto->ofproto_class->rule_destruct(rule);
1238 ofopgroup_submit(group);
1244 /* Starts the process of deleting all of the flows from all of ofproto's flow
1245 * tables and then reintroducing the flows required by in-band control and
1246 * fail-open. The process will complete in a later call to ofproto_run(). */
1248 ofproto_flush_flows(struct ofproto *ofproto)
1250 COVERAGE_INC(ofproto_flush);
1251 ofproto->state = S_FLUSH;
1255 reinit_ports(struct ofproto *p)
1257 struct ofproto_port_dump dump;
1258 struct sset devnames;
1259 struct ofport *ofport;
1260 struct ofproto_port ofproto_port;
1261 const char *devname;
1263 COVERAGE_INC(ofproto_reinit_ports);
1265 sset_init(&devnames);
1266 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1267 sset_add(&devnames, netdev_get_name(ofport->netdev));
1269 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1270 sset_add(&devnames, ofproto_port.name);
1273 SSET_FOR_EACH (devname, &devnames) {
1274 update_port(p, devname);
1276 sset_destroy(&devnames);
1279 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1280 * netdev cannot be opened. On success, also fills in 'opp'. */
1281 static struct netdev *
1282 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
1284 uint32_t curr, advertised, supported, peer;
1285 enum netdev_flags flags;
1286 struct netdev *netdev;
1289 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1291 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1292 "cannot be opened (%s)",
1293 ofproto_port->name, ofproto_port->ofp_port,
1294 ofproto_port->name, strerror(error));
1298 netdev_get_flags(netdev, &flags);
1299 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1301 opp->port_no = htons(ofproto_port->ofp_port);
1302 netdev_get_etheraddr(netdev, opp->hw_addr);
1303 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
1304 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1305 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1306 opp->curr = htonl(curr);
1307 opp->advertised = htonl(advertised);
1308 opp->supported = htonl(supported);
1309 opp->peer = htonl(peer);
1314 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1315 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1318 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1320 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1321 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1322 && a->state == b->state
1323 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1324 && a->curr == b->curr
1325 && a->advertised == b->advertised
1326 && a->supported == b->supported
1327 && a->peer == b->peer);
1330 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1331 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1332 * one with the same name or port number). */
1334 ofport_install(struct ofproto *p,
1335 struct netdev *netdev, const struct ofp_phy_port *opp)
1337 const char *netdev_name = netdev_get_name(netdev);
1338 struct ofport *ofport;
1342 /* Create ofport. */
1343 ofport = p->ofproto_class->port_alloc();
1348 ofport->ofproto = p;
1349 ofport->netdev = netdev;
1350 ofport->change_seq = netdev_change_seq(netdev);
1352 ofport->ofp_port = ntohs(opp->port_no);
1354 /* Add port to 'p'. */
1355 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1356 shash_add(&p->port_by_name, netdev_name, ofport);
1358 if (!netdev_get_mtu(netdev, &dev_mtu)) {
1359 set_internal_devs_mtu(p);
1360 ofport->mtu = dev_mtu;
1365 /* Let the ofproto_class initialize its private data. */
1366 error = p->ofproto_class->port_construct(ofport);
1370 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1374 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1375 p->name, netdev_name, strerror(error));
1377 ofport_destroy__(ofport);
1379 netdev_close(netdev);
1383 /* Removes 'ofport' from 'p' and destroys it. */
1385 ofport_remove(struct ofport *ofport)
1387 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1389 ofport_destroy(ofport);
1392 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1395 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1397 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1399 ofport_remove(port);
1403 /* Updates 'port' with new 'opp' description.
1405 * Does not handle a name or port number change. The caller must implement
1406 * such a change as a delete followed by an add. */
1408 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1410 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1411 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1412 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1413 port->opp.state = opp->state;
1414 port->opp.curr = opp->curr;
1415 port->opp.advertised = opp->advertised;
1416 port->opp.supported = opp->supported;
1417 port->opp.peer = opp->peer;
1419 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1422 /* Update OpenFlow 'state' in 'port' and notify controller. */
1424 ofproto_port_set_state(struct ofport *port, ovs_be32 state)
1426 if (port->opp.state != state) {
1427 port->opp.state = state;
1428 connmgr_send_port_status(port->ofproto->connmgr, &port->opp,
1434 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1436 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1438 if (port->ofproto->ofproto_class->set_realdev) {
1439 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1441 if (port->ofproto->ofproto_class->set_stp_port) {
1442 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1444 if (port->ofproto->ofproto_class->set_cfm) {
1445 port->ofproto->ofproto_class->set_cfm(port, NULL);
1447 if (port->ofproto->ofproto_class->bundle_remove) {
1448 port->ofproto->ofproto_class->bundle_remove(port);
1454 ofport_destroy__(struct ofport *port)
1456 struct ofproto *ofproto = port->ofproto;
1457 const char *name = netdev_get_name(port->netdev);
1459 hmap_remove(&ofproto->ports, &port->hmap_node);
1460 shash_delete(&ofproto->port_by_name,
1461 shash_find(&ofproto->port_by_name, name));
1463 netdev_close(port->netdev);
1464 ofproto->ofproto_class->port_dealloc(port);
1468 ofport_destroy(struct ofport *port)
1471 port->ofproto->ofproto_class->port_destruct(port);
1472 ofport_destroy__(port);
1477 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1479 struct ofport *port;
1481 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1482 hash_int(ofp_port, 0), &ofproto->ports) {
1483 if (port->ofp_port == ofp_port) {
1491 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1493 struct ofproto *ofproto = port->ofproto;
1496 if (ofproto->ofproto_class->port_get_stats) {
1497 error = ofproto->ofproto_class->port_get_stats(port, stats);
1506 update_port(struct ofproto *ofproto, const char *name)
1508 struct ofproto_port ofproto_port;
1509 struct ofp_phy_port opp;
1510 struct netdev *netdev;
1511 struct ofport *port;
1513 COVERAGE_INC(ofproto_update_port);
1515 /* Fetch 'name''s location and properties from the datapath. */
1516 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1517 ? ofport_open(&ofproto_port, &opp)
1520 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1521 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1522 struct netdev *old_netdev = port->netdev;
1525 /* 'name' hasn't changed location. Any properties changed? */
1526 if (!ofport_equal(&port->opp, &opp)) {
1527 ofport_modified(port, &opp);
1530 /* If this is a non-internal port and the MTU changed, check
1531 * if the datapath's MTU needs to be updated. */
1532 if (strcmp(netdev_get_type(netdev), "internal")
1533 && !netdev_get_mtu(netdev, &dev_mtu)
1534 && port->mtu != dev_mtu) {
1535 set_internal_devs_mtu(ofproto);
1536 port->mtu = dev_mtu;
1539 /* Install the newly opened netdev in case it has changed.
1540 * Don't close the old netdev yet in case port_modified has to
1541 * remove a retained reference to it.*/
1542 port->netdev = netdev;
1543 port->change_seq = netdev_change_seq(netdev);
1545 if (port->ofproto->ofproto_class->port_modified) {
1546 port->ofproto->ofproto_class->port_modified(port);
1549 netdev_close(old_netdev);
1551 /* If 'port' is nonnull then its name differs from 'name' and thus
1552 * we should delete it. If we think there's a port named 'name'
1553 * then its port number must be wrong now so delete it too. */
1555 ofport_remove(port);
1557 ofport_remove_with_name(ofproto, name);
1558 ofport_install(ofproto, netdev, &opp);
1561 /* Any port named 'name' is gone now. */
1562 ofport_remove_with_name(ofproto, name);
1564 ofproto_port_destroy(&ofproto_port);
1568 init_ports(struct ofproto *p)
1570 struct ofproto_port_dump dump;
1571 struct ofproto_port ofproto_port;
1573 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1574 uint16_t ofp_port = ofproto_port.ofp_port;
1575 if (ofproto_get_port(p, ofp_port)) {
1576 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1578 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1579 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1582 struct ofp_phy_port opp;
1583 struct netdev *netdev;
1585 netdev = ofport_open(&ofproto_port, &opp);
1587 ofport_install(p, netdev, &opp);
1595 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1596 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1598 find_min_mtu(struct ofproto *p)
1600 struct ofport *ofport;
1603 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1604 struct netdev *netdev = ofport->netdev;
1607 /* Skip any internal ports, since that's what we're trying to
1609 if (!strcmp(netdev_get_type(netdev), "internal")) {
1613 if (netdev_get_mtu(netdev, &dev_mtu)) {
1616 if (!mtu || dev_mtu < mtu) {
1621 return mtu ? mtu: ETH_PAYLOAD_MAX;
1624 /* Set the MTU of all datapath devices on 'p' to the minimum of the
1625 * non-datapath ports. */
1627 set_internal_devs_mtu(struct ofproto *p)
1629 struct ofport *ofport;
1630 int mtu = find_min_mtu(p);
1632 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1633 struct netdev *netdev = ofport->netdev;
1635 if (!strcmp(netdev_get_type(netdev), "internal")) {
1636 netdev_set_mtu(netdev, mtu);
1642 ofproto_rule_destroy__(struct rule *rule)
1644 free(rule->actions);
1645 rule->ofproto->ofproto_class->rule_dealloc(rule);
1648 /* This function allows an ofproto implementation to destroy any rules that
1649 * remain when its ->destruct() function is called. The caller must have
1650 * already uninitialized any derived members of 'rule' (step 5 described in the
1651 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1652 * This function implements steps 6 and 7.
1654 * This function should only be called from an ofproto implementation's
1655 * ->destruct() function. It is not suitable elsewhere. */
1657 ofproto_rule_destroy(struct rule *rule)
1659 assert(!rule->pending);
1660 oftable_remove_rule(rule);
1661 ofproto_rule_destroy__(rule);
1664 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1665 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1668 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1670 const union ofp_action *oa;
1673 if (out_port == OFPP_NONE) {
1676 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1677 if (action_outputs_to_port(oa, htons(out_port))) {
1684 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1685 * statistics appropriately. 'packet' must have at least sizeof(struct
1686 * ofp_packet_in) bytes of headroom.
1688 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1689 * with statistics for 'packet' either way.
1691 * Takes ownership of 'packet'. */
1693 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1697 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1699 flow_extract(packet, 0, 0, in_port, &flow);
1700 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1703 /* Returns true if 'rule' should be hidden from the controller.
1705 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1706 * (e.g. by in-band control) and are intentionally hidden from the
1709 rule_is_hidden(const struct rule *rule)
1711 return rule->cr.priority > UINT16_MAX;
1715 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1717 ofconn_send_reply(ofconn, make_echo_reply(oh));
1722 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1724 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1725 struct ofp_switch_features *osf;
1727 struct ofport *port;
1731 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1732 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1734 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1735 osf->datapath_id = htonll(ofproto->datapath_id);
1736 osf->n_buffers = htonl(pktbuf_capacity());
1737 osf->n_tables = ofproto->n_tables;
1738 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1739 OFPC_PORT_STATS | OFPC_QUEUE_STATS);
1741 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1743 osf->actions = htonl(actions);
1745 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1746 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1749 ofconn_send_reply(ofconn, buf);
1754 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1756 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1757 struct ofp_switch_config *osc;
1758 enum ofp_config_flags flags;
1762 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1763 flags = ofproto->frag_handling;
1764 if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
1765 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
1767 osc->flags = htons(flags);
1768 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1769 ofconn_send_reply(ofconn, buf);
1775 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1777 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1778 uint16_t flags = ntohs(osc->flags);
1780 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
1781 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1782 enum ofp_config_flags cur = ofproto->frag_handling;
1783 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
1785 assert((cur & OFPC_FRAG_MASK) == cur);
1787 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
1788 ofproto->frag_handling = next;
1790 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
1792 ofputil_frag_handling_to_string(next));
1796 ofconn_set_invalid_ttl_to_controller(ofconn,
1797 (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
1799 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1804 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1805 * error message code for the caller to propagate upward. Otherwise, returns
1808 * The log message mentions 'msg_type'. */
1810 reject_slave_controller(struct ofconn *ofconn)
1812 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1813 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1814 return OFPERR_OFPBRC_EPERM;
1821 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1823 struct ofproto *p = ofconn_get_ofproto(ofconn);
1824 struct ofp_packet_out *opo;
1825 struct ofpbuf payload, *buffer;
1826 union ofp_action *ofp_actions;
1827 struct ofpbuf request;
1829 size_t n_ofp_actions;
1833 COVERAGE_INC(ofproto_packet_out);
1835 error = reject_slave_controller(ofconn);
1840 /* Get ofp_packet_out. */
1841 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1842 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1845 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1846 &ofp_actions, &n_ofp_actions);
1852 if (opo->buffer_id != htonl(UINT32_MAX)) {
1853 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1855 if (error || !buffer) {
1864 /* Get in_port and partially validate it.
1866 * We don't know what range of ports the ofproto actually implements, but
1867 * we do know that only certain reserved ports (numbered OFPP_MAX and
1868 * above) are valid. */
1869 in_port = ntohs(opo->in_port);
1870 if (in_port >= OFPP_MAX && in_port != OFPP_LOCAL && in_port != OFPP_NONE) {
1871 return OFPERR_NXBRC_BAD_IN_PORT;
1874 /* Send out packet. */
1875 flow_extract(&payload, 0, 0, in_port, &flow);
1876 error = p->ofproto_class->packet_out(p, &payload, &flow,
1877 ofp_actions, n_ofp_actions);
1878 ofpbuf_delete(buffer);
1884 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1886 ovs_be32 old_config = port->opp.config;
1888 mask &= config ^ port->opp.config;
1889 if (mask & htonl(OFPPC_PORT_DOWN)) {
1890 if (config & htonl(OFPPC_PORT_DOWN)) {
1891 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1893 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1897 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1898 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1899 OFPPC_NO_PACKET_IN));
1900 if (port->opp.config != old_config) {
1901 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1906 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1908 struct ofproto *p = ofconn_get_ofproto(ofconn);
1909 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1910 struct ofport *port;
1913 error = reject_slave_controller(ofconn);
1918 port = ofproto_get_port(p, ntohs(opm->port_no));
1920 return OFPERR_OFPPMFC_BAD_PORT;
1921 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1922 return OFPERR_OFPPMFC_BAD_HW_ADDR;
1924 update_port_config(port, opm->config, opm->mask);
1925 if (opm->advertise) {
1926 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1933 handle_desc_stats_request(struct ofconn *ofconn,
1934 const struct ofp_stats_msg *request)
1936 struct ofproto *p = ofconn_get_ofproto(ofconn);
1937 struct ofp_desc_stats *ods;
1940 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
1941 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1942 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1943 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1944 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1945 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1946 ofconn_send_reply(ofconn, msg);
1952 handle_table_stats_request(struct ofconn *ofconn,
1953 const struct ofp_stats_msg *request)
1955 struct ofproto *p = ofconn_get_ofproto(ofconn);
1956 struct ofp_table_stats *ots;
1960 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
1962 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1963 for (i = 0; i < p->n_tables; i++) {
1964 ots[i].table_id = i;
1965 sprintf(ots[i].name, "table%zu", i);
1966 ots[i].wildcards = htonl(OFPFW_ALL);
1967 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1968 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
1971 p->ofproto_class->get_tables(p, ots);
1973 ofconn_send_reply(ofconn, msg);
1978 append_port_stat(struct ofport *port, struct list *replies)
1980 struct netdev_stats stats;
1981 struct ofp_port_stats *ops;
1983 /* Intentionally ignore return value, since errors will set
1984 * 'stats' to all-1s, which is correct for OpenFlow, and
1985 * netdev_get_stats() will log errors. */
1986 ofproto_port_get_stats(port, &stats);
1988 ops = ofputil_append_stats_reply(sizeof *ops, replies);
1989 ops->port_no = port->opp.port_no;
1990 memset(ops->pad, 0, sizeof ops->pad);
1991 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1992 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1993 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1994 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1995 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1996 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1997 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1998 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1999 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2000 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2001 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2002 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2006 handle_port_stats_request(struct ofconn *ofconn,
2007 const struct ofp_port_stats_request *psr)
2009 struct ofproto *p = ofconn_get_ofproto(ofconn);
2010 struct ofport *port;
2011 struct list replies;
2013 ofputil_start_stats_reply(&psr->osm, &replies);
2014 if (psr->port_no != htons(OFPP_NONE)) {
2015 port = ofproto_get_port(p, ntohs(psr->port_no));
2017 append_port_stat(port, &replies);
2020 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2021 append_port_stat(port, &replies);
2025 ofconn_send_replies(ofconn, &replies);
2030 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2032 long long int msecs = time_msec() - start;
2033 *sec = msecs / 1000;
2034 *nsec = (msecs % 1000) * (1000 * 1000);
2037 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2038 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2040 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2042 return (table_id == 0xff || table_id < ofproto->n_tables
2044 : OFPERR_NXBRC_BAD_TABLE_ID);
2048 static struct oftable *
2049 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
2051 if (table_id == 0xff) {
2052 return &ofproto->tables[0];
2053 } else if (table_id < ofproto->n_tables) {
2054 return &ofproto->tables[table_id];
2060 static struct oftable *
2061 next_matching_table(struct ofproto *ofproto,
2062 struct oftable *table, uint8_t table_id)
2064 return (table_id == 0xff && table < &ofproto->tables[ofproto->n_tables - 1]
2069 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2071 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2074 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2075 * only once, for that table.
2077 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2078 * entered at all. (Perhaps you should have validated TABLE_ID with
2079 * check_table_id().)
2081 * All parameters are evaluated multiple times.
2083 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2084 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2086 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2088 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2089 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2090 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2093 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2094 * to 'out_port' are included.
2096 * Hidden rules are always omitted.
2098 * Returns 0 on success, otherwise an OpenFlow error code. */
2100 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2101 const struct cls_rule *match,
2102 ovs_be64 cookie, ovs_be64 cookie_mask,
2103 uint16_t out_port, struct list *rules)
2105 struct oftable *table;
2108 error = check_table_id(ofproto, table_id);
2114 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2115 struct cls_cursor cursor;
2118 cls_cursor_init(&cursor, &table->cls, match);
2119 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2120 if (rule->pending) {
2121 return OFPROTO_POSTPONE;
2123 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2124 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2125 list_push_back(rules, &rule->ofproto_node);
2132 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2133 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2134 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2137 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2138 * to 'out_port' are included.
2140 * Hidden rules are always omitted.
2142 * Returns 0 on success, otherwise an OpenFlow error code. */
2144 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2145 const struct cls_rule *match,
2146 ovs_be64 cookie, ovs_be64 cookie_mask,
2147 uint16_t out_port, struct list *rules)
2149 struct oftable *table;
2152 error = check_table_id(ofproto, table_id);
2158 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2161 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2164 if (rule->pending) {
2165 return OFPROTO_POSTPONE;
2167 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2168 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2169 list_push_back(rules, &rule->ofproto_node);
2177 handle_flow_stats_request(struct ofconn *ofconn,
2178 const struct ofp_stats_msg *osm)
2180 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2181 struct ofputil_flow_stats_request fsr;
2182 struct list replies;
2187 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
2192 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2193 fsr.cookie, fsr.cookie_mask,
2194 fsr.out_port, &rules);
2199 ofputil_start_stats_reply(osm, &replies);
2200 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2201 struct ofputil_flow_stats fs;
2204 fs.cookie = rule->flow_cookie;
2205 fs.table_id = rule->table_id;
2206 calc_flow_duration__(rule->created, &fs.duration_sec,
2208 fs.idle_timeout = rule->idle_timeout;
2209 fs.hard_timeout = rule->hard_timeout;
2210 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2212 fs.actions = rule->actions;
2213 fs.n_actions = rule->n_actions;
2214 ofputil_append_flow_stats_reply(&fs, &replies);
2216 ofconn_send_replies(ofconn, &replies);
2222 flow_stats_ds(struct rule *rule, struct ds *results)
2224 uint64_t packet_count, byte_count;
2226 rule->ofproto->ofproto_class->rule_get_stats(rule,
2227 &packet_count, &byte_count);
2229 if (rule->table_id != 0) {
2230 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2232 ds_put_format(results, "duration=%llds, ",
2233 (time_msec() - rule->created) / 1000);
2234 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2235 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2236 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2237 cls_rule_format(&rule->cr, results);
2238 ds_put_char(results, ',');
2239 if (rule->n_actions > 0) {
2240 ofp_print_actions(results, rule->actions, rule->n_actions);
2242 ds_put_cstr(results, "drop");
2244 ds_put_cstr(results, "\n");
2247 /* Adds a pretty-printed description of all flows to 'results', including
2248 * hidden flows (e.g., set up by in-band control). */
2250 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2252 struct oftable *table;
2254 OFPROTO_FOR_EACH_TABLE (table, p) {
2255 struct cls_cursor cursor;
2258 cls_cursor_init(&cursor, &table->cls, NULL);
2259 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2260 flow_stats_ds(rule, results);
2265 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2266 * '*engine_type' and '*engine_id', respectively. */
2268 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2269 uint8_t *engine_type, uint8_t *engine_id)
2271 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2274 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
2275 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
2276 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
2278 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2280 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2281 return (ofport && ofproto->ofproto_class->get_cfm_fault
2282 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2286 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2287 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2288 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2289 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2291 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2292 uint16_t ofp_port, const uint64_t **rmps,
2295 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2299 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2300 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2306 handle_aggregate_stats_request(struct ofconn *ofconn,
2307 const struct ofp_stats_msg *osm)
2309 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2310 struct ofputil_flow_stats_request request;
2311 struct ofputil_aggregate_stats stats;
2312 bool unknown_packets, unknown_bytes;
2313 struct ofpbuf *reply;
2318 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2323 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2324 request.cookie, request.cookie_mask,
2325 request.out_port, &rules);
2330 memset(&stats, 0, sizeof stats);
2331 unknown_packets = unknown_bytes = false;
2332 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2333 uint64_t packet_count;
2334 uint64_t byte_count;
2336 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2339 if (packet_count == UINT64_MAX) {
2340 unknown_packets = true;
2342 stats.packet_count += packet_count;
2345 if (byte_count == UINT64_MAX) {
2346 unknown_bytes = true;
2348 stats.byte_count += byte_count;
2353 if (unknown_packets) {
2354 stats.packet_count = UINT64_MAX;
2356 if (unknown_bytes) {
2357 stats.byte_count = UINT64_MAX;
2360 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2361 ofconn_send_reply(ofconn, reply);
2366 struct queue_stats_cbdata {
2367 struct ofport *ofport;
2368 struct list replies;
2372 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2373 const struct netdev_queue_stats *stats)
2375 struct ofp_queue_stats *reply;
2377 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2378 reply->port_no = cbdata->ofport->opp.port_no;
2379 memset(reply->pad, 0, sizeof reply->pad);
2380 reply->queue_id = htonl(queue_id);
2381 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2382 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2383 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2387 handle_queue_stats_dump_cb(uint32_t queue_id,
2388 struct netdev_queue_stats *stats,
2391 struct queue_stats_cbdata *cbdata = cbdata_;
2393 put_queue_stats(cbdata, queue_id, stats);
2397 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2398 struct queue_stats_cbdata *cbdata)
2400 cbdata->ofport = port;
2401 if (queue_id == OFPQ_ALL) {
2402 netdev_dump_queue_stats(port->netdev,
2403 handle_queue_stats_dump_cb, cbdata);
2405 struct netdev_queue_stats stats;
2407 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2408 put_queue_stats(cbdata, queue_id, &stats);
2414 handle_queue_stats_request(struct ofconn *ofconn,
2415 const struct ofp_queue_stats_request *qsr)
2417 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2418 struct queue_stats_cbdata cbdata;
2419 struct ofport *port;
2420 unsigned int port_no;
2423 COVERAGE_INC(ofproto_queue_req);
2425 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2427 port_no = ntohs(qsr->port_no);
2428 queue_id = ntohl(qsr->queue_id);
2429 if (port_no == OFPP_ALL) {
2430 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2431 handle_queue_stats_for_port(port, queue_id, &cbdata);
2433 } else if (port_no < OFPP_MAX) {
2434 port = ofproto_get_port(ofproto, port_no);
2436 handle_queue_stats_for_port(port, queue_id, &cbdata);
2439 ofpbuf_list_delete(&cbdata.replies);
2440 return OFPERR_OFPQOFC_BAD_PORT;
2442 ofconn_send_replies(ofconn, &cbdata.replies);
2448 is_flow_deletion_pending(const struct ofproto *ofproto,
2449 const struct cls_rule *cls_rule,
2452 if (!hmap_is_empty(&ofproto->deletions)) {
2453 struct ofoperation *op;
2455 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2456 cls_rule_hash(cls_rule, table_id),
2457 &ofproto->deletions) {
2458 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2467 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2468 * in which no matching flow already exists in the flow table.
2470 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2471 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2472 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2473 * initiated now but may be retried later.
2475 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2478 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2479 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2481 struct oftable *table;
2482 struct ofopgroup *group;
2483 struct rule *victim;
2487 error = check_table_id(ofproto, fm->table_id);
2493 if (fm->table_id == 0xff) {
2495 if (ofproto->ofproto_class->rule_choose_table) {
2496 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2501 assert(table_id < ofproto->n_tables);
2502 table = &ofproto->tables[table_id];
2504 table = &ofproto->tables[0];
2506 } else if (fm->table_id < ofproto->n_tables) {
2507 table = &ofproto->tables[fm->table_id];
2509 return OFPERR_NXFMFC_BAD_TABLE_ID;
2512 /* Check for overlap, if requested. */
2513 if (fm->flags & OFPFF_CHECK_OVERLAP
2514 && classifier_rule_overlaps(&table->cls, &fm->cr)) {
2515 return OFPERR_OFPFMFC_OVERLAP;
2518 /* Serialize against pending deletion. */
2519 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2520 return OFPROTO_POSTPONE;
2523 /* Allocate new rule. */
2524 rule = ofproto->ofproto_class->rule_alloc();
2526 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2527 ofproto->name, strerror(error));
2530 rule->ofproto = ofproto;
2532 rule->pending = NULL;
2533 rule->flow_cookie = fm->cookie;
2534 rule->created = rule->modified = time_msec();
2535 rule->idle_timeout = fm->idle_timeout;
2536 rule->hard_timeout = fm->hard_timeout;
2537 rule->table_id = table - ofproto->tables;
2538 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2539 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2540 rule->n_actions = fm->n_actions;
2542 /* Insert new rule. */
2543 victim = oftable_replace_rule(rule);
2544 if (victim && victim->pending) {
2545 error = OFPROTO_POSTPONE;
2547 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2548 ofoperation_create(group, rule, OFOPERATION_ADD);
2549 rule->pending->victim = victim;
2551 error = ofproto->ofproto_class->rule_construct(rule);
2553 ofoperation_destroy(rule->pending);
2555 ofopgroup_submit(group);
2558 /* Back out if an error occurred. */
2560 oftable_substitute_rule(rule, victim);
2561 ofproto_rule_destroy__(rule);
2566 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2568 /* Modifies the rules listed in 'rules', changing their actions to match those
2571 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2574 * Returns 0 on success, otherwise an OpenFlow error code. */
2576 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2577 const struct ofputil_flow_mod *fm,
2578 const struct ofp_header *request, struct list *rules)
2580 struct ofopgroup *group;
2583 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2584 LIST_FOR_EACH (rule, ofproto_node, rules) {
2585 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2586 rule->actions, rule->n_actions)) {
2587 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2588 rule->pending->actions = rule->actions;
2589 rule->pending->n_actions = rule->n_actions;
2590 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2591 rule->n_actions = fm->n_actions;
2592 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2594 rule->modified = time_msec();
2596 rule->flow_cookie = fm->cookie;
2598 ofopgroup_submit(group);
2603 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
2606 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2609 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2610 const struct ofputil_flow_mod *fm,
2611 const struct ofp_header *request)
2616 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
2617 fm->cookie, fm->cookie_mask,
2619 return (error ? error
2620 : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
2621 : modify_flows__(ofproto, ofconn, fm, request, &rules));
2624 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2627 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2630 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2631 const struct ofputil_flow_mod *fm,
2632 const struct ofp_header *request)
2637 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
2638 fm->cookie, fm->cookie_mask,
2640 return (error ? error
2641 : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
2642 : list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
2643 fm, request, &rules)
2647 /* OFPFC_DELETE implementation. */
2649 /* Deletes the rules listed in 'rules'.
2651 * Returns 0 on success, otherwise an OpenFlow error code. */
2653 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2654 const struct ofp_header *request, struct list *rules)
2656 struct rule *rule, *next;
2657 struct ofopgroup *group;
2659 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
2660 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2661 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2663 ofoperation_create(group, rule, OFOPERATION_DELETE);
2664 oftable_remove_rule(rule);
2665 rule->ofproto->ofproto_class->rule_destruct(rule);
2667 ofopgroup_submit(group);
2672 /* Implements OFPFC_DELETE. */
2674 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2675 const struct ofputil_flow_mod *fm,
2676 const struct ofp_header *request)
2681 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
2682 fm->cookie, fm->cookie_mask,
2683 fm->out_port, &rules);
2684 return (error ? error
2685 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
2690 /* Implements OFPFC_DELETE_STRICT. */
2692 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2693 const struct ofputil_flow_mod *fm,
2694 const struct ofp_header *request)
2699 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
2700 fm->cookie, fm->cookie_mask,
2701 fm->out_port, &rules);
2702 return (error ? error
2703 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
2709 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2711 struct ofputil_flow_removed fr;
2713 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2718 fr.cookie = rule->flow_cookie;
2720 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2721 fr.idle_timeout = rule->idle_timeout;
2722 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2725 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2728 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2729 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2732 * ofproto implementation ->run() functions should use this function to expire
2733 * OpenFlow flows. */
2735 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2737 struct ofproto *ofproto = rule->ofproto;
2738 struct ofopgroup *group;
2740 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2742 ofproto_rule_send_removed(rule, reason);
2744 group = ofopgroup_create_unattached(ofproto);
2745 ofoperation_create(group, rule, OFOPERATION_DELETE);
2746 oftable_remove_rule(rule);
2747 rule->ofproto->ofproto_class->rule_destruct(rule);
2748 ofopgroup_submit(group);
2752 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2754 struct ofputil_flow_mod fm;
2757 error = reject_slave_controller(ofconn);
2762 error = ofputil_decode_flow_mod(&fm, oh,
2763 ofconn_get_flow_mod_table_id(ofconn));
2768 /* We do not support the emergency flow cache. It will hopefully get
2769 * dropped from OpenFlow in the near future. */
2770 if (fm.flags & OFPFF_EMERG) {
2771 /* There isn't a good fit for an error code, so just state that the
2772 * flow table is full. */
2773 return OFPERR_OFPFMFC_ALL_TABLES_FULL;
2776 return handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
2780 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
2781 const struct ofputil_flow_mod *fm,
2782 const struct ofp_header *oh)
2784 if (ofproto->n_pending >= 50) {
2785 assert(!list_is_empty(&ofproto->pending));
2786 return OFPROTO_POSTPONE;
2789 switch (fm->command) {
2791 return add_flow(ofproto, ofconn, fm, oh);
2794 return modify_flows_loose(ofproto, ofconn, fm, oh);
2796 case OFPFC_MODIFY_STRICT:
2797 return modify_flow_strict(ofproto, ofconn, fm, oh);
2800 return delete_flows_loose(ofproto, ofconn, fm, oh);
2802 case OFPFC_DELETE_STRICT:
2803 return delete_flow_strict(ofproto, ofconn, fm, oh);
2806 if (fm->command > 0xff) {
2807 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2808 "flow_mod_table_id extension is not enabled");
2810 return OFPERR_OFPFMFC_BAD_COMMAND;
2815 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2817 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2818 struct nx_role_request *reply;
2822 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2823 return OFPERR_OFPBRC_EPERM;
2826 role = ntohl(nrr->role);
2827 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2828 && role != NX_ROLE_SLAVE) {
2829 return OFPERR_NXBRC_BAD_ROLE;
2832 if (ofconn_get_role(ofconn) != role
2833 && ofconn_has_pending_opgroups(ofconn)) {
2834 return OFPROTO_POSTPONE;
2837 ofconn_set_role(ofconn, role);
2839 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2840 reply->role = htonl(role);
2841 ofconn_send_reply(ofconn, buf);
2847 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2848 const struct ofp_header *oh)
2850 const struct nx_flow_mod_table_id *msg
2851 = (const struct nx_flow_mod_table_id *) oh;
2853 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2858 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2860 const struct nx_set_flow_format *msg
2861 = (const struct nx_set_flow_format *) oh;
2864 format = ntohl(msg->format);
2865 if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
2866 return OFPERR_OFPBRC_EPERM;
2869 if (format != ofconn_get_flow_format(ofconn)
2870 && ofconn_has_pending_opgroups(ofconn)) {
2871 /* Avoid sending async messages in surprising flow format. */
2872 return OFPROTO_POSTPONE;
2875 ofconn_set_flow_format(ofconn, format);
2880 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
2881 const struct ofp_header *oh)
2883 const struct nx_set_packet_in_format *msg;
2886 msg = (const struct nx_set_packet_in_format *) oh;
2887 format = ntohl(msg->format);
2888 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
2889 return OFPERR_OFPBRC_EPERM;
2892 if (format != ofconn_get_packet_in_format(ofconn)
2893 && ofconn_has_pending_opgroups(ofconn)) {
2894 /* Avoid sending async message in surprsing packet in format. */
2895 return OFPROTO_POSTPONE;
2898 ofconn_set_packet_in_format(ofconn, format);
2903 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2905 struct ofp_header *ob;
2908 if (ofconn_has_pending_opgroups(ofconn)) {
2909 return OFPROTO_POSTPONE;
2912 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2913 ofconn_send_reply(ofconn, buf);
2918 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2920 const struct ofp_header *oh = msg->data;
2921 const struct ofputil_msg_type *type;
2924 error = ofputil_decode_msg_type(oh, &type);
2929 switch (ofputil_msg_type_code(type)) {
2930 /* OpenFlow requests. */
2931 case OFPUTIL_OFPT_ECHO_REQUEST:
2932 return handle_echo_request(ofconn, oh);
2934 case OFPUTIL_OFPT_FEATURES_REQUEST:
2935 return handle_features_request(ofconn, oh);
2937 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2938 return handle_get_config_request(ofconn, oh);
2940 case OFPUTIL_OFPT_SET_CONFIG:
2941 return handle_set_config(ofconn, msg->data);
2943 case OFPUTIL_OFPT_PACKET_OUT:
2944 return handle_packet_out(ofconn, oh);
2946 case OFPUTIL_OFPT_PORT_MOD:
2947 return handle_port_mod(ofconn, oh);
2949 case OFPUTIL_OFPT_FLOW_MOD:
2950 return handle_flow_mod(ofconn, oh);
2952 case OFPUTIL_OFPT_BARRIER_REQUEST:
2953 return handle_barrier_request(ofconn, oh);
2955 /* OpenFlow replies. */
2956 case OFPUTIL_OFPT_ECHO_REPLY:
2959 /* Nicira extension requests. */
2960 case OFPUTIL_NXT_ROLE_REQUEST:
2961 return handle_role_request(ofconn, oh);
2963 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2964 return handle_nxt_flow_mod_table_id(ofconn, oh);
2966 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2967 return handle_nxt_set_flow_format(ofconn, oh);
2969 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
2970 return handle_nxt_set_packet_in_format(ofconn, oh);
2972 case OFPUTIL_NXT_FLOW_MOD:
2973 return handle_flow_mod(ofconn, oh);
2975 /* Statistics requests. */
2976 case OFPUTIL_OFPST_DESC_REQUEST:
2977 return handle_desc_stats_request(ofconn, msg->data);
2979 case OFPUTIL_OFPST_FLOW_REQUEST:
2980 case OFPUTIL_NXST_FLOW_REQUEST:
2981 return handle_flow_stats_request(ofconn, msg->data);
2983 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2984 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2985 return handle_aggregate_stats_request(ofconn, msg->data);
2987 case OFPUTIL_OFPST_TABLE_REQUEST:
2988 return handle_table_stats_request(ofconn, msg->data);
2990 case OFPUTIL_OFPST_PORT_REQUEST:
2991 return handle_port_stats_request(ofconn, msg->data);
2993 case OFPUTIL_OFPST_QUEUE_REQUEST:
2994 return handle_queue_stats_request(ofconn, msg->data);
2996 case OFPUTIL_MSG_INVALID:
2997 case OFPUTIL_OFPT_HELLO:
2998 case OFPUTIL_OFPT_ERROR:
2999 case OFPUTIL_OFPT_FEATURES_REPLY:
3000 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3001 case OFPUTIL_OFPT_PACKET_IN:
3002 case OFPUTIL_OFPT_FLOW_REMOVED:
3003 case OFPUTIL_OFPT_PORT_STATUS:
3004 case OFPUTIL_OFPT_BARRIER_REPLY:
3005 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3006 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3007 case OFPUTIL_OFPST_DESC_REPLY:
3008 case OFPUTIL_OFPST_FLOW_REPLY:
3009 case OFPUTIL_OFPST_QUEUE_REPLY:
3010 case OFPUTIL_OFPST_PORT_REPLY:
3011 case OFPUTIL_OFPST_TABLE_REPLY:
3012 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3013 case OFPUTIL_NXT_ROLE_REPLY:
3014 case OFPUTIL_NXT_FLOW_REMOVED:
3015 case OFPUTIL_NXT_PACKET_IN:
3016 case OFPUTIL_NXST_FLOW_REPLY:
3017 case OFPUTIL_NXST_AGGREGATE_REPLY:
3019 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
3020 return OFPERR_OFPBRC_BAD_STAT;
3022 return OFPERR_OFPBRC_BAD_TYPE;
3028 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3030 int error = handle_openflow__(ofconn, ofp_msg);
3031 if (error && error != OFPROTO_POSTPONE) {
3032 ofconn_send_error(ofconn, ofp_msg->data, error);
3034 COVERAGE_INC(ofproto_recv_openflow);
3035 return error != OFPROTO_POSTPONE;
3038 /* Asynchronous operations. */
3040 /* Creates and returns a new ofopgroup that is not associated with any
3041 * OpenFlow connection.
3043 * The caller should add operations to the returned group with
3044 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3045 static struct ofopgroup *
3046 ofopgroup_create_unattached(struct ofproto *ofproto)
3048 struct ofopgroup *group = xzalloc(sizeof *group);
3049 group->ofproto = ofproto;
3050 list_init(&group->ofproto_node);
3051 list_init(&group->ops);
3052 list_init(&group->ofconn_node);
3056 /* Creates and returns a new ofopgroup for 'ofproto'.
3058 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3059 * connection. The 'request' and 'buffer_id' arguments are ignored.
3061 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3062 * If the ofopgroup eventually fails, then the error reply will include
3063 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3064 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3066 * The caller should add operations to the returned group with
3067 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3068 static struct ofopgroup *
3069 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3070 const struct ofp_header *request, uint32_t buffer_id)
3072 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3074 size_t request_len = ntohs(request->length);
3076 assert(ofconn_get_ofproto(ofconn) == ofproto);
3078 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3079 group->ofconn = ofconn;
3080 group->request = xmemdup(request, MIN(request_len, 64));
3081 group->buffer_id = buffer_id;
3086 /* Submits 'group' for processing.
3088 * If 'group' contains no operations (e.g. none were ever added, or all of the
3089 * ones that were added completed synchronously), then it is destroyed
3090 * immediately. Otherwise it is added to the ofproto's list of pending
3093 ofopgroup_submit(struct ofopgroup *group)
3095 if (list_is_empty(&group->ops)) {
3096 ofopgroup_destroy(group);
3098 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3099 group->ofproto->n_pending++;
3104 ofopgroup_destroy(struct ofopgroup *group)
3106 assert(list_is_empty(&group->ops));
3107 if (!list_is_empty(&group->ofproto_node)) {
3108 assert(group->ofproto->n_pending > 0);
3109 group->ofproto->n_pending--;
3110 list_remove(&group->ofproto_node);
3112 if (!list_is_empty(&group->ofconn_node)) {
3113 list_remove(&group->ofconn_node);
3115 ofconn_send_error(group->ofconn, group->request, group->error);
3117 connmgr_retry(group->ofproto->connmgr);
3119 free(group->request);
3123 /* Initiates a new operation on 'rule', of the specified 'type', within
3124 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
3126 ofoperation_create(struct ofopgroup *group, struct rule *rule,
3127 enum ofoperation_type type)
3129 struct ofoperation *op;
3131 assert(!rule->pending);
3133 op = rule->pending = xzalloc(sizeof *op);
3135 list_push_back(&group->ops, &op->group_node);
3139 op->flow_cookie = rule->flow_cookie;
3141 if (type == OFOPERATION_DELETE) {
3142 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
3143 cls_rule_hash(&rule->cr, rule->table_id));
3148 ofoperation_destroy(struct ofoperation *op)
3150 struct ofopgroup *group = op->group;
3153 op->rule->pending = NULL;
3155 if (op->type == OFOPERATION_DELETE) {
3156 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
3158 list_remove(&op->group_node);
3162 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
3163 ofopgroup_destroy(group);
3167 /* Indicates that 'op' completed with status 'error', which is either 0 to
3168 * indicate success or an OpenFlow error code on failure.
3170 * If 'error' is 0, indicating success, the operation will be committed
3171 * permanently to the flow table. There is one interesting subcase:
3173 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3174 * the flow table (the "victim" rule) by a new one, then the caller must
3175 * have uninitialized any derived state in the victim rule, as in step 5 in
3176 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3177 * performs steps 6 and 7 for the victim rule, most notably by calling its
3178 * ->rule_dealloc() function.
3180 * If 'error' is nonzero, then generally the operation will be rolled back:
3182 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3183 * restores the original rule. The caller must have uninitialized any
3184 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3185 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3186 * and 7 for the new rule, calling its ->rule_dealloc() function.
3188 * - If 'op' is a "modify flow" operation, ofproto restores the original
3191 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3192 * allowed to fail. It must always succeed.
3194 * Please see the large comment in ofproto/ofproto-provider.h titled
3195 * "Asynchronous Operation Support" for more information. */
3197 ofoperation_complete(struct ofoperation *op, enum ofperr error)
3199 struct ofopgroup *group = op->group;
3200 struct rule *rule = op->rule;
3201 struct ofproto *ofproto = rule->ofproto;
3203 assert(rule->pending == op);
3204 assert(op->status < 0);
3208 && op->type != OFOPERATION_DELETE
3210 && group->buffer_id != UINT32_MAX
3211 && list_is_singleton(&op->group_node)) {
3212 struct ofpbuf *packet;
3215 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3219 error = rule_execute(rule, in_port, packet);
3222 if (!group->error) {
3223 group->error = error;
3227 case OFOPERATION_ADD:
3230 ofproto_rule_destroy__(op->victim);
3232 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3233 == htons(VLAN_VID_MASK)) {
3234 if (ofproto->vlan_bitmap) {
3235 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3237 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3238 bitmap_set1(ofproto->vlan_bitmap, vid);
3239 ofproto->vlans_changed = true;
3242 ofproto->vlans_changed = true;
3246 oftable_substitute_rule(rule, op->victim);
3247 ofproto_rule_destroy__(rule);
3252 case OFOPERATION_DELETE:
3254 ofproto_rule_destroy__(rule);
3258 case OFOPERATION_MODIFY:
3260 rule->modified = time_msec();
3262 free(rule->actions);
3263 rule->actions = op->actions;
3264 rule->n_actions = op->n_actions;
3272 ofoperation_destroy(op);
3276 ofoperation_get_victim(struct ofoperation *op)
3278 assert(op->type == OFOPERATION_ADD);
3283 pick_datapath_id(const struct ofproto *ofproto)
3285 const struct ofport *port;
3287 port = ofproto_get_port(ofproto, OFPP_LOCAL);
3289 uint8_t ea[ETH_ADDR_LEN];
3292 error = netdev_get_etheraddr(port->netdev, ea);
3294 return eth_addr_to_uint64(ea);
3296 VLOG_WARN("could not get MAC address for %s (%s)",
3297 netdev_get_name(port->netdev), strerror(error));
3299 return ofproto->fallback_dpid;
3303 pick_fallback_dpid(void)
3305 uint8_t ea[ETH_ADDR_LEN];
3306 eth_addr_nicira_random(ea);
3307 return eth_addr_to_uint64(ea);
3312 /* Initializes 'table'. */
3314 oftable_init(struct oftable *table)
3316 classifier_init(&table->cls);
3319 /* Destroys 'table'.
3321 * The caller is responsible for freeing 'table' itself. */
3323 oftable_destroy(struct oftable *table)
3325 assert(classifier_is_empty(&table->cls));
3326 classifier_destroy(&table->cls);
3329 /* Removes 'rule' from the oftable that contains it. */
3331 oftable_remove_rule(struct rule *rule)
3333 struct ofproto *ofproto = rule->ofproto;
3334 struct oftable *table = &ofproto->tables[rule->table_id];
3336 classifier_remove(&table->cls, &rule->cr);
3339 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
3340 * oftable that has an identical cls_rule. Returns the rule that was removed,
3341 * if any, and otherwise NULL. */
3342 static struct rule *
3343 oftable_replace_rule(struct rule *rule)
3345 struct ofproto *ofproto = rule->ofproto;
3346 struct oftable *table = &ofproto->tables[rule->table_id];
3348 return rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
3351 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
3353 oftable_substitute_rule(struct rule *old, struct rule *new)
3356 oftable_replace_rule(new);
3358 oftable_remove_rule(old);
3362 /* unixctl commands. */
3365 ofproto_lookup(const char *name)
3367 struct ofproto *ofproto;
3369 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
3371 if (!strcmp(ofproto->name, name)) {
3379 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
3380 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
3382 struct ofproto *ofproto;
3386 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
3387 ds_put_format(&results, "%s\n", ofproto->name);
3389 unixctl_command_reply(conn, 200, ds_cstr(&results));
3390 ds_destroy(&results);
3394 ofproto_unixctl_init(void)
3396 static bool registered;
3402 unixctl_command_register("ofproto/list", "", 0, 0,
3403 ofproto_unixctl_list, NULL);
3406 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
3408 * This is deprecated. It is only for compatibility with broken device drivers
3409 * in old versions of Linux that do not properly support VLANs when VLAN
3410 * devices are not used. When broken device drivers are no longer in
3411 * widespread use, we will delete these interfaces. */
3413 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
3414 * (exactly) by an OpenFlow rule in 'ofproto'. */
3416 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
3418 const struct oftable *oftable;
3420 free(ofproto->vlan_bitmap);
3421 ofproto->vlan_bitmap = bitmap_allocate(4096);
3422 ofproto->vlans_changed = false;
3424 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
3425 const struct cls_table *table;
3427 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
3428 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3429 == htons(VLAN_VID_MASK)) {
3430 const struct cls_rule *rule;
3432 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
3433 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
3434 bitmap_set1(vlan_bitmap, vid);
3435 bitmap_set1(ofproto->vlan_bitmap, vid);
3442 /* Returns true if new VLANs have come into use by the flow table since the
3443 * last call to ofproto_get_vlan_usage().
3445 * We don't track when old VLANs stop being used. */
3447 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
3449 return ofproto->vlans_changed;
3452 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
3453 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
3454 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
3455 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
3456 * then the VLAN device is un-enslaved. */
3458 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
3459 uint16_t realdev_ofp_port, int vid)
3461 struct ofport *ofport;
3464 assert(vlandev_ofp_port != realdev_ofp_port);
3466 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
3468 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
3469 ofproto->name, vlandev_ofp_port);
3473 if (!ofproto->ofproto_class->set_realdev) {
3474 if (!vlandev_ofp_port) {
3477 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
3481 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
3483 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
3484 ofproto->name, vlandev_ofp_port,
3485 netdev_get_name(ofport->netdev), strerror(error));