]> git.proxmox.com Git - ovs.git/blob - ofproto/ofproto.c
rule_ofpacts: keep datapath meter_id.
[ovs.git] / ofproto / ofproto.c
1 /*
2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
4 *
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:
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
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.
16 */
17
18 #include <config.h>
19 #include "ofproto.h"
20 #include <errno.h>
21 #include <inttypes.h>
22 #include <stdbool.h>
23 #include <stdlib.h>
24 #include <unistd.h>
25 #include "bitmap.h"
26 #include "byte-order.h"
27 #include "classifier.h"
28 #include "connmgr.h"
29 #include "coverage.h"
30 #include "dynamic-string.h"
31 #include "hash.h"
32 #include "hmap.h"
33 #include "meta-flow.h"
34 #include "netdev.h"
35 #include "nx-match.h"
36 #include "ofp-actions.h"
37 #include "ofp-errors.h"
38 #include "ofp-msgs.h"
39 #include "ofp-print.h"
40 #include "ofp-util.h"
41 #include "ofpbuf.h"
42 #include "ofproto-provider.h"
43 #include "openflow/nicira-ext.h"
44 #include "openflow/openflow.h"
45 #include "packets.h"
46 #include "pinsched.h"
47 #include "pktbuf.h"
48 #include "poll-loop.h"
49 #include "random.h"
50 #include "shash.h"
51 #include "simap.h"
52 #include "sset.h"
53 #include "timeval.h"
54 #include "unaligned.h"
55 #include "unixctl.h"
56 #include "vlog.h"
57
58 VLOG_DEFINE_THIS_MODULE(ofproto);
59
60 COVERAGE_DEFINE(ofproto_error);
61 COVERAGE_DEFINE(ofproto_flush);
62 COVERAGE_DEFINE(ofproto_no_packet_in);
63 COVERAGE_DEFINE(ofproto_packet_out);
64 COVERAGE_DEFINE(ofproto_queue_req);
65 COVERAGE_DEFINE(ofproto_recv_openflow);
66 COVERAGE_DEFINE(ofproto_reinit_ports);
67 COVERAGE_DEFINE(ofproto_uninstallable);
68 COVERAGE_DEFINE(ofproto_update_port);
69
70 enum ofproto_state {
71 S_OPENFLOW, /* Processing OpenFlow commands. */
72 S_EVICT, /* Evicting flows from over-limit tables. */
73 S_FLUSH, /* Deleting all flow table rules. */
74 };
75
76 enum ofoperation_type {
77 OFOPERATION_ADD,
78 OFOPERATION_DELETE,
79 OFOPERATION_MODIFY,
80 OFOPERATION_REPLACE
81 };
82
83 /* A single OpenFlow request can execute any number of operations. The
84 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
85 * ofconn to which an error reply should be sent if necessary.
86 *
87 * ofproto initiates some operations internally. These operations are still
88 * assigned to groups but will not have an associated ofconn. */
89 struct ofopgroup {
90 struct ofproto *ofproto; /* Owning ofproto. */
91 struct list ofproto_node; /* In ofproto's "pending" list. */
92 struct list ops; /* List of "struct ofoperation"s. */
93 int n_running; /* Number of ops still pending. */
94
95 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * packet on success.
97 *
98 * If list_is_empty(ofconn_node) then this ofopgroup never had an
99 * associated ofconn or its ofconn's connection dropped after it initiated
100 * the operation. In the latter case 'ofconn' is a wild pointer that
101 * refers to freed memory, so the 'ofconn' member must be used only if
102 * !list_is_empty(ofconn_node).
103 */
104 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
105 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
106 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
107 uint32_t buffer_id; /* Buffer id from original request. */
108 };
109
110 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
111 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
112 const struct ofp_header *,
113 uint32_t buffer_id);
114 static void ofopgroup_submit(struct ofopgroup *);
115 static void ofopgroup_complete(struct ofopgroup *);
116
117 /* A single flow table operation. */
118 struct ofoperation {
119 struct ofopgroup *group; /* Owning group. */
120 struct list group_node; /* In ofopgroup's "ops" list. */
121 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
122 struct rule *rule; /* Rule being operated upon. */
123 enum ofoperation_type type; /* Type of operation. */
124
125 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
126 * are changing. */
127 struct rule_actions *actions;
128
129 /* OFOPERATION_DELETE. */
130 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
131
132 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
133 uint16_t idle_timeout; /* Rule's old idle timeout. */
134 uint16_t hard_timeout; /* Rule's old hard timeout. */
135 enum ofputil_flow_mod_flags flags; /* Rule's old flags. */
136 enum ofperr error; /* 0 if no error. */
137 };
138
139 static struct ofoperation *ofoperation_create(struct ofopgroup *,
140 struct rule *,
141 enum ofoperation_type,
142 enum ofp_flow_removed_reason);
143 static void ofoperation_destroy(struct ofoperation *);
144
145 /* oftable. */
146 static void oftable_init(struct oftable *);
147 static void oftable_destroy(struct oftable *);
148
149 static void oftable_set_name(struct oftable *, const char *name);
150
151 static void oftable_disable_eviction(struct oftable *);
152 static void oftable_enable_eviction(struct oftable *,
153 const struct mf_subfield *fields,
154 size_t n_fields);
155
156 static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex);
157 static void oftable_remove_rule__(struct ofproto *, struct rule *)
158 OVS_REQUIRES(ofproto_mutex);
159 static void oftable_insert_rule(struct rule *);
160
161 /* A set of rules within a single OpenFlow table (oftable) that have the same
162 * values for the oftable's eviction_fields. A rule to be evicted, when one is
163 * needed, is taken from the eviction group that contains the greatest number
164 * of rules.
165 *
166 * An oftable owns any number of eviction groups, each of which contains any
167 * number of rules.
168 *
169 * Membership in an eviction group is imprecise, based on the hash of the
170 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
171 * That is, if two rules have different eviction_fields, but those
172 * eviction_fields hash to the same value, then they will belong to the same
173 * eviction_group anyway.
174 *
175 * (When eviction is not enabled on an oftable, we don't track any eviction
176 * groups, to save time and space.) */
177 struct eviction_group {
178 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
179 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
180 struct heap rules; /* Contains "struct rule"s. */
181 };
182
183 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep);
184 static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
185 static uint32_t rule_eviction_priority(struct rule *);
186 static void eviction_group_add_rule(struct rule *);
187 static void eviction_group_remove_rule(struct rule *);
188
189 /* Criteria that flow_mod and other operations use for selecting rules on
190 * which to operate. */
191 struct rule_criteria {
192 /* An OpenFlow table or 255 for all tables. */
193 uint8_t table_id;
194
195 /* OpenFlow matching criteria. Interpreted different in "loose" way by
196 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
197 * defined in the OpenFlow spec. */
198 struct cls_rule cr;
199
200 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
201 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
202 * The rule will not be selected if M is 1 and B != C. */
203 ovs_be64 cookie;
204 ovs_be64 cookie_mask;
205
206 /* Selection based on actions within a rule:
207 *
208 * If out_port != OFPP_ANY, selects only rules that output to out_port.
209 * If out_group != OFPG_ALL, select only rules that output to out_group. */
210 ofp_port_t out_port;
211 uint32_t out_group;
212 };
213
214 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
215 const struct match *match,
216 unsigned int priority,
217 ovs_be64 cookie, ovs_be64 cookie_mask,
218 ofp_port_t out_port, uint32_t out_group);
219 static void rule_criteria_destroy(struct rule_criteria *);
220
221 /* A packet that needs to be passed to rule_execute().
222 *
223 * (We can't do this immediately from ofopgroup_complete() because that holds
224 * ofproto_mutex, which rule_execute() needs released.) */
225 struct rule_execute {
226 struct list list_node; /* In struct ofproto's "rule_executes" list. */
227 struct rule *rule; /* Owns a reference to the rule. */
228 ofp_port_t in_port;
229 struct ofpbuf *packet; /* Owns the packet. */
230 };
231
232 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
233 static void destroy_rule_executes(struct ofproto *);
234
235 /* ofport. */
236 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
237 static void ofport_destroy(struct ofport *);
238
239 static void update_port(struct ofproto *, const char *devname);
240 static int init_ports(struct ofproto *);
241 static void reinit_ports(struct ofproto *);
242
243 /* rule. */
244 static void ofproto_rule_destroy__(struct rule *);
245 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
246 static bool rule_is_modifiable(const struct rule *);
247
248 /* OpenFlow. */
249 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
250 struct ofputil_flow_mod *,
251 const struct ofp_header *);
252 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
253 struct ofputil_flow_mod *,
254 const struct ofp_header *,
255 const struct rule_collection *);
256 static void delete_flow__(struct rule *rule, struct ofopgroup *,
257 enum ofp_flow_removed_reason)
258 OVS_REQUIRES(ofproto_mutex);
259 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
260 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
261 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
262 struct ofputil_flow_mod *,
263 const struct ofp_header *)
264 OVS_EXCLUDED(ofproto_mutex);
265 static void calc_duration(long long int start, long long int now,
266 uint32_t *sec, uint32_t *nsec);
267
268 /* ofproto. */
269 static uint64_t pick_datapath_id(const struct ofproto *);
270 static uint64_t pick_fallback_dpid(void);
271 static void ofproto_destroy__(struct ofproto *);
272 static void update_mtu(struct ofproto *, struct ofport *);
273 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
274
275 /* unixctl. */
276 static void ofproto_unixctl_init(void);
277
278 /* All registered ofproto classes, in probe order. */
279 static const struct ofproto_class **ofproto_classes;
280 static size_t n_ofproto_classes;
281 static size_t allocated_ofproto_classes;
282
283 /* Global lock that protects all flow table operations. */
284 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
285
286 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
287 unsigned n_handler_threads;
288 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
289
290 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
291 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
292
293 /* Initial mappings of port to OpenFlow number mappings. */
294 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
295
296 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
297
298 /* The default value of true waits for flow restore. */
299 static bool flow_restore_wait = true;
300
301 /* Must be called to initialize the ofproto library.
302 *
303 * The caller may pass in 'iface_hints', which contains an shash of
304 * "iface_hint" elements indexed by the interface's name. The provider
305 * may use these hints to describe the startup configuration in order to
306 * reinitialize its state. The caller owns the provided data, so a
307 * provider will make copies of anything required. An ofproto provider
308 * will remove any existing state that is not described by the hint, and
309 * may choose to remove it all. */
310 void
311 ofproto_init(const struct shash *iface_hints)
312 {
313 struct shash_node *node;
314 size_t i;
315
316 ofproto_class_register(&ofproto_dpif_class);
317
318 /* Make a local copy, since we don't own 'iface_hints' elements. */
319 SHASH_FOR_EACH(node, iface_hints) {
320 const struct iface_hint *orig_hint = node->data;
321 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
322 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
323
324 new_hint->br_name = xstrdup(orig_hint->br_name);
325 new_hint->br_type = xstrdup(br_type);
326 new_hint->ofp_port = orig_hint->ofp_port;
327
328 shash_add(&init_ofp_ports, node->name, new_hint);
329 }
330
331 for (i = 0; i < n_ofproto_classes; i++) {
332 ofproto_classes[i]->init(&init_ofp_ports);
333 }
334 }
335
336 /* 'type' should be a normalized datapath type, as returned by
337 * ofproto_normalize_type(). Returns the corresponding ofproto_class
338 * structure, or a null pointer if there is none registered for 'type'. */
339 static const struct ofproto_class *
340 ofproto_class_find__(const char *type)
341 {
342 size_t i;
343
344 for (i = 0; i < n_ofproto_classes; i++) {
345 const struct ofproto_class *class = ofproto_classes[i];
346 struct sset types;
347 bool found;
348
349 sset_init(&types);
350 class->enumerate_types(&types);
351 found = sset_contains(&types, type);
352 sset_destroy(&types);
353
354 if (found) {
355 return class;
356 }
357 }
358 VLOG_WARN("unknown datapath type %s", type);
359 return NULL;
360 }
361
362 /* Registers a new ofproto class. After successful registration, new ofprotos
363 * of that type can be created using ofproto_create(). */
364 int
365 ofproto_class_register(const struct ofproto_class *new_class)
366 {
367 size_t i;
368
369 for (i = 0; i < n_ofproto_classes; i++) {
370 if (ofproto_classes[i] == new_class) {
371 return EEXIST;
372 }
373 }
374
375 if (n_ofproto_classes >= allocated_ofproto_classes) {
376 ofproto_classes = x2nrealloc(ofproto_classes,
377 &allocated_ofproto_classes,
378 sizeof *ofproto_classes);
379 }
380 ofproto_classes[n_ofproto_classes++] = new_class;
381 return 0;
382 }
383
384 /* Unregisters a datapath provider. 'type' must have been previously
385 * registered and not currently be in use by any ofprotos. After
386 * unregistration new datapaths of that type cannot be opened using
387 * ofproto_create(). */
388 int
389 ofproto_class_unregister(const struct ofproto_class *class)
390 {
391 size_t i;
392
393 for (i = 0; i < n_ofproto_classes; i++) {
394 if (ofproto_classes[i] == class) {
395 for (i++; i < n_ofproto_classes; i++) {
396 ofproto_classes[i - 1] = ofproto_classes[i];
397 }
398 n_ofproto_classes--;
399 return 0;
400 }
401 }
402 VLOG_WARN("attempted to unregister an ofproto class that is not "
403 "registered");
404 return EAFNOSUPPORT;
405 }
406
407 /* Clears 'types' and enumerates all registered ofproto types into it. The
408 * caller must first initialize the sset. */
409 void
410 ofproto_enumerate_types(struct sset *types)
411 {
412 size_t i;
413
414 for (i = 0; i < n_ofproto_classes; i++) {
415 ofproto_classes[i]->enumerate_types(types);
416 }
417 }
418
419 /* Returns the fully spelled out name for the given ofproto 'type'.
420 *
421 * Normalized type string can be compared with strcmp(). Unnormalized type
422 * string might be the same even if they have different spellings. */
423 const char *
424 ofproto_normalize_type(const char *type)
425 {
426 return type && type[0] ? type : "system";
427 }
428
429 /* Clears 'names' and enumerates the names of all known created ofprotos with
430 * the given 'type'. The caller must first initialize the sset. Returns 0 if
431 * successful, otherwise a positive errno value.
432 *
433 * Some kinds of datapaths might not be practically enumerable. This is not
434 * considered an error. */
435 int
436 ofproto_enumerate_names(const char *type, struct sset *names)
437 {
438 const struct ofproto_class *class = ofproto_class_find__(type);
439 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
440 }
441
442 int
443 ofproto_create(const char *datapath_name, const char *datapath_type,
444 struct ofproto **ofprotop)
445 {
446 const struct ofproto_class *class;
447 struct ofproto *ofproto;
448 int error;
449 int i;
450
451 *ofprotop = NULL;
452
453 ofproto_unixctl_init();
454
455 datapath_type = ofproto_normalize_type(datapath_type);
456 class = ofproto_class_find__(datapath_type);
457 if (!class) {
458 VLOG_WARN("could not create datapath %s of unknown type %s",
459 datapath_name, datapath_type);
460 return EAFNOSUPPORT;
461 }
462
463 ofproto = class->alloc();
464 if (!ofproto) {
465 VLOG_ERR("failed to allocate datapath %s of type %s",
466 datapath_name, datapath_type);
467 return ENOMEM;
468 }
469
470 /* Initialize. */
471 ovs_mutex_lock(&ofproto_mutex);
472 memset(ofproto, 0, sizeof *ofproto);
473 ofproto->ofproto_class = class;
474 ofproto->name = xstrdup(datapath_name);
475 ofproto->type = xstrdup(datapath_type);
476 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
477 hash_string(ofproto->name, 0));
478 ofproto->datapath_id = 0;
479 ofproto->forward_bpdu = false;
480 ofproto->fallback_dpid = pick_fallback_dpid();
481 ofproto->mfr_desc = NULL;
482 ofproto->hw_desc = NULL;
483 ofproto->sw_desc = NULL;
484 ofproto->serial_desc = NULL;
485 ofproto->dp_desc = NULL;
486 ofproto->frag_handling = OFPC_FRAG_NORMAL;
487 hmap_init(&ofproto->ports);
488 shash_init(&ofproto->port_by_name);
489 simap_init(&ofproto->ofp_requests);
490 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
491 ofproto->eviction_group_timer = LLONG_MIN;
492 ofproto->tables = NULL;
493 ofproto->n_tables = 0;
494 hindex_init(&ofproto->cookies);
495 list_init(&ofproto->expirable);
496 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
497 ofproto->state = S_OPENFLOW;
498 list_init(&ofproto->pending);
499 ofproto->n_pending = 0;
500 hmap_init(&ofproto->deletions);
501 guarded_list_init(&ofproto->rule_executes);
502 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
503 ofproto->first_op = ofproto->last_op = LLONG_MIN;
504 ofproto->next_op_report = LLONG_MAX;
505 ofproto->op_backoff = LLONG_MIN;
506 ofproto->vlan_bitmap = NULL;
507 ofproto->vlans_changed = false;
508 ofproto->min_mtu = INT_MAX;
509 ovs_rwlock_init(&ofproto->groups_rwlock);
510 hmap_init(&ofproto->groups);
511 ovs_mutex_unlock(&ofproto_mutex);
512
513 error = ofproto->ofproto_class->construct(ofproto);
514 if (error) {
515 VLOG_ERR("failed to open datapath %s: %s",
516 datapath_name, ovs_strerror(error));
517 ofproto_destroy__(ofproto);
518 return error;
519 }
520
521 /* The "max_ports" member should have been set by ->construct(ofproto).
522 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
523 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
524 bitmap_set1(ofproto->ofp_port_ids, 0);
525
526 /* Check that hidden tables, if any, are at the end. */
527 ovs_assert(ofproto->n_tables);
528 for (i = 0; i + 1 < ofproto->n_tables; i++) {
529 enum oftable_flags flags = ofproto->tables[i].flags;
530 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
531
532 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
533 }
534
535 ofproto->datapath_id = pick_datapath_id(ofproto);
536 init_ports(ofproto);
537
538 /* Initialize meters table. */
539 if (ofproto->ofproto_class->meter_get_features) {
540 ofproto->ofproto_class->meter_get_features(ofproto,
541 &ofproto->meter_features);
542 } else {
543 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
544 }
545 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
546 * sizeof(struct meter *));
547
548 *ofprotop = ofproto;
549 return 0;
550 }
551
552 /* Must be called (only) by an ofproto implementation in its constructor
553 * function. See the large comment on 'construct' in struct ofproto_class for
554 * details. */
555 void
556 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
557 {
558 struct oftable *table;
559
560 ovs_assert(!ofproto->n_tables);
561 ovs_assert(n_tables >= 1 && n_tables <= 255);
562
563 ofproto->n_tables = n_tables;
564 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
565 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
566 oftable_init(table);
567 }
568 }
569
570 /* To be optionally called (only) by an ofproto implementation in its
571 * constructor function. See the large comment on 'construct' in struct
572 * ofproto_class for details.
573 *
574 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
575 * will then ensure that actions passed into the ofproto implementation will
576 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
577 * function is not called, there will be no such restriction.
578 *
579 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
580 * the 'max_ports' restriction. */
581 void
582 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
583 {
584 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
585 ofproto->max_ports = max_ports;
586 }
587
588 uint64_t
589 ofproto_get_datapath_id(const struct ofproto *ofproto)
590 {
591 return ofproto->datapath_id;
592 }
593
594 void
595 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
596 {
597 uint64_t old_dpid = p->datapath_id;
598 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
599 if (p->datapath_id != old_dpid) {
600 /* Force all active connections to reconnect, since there is no way to
601 * notify a controller that the datapath ID has changed. */
602 ofproto_reconnect_controllers(p);
603 }
604 }
605
606 void
607 ofproto_set_controllers(struct ofproto *p,
608 const struct ofproto_controller *controllers,
609 size_t n_controllers, uint32_t allowed_versions)
610 {
611 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
612 allowed_versions);
613 }
614
615 void
616 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
617 {
618 connmgr_set_fail_mode(p->connmgr, fail_mode);
619 }
620
621 /* Drops the connections between 'ofproto' and all of its controllers, forcing
622 * them to reconnect. */
623 void
624 ofproto_reconnect_controllers(struct ofproto *ofproto)
625 {
626 connmgr_reconnect(ofproto->connmgr);
627 }
628
629 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
630 * in-band control should guarantee access, in the same way that in-band
631 * control guarantees access to OpenFlow controllers. */
632 void
633 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
634 const struct sockaddr_in *extras, size_t n)
635 {
636 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
637 }
638
639 /* Sets the OpenFlow queue used by flows set up by in-band control on
640 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
641 * flows will use the default queue. */
642 void
643 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
644 {
645 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
646 }
647
648 /* Sets the number of flows at which eviction from the kernel flow table
649 * will occur. */
650 void
651 ofproto_set_flow_eviction_threshold(unsigned threshold)
652 {
653 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
654 threshold);
655 }
656
657 /* Sets the path for handling flow misses. */
658 void
659 ofproto_set_flow_miss_model(unsigned model)
660 {
661 flow_miss_model = model;
662 }
663
664 /* If forward_bpdu is true, the NORMAL action will forward frames with
665 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
666 * the NORMAL action will drop these frames. */
667 void
668 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
669 {
670 bool old_val = ofproto->forward_bpdu;
671 ofproto->forward_bpdu = forward_bpdu;
672 if (old_val != ofproto->forward_bpdu) {
673 if (ofproto->ofproto_class->forward_bpdu_changed) {
674 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
675 }
676 }
677 }
678
679 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
680 * 'idle_time', in seconds, and the maximum number of MAC table entries to
681 * 'max_entries'. */
682 void
683 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
684 size_t max_entries)
685 {
686 if (ofproto->ofproto_class->set_mac_table_config) {
687 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
688 max_entries);
689 }
690 }
691
692 /* Sets number of upcall handler threads. The default is
693 * (number of online cores - 2). */
694 void
695 ofproto_set_n_handler_threads(unsigned limit)
696 {
697 if (limit) {
698 n_handler_threads = limit;
699 } else {
700 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
701 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
702 }
703 }
704
705 void
706 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
707 {
708 free(p->dp_desc);
709 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
710 }
711
712 int
713 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
714 {
715 return connmgr_set_snoops(ofproto->connmgr, snoops);
716 }
717
718 int
719 ofproto_set_netflow(struct ofproto *ofproto,
720 const struct netflow_options *nf_options)
721 {
722 if (nf_options && sset_is_empty(&nf_options->collectors)) {
723 nf_options = NULL;
724 }
725
726 if (ofproto->ofproto_class->set_netflow) {
727 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
728 } else {
729 return nf_options ? EOPNOTSUPP : 0;
730 }
731 }
732
733 int
734 ofproto_set_sflow(struct ofproto *ofproto,
735 const struct ofproto_sflow_options *oso)
736 {
737 if (oso && sset_is_empty(&oso->targets)) {
738 oso = NULL;
739 }
740
741 if (ofproto->ofproto_class->set_sflow) {
742 return ofproto->ofproto_class->set_sflow(ofproto, oso);
743 } else {
744 return oso ? EOPNOTSUPP : 0;
745 }
746 }
747
748 int
749 ofproto_set_ipfix(struct ofproto *ofproto,
750 const struct ofproto_ipfix_bridge_exporter_options *bo,
751 const struct ofproto_ipfix_flow_exporter_options *fo,
752 size_t n_fo)
753 {
754 if (ofproto->ofproto_class->set_ipfix) {
755 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
756 } else {
757 return (bo || fo) ? EOPNOTSUPP : 0;
758 }
759 }
760
761 void
762 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
763 {
764 flow_restore_wait = flow_restore_wait_db;
765 }
766
767 bool
768 ofproto_get_flow_restore_wait(void)
769 {
770 return flow_restore_wait;
771 }
772
773 \f
774 /* Spanning Tree Protocol (STP) configuration. */
775
776 /* Configures STP on 'ofproto' using the settings defined in 's'. If
777 * 's' is NULL, disables STP.
778 *
779 * Returns 0 if successful, otherwise a positive errno value. */
780 int
781 ofproto_set_stp(struct ofproto *ofproto,
782 const struct ofproto_stp_settings *s)
783 {
784 return (ofproto->ofproto_class->set_stp
785 ? ofproto->ofproto_class->set_stp(ofproto, s)
786 : EOPNOTSUPP);
787 }
788
789 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
790 * 'enabled' member of 's' is false, then the other members are not
791 * meaningful.
792 *
793 * Returns 0 if successful, otherwise a positive errno value. */
794 int
795 ofproto_get_stp_status(struct ofproto *ofproto,
796 struct ofproto_stp_status *s)
797 {
798 return (ofproto->ofproto_class->get_stp_status
799 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
800 : EOPNOTSUPP);
801 }
802
803 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
804 * in 's'. The caller is responsible for assigning STP port numbers
805 * (using the 'port_num' member in the range of 1 through 255, inclusive)
806 * and ensuring there are no duplicates. If the 's' is NULL, then STP
807 * is disabled on the port.
808 *
809 * Returns 0 if successful, otherwise a positive errno value.*/
810 int
811 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
812 const struct ofproto_port_stp_settings *s)
813 {
814 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
815 if (!ofport) {
816 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
817 ofproto->name, ofp_port);
818 return ENODEV;
819 }
820
821 return (ofproto->ofproto_class->set_stp_port
822 ? ofproto->ofproto_class->set_stp_port(ofport, s)
823 : EOPNOTSUPP);
824 }
825
826 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
827 * 's'. If the 'enabled' member in 's' is false, then the other members
828 * are not meaningful.
829 *
830 * Returns 0 if successful, otherwise a positive errno value.*/
831 int
832 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
833 struct ofproto_port_stp_status *s)
834 {
835 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
836 if (!ofport) {
837 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
838 "port %"PRIu16, ofproto->name, ofp_port);
839 return ENODEV;
840 }
841
842 return (ofproto->ofproto_class->get_stp_port_status
843 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
844 : EOPNOTSUPP);
845 }
846 \f
847 /* Queue DSCP configuration. */
848
849 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
850 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
851 * to implement QoS. Instead, it is used to implement features which require
852 * knowledge of what queues exist on a port, and some basic information about
853 * them.
854 *
855 * Returns 0 if successful, otherwise a positive errno value. */
856 int
857 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
858 const struct ofproto_port_queue *queues,
859 size_t n_queues)
860 {
861 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
862
863 if (!ofport) {
864 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
865 ofproto->name, ofp_port);
866 return ENODEV;
867 }
868
869 return (ofproto->ofproto_class->set_queues
870 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
871 : EOPNOTSUPP);
872 }
873 \f
874 /* Connectivity Fault Management configuration. */
875
876 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
877 void
878 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
879 {
880 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
881 if (ofport && ofproto->ofproto_class->set_cfm) {
882 ofproto->ofproto_class->set_cfm(ofport, NULL);
883 }
884 }
885
886 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
887 * basic configuration from the configuration members in 'cfm', and the remote
888 * maintenance point ID from remote_mpid. Ignores the statistics members of
889 * 'cfm'.
890 *
891 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
892 void
893 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
894 const struct cfm_settings *s)
895 {
896 struct ofport *ofport;
897 int error;
898
899 ofport = ofproto_get_port(ofproto, ofp_port);
900 if (!ofport) {
901 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
902 ofproto->name, ofp_port);
903 return;
904 }
905
906 /* XXX: For configuration simplicity, we only support one remote_mpid
907 * outside of the CFM module. It's not clear if this is the correct long
908 * term solution or not. */
909 error = (ofproto->ofproto_class->set_cfm
910 ? ofproto->ofproto_class->set_cfm(ofport, s)
911 : EOPNOTSUPP);
912 if (error) {
913 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
914 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
915 ovs_strerror(error));
916 }
917 }
918
919 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
920 * 'ofproto' does not have a port 'ofp_port'. */
921 void
922 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
923 const struct smap *cfg)
924 {
925 struct ofport *ofport;
926 int error;
927
928 ofport = ofproto_get_port(ofproto, ofp_port);
929 if (!ofport) {
930 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
931 ofproto->name, ofp_port);
932 return;
933 }
934
935 error = (ofproto->ofproto_class->set_bfd
936 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
937 : EOPNOTSUPP);
938 if (error) {
939 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
940 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
941 ovs_strerror(error));
942 }
943 }
944
945 /* Populates 'status' with key value pairs indicating the status of the BFD
946 * session on 'ofp_port'. This information is intended to be populated in the
947 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
948 * 'ofproto'. */
949 int
950 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
951 struct smap *status)
952 {
953 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
954 return (ofport && ofproto->ofproto_class->get_bfd_status
955 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
956 : EOPNOTSUPP);
957 }
958
959 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
960 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
961 * 0 if LACP partner information is not current (generally indicating a
962 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
963 int
964 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
965 {
966 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
967 return (ofport && ofproto->ofproto_class->port_is_lacp_current
968 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
969 : -1);
970 }
971 \f
972 /* Bundles. */
973
974 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
975 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
976 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
977 * configuration plus, if there is more than one slave, a bonding
978 * configuration.
979 *
980 * If 'aux' is already registered then this function updates its configuration
981 * to 's'. Otherwise, this function registers a new bundle.
982 *
983 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
984 * port. */
985 int
986 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
987 const struct ofproto_bundle_settings *s)
988 {
989 return (ofproto->ofproto_class->bundle_set
990 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
991 : EOPNOTSUPP);
992 }
993
994 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
995 * If no such bundle has been registered, this has no effect. */
996 int
997 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
998 {
999 return ofproto_bundle_register(ofproto, aux, NULL);
1000 }
1001
1002 \f
1003 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1004 * If 'aux' is already registered then this function updates its configuration
1005 * to 's'. Otherwise, this function registers a new mirror. */
1006 int
1007 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1008 const struct ofproto_mirror_settings *s)
1009 {
1010 return (ofproto->ofproto_class->mirror_set
1011 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1012 : EOPNOTSUPP);
1013 }
1014
1015 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1016 * If no mirror has been registered, this has no effect. */
1017 int
1018 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1019 {
1020 return ofproto_mirror_register(ofproto, aux, NULL);
1021 }
1022
1023 /* Retrieves statistics from mirror associated with client data pointer
1024 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1025 * 'bytes', respectively. If a particular counters is not supported,
1026 * the appropriate argument is set to UINT64_MAX. */
1027 int
1028 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1029 uint64_t *packets, uint64_t *bytes)
1030 {
1031 if (!ofproto->ofproto_class->mirror_get_stats) {
1032 *packets = *bytes = UINT64_MAX;
1033 return EOPNOTSUPP;
1034 }
1035
1036 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1037 packets, bytes);
1038 }
1039
1040 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1041 * which all packets are flooded, instead of using MAC learning. If
1042 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1043 *
1044 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1045 * port. */
1046 int
1047 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1048 {
1049 return (ofproto->ofproto_class->set_flood_vlans
1050 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1051 : EOPNOTSUPP);
1052 }
1053
1054 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1055 * output for a mirror. */
1056 bool
1057 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1058 {
1059 return (ofproto->ofproto_class->is_mirror_output_bundle
1060 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1061 : false);
1062 }
1063 \f
1064 /* Configuration of OpenFlow tables. */
1065
1066 /* Returns the number of OpenFlow tables in 'ofproto'. */
1067 int
1068 ofproto_get_n_tables(const struct ofproto *ofproto)
1069 {
1070 return ofproto->n_tables;
1071 }
1072
1073 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1074 * settings from 's'. 'table_id' must be in the range 0 through the number of
1075 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1076 *
1077 * For read-only tables, only the name may be configured. */
1078 void
1079 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1080 const struct ofproto_table_settings *s)
1081 {
1082 struct oftable *table;
1083
1084 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1085 table = &ofproto->tables[table_id];
1086
1087 oftable_set_name(table, s->name);
1088
1089 if (table->flags & OFTABLE_READONLY) {
1090 return;
1091 }
1092
1093 if (s->groups) {
1094 oftable_enable_eviction(table, s->groups, s->n_groups);
1095 } else {
1096 oftable_disable_eviction(table);
1097 }
1098
1099 table->max_flows = s->max_flows;
1100 ovs_rwlock_rdlock(&table->cls.rwlock);
1101 if (classifier_count(&table->cls) > table->max_flows
1102 && table->eviction_fields) {
1103 /* 'table' contains more flows than allowed. We might not be able to
1104 * evict them right away because of the asynchronous nature of flow
1105 * table changes. Schedule eviction for later. */
1106 switch (ofproto->state) {
1107 case S_OPENFLOW:
1108 ofproto->state = S_EVICT;
1109 break;
1110 case S_EVICT:
1111 case S_FLUSH:
1112 /* We're already deleting flows, nothing more to do. */
1113 break;
1114 }
1115 }
1116 ovs_rwlock_unlock(&table->cls.rwlock);
1117 }
1118 \f
1119 bool
1120 ofproto_has_snoops(const struct ofproto *ofproto)
1121 {
1122 return connmgr_has_snoops(ofproto->connmgr);
1123 }
1124
1125 void
1126 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1127 {
1128 connmgr_get_snoops(ofproto->connmgr, snoops);
1129 }
1130
1131 static void
1132 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule)
1133 OVS_REQUIRES(ofproto_mutex)
1134 {
1135 struct ofopgroup *group;
1136
1137 ovs_assert(!rule->pending);
1138
1139 group = ofopgroup_create_unattached(ofproto);
1140 delete_flow__(rule, group, OFPRR_DELETE);
1141 ofopgroup_submit(group);
1142 }
1143
1144 /* Deletes 'rule' from 'cls' within 'ofproto'.
1145 *
1146 * Within an ofproto implementation, this function allows an ofproto
1147 * implementation to destroy any rules that remain when its ->destruct()
1148 * function is called. This function is not suitable for use elsewhere in an
1149 * ofproto implementation.
1150 *
1151 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1152 * Cycle" in ofproto-provider.h. */
1153 void
1154 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1155 OVS_EXCLUDED(ofproto_mutex)
1156 {
1157 struct ofopgroup *group;
1158
1159 ovs_mutex_lock(&ofproto_mutex);
1160 ovs_assert(!rule->pending);
1161
1162 group = ofopgroup_create_unattached(ofproto);
1163 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1164 oftable_remove_rule__(ofproto, rule);
1165 ofproto->ofproto_class->rule_delete(rule);
1166 ofopgroup_submit(group);
1167
1168 ovs_mutex_unlock(&ofproto_mutex);
1169 }
1170
1171 static void
1172 ofproto_flush__(struct ofproto *ofproto)
1173 OVS_EXCLUDED(ofproto_mutex)
1174 {
1175 struct oftable *table;
1176
1177 if (ofproto->ofproto_class->flush) {
1178 ofproto->ofproto_class->flush(ofproto);
1179 }
1180
1181 ovs_mutex_lock(&ofproto_mutex);
1182 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1183 struct rule *rule, *next_rule;
1184 struct cls_cursor cursor;
1185
1186 if (table->flags & OFTABLE_HIDDEN) {
1187 continue;
1188 }
1189
1190 ovs_rwlock_rdlock(&table->cls.rwlock);
1191 cls_cursor_init(&cursor, &table->cls, NULL);
1192 ovs_rwlock_unlock(&table->cls.rwlock);
1193 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1194 if (!rule->pending) {
1195 ofproto_rule_delete__(ofproto, rule);
1196 }
1197 }
1198 }
1199 ovs_mutex_unlock(&ofproto_mutex);
1200 }
1201
1202 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1203
1204 static void
1205 ofproto_destroy__(struct ofproto *ofproto)
1206 OVS_EXCLUDED(ofproto_mutex)
1207 {
1208 struct oftable *table;
1209
1210 ovs_assert(list_is_empty(&ofproto->pending));
1211
1212 destroy_rule_executes(ofproto);
1213 guarded_list_destroy(&ofproto->rule_executes);
1214
1215 delete_group(ofproto, OFPG_ALL);
1216 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1217 hmap_destroy(&ofproto->groups);
1218
1219 connmgr_destroy(ofproto->connmgr);
1220
1221 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1222 free(ofproto->name);
1223 free(ofproto->type);
1224 free(ofproto->mfr_desc);
1225 free(ofproto->hw_desc);
1226 free(ofproto->sw_desc);
1227 free(ofproto->serial_desc);
1228 free(ofproto->dp_desc);
1229 hmap_destroy(&ofproto->ports);
1230 shash_destroy(&ofproto->port_by_name);
1231 bitmap_free(ofproto->ofp_port_ids);
1232 simap_destroy(&ofproto->ofp_requests);
1233
1234 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1235 oftable_destroy(table);
1236 }
1237 free(ofproto->tables);
1238
1239 hmap_destroy(&ofproto->deletions);
1240
1241 free(ofproto->vlan_bitmap);
1242
1243 ofproto->ofproto_class->dealloc(ofproto);
1244 }
1245
1246 void
1247 ofproto_destroy(struct ofproto *p)
1248 OVS_EXCLUDED(ofproto_mutex)
1249 {
1250 struct ofport *ofport, *next_ofport;
1251
1252 if (!p) {
1253 return;
1254 }
1255
1256 if (p->meters) {
1257 meter_delete(p, 1, p->meter_features.max_meters);
1258 p->meter_features.max_meters = 0;
1259 free(p->meters);
1260 p->meters = NULL;
1261 }
1262
1263 ofproto_flush__(p);
1264 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1265 ofport_destroy(ofport);
1266 }
1267
1268 p->ofproto_class->destruct(p);
1269 ofproto_destroy__(p);
1270 }
1271
1272 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1273 * kernel datapath, for example, this destroys the datapath in the kernel, and
1274 * with the netdev-based datapath, it tears down the data structures that
1275 * represent the datapath.
1276 *
1277 * The datapath should not be currently open as an ofproto. */
1278 int
1279 ofproto_delete(const char *name, const char *type)
1280 {
1281 const struct ofproto_class *class = ofproto_class_find__(type);
1282 return (!class ? EAFNOSUPPORT
1283 : !class->del ? EACCES
1284 : class->del(type, name));
1285 }
1286
1287 static void
1288 process_port_change(struct ofproto *ofproto, int error, char *devname)
1289 {
1290 if (error == ENOBUFS) {
1291 reinit_ports(ofproto);
1292 } else if (!error) {
1293 update_port(ofproto, devname);
1294 free(devname);
1295 }
1296 }
1297
1298 int
1299 ofproto_type_run(const char *datapath_type)
1300 {
1301 const struct ofproto_class *class;
1302 int error;
1303
1304 datapath_type = ofproto_normalize_type(datapath_type);
1305 class = ofproto_class_find__(datapath_type);
1306
1307 error = class->type_run ? class->type_run(datapath_type) : 0;
1308 if (error && error != EAGAIN) {
1309 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1310 datapath_type, ovs_strerror(error));
1311 }
1312 return error;
1313 }
1314
1315 int
1316 ofproto_type_run_fast(const char *datapath_type)
1317 {
1318 const struct ofproto_class *class;
1319 int error;
1320
1321 datapath_type = ofproto_normalize_type(datapath_type);
1322 class = ofproto_class_find__(datapath_type);
1323
1324 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1325 if (error && error != EAGAIN) {
1326 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1327 datapath_type, ovs_strerror(error));
1328 }
1329 return error;
1330 }
1331
1332 void
1333 ofproto_type_wait(const char *datapath_type)
1334 {
1335 const struct ofproto_class *class;
1336
1337 datapath_type = ofproto_normalize_type(datapath_type);
1338 class = ofproto_class_find__(datapath_type);
1339
1340 if (class->type_wait) {
1341 class->type_wait(datapath_type);
1342 }
1343 }
1344
1345 static bool
1346 any_pending_ops(const struct ofproto *p)
1347 OVS_EXCLUDED(ofproto_mutex)
1348 {
1349 bool b;
1350
1351 ovs_mutex_lock(&ofproto_mutex);
1352 b = !list_is_empty(&p->pending);
1353 ovs_mutex_unlock(&ofproto_mutex);
1354
1355 return b;
1356 }
1357
1358 int
1359 ofproto_run(struct ofproto *p)
1360 {
1361 struct sset changed_netdevs;
1362 const char *changed_netdev;
1363 struct ofport *ofport;
1364 int error;
1365
1366 error = p->ofproto_class->run(p);
1367 if (error && error != EAGAIN) {
1368 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1369 }
1370
1371 run_rule_executes(p);
1372
1373 /* Restore the eviction group heap invariant occasionally. */
1374 if (p->eviction_group_timer < time_msec()) {
1375 size_t i;
1376
1377 p->eviction_group_timer = time_msec() + 1000;
1378
1379 for (i = 0; i < p->n_tables; i++) {
1380 struct oftable *table = &p->tables[i];
1381 struct eviction_group *evg;
1382 struct cls_cursor cursor;
1383 struct rule *rule;
1384
1385 if (!table->eviction_fields) {
1386 continue;
1387 }
1388
1389 ovs_mutex_lock(&ofproto_mutex);
1390 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1391 heap_rebuild(&evg->rules);
1392 }
1393
1394 ovs_rwlock_rdlock(&table->cls.rwlock);
1395 cls_cursor_init(&cursor, &table->cls, NULL);
1396 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1397 if (!rule->eviction_group
1398 && (rule->idle_timeout || rule->hard_timeout)) {
1399 eviction_group_add_rule(rule);
1400 }
1401 }
1402 ovs_rwlock_unlock(&table->cls.rwlock);
1403 ovs_mutex_unlock(&ofproto_mutex);
1404 }
1405 }
1406
1407 if (p->ofproto_class->port_poll) {
1408 char *devname;
1409
1410 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1411 process_port_change(p, error, devname);
1412 }
1413 }
1414
1415 /* Update OpenFlow port status for any port whose netdev has changed.
1416 *
1417 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1418 * destroyed, so it's not safe to update ports directly from the
1419 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1420 * need this two-phase approach. */
1421 sset_init(&changed_netdevs);
1422 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1423 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1424 if (ofport->change_seq != change_seq) {
1425 ofport->change_seq = change_seq;
1426 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1427 }
1428 }
1429 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1430 update_port(p, changed_netdev);
1431 }
1432 sset_destroy(&changed_netdevs);
1433
1434 switch (p->state) {
1435 case S_OPENFLOW:
1436 connmgr_run(p->connmgr, handle_openflow);
1437 break;
1438
1439 case S_EVICT:
1440 connmgr_run(p->connmgr, NULL);
1441 ofproto_evict(p);
1442 if (!any_pending_ops(p)) {
1443 p->state = S_OPENFLOW;
1444 }
1445 break;
1446
1447 case S_FLUSH:
1448 connmgr_run(p->connmgr, NULL);
1449 ofproto_flush__(p);
1450 if (!any_pending_ops(p)) {
1451 connmgr_flushed(p->connmgr);
1452 p->state = S_OPENFLOW;
1453 }
1454 break;
1455
1456 default:
1457 NOT_REACHED();
1458 }
1459
1460 if (time_msec() >= p->next_op_report) {
1461 long long int ago = (time_msec() - p->first_op) / 1000;
1462 long long int interval = (p->last_op - p->first_op) / 1000;
1463 struct ds s;
1464
1465 ds_init(&s);
1466 ds_put_format(&s, "%d flow_mods ",
1467 p->n_add + p->n_delete + p->n_modify);
1468 if (interval == ago) {
1469 ds_put_format(&s, "in the last %lld s", ago);
1470 } else if (interval) {
1471 ds_put_format(&s, "in the %lld s starting %lld s ago",
1472 interval, ago);
1473 } else {
1474 ds_put_format(&s, "%lld s ago", ago);
1475 }
1476
1477 ds_put_cstr(&s, " (");
1478 if (p->n_add) {
1479 ds_put_format(&s, "%d adds, ", p->n_add);
1480 }
1481 if (p->n_delete) {
1482 ds_put_format(&s, "%d deletes, ", p->n_delete);
1483 }
1484 if (p->n_modify) {
1485 ds_put_format(&s, "%d modifications, ", p->n_modify);
1486 }
1487 s.length -= 2;
1488 ds_put_char(&s, ')');
1489
1490 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1491 ds_destroy(&s);
1492
1493 p->n_add = p->n_delete = p->n_modify = 0;
1494 p->next_op_report = LLONG_MAX;
1495 }
1496
1497 return error;
1498 }
1499
1500 /* Performs periodic activity required by 'ofproto' that needs to be done
1501 * with the least possible latency.
1502 *
1503 * It makes sense to call this function a couple of times per poll loop, to
1504 * provide a significant performance boost on some benchmarks with the
1505 * ofproto-dpif implementation. */
1506 int
1507 ofproto_run_fast(struct ofproto *p)
1508 {
1509 int error;
1510
1511 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1512 if (error && error != EAGAIN) {
1513 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1514 p->name, ovs_strerror(error));
1515 }
1516 return error;
1517 }
1518
1519 void
1520 ofproto_wait(struct ofproto *p)
1521 {
1522 struct ofport *ofport;
1523
1524 p->ofproto_class->wait(p);
1525 if (p->ofproto_class->port_poll_wait) {
1526 p->ofproto_class->port_poll_wait(p);
1527 }
1528
1529 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1530 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1531 poll_immediate_wake();
1532 }
1533 }
1534
1535 switch (p->state) {
1536 case S_OPENFLOW:
1537 connmgr_wait(p->connmgr, true);
1538 break;
1539
1540 case S_EVICT:
1541 case S_FLUSH:
1542 connmgr_wait(p->connmgr, false);
1543 if (!any_pending_ops(p)) {
1544 poll_immediate_wake();
1545 }
1546 break;
1547 }
1548 }
1549
1550 bool
1551 ofproto_is_alive(const struct ofproto *p)
1552 {
1553 return connmgr_has_controllers(p->connmgr);
1554 }
1555
1556 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1557 * memory_report(). */
1558 void
1559 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1560 {
1561 const struct oftable *table;
1562 unsigned int n_rules;
1563
1564 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1565
1566 ovs_mutex_lock(&ofproto_mutex);
1567 simap_increase(usage, "ops",
1568 ofproto->n_pending + hmap_count(&ofproto->deletions));
1569 ovs_mutex_unlock(&ofproto_mutex);
1570
1571 n_rules = 0;
1572 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1573 ovs_rwlock_rdlock(&table->cls.rwlock);
1574 n_rules += classifier_count(&table->cls);
1575 ovs_rwlock_unlock(&table->cls.rwlock);
1576 }
1577 simap_increase(usage, "rules", n_rules);
1578
1579 if (ofproto->ofproto_class->get_memory_usage) {
1580 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1581 }
1582
1583 connmgr_get_memory_usage(ofproto->connmgr, usage);
1584 }
1585
1586 void
1587 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1588 struct shash *info)
1589 {
1590 connmgr_get_controller_info(ofproto->connmgr, info);
1591 }
1592
1593 void
1594 ofproto_free_ofproto_controller_info(struct shash *info)
1595 {
1596 connmgr_free_controller_info(info);
1597 }
1598
1599 /* Makes a deep copy of 'old' into 'port'. */
1600 void
1601 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1602 {
1603 port->name = xstrdup(old->name);
1604 port->type = xstrdup(old->type);
1605 port->ofp_port = old->ofp_port;
1606 }
1607
1608 /* Frees memory allocated to members of 'ofproto_port'.
1609 *
1610 * Do not call this function on an ofproto_port obtained from
1611 * ofproto_port_dump_next(): that function retains ownership of the data in the
1612 * ofproto_port. */
1613 void
1614 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1615 {
1616 free(ofproto_port->name);
1617 free(ofproto_port->type);
1618 }
1619
1620 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1621 *
1622 * This function provides no status indication. An error status for the entire
1623 * dump operation is provided when it is completed by calling
1624 * ofproto_port_dump_done().
1625 */
1626 void
1627 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1628 const struct ofproto *ofproto)
1629 {
1630 dump->ofproto = ofproto;
1631 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1632 &dump->state);
1633 }
1634
1635 /* Attempts to retrieve another port from 'dump', which must have been created
1636 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1637 * 'port' and returns true. On failure, returns false.
1638 *
1639 * Failure might indicate an actual error or merely that the last port has been
1640 * dumped. An error status for the entire dump operation is provided when it
1641 * is completed by calling ofproto_port_dump_done().
1642 *
1643 * The ofproto owns the data stored in 'port'. It will remain valid until at
1644 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1645 * ofproto_port_dump_done(). */
1646 bool
1647 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1648 struct ofproto_port *port)
1649 {
1650 const struct ofproto *ofproto = dump->ofproto;
1651
1652 if (dump->error) {
1653 return false;
1654 }
1655
1656 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1657 port);
1658 if (dump->error) {
1659 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1660 return false;
1661 }
1662 return true;
1663 }
1664
1665 /* Completes port table dump operation 'dump', which must have been created
1666 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1667 * error-free, otherwise a positive errno value describing the problem. */
1668 int
1669 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1670 {
1671 const struct ofproto *ofproto = dump->ofproto;
1672 if (!dump->error) {
1673 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1674 dump->state);
1675 }
1676 return dump->error == EOF ? 0 : dump->error;
1677 }
1678
1679 /* Returns the type to pass to netdev_open() when a datapath of type
1680 * 'datapath_type' has a port of type 'port_type', for a few special
1681 * cases when a netdev type differs from a port type. For example, when
1682 * using the userspace datapath, a port of type "internal" needs to be
1683 * opened as "tap".
1684 *
1685 * Returns either 'type' itself or a string literal, which must not be
1686 * freed. */
1687 const char *
1688 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1689 {
1690 const struct ofproto_class *class;
1691
1692 datapath_type = ofproto_normalize_type(datapath_type);
1693 class = ofproto_class_find__(datapath_type);
1694 if (!class) {
1695 return port_type;
1696 }
1697
1698 return (class->port_open_type
1699 ? class->port_open_type(datapath_type, port_type)
1700 : port_type);
1701 }
1702
1703 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1704 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1705 * the port's OpenFlow port number.
1706 *
1707 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1708 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1709 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1710 * 'ofp_portp' is non-null). */
1711 int
1712 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1713 ofp_port_t *ofp_portp)
1714 {
1715 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1716 int error;
1717
1718 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1719 if (!error) {
1720 const char *netdev_name = netdev_get_name(netdev);
1721
1722 simap_put(&ofproto->ofp_requests, netdev_name,
1723 ofp_to_u16(ofp_port));
1724 update_port(ofproto, netdev_name);
1725 }
1726 if (ofp_portp) {
1727 struct ofproto_port ofproto_port;
1728
1729 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1730 &ofproto_port);
1731 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1732 ofproto_port_destroy(&ofproto_port);
1733 }
1734 return error;
1735 }
1736
1737 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1738 * initializes '*port' appropriately; on failure, returns a positive errno
1739 * value.
1740 *
1741 * The caller owns the data in 'ofproto_port' and must free it with
1742 * ofproto_port_destroy() when it is no longer needed. */
1743 int
1744 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1745 struct ofproto_port *port)
1746 {
1747 int error;
1748
1749 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1750 if (error) {
1751 memset(port, 0, sizeof *port);
1752 }
1753 return error;
1754 }
1755
1756 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1757 * Returns 0 if successful, otherwise a positive errno. */
1758 int
1759 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1760 {
1761 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1762 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1763 struct simap_node *ofp_request_node;
1764 int error;
1765
1766 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1767 if (ofp_request_node) {
1768 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1769 }
1770
1771 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1772 if (!error && ofport) {
1773 /* 'name' is the netdev's name and update_port() is going to close the
1774 * netdev. Just in case update_port() refers to 'name' after it
1775 * destroys 'ofport', make a copy of it around the update_port()
1776 * call. */
1777 char *devname = xstrdup(name);
1778 update_port(ofproto, devname);
1779 free(devname);
1780 }
1781 return error;
1782 }
1783
1784 static int
1785 simple_flow_mod(struct ofproto *ofproto,
1786 const struct match *match, unsigned int priority,
1787 const struct ofpact *ofpacts, size_t ofpacts_len,
1788 enum ofp_flow_mod_command command)
1789 {
1790 struct ofputil_flow_mod fm;
1791
1792 memset(&fm, 0, sizeof fm);
1793 fm.match = *match;
1794 fm.priority = priority;
1795 fm.cookie = 0;
1796 fm.new_cookie = 0;
1797 fm.modify_cookie = false;
1798 fm.table_id = 0;
1799 fm.command = command;
1800 fm.idle_timeout = 0;
1801 fm.hard_timeout = 0;
1802 fm.buffer_id = UINT32_MAX;
1803 fm.out_port = OFPP_ANY;
1804 fm.out_group = OFPG_ANY;
1805 fm.flags = 0;
1806 fm.ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1807 fm.ofpacts_len = ofpacts_len;
1808
1809 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1810 }
1811
1812 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1813 * performs the 'n_actions' actions in 'actions'. The new flow will not
1814 * timeout.
1815 *
1816 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1817 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1818 * controllers; otherwise, it will be hidden.
1819 *
1820 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1821 *
1822 * This is a helper function for in-band control and fail-open. */
1823 void
1824 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1825 unsigned int priority,
1826 const struct ofpact *ofpacts, size_t ofpacts_len)
1827 OVS_EXCLUDED(ofproto_mutex)
1828 {
1829 const struct rule *rule;
1830 bool must_add;
1831
1832 /* First do a cheap check whether the rule we're looking for already exists
1833 * with the actions that we want. If it does, then we're done. */
1834 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1835 rule = rule_from_cls_rule(classifier_find_match_exactly(
1836 &ofproto->tables[0].cls, match, priority));
1837 if (rule) {
1838 ovs_mutex_lock(&rule->mutex);
1839 must_add = !ofpacts_equal(rule->actions->ofpacts,
1840 rule->actions->ofpacts_len,
1841 ofpacts, ofpacts_len);
1842 ovs_mutex_unlock(&rule->mutex);
1843 } else {
1844 must_add = true;
1845 }
1846 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1847
1848 /* If there's no such rule or the rule doesn't have the actions we want,
1849 * fall back to a executing a full flow mod. We can't optimize this at
1850 * all because we didn't take enough locks above to ensure that the flow
1851 * table didn't already change beneath us. */
1852 if (must_add) {
1853 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1854 OFPFC_MODIFY_STRICT);
1855 }
1856 }
1857
1858 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1859 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1860 * operation cannot be initiated now but may be retried later.
1861 *
1862 * This is a helper function for in-band control and fail-open and the "learn"
1863 * action. */
1864 int
1865 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1866 OVS_EXCLUDED(ofproto_mutex)
1867 {
1868 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1869 }
1870
1871 /* Searches for a rule with matching criteria exactly equal to 'target' in
1872 * ofproto's table 0 and, if it finds one, deletes it.
1873 *
1874 * This is a helper function for in-band control and fail-open. */
1875 bool
1876 ofproto_delete_flow(struct ofproto *ofproto,
1877 const struct match *target, unsigned int priority)
1878 OVS_EXCLUDED(ofproto_mutex)
1879 {
1880 struct classifier *cls = &ofproto->tables[0].cls;
1881 struct rule *rule;
1882
1883 /* First do a cheap check whether the rule we're looking for has already
1884 * been deleted. If so, then we're done. */
1885 ovs_rwlock_rdlock(&cls->rwlock);
1886 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1887 priority));
1888 ovs_rwlock_unlock(&cls->rwlock);
1889 if (!rule) {
1890 return true;
1891 }
1892
1893 /* Fall back to a executing a full flow mod. We can't optimize this at all
1894 * because we didn't take enough locks above to ensure that the flow table
1895 * didn't already change beneath us. */
1896 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1897 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1898 }
1899
1900 /* Starts the process of deleting all of the flows from all of ofproto's flow
1901 * tables and then reintroducing the flows required by in-band control and
1902 * fail-open. The process will complete in a later call to ofproto_run(). */
1903 void
1904 ofproto_flush_flows(struct ofproto *ofproto)
1905 {
1906 COVERAGE_INC(ofproto_flush);
1907 ofproto->state = S_FLUSH;
1908 }
1909 \f
1910 static void
1911 reinit_ports(struct ofproto *p)
1912 {
1913 struct ofproto_port_dump dump;
1914 struct sset devnames;
1915 struct ofport *ofport;
1916 struct ofproto_port ofproto_port;
1917 const char *devname;
1918
1919 COVERAGE_INC(ofproto_reinit_ports);
1920
1921 sset_init(&devnames);
1922 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1923 sset_add(&devnames, netdev_get_name(ofport->netdev));
1924 }
1925 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1926 sset_add(&devnames, ofproto_port.name);
1927 }
1928
1929 SSET_FOR_EACH (devname, &devnames) {
1930 update_port(p, devname);
1931 }
1932 sset_destroy(&devnames);
1933 }
1934
1935 static ofp_port_t
1936 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1937 {
1938 uint16_t port_idx;
1939
1940 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1941 port_idx = port_idx ? port_idx : UINT16_MAX;
1942
1943 if (port_idx >= ofproto->max_ports
1944 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1945 uint16_t end_port_no = ofproto->alloc_port_no;
1946
1947 /* Search for a free OpenFlow port number. We try not to
1948 * immediately reuse them to prevent problems due to old
1949 * flows. */
1950 for (;;) {
1951 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1952 ofproto->alloc_port_no = 0;
1953 }
1954 if (!bitmap_is_set(ofproto->ofp_port_ids,
1955 ofproto->alloc_port_no)) {
1956 port_idx = ofproto->alloc_port_no;
1957 break;
1958 }
1959 if (ofproto->alloc_port_no == end_port_no) {
1960 return OFPP_NONE;
1961 }
1962 }
1963 }
1964 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1965 return u16_to_ofp(port_idx);
1966 }
1967
1968 static void
1969 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1970 {
1971 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
1972 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1973 }
1974 }
1975
1976 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1977 * pointer if the netdev cannot be opened. On success, also fills in
1978 * 'opp'. */
1979 static struct netdev *
1980 ofport_open(struct ofproto *ofproto,
1981 struct ofproto_port *ofproto_port,
1982 struct ofputil_phy_port *pp)
1983 {
1984 enum netdev_flags flags;
1985 struct netdev *netdev;
1986 int error;
1987
1988 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1989 if (error) {
1990 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1991 "cannot be opened (%s)",
1992 ofproto->name,
1993 ofproto_port->name, ofproto_port->ofp_port,
1994 ofproto_port->name, ovs_strerror(error));
1995 return NULL;
1996 }
1997
1998 if (ofproto_port->ofp_port == OFPP_NONE) {
1999 if (!strcmp(ofproto->name, ofproto_port->name)) {
2000 ofproto_port->ofp_port = OFPP_LOCAL;
2001 } else {
2002 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2003 ofproto_port->name);
2004 }
2005 }
2006 pp->port_no = ofproto_port->ofp_port;
2007 netdev_get_etheraddr(netdev, pp->hw_addr);
2008 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2009 netdev_get_flags(netdev, &flags);
2010 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2011 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2012 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2013 &pp->supported, &pp->peer);
2014 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2015 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2016
2017 return netdev;
2018 }
2019
2020 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2021 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2022 * disregarded. */
2023 static bool
2024 ofport_equal(const struct ofputil_phy_port *a,
2025 const struct ofputil_phy_port *b)
2026 {
2027 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2028 && a->state == b->state
2029 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2030 && a->curr == b->curr
2031 && a->advertised == b->advertised
2032 && a->supported == b->supported
2033 && a->peer == b->peer
2034 && a->curr_speed == b->curr_speed
2035 && a->max_speed == b->max_speed);
2036 }
2037
2038 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2039 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2040 * one with the same name or port number). */
2041 static void
2042 ofport_install(struct ofproto *p,
2043 struct netdev *netdev, const struct ofputil_phy_port *pp)
2044 {
2045 const char *netdev_name = netdev_get_name(netdev);
2046 struct ofport *ofport;
2047 int error;
2048
2049 /* Create ofport. */
2050 ofport = p->ofproto_class->port_alloc();
2051 if (!ofport) {
2052 error = ENOMEM;
2053 goto error;
2054 }
2055 ofport->ofproto = p;
2056 ofport->netdev = netdev;
2057 ofport->change_seq = netdev_change_seq(netdev);
2058 ofport->pp = *pp;
2059 ofport->ofp_port = pp->port_no;
2060 ofport->created = time_msec();
2061
2062 /* Add port to 'p'. */
2063 hmap_insert(&p->ports, &ofport->hmap_node,
2064 hash_ofp_port(ofport->ofp_port));
2065 shash_add(&p->port_by_name, netdev_name, ofport);
2066
2067 update_mtu(p, ofport);
2068
2069 /* Let the ofproto_class initialize its private data. */
2070 error = p->ofproto_class->port_construct(ofport);
2071 if (error) {
2072 goto error;
2073 }
2074 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2075 return;
2076
2077 error:
2078 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2079 p->name, netdev_name, ovs_strerror(error));
2080 if (ofport) {
2081 ofport_destroy__(ofport);
2082 } else {
2083 netdev_close(netdev);
2084 }
2085 }
2086
2087 /* Removes 'ofport' from 'p' and destroys it. */
2088 static void
2089 ofport_remove(struct ofport *ofport)
2090 {
2091 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2092 OFPPR_DELETE);
2093 ofport_destroy(ofport);
2094 }
2095
2096 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2097 * destroys it. */
2098 static void
2099 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2100 {
2101 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2102 if (port) {
2103 ofport_remove(port);
2104 }
2105 }
2106
2107 /* Updates 'port' with new 'pp' description.
2108 *
2109 * Does not handle a name or port number change. The caller must implement
2110 * such a change as a delete followed by an add. */
2111 static void
2112 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2113 {
2114 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2115 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2116 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2117 port->pp.state = pp->state;
2118 port->pp.curr = pp->curr;
2119 port->pp.advertised = pp->advertised;
2120 port->pp.supported = pp->supported;
2121 port->pp.peer = pp->peer;
2122 port->pp.curr_speed = pp->curr_speed;
2123 port->pp.max_speed = pp->max_speed;
2124
2125 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2126 }
2127
2128 /* Update OpenFlow 'state' in 'port' and notify controller. */
2129 void
2130 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2131 {
2132 if (port->pp.state != state) {
2133 port->pp.state = state;
2134 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2135 OFPPR_MODIFY);
2136 }
2137 }
2138
2139 void
2140 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2141 {
2142 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2143 if (port) {
2144 if (port->ofproto->ofproto_class->set_realdev) {
2145 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2146 }
2147 if (port->ofproto->ofproto_class->set_stp_port) {
2148 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2149 }
2150 if (port->ofproto->ofproto_class->set_cfm) {
2151 port->ofproto->ofproto_class->set_cfm(port, NULL);
2152 }
2153 if (port->ofproto->ofproto_class->bundle_remove) {
2154 port->ofproto->ofproto_class->bundle_remove(port);
2155 }
2156 }
2157 }
2158
2159 static void
2160 ofport_destroy__(struct ofport *port)
2161 {
2162 struct ofproto *ofproto = port->ofproto;
2163 const char *name = netdev_get_name(port->netdev);
2164
2165 hmap_remove(&ofproto->ports, &port->hmap_node);
2166 shash_delete(&ofproto->port_by_name,
2167 shash_find(&ofproto->port_by_name, name));
2168
2169 netdev_close(port->netdev);
2170 ofproto->ofproto_class->port_dealloc(port);
2171 }
2172
2173 static void
2174 ofport_destroy(struct ofport *port)
2175 {
2176 if (port) {
2177 dealloc_ofp_port(port->ofproto, port->ofp_port);
2178 port->ofproto->ofproto_class->port_destruct(port);
2179 ofport_destroy__(port);
2180 }
2181 }
2182
2183 struct ofport *
2184 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2185 {
2186 struct ofport *port;
2187
2188 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2189 &ofproto->ports) {
2190 if (port->ofp_port == ofp_port) {
2191 return port;
2192 }
2193 }
2194 return NULL;
2195 }
2196
2197 int
2198 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2199 {
2200 struct ofproto *ofproto = port->ofproto;
2201 int error;
2202
2203 if (ofproto->ofproto_class->port_get_stats) {
2204 error = ofproto->ofproto_class->port_get_stats(port, stats);
2205 } else {
2206 error = EOPNOTSUPP;
2207 }
2208
2209 return error;
2210 }
2211
2212 static void
2213 update_port(struct ofproto *ofproto, const char *name)
2214 {
2215 struct ofproto_port ofproto_port;
2216 struct ofputil_phy_port pp;
2217 struct netdev *netdev;
2218 struct ofport *port;
2219
2220 COVERAGE_INC(ofproto_update_port);
2221
2222 /* Fetch 'name''s location and properties from the datapath. */
2223 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2224 ? ofport_open(ofproto, &ofproto_port, &pp)
2225 : NULL);
2226
2227 if (netdev) {
2228 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2229 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2230 struct netdev *old_netdev = port->netdev;
2231
2232 /* 'name' hasn't changed location. Any properties changed? */
2233 if (!ofport_equal(&port->pp, &pp)) {
2234 ofport_modified(port, &pp);
2235 }
2236
2237 update_mtu(ofproto, port);
2238
2239 /* Install the newly opened netdev in case it has changed.
2240 * Don't close the old netdev yet in case port_modified has to
2241 * remove a retained reference to it.*/
2242 port->netdev = netdev;
2243 port->change_seq = netdev_change_seq(netdev);
2244
2245 if (port->ofproto->ofproto_class->port_modified) {
2246 port->ofproto->ofproto_class->port_modified(port);
2247 }
2248
2249 netdev_close(old_netdev);
2250 } else {
2251 /* If 'port' is nonnull then its name differs from 'name' and thus
2252 * we should delete it. If we think there's a port named 'name'
2253 * then its port number must be wrong now so delete it too. */
2254 if (port) {
2255 ofport_remove(port);
2256 }
2257 ofport_remove_with_name(ofproto, name);
2258 ofport_install(ofproto, netdev, &pp);
2259 }
2260 } else {
2261 /* Any port named 'name' is gone now. */
2262 ofport_remove_with_name(ofproto, name);
2263 }
2264 ofproto_port_destroy(&ofproto_port);
2265 }
2266
2267 static int
2268 init_ports(struct ofproto *p)
2269 {
2270 struct ofproto_port_dump dump;
2271 struct ofproto_port ofproto_port;
2272 struct shash_node *node, *next;
2273
2274 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2275 const char *name = ofproto_port.name;
2276
2277 if (shash_find(&p->port_by_name, name)) {
2278 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2279 p->name, name);
2280 } else {
2281 struct ofputil_phy_port pp;
2282 struct netdev *netdev;
2283
2284 /* Check if an OpenFlow port number had been requested. */
2285 node = shash_find(&init_ofp_ports, name);
2286 if (node) {
2287 const struct iface_hint *iface_hint = node->data;
2288 simap_put(&p->ofp_requests, name,
2289 ofp_to_u16(iface_hint->ofp_port));
2290 }
2291
2292 netdev = ofport_open(p, &ofproto_port, &pp);
2293 if (netdev) {
2294 ofport_install(p, netdev, &pp);
2295 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2296 p->alloc_port_no = MAX(p->alloc_port_no,
2297 ofp_to_u16(ofproto_port.ofp_port));
2298 }
2299 }
2300 }
2301 }
2302
2303 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2304 struct iface_hint *iface_hint = node->data;
2305
2306 if (!strcmp(iface_hint->br_name, p->name)) {
2307 free(iface_hint->br_name);
2308 free(iface_hint->br_type);
2309 free(iface_hint);
2310 shash_delete(&init_ofp_ports, node);
2311 }
2312 }
2313
2314 return 0;
2315 }
2316
2317 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2318 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2319 static int
2320 find_min_mtu(struct ofproto *p)
2321 {
2322 struct ofport *ofport;
2323 int mtu = 0;
2324
2325 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2326 struct netdev *netdev = ofport->netdev;
2327 int dev_mtu;
2328
2329 /* Skip any internal ports, since that's what we're trying to
2330 * set. */
2331 if (!strcmp(netdev_get_type(netdev), "internal")) {
2332 continue;
2333 }
2334
2335 if (netdev_get_mtu(netdev, &dev_mtu)) {
2336 continue;
2337 }
2338 if (!mtu || dev_mtu < mtu) {
2339 mtu = dev_mtu;
2340 }
2341 }
2342
2343 return mtu ? mtu: ETH_PAYLOAD_MAX;
2344 }
2345
2346 /* Update MTU of all datapath devices on 'p' to the minimum of the
2347 * non-datapath ports in event of 'port' added or changed. */
2348 static void
2349 update_mtu(struct ofproto *p, struct ofport *port)
2350 {
2351 struct ofport *ofport;
2352 struct netdev *netdev = port->netdev;
2353 int dev_mtu, old_min;
2354
2355 if (netdev_get_mtu(netdev, &dev_mtu)) {
2356 port->mtu = 0;
2357 return;
2358 }
2359 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2360 if (dev_mtu > p->min_mtu) {
2361 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2362 dev_mtu = p->min_mtu;
2363 }
2364 }
2365 port->mtu = dev_mtu;
2366 return;
2367 }
2368
2369 /* For non-internal port find new min mtu. */
2370 old_min = p->min_mtu;
2371 port->mtu = dev_mtu;
2372 p->min_mtu = find_min_mtu(p);
2373 if (p->min_mtu == old_min) {
2374 return;
2375 }
2376
2377 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2378 struct netdev *netdev = ofport->netdev;
2379
2380 if (!strcmp(netdev_get_type(netdev), "internal")) {
2381 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2382 ofport->mtu = p->min_mtu;
2383 }
2384 }
2385 }
2386 }
2387 \f
2388 void
2389 ofproto_rule_ref(struct rule *rule)
2390 {
2391 if (rule) {
2392 unsigned int orig;
2393
2394 atomic_add(&rule->ref_count, 1, &orig);
2395 ovs_assert(orig != 0);
2396 }
2397 }
2398
2399 void
2400 ofproto_rule_unref(struct rule *rule)
2401 {
2402 if (rule) {
2403 unsigned int orig;
2404
2405 atomic_sub(&rule->ref_count, 1, &orig);
2406 if (orig == 1) {
2407 rule->ofproto->ofproto_class->rule_destruct(rule);
2408 ofproto_rule_destroy__(rule);
2409 } else {
2410 ovs_assert(orig != 0);
2411 }
2412 }
2413 }
2414
2415 struct rule_actions *
2416 rule_get_actions(const struct rule *rule)
2417 OVS_EXCLUDED(rule->mutex)
2418 {
2419 struct rule_actions *actions;
2420
2421 ovs_mutex_lock(&rule->mutex);
2422 actions = rule_get_actions__(rule);
2423 ovs_mutex_unlock(&rule->mutex);
2424
2425 return actions;
2426 }
2427
2428 struct rule_actions *
2429 rule_get_actions__(const struct rule *rule)
2430 OVS_REQUIRES(rule->mutex)
2431 {
2432 rule_actions_ref(rule->actions);
2433 return rule->actions;
2434 }
2435
2436 static void
2437 ofproto_rule_destroy__(struct rule *rule)
2438 OVS_NO_THREAD_SAFETY_ANALYSIS
2439 {
2440 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2441 rule_actions_unref(rule->actions);
2442 ovs_mutex_destroy(&rule->mutex);
2443 rule->ofproto->ofproto_class->rule_dealloc(rule);
2444 }
2445
2446 static uint32_t get_provider_meter_id(const struct ofproto *,
2447 uint32_t of_meter_id);
2448
2449 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2450 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2451 struct rule_actions *
2452 rule_actions_create(const struct ofproto *ofproto,
2453 const struct ofpact *ofpacts, size_t ofpacts_len)
2454 {
2455 struct rule_actions *actions;
2456
2457 actions = xmalloc(sizeof *actions);
2458 atomic_init(&actions->ref_count, 1);
2459 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2460 actions->ofpacts_len = ofpacts_len;
2461 actions->provider_meter_id
2462 = get_provider_meter_id(ofproto,
2463 ofpacts_get_meter(ofpacts, ofpacts_len));
2464
2465 return actions;
2466 }
2467
2468 /* Increments 'actions''s ref_count. */
2469 void
2470 rule_actions_ref(struct rule_actions *actions)
2471 {
2472 if (actions) {
2473 unsigned int orig;
2474
2475 atomic_add(&actions->ref_count, 1, &orig);
2476 ovs_assert(orig != 0);
2477 }
2478 }
2479
2480 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2481 * reaches 0. */
2482 void
2483 rule_actions_unref(struct rule_actions *actions)
2484 {
2485 if (actions) {
2486 unsigned int orig;
2487
2488 atomic_sub(&actions->ref_count, 1, &orig);
2489 if (orig == 1) {
2490 free(actions);
2491 } else {
2492 ovs_assert(orig != 0);
2493 }
2494 }
2495 }
2496
2497 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2498 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2499 static bool
2500 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2501 OVS_REQUIRES(ofproto_mutex)
2502 {
2503 return (port == OFPP_ANY
2504 || ofpacts_output_to_port(rule->actions->ofpacts,
2505 rule->actions->ofpacts_len, port));
2506 }
2507
2508 /* Returns true if 'rule' has group and equals group_id. */
2509 static bool
2510 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2511 OVS_REQUIRES(ofproto_mutex)
2512 {
2513 return (group_id == OFPG11_ANY
2514 || ofpacts_output_to_group(rule->actions->ofpacts,
2515 rule->actions->ofpacts_len, group_id));
2516 }
2517
2518 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2519 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2520 bool
2521 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2522 OVS_REQUIRES(ofproto_mutex)
2523 {
2524 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2525 return true;
2526 }
2527
2528 switch (op->type) {
2529 case OFOPERATION_ADD:
2530 case OFOPERATION_DELETE:
2531 return false;
2532
2533 case OFOPERATION_MODIFY:
2534 case OFOPERATION_REPLACE:
2535 return ofpacts_output_to_port(op->actions->ofpacts,
2536 op->actions->ofpacts_len, out_port);
2537 }
2538
2539 NOT_REACHED();
2540 }
2541
2542 static void
2543 rule_execute_destroy(struct rule_execute *e)
2544 {
2545 ofproto_rule_unref(e->rule);
2546 list_remove(&e->list_node);
2547 free(e);
2548 }
2549
2550 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2551 * by passing them to the ofproto provider. */
2552 static void
2553 run_rule_executes(struct ofproto *ofproto)
2554 OVS_EXCLUDED(ofproto_mutex)
2555 {
2556 struct rule_execute *e, *next;
2557 struct list executes;
2558
2559 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2560 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2561 union flow_in_port in_port_;
2562 struct flow flow;
2563
2564 in_port_.ofp_port = e->in_port;
2565 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2566 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2567
2568 rule_execute_destroy(e);
2569 }
2570 }
2571
2572 /* Destroys and discards all "rule_execute" operations queued up in
2573 * ofproto->rule_executes. */
2574 static void
2575 destroy_rule_executes(struct ofproto *ofproto)
2576 {
2577 struct rule_execute *e, *next;
2578 struct list executes;
2579
2580 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2581 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2582 ofpbuf_delete(e->packet);
2583 rule_execute_destroy(e);
2584 }
2585 }
2586
2587 /* Returns true if 'rule' should be hidden from the controller.
2588 *
2589 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2590 * (e.g. by in-band control) and are intentionally hidden from the
2591 * controller. */
2592 static bool
2593 ofproto_rule_is_hidden(const struct rule *rule)
2594 {
2595 return rule->cr.priority > UINT16_MAX;
2596 }
2597
2598 static enum oftable_flags
2599 rule_get_flags(const struct rule *rule)
2600 {
2601 return rule->ofproto->tables[rule->table_id].flags;
2602 }
2603
2604 static bool
2605 rule_is_modifiable(const struct rule *rule)
2606 {
2607 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2608 }
2609 \f
2610 static enum ofperr
2611 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2612 {
2613 ofconn_send_reply(ofconn, make_echo_reply(oh));
2614 return 0;
2615 }
2616
2617 static enum ofperr
2618 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2619 {
2620 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2621 struct ofputil_switch_features features;
2622 struct ofport *port;
2623 bool arp_match_ip;
2624 struct ofpbuf *b;
2625 int n_tables;
2626 int i;
2627
2628 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2629 &features.actions);
2630 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2631
2632 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2633 * if present, are always at the end.) */
2634 n_tables = ofproto->n_tables;
2635 for (i = 0; i < ofproto->n_tables; i++) {
2636 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2637 n_tables = i;
2638 break;
2639 }
2640 }
2641
2642 features.datapath_id = ofproto->datapath_id;
2643 features.n_buffers = pktbuf_capacity();
2644 features.n_tables = n_tables;
2645 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2646 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2647 if (arp_match_ip) {
2648 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2649 }
2650 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2651 features.auxiliary_id = 0;
2652 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2653 oh->xid);
2654 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2655 ofputil_put_switch_features_port(&port->pp, b);
2656 }
2657
2658 ofconn_send_reply(ofconn, b);
2659 return 0;
2660 }
2661
2662 static enum ofperr
2663 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2664 {
2665 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2666 struct ofp_switch_config *osc;
2667 enum ofp_config_flags flags;
2668 struct ofpbuf *buf;
2669
2670 /* Send reply. */
2671 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2672 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2673 flags = ofproto->frag_handling;
2674 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2675 if (oh->version < OFP13_VERSION
2676 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2677 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2678 }
2679 osc->flags = htons(flags);
2680 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2681 ofconn_send_reply(ofconn, buf);
2682
2683 return 0;
2684 }
2685
2686 static enum ofperr
2687 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2688 {
2689 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2690 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2691 uint16_t flags = ntohs(osc->flags);
2692
2693 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2694 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2695 enum ofp_config_flags cur = ofproto->frag_handling;
2696 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2697
2698 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2699 if (cur != next) {
2700 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2701 ofproto->frag_handling = next;
2702 } else {
2703 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2704 ofproto->name,
2705 ofputil_frag_handling_to_string(next));
2706 }
2707 }
2708 }
2709 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2710 ofconn_set_invalid_ttl_to_controller(ofconn,
2711 (oh->version < OFP13_VERSION
2712 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2713
2714 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2715
2716 return 0;
2717 }
2718
2719 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2720 * error message code for the caller to propagate upward. Otherwise, returns
2721 * 0.
2722 *
2723 * The log message mentions 'msg_type'. */
2724 static enum ofperr
2725 reject_slave_controller(struct ofconn *ofconn)
2726 {
2727 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2728 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2729 return OFPERR_OFPBRC_EPERM;
2730 } else {
2731 return 0;
2732 }
2733 }
2734
2735 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2736 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2737 * 'flow' may be temporarily modified, but is restored at return.
2738 */
2739 static enum ofperr
2740 ofproto_check_ofpacts(struct ofproto *ofproto,
2741 const struct ofpact ofpacts[], size_t ofpacts_len,
2742 struct flow *flow, uint8_t table_id)
2743 {
2744 enum ofperr error;
2745 uint32_t mid;
2746
2747 error = ofpacts_check(ofpacts, ofpacts_len, flow,
2748 u16_to_ofp(ofproto->max_ports), table_id);
2749 if (error) {
2750 return error;
2751 }
2752
2753 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2754 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2755 return OFPERR_OFPMMFC_INVALID_METER;
2756 }
2757 return 0;
2758 }
2759
2760 static enum ofperr
2761 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2762 {
2763 struct ofproto *p = ofconn_get_ofproto(ofconn);
2764 struct ofputil_packet_out po;
2765 struct ofpbuf *payload;
2766 uint64_t ofpacts_stub[1024 / 8];
2767 struct ofpbuf ofpacts;
2768 struct flow flow;
2769 union flow_in_port in_port_;
2770 enum ofperr error;
2771
2772 COVERAGE_INC(ofproto_packet_out);
2773
2774 error = reject_slave_controller(ofconn);
2775 if (error) {
2776 goto exit;
2777 }
2778
2779 /* Decode message. */
2780 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2781 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2782 if (error) {
2783 goto exit_free_ofpacts;
2784 }
2785 if (ofp_to_u16(po.in_port) >= p->max_ports
2786 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2787 error = OFPERR_OFPBRC_BAD_PORT;
2788 goto exit_free_ofpacts;
2789 }
2790
2791
2792 /* Get payload. */
2793 if (po.buffer_id != UINT32_MAX) {
2794 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2795 if (error || !payload) {
2796 goto exit_free_ofpacts;
2797 }
2798 } else {
2799 /* Ensure that the L3 header is 32-bit aligned. */
2800 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2801 }
2802
2803 /* Verify actions against packet, then send packet if successful. */
2804 in_port_.ofp_port = po.in_port;
2805 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2806 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2807 if (!error) {
2808 error = p->ofproto_class->packet_out(p, payload, &flow,
2809 po.ofpacts, po.ofpacts_len);
2810 }
2811 ofpbuf_delete(payload);
2812
2813 exit_free_ofpacts:
2814 ofpbuf_uninit(&ofpacts);
2815 exit:
2816 return error;
2817 }
2818
2819 static void
2820 update_port_config(struct ofport *port,
2821 enum ofputil_port_config config,
2822 enum ofputil_port_config mask)
2823 {
2824 enum ofputil_port_config old_config = port->pp.config;
2825 enum ofputil_port_config toggle;
2826
2827 toggle = (config ^ port->pp.config) & mask;
2828 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2829 if (config & OFPUTIL_PC_PORT_DOWN) {
2830 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2831 } else {
2832 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2833 }
2834 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2835 }
2836
2837 port->pp.config ^= toggle;
2838 if (port->pp.config != old_config) {
2839 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2840 }
2841 }
2842
2843 static enum ofperr
2844 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2845 {
2846 struct ofproto *p = ofconn_get_ofproto(ofconn);
2847 struct ofputil_port_mod pm;
2848 struct ofport *port;
2849 enum ofperr error;
2850
2851 error = reject_slave_controller(ofconn);
2852 if (error) {
2853 return error;
2854 }
2855
2856 error = ofputil_decode_port_mod(oh, &pm);
2857 if (error) {
2858 return error;
2859 }
2860
2861 port = ofproto_get_port(p, pm.port_no);
2862 if (!port) {
2863 return OFPERR_OFPPMFC_BAD_PORT;
2864 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2865 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2866 } else {
2867 update_port_config(port, pm.config, pm.mask);
2868 if (pm.advertise) {
2869 netdev_set_advertisements(port->netdev, pm.advertise);
2870 }
2871 }
2872 return 0;
2873 }
2874
2875 static enum ofperr
2876 handle_desc_stats_request(struct ofconn *ofconn,
2877 const struct ofp_header *request)
2878 {
2879 static const char *default_mfr_desc = "Nicira, Inc.";
2880 static const char *default_hw_desc = "Open vSwitch";
2881 static const char *default_sw_desc = VERSION;
2882 static const char *default_serial_desc = "None";
2883 static const char *default_dp_desc = "None";
2884
2885 struct ofproto *p = ofconn_get_ofproto(ofconn);
2886 struct ofp_desc_stats *ods;
2887 struct ofpbuf *msg;
2888
2889 msg = ofpraw_alloc_stats_reply(request, 0);
2890 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2891 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2892 sizeof ods->mfr_desc);
2893 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2894 sizeof ods->hw_desc);
2895 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2896 sizeof ods->sw_desc);
2897 ovs_strlcpy(ods->serial_num,
2898 p->serial_desc ? p->serial_desc : default_serial_desc,
2899 sizeof ods->serial_num);
2900 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2901 sizeof ods->dp_desc);
2902 ofconn_send_reply(ofconn, msg);
2903
2904 return 0;
2905 }
2906
2907 static enum ofperr
2908 handle_table_stats_request(struct ofconn *ofconn,
2909 const struct ofp_header *request)
2910 {
2911 struct ofproto *p = ofconn_get_ofproto(ofconn);
2912 struct ofp12_table_stats *ots;
2913 struct ofpbuf *msg;
2914 int n_tables;
2915 size_t i;
2916
2917 /* Set up default values.
2918 *
2919 * ofp12_table_stats is used as a generic structure as
2920 * it is able to hold all the fields for ofp10_table_stats
2921 * and ofp11_table_stats (and of course itself).
2922 */
2923 ots = xcalloc(p->n_tables, sizeof *ots);
2924 for (i = 0; i < p->n_tables; i++) {
2925 ots[i].table_id = i;
2926 sprintf(ots[i].name, "table%zu", i);
2927 ots[i].match = htonll(OFPXMT13_MASK);
2928 ots[i].wildcards = htonll(OFPXMT13_MASK);
2929 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2930 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2931 ots[i].write_setfields = htonll(OFPXMT13_MASK);
2932 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
2933 ots[i].metadata_match = htonll(UINT64_MAX);
2934 ots[i].metadata_write = htonll(UINT64_MAX);
2935 ots[i].instructions = htonl(OFPIT11_ALL);
2936 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2937 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2938 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
2939 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2940 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
2941 }
2942
2943 p->ofproto_class->get_tables(p, ots);
2944
2945 /* Post-process the tables, dropping hidden tables. */
2946 n_tables = p->n_tables;
2947 for (i = 0; i < p->n_tables; i++) {
2948 const struct oftable *table = &p->tables[i];
2949
2950 if (table->flags & OFTABLE_HIDDEN) {
2951 n_tables = i;
2952 break;
2953 }
2954
2955 if (table->name) {
2956 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2957 }
2958
2959 if (table->max_flows < ntohl(ots[i].max_entries)) {
2960 ots[i].max_entries = htonl(table->max_flows);
2961 }
2962 }
2963
2964 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2965 ofconn_send_reply(ofconn, msg);
2966
2967 free(ots);
2968
2969 return 0;
2970 }
2971
2972 static void
2973 append_port_stat(struct ofport *port, struct list *replies)
2974 {
2975 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2976
2977 calc_duration(port->created, time_msec(),
2978 &ops.duration_sec, &ops.duration_nsec);
2979
2980 /* Intentionally ignore return value, since errors will set
2981 * 'stats' to all-1s, which is correct for OpenFlow, and
2982 * netdev_get_stats() will log errors. */
2983 ofproto_port_get_stats(port, &ops.stats);
2984
2985 ofputil_append_port_stat(replies, &ops);
2986 }
2987
2988 static enum ofperr
2989 handle_port_stats_request(struct ofconn *ofconn,
2990 const struct ofp_header *request)
2991 {
2992 struct ofproto *p = ofconn_get_ofproto(ofconn);
2993 struct ofport *port;
2994 struct list replies;
2995 ofp_port_t port_no;
2996 enum ofperr error;
2997
2998 error = ofputil_decode_port_stats_request(request, &port_no);
2999 if (error) {
3000 return error;
3001 }
3002
3003 ofpmp_init(&replies, request);
3004 if (port_no != OFPP_ANY) {
3005 port = ofproto_get_port(p, port_no);
3006 if (port) {
3007 append_port_stat(port, &replies);
3008 }
3009 } else {
3010 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3011 append_port_stat(port, &replies);
3012 }
3013 }
3014
3015 ofconn_send_replies(ofconn, &replies);
3016 return 0;
3017 }
3018
3019 static enum ofperr
3020 handle_port_desc_stats_request(struct ofconn *ofconn,
3021 const struct ofp_header *request)
3022 {
3023 struct ofproto *p = ofconn_get_ofproto(ofconn);
3024 enum ofp_version version;
3025 struct ofport *port;
3026 struct list replies;
3027
3028 ofpmp_init(&replies, request);
3029
3030 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3031 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3032 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3033 }
3034
3035 ofconn_send_replies(ofconn, &replies);
3036 return 0;
3037 }
3038
3039 static uint32_t
3040 hash_cookie(ovs_be64 cookie)
3041 {
3042 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3043 (OVS_FORCE uint64_t)cookie);
3044 }
3045
3046 static void
3047 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3048 OVS_REQUIRES(ofproto_mutex)
3049 {
3050 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3051 hash_cookie(rule->flow_cookie));
3052 }
3053
3054 static void
3055 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3056 OVS_REQUIRES(ofproto_mutex)
3057 {
3058 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3059 }
3060
3061 static void
3062 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3063 ovs_be64 new_cookie)
3064 OVS_REQUIRES(ofproto_mutex)
3065 {
3066 if (new_cookie != rule->flow_cookie) {
3067 cookies_remove(ofproto, rule);
3068
3069 ovs_mutex_lock(&rule->mutex);
3070 rule->flow_cookie = new_cookie;
3071 ovs_mutex_unlock(&rule->mutex);
3072
3073 cookies_insert(ofproto, rule);
3074 }
3075 }
3076
3077 static void
3078 calc_duration(long long int start, long long int now,
3079 uint32_t *sec, uint32_t *nsec)
3080 {
3081 long long int msecs = now - start;
3082 *sec = msecs / 1000;
3083 *nsec = (msecs % 1000) * (1000 * 1000);
3084 }
3085
3086 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3087 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3088 static enum ofperr
3089 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3090 {
3091 return (table_id == 0xff || table_id < ofproto->n_tables
3092 ? 0
3093 : OFPERR_OFPBRC_BAD_TABLE_ID);
3094
3095 }
3096
3097 static struct oftable *
3098 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3099 {
3100 struct oftable *table;
3101
3102 for (table = &ofproto->tables[table_id];
3103 table < &ofproto->tables[ofproto->n_tables];
3104 table++) {
3105 if (!(table->flags & OFTABLE_HIDDEN)) {
3106 return table;
3107 }
3108 }
3109
3110 return NULL;
3111 }
3112
3113 static struct oftable *
3114 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3115 {
3116 if (table_id == 0xff) {
3117 return next_visible_table(ofproto, 0);
3118 } else if (table_id < ofproto->n_tables) {
3119 return &ofproto->tables[table_id];
3120 } else {
3121 return NULL;
3122 }
3123 }
3124
3125 static struct oftable *
3126 next_matching_table(const struct ofproto *ofproto,
3127 const struct oftable *table, uint8_t table_id)
3128 {
3129 return (table_id == 0xff
3130 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3131 : NULL);
3132 }
3133
3134 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3135 *
3136 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3137 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3138 *
3139 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3140 * only once, for that table. (This can be used to access tables marked
3141 * OFTABLE_HIDDEN.)
3142 *
3143 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3144 * entered at all. (Perhaps you should have validated TABLE_ID with
3145 * check_table_id().)
3146 *
3147 * All parameters are evaluated multiple times.
3148 */
3149 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3150 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3151 (TABLE) != NULL; \
3152 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3153
3154 /* Initializes 'criteria' in a straightforward way based on the other
3155 * parameters.
3156 *
3157 * For "loose" matching, the 'priority' parameter is unimportant and may be
3158 * supplied as 0. */
3159 static void
3160 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3161 const struct match *match, unsigned int priority,
3162 ovs_be64 cookie, ovs_be64 cookie_mask,
3163 ofp_port_t out_port, uint32_t out_group)
3164 {
3165 criteria->table_id = table_id;
3166 cls_rule_init(&criteria->cr, match, priority);
3167 criteria->cookie = cookie;
3168 criteria->cookie_mask = cookie_mask;
3169 criteria->out_port = out_port;
3170 criteria->out_group = out_group;
3171 }
3172
3173 static void
3174 rule_criteria_destroy(struct rule_criteria *criteria)
3175 {
3176 cls_rule_destroy(&criteria->cr);
3177 }
3178
3179 void
3180 rule_collection_init(struct rule_collection *rules)
3181 {
3182 rules->rules = rules->stub;
3183 rules->n = 0;
3184 rules->capacity = ARRAY_SIZE(rules->stub);
3185 }
3186
3187 void
3188 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3189 {
3190 if (rules->n >= rules->capacity) {
3191 size_t old_size, new_size;
3192
3193 old_size = rules->capacity * sizeof *rules->rules;
3194 rules->capacity *= 2;
3195 new_size = rules->capacity * sizeof *rules->rules;
3196
3197 if (rules->rules == rules->stub) {
3198 rules->rules = xmalloc(new_size);
3199 memcpy(rules->rules, rules->stub, old_size);
3200 } else {
3201 rules->rules = xrealloc(rules->rules, new_size);
3202 }
3203 }
3204
3205 rules->rules[rules->n++] = rule;
3206 }
3207
3208 void
3209 rule_collection_ref(struct rule_collection *rules)
3210 OVS_REQUIRES(ofproto_mutex)
3211 {
3212 size_t i;
3213
3214 for (i = 0; i < rules->n; i++) {
3215 ofproto_rule_ref(rules->rules[i]);
3216 }
3217 }
3218
3219 void
3220 rule_collection_unref(struct rule_collection *rules)
3221 {
3222 size_t i;
3223
3224 for (i = 0; i < rules->n; i++) {
3225 ofproto_rule_unref(rules->rules[i]);
3226 }
3227 }
3228
3229 void
3230 rule_collection_destroy(struct rule_collection *rules)
3231 {
3232 if (rules->rules != rules->stub) {
3233 free(rules->rules);
3234 }
3235 }
3236
3237 static enum ofperr
3238 collect_rule(struct rule *rule, const struct rule_criteria *c,
3239 struct rule_collection *rules)
3240 OVS_REQUIRES(ofproto_mutex)
3241 {
3242 if (ofproto_rule_is_hidden(rule)) {
3243 return 0;
3244 } else if (rule->pending) {
3245 return OFPROTO_POSTPONE;
3246 } else {
3247 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3248 && ofproto_rule_has_out_port(rule, c->out_port)
3249 && ofproto_rule_has_out_group(rule, c->out_group)
3250 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3251 rule_collection_add(rules, rule);
3252 }
3253 return 0;
3254 }
3255 }
3256
3257 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3258 * on classifiers rules are done in the "loose" way required for OpenFlow
3259 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3260 * 'rules'.
3261 *
3262 * Hidden rules are always omitted.
3263 *
3264 * Returns 0 on success, otherwise an OpenFlow error code. */
3265 static enum ofperr
3266 collect_rules_loose(struct ofproto *ofproto,
3267 const struct rule_criteria *criteria,
3268 struct rule_collection *rules)
3269 OVS_REQUIRES(ofproto_mutex)
3270 {
3271 struct oftable *table;
3272 enum ofperr error;
3273
3274 rule_collection_init(rules);
3275
3276 error = check_table_id(ofproto, criteria->table_id);
3277 if (error) {
3278 goto exit;
3279 }
3280
3281 if (criteria->cookie_mask == htonll(UINT64_MAX)) {
3282 struct rule *rule;
3283
3284 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3285 hash_cookie(criteria->cookie),
3286 &ofproto->cookies) {
3287 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3288 error = collect_rule(rule, criteria, rules);
3289 if (error) {
3290 break;
3291 }
3292 }
3293 }
3294 } else {
3295 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3296 struct cls_cursor cursor;
3297 struct rule *rule;
3298
3299 ovs_rwlock_rdlock(&table->cls.rwlock);
3300 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3301 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3302 error = collect_rule(rule, criteria, rules);
3303 if (error) {
3304 break;
3305 }
3306 }
3307 ovs_rwlock_unlock(&table->cls.rwlock);
3308 }
3309 }
3310
3311 exit:
3312 if (error) {
3313 rule_collection_destroy(rules);
3314 }
3315 return error;
3316 }
3317
3318 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3319 * on classifiers rules are done in the "strict" way required for OpenFlow
3320 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3321 * rules on list 'rules'.
3322 *
3323 * Hidden rules are always omitted.
3324 *
3325 * Returns 0 on success, otherwise an OpenFlow error code. */
3326 static enum ofperr
3327 collect_rules_strict(struct ofproto *ofproto,
3328 const struct rule_criteria *criteria,
3329 struct rule_collection *rules)
3330 OVS_REQUIRES(ofproto_mutex)
3331 {
3332 struct oftable *table;
3333 int error;
3334
3335 rule_collection_init(rules);
3336
3337 error = check_table_id(ofproto, criteria->table_id);
3338 if (error) {
3339 goto exit;
3340 }
3341
3342 if (criteria->cookie_mask == htonll(UINT64_MAX)) {
3343 struct rule *rule;
3344
3345 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3346 hash_cookie(criteria->cookie),
3347 &ofproto->cookies) {
3348 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3349 error = collect_rule(rule, criteria, rules);
3350 if (error) {
3351 break;
3352 }
3353 }
3354 }
3355 } else {
3356 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3357 struct rule *rule;
3358
3359 ovs_rwlock_rdlock(&table->cls.rwlock);
3360 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3361 &table->cls, &criteria->cr));
3362 ovs_rwlock_unlock(&table->cls.rwlock);
3363 if (rule) {
3364 error = collect_rule(rule, criteria, rules);
3365 if (error) {
3366 break;
3367 }
3368 }
3369 }
3370 }
3371
3372 exit:
3373 if (error) {
3374 rule_collection_destroy(rules);
3375 }
3376 return error;
3377 }
3378
3379 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3380 * forced into the range of a uint16_t. */
3381 static int
3382 age_secs(long long int age_ms)
3383 {
3384 return (age_ms < 0 ? 0
3385 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3386 : (unsigned int) age_ms / 1000);
3387 }
3388
3389 static enum ofperr
3390 handle_flow_stats_request(struct ofconn *ofconn,
3391 const struct ofp_header *request)
3392 OVS_EXCLUDED(ofproto_mutex)
3393 {
3394 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3395 struct ofputil_flow_stats_request fsr;
3396 struct rule_criteria criteria;
3397 struct rule_collection rules;
3398 struct list replies;
3399 enum ofperr error;
3400 size_t i;
3401
3402 error = ofputil_decode_flow_stats_request(&fsr, request);
3403 if (error) {
3404 return error;
3405 }
3406
3407 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3408 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3409
3410 ovs_mutex_lock(&ofproto_mutex);
3411 error = collect_rules_loose(ofproto, &criteria, &rules);
3412 rule_criteria_destroy(&criteria);
3413 if (!error) {
3414 rule_collection_ref(&rules);
3415 }
3416 ovs_mutex_unlock(&ofproto_mutex);
3417
3418 if (error) {
3419 return error;
3420 }
3421
3422 ofpmp_init(&replies, request);
3423 for (i = 0; i < rules.n; i++) {
3424 struct rule *rule = rules.rules[i];
3425 long long int now = time_msec();
3426 struct ofputil_flow_stats fs;
3427 long long int created, used, modified;
3428 struct rule_actions *actions;
3429 enum ofputil_flow_mod_flags flags;
3430
3431 ovs_mutex_lock(&rule->mutex);
3432 fs.cookie = rule->flow_cookie;
3433 fs.idle_timeout = rule->idle_timeout;
3434 fs.hard_timeout = rule->hard_timeout;
3435 created = rule->created;
3436 used = rule->used;
3437 modified = rule->modified;
3438 actions = rule_get_actions__(rule);
3439 flags = rule->flags;
3440 ovs_mutex_unlock(&rule->mutex);
3441
3442 minimatch_expand(&rule->cr.match, &fs.match);
3443 fs.table_id = rule->table_id;
3444 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3445 fs.priority = rule->cr.priority;
3446 fs.idle_age = age_secs(now - used);
3447 fs.hard_age = age_secs(now - modified);
3448 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3449 &fs.byte_count);
3450 fs.ofpacts = actions->ofpacts;
3451 fs.ofpacts_len = actions->ofpacts_len;
3452
3453 fs.flags = flags;
3454 ofputil_append_flow_stats_reply(&fs, &replies);
3455
3456 rule_actions_unref(actions);
3457 }
3458
3459 rule_collection_unref(&rules);
3460 rule_collection_destroy(&rules);
3461
3462 ofconn_send_replies(ofconn, &replies);
3463
3464 return 0;
3465 }
3466
3467 static void
3468 flow_stats_ds(struct rule *rule, struct ds *results)
3469 {
3470 uint64_t packet_count, byte_count;
3471 struct rule_actions *actions;
3472 long long int created;
3473
3474 rule->ofproto->ofproto_class->rule_get_stats(rule,
3475 &packet_count, &byte_count);
3476
3477 ovs_mutex_lock(&rule->mutex);
3478 actions = rule_get_actions__(rule);
3479 created = rule->created;
3480 ovs_mutex_unlock(&rule->mutex);
3481
3482 if (rule->table_id != 0) {
3483 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3484 }
3485 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3486 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3487 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3488 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3489 cls_rule_format(&rule->cr, results);
3490 ds_put_char(results, ',');
3491
3492 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3493
3494 ds_put_cstr(results, "\n");
3495
3496 rule_actions_unref(actions);
3497 }
3498
3499 /* Adds a pretty-printed description of all flows to 'results', including
3500 * hidden flows (e.g., set up by in-band control). */
3501 void
3502 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3503 {
3504 struct oftable *table;
3505
3506 OFPROTO_FOR_EACH_TABLE (table, p) {
3507 struct cls_cursor cursor;
3508 struct rule *rule;
3509
3510 ovs_rwlock_rdlock(&table->cls.rwlock);
3511 cls_cursor_init(&cursor, &table->cls, NULL);
3512 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3513 flow_stats_ds(rule, results);
3514 }
3515 ovs_rwlock_unlock(&table->cls.rwlock);
3516 }
3517 }
3518
3519 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3520 * '*engine_type' and '*engine_id', respectively. */
3521 void
3522 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3523 uint8_t *engine_type, uint8_t *engine_id)
3524 {
3525 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3526 }
3527
3528 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3529 * true if the port's CFM status was successfully stored into '*status'.
3530 * Returns false if the port did not have CFM configured, in which case
3531 * '*status' is indeterminate.
3532 *
3533 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3534 bool
3535 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3536 struct ofproto_cfm_status *status)
3537 {
3538 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3539 return (ofport
3540 && ofproto->ofproto_class->get_cfm_status
3541 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3542 }
3543
3544 static enum ofperr
3545 handle_aggregate_stats_request(struct ofconn *ofconn,
3546 const struct ofp_header *oh)
3547 OVS_EXCLUDED(ofproto_mutex)
3548 {
3549 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3550 struct ofputil_flow_stats_request request;
3551 struct ofputil_aggregate_stats stats;
3552 bool unknown_packets, unknown_bytes;
3553 struct rule_criteria criteria;
3554 struct rule_collection rules;
3555 struct ofpbuf *reply;
3556 enum ofperr error;
3557 size_t i;
3558
3559 error = ofputil_decode_flow_stats_request(&request, oh);
3560 if (error) {
3561 return error;
3562 }
3563
3564 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3565 request.cookie, request.cookie_mask,
3566 request.out_port, request.out_group);
3567
3568 ovs_mutex_lock(&ofproto_mutex);
3569 error = collect_rules_loose(ofproto, &criteria, &rules);
3570 rule_criteria_destroy(&criteria);
3571 if (!error) {
3572 rule_collection_ref(&rules);
3573 }
3574 ovs_mutex_unlock(&ofproto_mutex);
3575
3576 if (error) {
3577 return error;
3578 }
3579
3580 memset(&stats, 0, sizeof stats);
3581 unknown_packets = unknown_bytes = false;
3582 for (i = 0; i < rules.n; i++) {
3583 struct rule *rule = rules.rules[i];
3584 uint64_t packet_count;
3585 uint64_t byte_count;
3586
3587 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3588 &byte_count);
3589
3590 if (packet_count == UINT64_MAX) {
3591 unknown_packets = true;
3592 } else {
3593 stats.packet_count += packet_count;
3594 }
3595
3596 if (byte_count == UINT64_MAX) {
3597 unknown_bytes = true;
3598 } else {
3599 stats.byte_count += byte_count;
3600 }
3601
3602 stats.flow_count++;
3603 }
3604 if (unknown_packets) {
3605 stats.packet_count = UINT64_MAX;
3606 }
3607 if (unknown_bytes) {
3608 stats.byte_count = UINT64_MAX;
3609 }
3610
3611 rule_collection_unref(&rules);
3612 rule_collection_destroy(&rules);
3613
3614 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3615 ofconn_send_reply(ofconn, reply);
3616
3617 return 0;
3618 }
3619
3620 struct queue_stats_cbdata {
3621 struct ofport *ofport;
3622 struct list replies;
3623 long long int now;
3624 };
3625
3626 static void
3627 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3628 const struct netdev_queue_stats *stats)
3629 {
3630 struct ofputil_queue_stats oqs;
3631
3632 oqs.port_no = cbdata->ofport->pp.port_no;
3633 oqs.queue_id = queue_id;
3634 oqs.tx_bytes = stats->tx_bytes;
3635 oqs.tx_packets = stats->tx_packets;
3636 oqs.tx_errors = stats->tx_errors;
3637 if (stats->created != LLONG_MIN) {
3638 calc_duration(stats->created, cbdata->now,
3639 &oqs.duration_sec, &oqs.duration_nsec);
3640 } else {
3641 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3642 }
3643 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3644 }
3645
3646 static void
3647 handle_queue_stats_dump_cb(uint32_t queue_id,
3648 struct netdev_queue_stats *stats,
3649 void *cbdata_)
3650 {
3651 struct queue_stats_cbdata *cbdata = cbdata_;
3652
3653 put_queue_stats(cbdata, queue_id, stats);
3654 }
3655
3656 static enum ofperr
3657 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3658 struct queue_stats_cbdata *cbdata)
3659 {
3660 cbdata->ofport = port;
3661 if (queue_id == OFPQ_ALL) {
3662 netdev_dump_queue_stats(port->netdev,
3663 handle_queue_stats_dump_cb, cbdata);
3664 } else {
3665 struct netdev_queue_stats stats;
3666
3667 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3668 put_queue_stats(cbdata, queue_id, &stats);
3669 } else {
3670 return OFPERR_OFPQOFC_BAD_QUEUE;
3671 }
3672 }
3673 return 0;
3674 }
3675
3676 static enum ofperr
3677 handle_queue_stats_request(struct ofconn *ofconn,
3678 const struct ofp_header *rq)
3679 {
3680 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3681 struct queue_stats_cbdata cbdata;
3682 struct ofport *port;
3683 enum ofperr error;
3684 struct ofputil_queue_stats_request oqsr;
3685
3686 COVERAGE_INC(ofproto_queue_req);
3687
3688 ofpmp_init(&cbdata.replies, rq);
3689 cbdata.now = time_msec();
3690
3691 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3692 if (error) {
3693 return error;
3694 }
3695
3696 if (oqsr.port_no == OFPP_ANY) {
3697 error = OFPERR_OFPQOFC_BAD_QUEUE;
3698 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3699 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3700 error = 0;
3701 }
3702 }
3703 } else {
3704 port = ofproto_get_port(ofproto, oqsr.port_no);
3705 error = (port
3706 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3707 : OFPERR_OFPQOFC_BAD_PORT);
3708 }
3709 if (!error) {
3710 ofconn_send_replies(ofconn, &cbdata.replies);
3711 } else {
3712 ofpbuf_list_delete(&cbdata.replies);
3713 }
3714
3715 return error;
3716 }
3717
3718 static bool
3719 is_flow_deletion_pending(const struct ofproto *ofproto,
3720 const struct cls_rule *cls_rule,
3721 uint8_t table_id)
3722 OVS_REQUIRES(ofproto_mutex)
3723 {
3724 if (!hmap_is_empty(&ofproto->deletions)) {
3725 struct ofoperation *op;
3726
3727 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3728 cls_rule_hash(cls_rule, table_id),
3729 &ofproto->deletions) {
3730 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3731 return true;
3732 }
3733 }
3734 }
3735
3736 return false;
3737 }
3738
3739 static bool
3740 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3741 OVS_REQUIRES(ofproto_mutex)
3742 OVS_NO_THREAD_SAFETY_ANALYSIS
3743 {
3744 return classifier_count(&table->cls) + extra_space > table->max_flows;
3745 }
3746
3747 static enum ofperr
3748 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3749 unsigned int extra_space)
3750 OVS_REQUIRES(ofproto_mutex)
3751 {
3752 while (should_evict_a_rule(table, extra_space)) {
3753 struct rule *rule;
3754
3755 if (!choose_rule_to_evict(table, &rule)) {
3756 return OFPERR_OFPFMFC_TABLE_FULL;
3757 } else if (rule->pending) {
3758 return OFPROTO_POSTPONE;
3759 } else {
3760 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3761 delete_flow__(rule, group, OFPRR_EVICTION);
3762 ofopgroup_submit(group);
3763 }
3764 }
3765
3766 return 0;
3767 }
3768
3769 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3770 * in which no matching flow already exists in the flow table.
3771 *
3772 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3773 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3774 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3775 * initiated now but may be retried later.
3776 *
3777 * The caller retains ownership of 'fm->ofpacts'.
3778 *
3779 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3780 * if any. */
3781 static enum ofperr
3782 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3783 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3784 OVS_REQUIRES(ofproto_mutex)
3785 {
3786 struct oftable *table;
3787 struct ofopgroup *group;
3788 struct cls_rule cr;
3789 struct rule *rule;
3790 uint8_t table_id;
3791 int error;
3792
3793 error = check_table_id(ofproto, fm->table_id);
3794 if (error) {
3795 return error;
3796 }
3797
3798 /* Pick table. */
3799 if (fm->table_id == 0xff) {
3800 if (ofproto->ofproto_class->rule_choose_table) {
3801 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3802 &fm->match,
3803 &table_id);
3804 if (error) {
3805 return error;
3806 }
3807 ovs_assert(table_id < ofproto->n_tables);
3808 } else {
3809 table_id = 0;
3810 }
3811 } else if (fm->table_id < ofproto->n_tables) {
3812 table_id = fm->table_id;
3813 } else {
3814 return OFPERR_OFPBRC_BAD_TABLE_ID;
3815 }
3816
3817 table = &ofproto->tables[table_id];
3818
3819 if (table->flags & OFTABLE_READONLY) {
3820 return OFPERR_OFPBRC_EPERM;
3821 }
3822
3823 cls_rule_init(&cr, &fm->match, fm->priority);
3824
3825 /* Transform "add" into "modify" if there's an existing identical flow. */
3826 ovs_rwlock_rdlock(&table->cls.rwlock);
3827 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3828 ovs_rwlock_unlock(&table->cls.rwlock);
3829 if (rule) {
3830 cls_rule_destroy(&cr);
3831 if (!rule_is_modifiable(rule)) {
3832 return OFPERR_OFPBRC_EPERM;
3833 } else if (rule->pending) {
3834 return OFPROTO_POSTPONE;
3835 } else {
3836 struct rule_collection rules;
3837
3838 rule_collection_init(&rules);
3839 rule_collection_add(&rules, rule);
3840 fm->modify_cookie = true;
3841 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3842 rule_collection_destroy(&rules);
3843
3844 return error;
3845 }
3846 }
3847
3848 /* Verify actions. */
3849 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3850 &fm->match.flow, table_id);
3851 if (error) {
3852 cls_rule_destroy(&cr);
3853 return error;
3854 }
3855
3856 /* Serialize against pending deletion. */
3857 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3858 cls_rule_destroy(&cr);
3859 return OFPROTO_POSTPONE;
3860 }
3861
3862 /* Check for overlap, if requested. */
3863 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3864 bool overlaps;
3865
3866 ovs_rwlock_rdlock(&table->cls.rwlock);
3867 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3868 ovs_rwlock_unlock(&table->cls.rwlock);
3869
3870 if (overlaps) {
3871 cls_rule_destroy(&cr);
3872 return OFPERR_OFPFMFC_OVERLAP;
3873 }
3874 }
3875
3876 /* If necessary, evict an existing rule to clear out space. */
3877 error = evict_rules_from_table(ofproto, table, 1);
3878 if (error) {
3879 cls_rule_destroy(&cr);
3880 return error;
3881 }
3882
3883 /* Allocate new rule. */
3884 rule = ofproto->ofproto_class->rule_alloc();
3885 if (!rule) {
3886 cls_rule_destroy(&cr);
3887 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3888 ofproto->name, ovs_strerror(error));
3889 return ENOMEM;
3890 }
3891
3892 /* Initialize base state. */
3893 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
3894 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
3895 atomic_init(&rule->ref_count, 1);
3896 rule->pending = NULL;
3897 rule->flow_cookie = fm->new_cookie;
3898 rule->created = rule->modified = rule->used = time_msec();
3899
3900 ovs_mutex_init(&rule->mutex);
3901 ovs_mutex_lock(&rule->mutex);
3902 rule->idle_timeout = fm->idle_timeout;
3903 rule->hard_timeout = fm->hard_timeout;
3904 ovs_mutex_unlock(&rule->mutex);
3905
3906 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
3907 rule->flags = fm->flags & OFPUTIL_FF_STATE;
3908 rule->actions = rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len);
3909 list_init(&rule->meter_list_node);
3910 rule->eviction_group = NULL;
3911 list_init(&rule->expirable);
3912 rule->monitor_flags = 0;
3913 rule->add_seqno = 0;
3914 rule->modify_seqno = 0;
3915
3916 /* Construct rule, initializing derived state. */
3917 error = ofproto->ofproto_class->rule_construct(rule);
3918 if (error) {
3919 ofproto_rule_destroy__(rule);
3920 return error;
3921 }
3922
3923 /* Insert rule. */
3924 oftable_insert_rule(rule);
3925
3926 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3927 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3928 ofproto->ofproto_class->rule_insert(rule);
3929 ofopgroup_submit(group);
3930
3931 return error;
3932 }
3933 \f
3934 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3935
3936 /* Modifies the rules listed in 'rules', changing their actions to match those
3937 * in 'fm'.
3938 *
3939 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3940 * if any.
3941 *
3942 * Returns 0 on success, otherwise an OpenFlow error code. */
3943 static enum ofperr
3944 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3945 struct ofputil_flow_mod *fm, const struct ofp_header *request,
3946 const struct rule_collection *rules)
3947 OVS_REQUIRES(ofproto_mutex)
3948 {
3949 enum ofoperation_type type;
3950 struct ofopgroup *group;
3951 enum ofperr error;
3952 size_t i;
3953
3954 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
3955 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3956 error = OFPERR_OFPBRC_EPERM;
3957 for (i = 0; i < rules->n; i++) {
3958 struct rule *rule = rules->rules[i];
3959 struct ofoperation *op;
3960 bool actions_changed;
3961 bool reset_counters;
3962
3963 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
3964
3965 if (rule_is_modifiable(rule)) {
3966 /* At least one rule is modifiable, don't report EPERM error. */
3967 error = 0;
3968 } else {
3969 continue;
3970 }
3971
3972 /* Verify actions. */
3973 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
3974 u16_to_ofp(ofproto->max_ports), rule->table_id);
3975 if (error) {
3976 return error;
3977 }
3978
3979 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3980 rule->actions->ofpacts,
3981 rule->actions->ofpacts_len);
3982
3983 op = ofoperation_create(group, rule, type, 0);
3984
3985 if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
3986 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3987 }
3988 if (type == OFOPERATION_REPLACE) {
3989 ovs_mutex_lock(&rule->mutex);
3990 rule->idle_timeout = fm->idle_timeout;
3991 rule->hard_timeout = fm->hard_timeout;
3992 ovs_mutex_unlock(&rule->mutex);
3993
3994 rule->flags = fm->flags & OFPUTIL_FF_STATE;
3995 if (fm->idle_timeout || fm->hard_timeout) {
3996 if (!rule->eviction_group) {
3997 eviction_group_add_rule(rule);
3998 }
3999 } else {
4000 eviction_group_remove_rule(rule);
4001 }
4002 }
4003
4004 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4005 if (actions_changed || reset_counters) {
4006 struct rule_actions *new_actions;
4007
4008 op->actions = rule->actions;
4009 new_actions = rule_actions_create(ofproto,
4010 fm->ofpacts, fm->ofpacts_len);
4011
4012 ovs_mutex_lock(&rule->mutex);
4013 rule->actions = new_actions;
4014 ovs_mutex_unlock(&rule->mutex);
4015
4016 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4017 reset_counters);
4018 } else {
4019 ofoperation_complete(op, 0);
4020 }
4021 }
4022 ofopgroup_submit(group);
4023
4024 return error;
4025 }
4026
4027 static enum ofperr
4028 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4029 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4030 OVS_REQUIRES(ofproto_mutex)
4031 {
4032 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
4033 return 0;
4034 }
4035 return add_flow(ofproto, ofconn, fm, request);
4036 }
4037
4038 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4039 * failure.
4040 *
4041 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4042 * if any. */
4043 static enum ofperr
4044 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4045 struct ofputil_flow_mod *fm,
4046 const struct ofp_header *request)
4047 OVS_REQUIRES(ofproto_mutex)
4048 {
4049 struct rule_criteria criteria;
4050 struct rule_collection rules;
4051 int error;
4052
4053 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4054 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4055 error = collect_rules_loose(ofproto, &criteria, &rules);
4056 rule_criteria_destroy(&criteria);
4057
4058 if (!error) {
4059 error = (rules.n > 0
4060 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4061 : modify_flows_add(ofproto, ofconn, fm, request));
4062 }
4063
4064 rule_collection_destroy(&rules);
4065
4066 return error;
4067 }
4068
4069 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4070 * code on failure.
4071 *
4072 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4073 * if any. */
4074 static enum ofperr
4075 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4076 struct ofputil_flow_mod *fm,
4077 const struct ofp_header *request)
4078 OVS_REQUIRES(ofproto_mutex)
4079 {
4080 struct rule_criteria criteria;
4081 struct rule_collection rules;
4082 int error;
4083
4084 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4085 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4086 error = collect_rules_strict(ofproto, &criteria, &rules);
4087 rule_criteria_destroy(&criteria);
4088
4089 if (!error) {
4090 if (rules.n == 0) {
4091 error = modify_flows_add(ofproto, ofconn, fm, request);
4092 } else if (rules.n == 1) {
4093 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4094 }
4095 }
4096
4097 rule_collection_destroy(&rules);
4098
4099 return error;
4100 }
4101 \f
4102 /* OFPFC_DELETE implementation. */
4103
4104 static void
4105 delete_flow__(struct rule *rule, struct ofopgroup *group,
4106 enum ofp_flow_removed_reason reason)
4107 OVS_REQUIRES(ofproto_mutex)
4108 {
4109 struct ofproto *ofproto = rule->ofproto;
4110
4111 ofproto_rule_send_removed(rule, reason);
4112
4113 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4114 oftable_remove_rule(rule);
4115 ofproto->ofproto_class->rule_delete(rule);
4116 }
4117
4118 /* Deletes the rules listed in 'rules'.
4119 *
4120 * Returns 0 on success, otherwise an OpenFlow error code. */
4121 static enum ofperr
4122 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4123 const struct ofp_header *request,
4124 const struct rule_collection *rules,
4125 enum ofp_flow_removed_reason reason)
4126 OVS_REQUIRES(ofproto_mutex)
4127 {
4128 struct ofopgroup *group;
4129 size_t i;
4130
4131 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4132 for (i = 0; i < rules->n; i++) {
4133 delete_flow__(rules->rules[i], group, reason);
4134 }
4135 ofopgroup_submit(group);
4136
4137 return 0;
4138 }
4139
4140 /* Implements OFPFC_DELETE. */
4141 static enum ofperr
4142 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4143 const struct ofputil_flow_mod *fm,
4144 const struct ofp_header *request)
4145 OVS_REQUIRES(ofproto_mutex)
4146 {
4147 struct rule_criteria criteria;
4148 struct rule_collection rules;
4149 enum ofperr error;
4150
4151 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4152 fm->cookie, fm->cookie_mask,
4153 fm->out_port, fm->out_group);
4154 error = collect_rules_loose(ofproto, &criteria, &rules);
4155 rule_criteria_destroy(&criteria);
4156
4157 if (!error && rules.n > 0) {
4158 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4159 }
4160 rule_collection_destroy(&rules);
4161
4162 return error;
4163 }
4164
4165 /* Implements OFPFC_DELETE_STRICT. */
4166 static enum ofperr
4167 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4168 const struct ofputil_flow_mod *fm,
4169 const struct ofp_header *request)
4170 OVS_REQUIRES(ofproto_mutex)
4171 {
4172 struct rule_criteria criteria;
4173 struct rule_collection rules;
4174 enum ofperr error;
4175
4176 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4177 fm->cookie, fm->cookie_mask,
4178 fm->out_port, fm->out_group);
4179 error = collect_rules_strict(ofproto, &criteria, &rules);
4180 rule_criteria_destroy(&criteria);
4181
4182 if (!error && rules.n > 0) {
4183 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4184 }
4185 rule_collection_destroy(&rules);
4186
4187 return error;
4188 }
4189
4190 static void
4191 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4192 OVS_REQUIRES(ofproto_mutex)
4193 {
4194 struct ofputil_flow_removed fr;
4195
4196 if (ofproto_rule_is_hidden(rule) ||
4197 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4198 return;
4199 }
4200
4201 minimatch_expand(&rule->cr.match, &fr.match);
4202 fr.priority = rule->cr.priority;
4203 fr.cookie = rule->flow_cookie;
4204 fr.reason = reason;
4205 fr.table_id = rule->table_id;
4206 calc_duration(rule->created, time_msec(),
4207 &fr.duration_sec, &fr.duration_nsec);
4208 ovs_mutex_lock(&rule->mutex);
4209 fr.idle_timeout = rule->idle_timeout;
4210 fr.hard_timeout = rule->hard_timeout;
4211 ovs_mutex_unlock(&rule->mutex);
4212 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4213 &fr.byte_count);
4214
4215 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4216 }
4217
4218 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4219 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4220 * ofproto.
4221 *
4222 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4223 * NULL).
4224 *
4225 * ofproto implementation ->run() functions should use this function to expire
4226 * OpenFlow flows. */
4227 void
4228 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4229 OVS_REQUIRES(ofproto_mutex)
4230 {
4231 struct ofproto *ofproto = rule->ofproto;
4232
4233 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4234 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4235
4236 ofproto_rule_send_removed(rule, reason);
4237 ofproto_rule_delete__(ofproto, rule);
4238 }
4239
4240 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4241 * means "infinite". */
4242 static void
4243 reduce_timeout(uint16_t max, uint16_t *timeout)
4244 {
4245 if (max && (!*timeout || *timeout > max)) {
4246 *timeout = max;
4247 }
4248 }
4249
4250 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4251 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4252 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4253 *
4254 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4255 void
4256 ofproto_rule_reduce_timeouts(struct rule *rule,
4257 uint16_t idle_timeout, uint16_t hard_timeout)
4258 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4259 {
4260 if (!idle_timeout && !hard_timeout) {
4261 return;
4262 }
4263
4264 ovs_mutex_lock(&ofproto_mutex);
4265 if (list_is_empty(&rule->expirable)) {
4266 list_insert(&rule->ofproto->expirable, &rule->expirable);
4267 }
4268 ovs_mutex_unlock(&ofproto_mutex);
4269
4270 ovs_mutex_lock(&rule->mutex);
4271 reduce_timeout(idle_timeout, &rule->idle_timeout);
4272 reduce_timeout(hard_timeout, &rule->hard_timeout);
4273 ovs_mutex_unlock(&rule->mutex);
4274 }
4275 \f
4276 static enum ofperr
4277 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4278 OVS_EXCLUDED(ofproto_mutex)
4279 {
4280 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4281 struct ofputil_flow_mod fm;
4282 uint64_t ofpacts_stub[1024 / 8];
4283 struct ofpbuf ofpacts;
4284 enum ofperr error;
4285 long long int now;
4286
4287 error = reject_slave_controller(ofconn);
4288 if (error) {
4289 goto exit;
4290 }
4291
4292 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4293 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4294 &ofpacts);
4295 if (!error) {
4296 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4297 }
4298 if (error) {
4299 goto exit_free_ofpacts;
4300 }
4301
4302 /* Record the operation for logging a summary report. */
4303 switch (fm.command) {
4304 case OFPFC_ADD:
4305 ofproto->n_add++;
4306 break;
4307
4308 case OFPFC_MODIFY:
4309 case OFPFC_MODIFY_STRICT:
4310 ofproto->n_modify++;
4311 break;
4312
4313 case OFPFC_DELETE:
4314 case OFPFC_DELETE_STRICT:
4315 ofproto->n_delete++;
4316 break;
4317 }
4318
4319 now = time_msec();
4320 if (ofproto->next_op_report == LLONG_MAX) {
4321 ofproto->first_op = now;
4322 ofproto->next_op_report = MAX(now + 10 * 1000,
4323 ofproto->op_backoff);
4324 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4325 }
4326 ofproto->last_op = now;
4327
4328 exit_free_ofpacts:
4329 ofpbuf_uninit(&ofpacts);
4330 exit:
4331 return error;
4332 }
4333
4334 static enum ofperr
4335 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4336 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4337 OVS_EXCLUDED(ofproto_mutex)
4338 {
4339 enum ofperr error;
4340
4341 ovs_mutex_lock(&ofproto_mutex);
4342 if (ofproto->n_pending < 50) {
4343 switch (fm->command) {
4344 case OFPFC_ADD:
4345 error = add_flow(ofproto, ofconn, fm, oh);
4346 break;
4347
4348 case OFPFC_MODIFY:
4349 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4350 break;
4351
4352 case OFPFC_MODIFY_STRICT:
4353 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4354 break;
4355
4356 case OFPFC_DELETE:
4357 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4358 break;
4359
4360 case OFPFC_DELETE_STRICT:
4361 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4362 break;
4363
4364 default:
4365 if (fm->command > 0xff) {
4366 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4367 "flow_mod_table_id extension is not enabled",
4368 ofproto->name);
4369 }
4370 error = OFPERR_OFPFMFC_BAD_COMMAND;
4371 break;
4372 }
4373 } else {
4374 ovs_assert(!list_is_empty(&ofproto->pending));
4375 error = OFPROTO_POSTPONE;
4376 }
4377 ovs_mutex_unlock(&ofproto_mutex);
4378
4379 run_rule_executes(ofproto);
4380 return error;
4381 }
4382
4383 static enum ofperr
4384 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4385 {
4386 struct ofputil_role_request request;
4387 struct ofputil_role_request reply;
4388 struct ofpbuf *buf;
4389 enum ofperr error;
4390
4391 error = ofputil_decode_role_message(oh, &request);
4392 if (error) {
4393 return error;
4394 }
4395
4396 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4397 if (ofconn_get_role(ofconn) != request.role
4398 && ofconn_has_pending_opgroups(ofconn)) {
4399 return OFPROTO_POSTPONE;
4400 }
4401
4402 if (request.have_generation_id
4403 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4404 return OFPERR_OFPRRFC_STALE;
4405 }
4406
4407 ofconn_set_role(ofconn, request.role);
4408 }
4409
4410 reply.role = ofconn_get_role(ofconn);
4411 reply.have_generation_id = ofconn_get_master_election_id(
4412 ofconn, &reply.generation_id);
4413 buf = ofputil_encode_role_reply(oh, &reply);
4414 ofconn_send_reply(ofconn, buf);
4415
4416 return 0;
4417 }
4418
4419 static enum ofperr
4420 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4421 const struct ofp_header *oh)
4422 {
4423 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4424 enum ofputil_protocol cur, next;
4425
4426 cur = ofconn_get_protocol(ofconn);
4427 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4428 ofconn_set_protocol(ofconn, next);
4429
4430 return 0;
4431 }
4432
4433 static enum ofperr
4434 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4435 {
4436 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4437 enum ofputil_protocol cur, next;
4438 enum ofputil_protocol next_base;
4439
4440 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4441 if (!next_base) {
4442 return OFPERR_OFPBRC_EPERM;
4443 }
4444
4445 cur = ofconn_get_protocol(ofconn);
4446 next = ofputil_protocol_set_base(cur, next_base);
4447 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4448 /* Avoid sending async messages in surprising protocol. */
4449 return OFPROTO_POSTPONE;
4450 }
4451
4452 ofconn_set_protocol(ofconn, next);
4453 return 0;
4454 }
4455
4456 static enum ofperr
4457 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4458 const struct ofp_header *oh)
4459 {
4460 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4461 uint32_t format;
4462
4463 format = ntohl(msg->format);
4464 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4465 return OFPERR_OFPBRC_EPERM;
4466 }
4467
4468 if (format != ofconn_get_packet_in_format(ofconn)
4469 && ofconn_has_pending_opgroups(ofconn)) {
4470 /* Avoid sending async message in surprsing packet in format. */
4471 return OFPROTO_POSTPONE;
4472 }
4473
4474 ofconn_set_packet_in_format(ofconn, format);
4475 return 0;
4476 }
4477
4478 static enum ofperr
4479 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4480 {
4481 const struct nx_async_config *msg = ofpmsg_body(oh);
4482 uint32_t master[OAM_N_TYPES];
4483 uint32_t slave[OAM_N_TYPES];
4484
4485 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4486 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4487 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4488
4489 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4490 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4491 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4492
4493 ofconn_set_async_config(ofconn, master, slave);
4494 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4495 !ofconn_get_miss_send_len(ofconn)) {
4496 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4497 }
4498
4499 return 0;
4500 }
4501
4502 static enum ofperr
4503 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4504 {
4505 struct ofpbuf *buf;
4506 uint32_t master[OAM_N_TYPES];
4507 uint32_t slave[OAM_N_TYPES];
4508 struct nx_async_config *msg;
4509
4510 ofconn_get_async_config(ofconn, master, slave);
4511 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4512 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4513
4514 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4515 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4516 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4517
4518 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4519 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4520 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4521
4522 ofconn_send_reply(ofconn, buf);
4523
4524 return 0;
4525 }
4526
4527 static enum ofperr
4528 handle_nxt_set_controller_id(struct ofconn *ofconn,
4529 const struct ofp_header *oh)
4530 {
4531 const struct nx_controller_id *nci = ofpmsg_body(oh);
4532
4533 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4534 return OFPERR_NXBRC_MUST_BE_ZERO;
4535 }
4536
4537 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4538 return 0;
4539 }
4540
4541 static enum ofperr
4542 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4543 {
4544 struct ofpbuf *buf;
4545
4546 if (ofconn_has_pending_opgroups(ofconn)) {
4547 return OFPROTO_POSTPONE;
4548 }
4549
4550 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4551 ? OFPRAW_OFPT10_BARRIER_REPLY
4552 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4553 ofconn_send_reply(ofconn, buf);
4554 return 0;
4555 }
4556
4557 static void
4558 ofproto_compose_flow_refresh_update(const struct rule *rule,
4559 enum nx_flow_monitor_flags flags,
4560 struct list *msgs)
4561 OVS_REQUIRES(ofproto_mutex)
4562 {
4563 struct ofoperation *op = rule->pending;
4564 const struct rule_actions *actions;
4565 struct ofputil_flow_update fu;
4566 struct match match;
4567
4568 if (op && op->type == OFOPERATION_ADD) {
4569 /* We'll report the final flow when the operation completes. Reporting
4570 * it now would cause a duplicate report later. */
4571 return;
4572 }
4573
4574 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4575 ? NXFME_ADDED : NXFME_MODIFIED);
4576 fu.reason = 0;
4577 ovs_mutex_lock(&rule->mutex);
4578 fu.idle_timeout = rule->idle_timeout;
4579 fu.hard_timeout = rule->hard_timeout;
4580 ovs_mutex_unlock(&rule->mutex);
4581 fu.table_id = rule->table_id;
4582 fu.cookie = rule->flow_cookie;
4583 minimatch_expand(&rule->cr.match, &match);
4584 fu.match = &match;
4585 fu.priority = rule->cr.priority;
4586
4587 if (!(flags & NXFMF_ACTIONS)) {
4588 actions = NULL;
4589 } else if (!op) {
4590 actions = rule->actions;
4591 } else {
4592 /* An operation is in progress. Use the previous version of the flow's
4593 * actions, so that when the operation commits we report the change. */
4594 switch (op->type) {
4595 case OFOPERATION_ADD:
4596 NOT_REACHED();
4597
4598 case OFOPERATION_MODIFY:
4599 case OFOPERATION_REPLACE:
4600 actions = op->actions ? op->actions : rule->actions;
4601 break;
4602
4603 case OFOPERATION_DELETE:
4604 actions = rule->actions;
4605 break;
4606
4607 default:
4608 NOT_REACHED();
4609 }
4610 }
4611 fu.ofpacts = actions ? actions->ofpacts : NULL;
4612 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4613
4614 if (list_is_empty(msgs)) {
4615 ofputil_start_flow_update(msgs);
4616 }
4617 ofputil_append_flow_update(&fu, msgs);
4618 }
4619
4620 void
4621 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4622 struct list *msgs)
4623 OVS_REQUIRES(ofproto_mutex)
4624 {
4625 size_t i;
4626
4627 for (i = 0; i < rules->n; i++) {
4628 struct rule *rule = rules->rules[i];
4629 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4630 rule->monitor_flags = 0;
4631
4632 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4633 }
4634 }
4635
4636 static void
4637 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4638 struct rule *rule, uint64_t seqno,
4639 struct rule_collection *rules)
4640 OVS_REQUIRES(ofproto_mutex)
4641 {
4642 enum nx_flow_monitor_flags update;
4643
4644 if (ofproto_rule_is_hidden(rule)) {
4645 return;
4646 }
4647
4648 if (!(rule->pending
4649 ? ofoperation_has_out_port(rule->pending, m->out_port)
4650 : ofproto_rule_has_out_port(rule, m->out_port))) {
4651 return;
4652 }
4653
4654 if (seqno) {
4655 if (rule->add_seqno > seqno) {
4656 update = NXFMF_ADD | NXFMF_MODIFY;
4657 } else if (rule->modify_seqno > seqno) {
4658 update = NXFMF_MODIFY;
4659 } else {
4660 return;
4661 }
4662
4663 if (!(m->flags & update)) {
4664 return;
4665 }
4666 } else {
4667 update = NXFMF_INITIAL;
4668 }
4669
4670 if (!rule->monitor_flags) {
4671 rule_collection_add(rules, rule);
4672 }
4673 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4674 }
4675
4676 static void
4677 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4678 uint64_t seqno,
4679 struct rule_collection *rules)
4680 OVS_REQUIRES(ofproto_mutex)
4681 {
4682 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4683 const struct ofoperation *op;
4684 const struct oftable *table;
4685 struct cls_rule target;
4686
4687 cls_rule_init_from_minimatch(&target, &m->match, 0);
4688 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4689 struct cls_cursor cursor;
4690 struct rule *rule;
4691
4692 ovs_rwlock_rdlock(&table->cls.rwlock);
4693 cls_cursor_init(&cursor, &table->cls, &target);
4694 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4695 ovs_assert(!rule->pending); /* XXX */
4696 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4697 }
4698 ovs_rwlock_unlock(&table->cls.rwlock);
4699 }
4700
4701 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4702 struct rule *rule = op->rule;
4703
4704 if (((m->table_id == 0xff
4705 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4706 : m->table_id == rule->table_id))
4707 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4708 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4709 }
4710 }
4711 cls_rule_destroy(&target);
4712 }
4713
4714 static void
4715 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4716 struct rule_collection *rules)
4717 OVS_REQUIRES(ofproto_mutex)
4718 {
4719 if (m->flags & NXFMF_INITIAL) {
4720 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4721 }
4722 }
4723
4724 void
4725 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4726 uint64_t seqno, struct rule_collection *rules)
4727 OVS_REQUIRES(ofproto_mutex)
4728 {
4729 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4730 }
4731
4732 static enum ofperr
4733 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4734 OVS_EXCLUDED(ofproto_mutex)
4735 {
4736 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4737 struct ofmonitor **monitors;
4738 size_t n_monitors, allocated_monitors;
4739 struct rule_collection rules;
4740 struct list replies;
4741 enum ofperr error;
4742 struct ofpbuf b;
4743 size_t i;
4744
4745 error = 0;
4746 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4747 monitors = NULL;
4748 n_monitors = allocated_monitors = 0;
4749
4750 ovs_mutex_lock(&ofproto_mutex);
4751 for (;;) {
4752 struct ofputil_flow_monitor_request request;
4753 struct ofmonitor *m;
4754 int retval;
4755
4756 retval = ofputil_decode_flow_monitor_request(&request, &b);
4757 if (retval == EOF) {
4758 break;
4759 } else if (retval) {
4760 error = retval;
4761 goto error;
4762 }
4763
4764 if (request.table_id != 0xff
4765 && request.table_id >= ofproto->n_tables) {
4766 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4767 goto error;
4768 }
4769
4770 error = ofmonitor_create(&request, ofconn, &m);
4771 if (error) {
4772 goto error;
4773 }
4774
4775 if (n_monitors >= allocated_monitors) {
4776 monitors = x2nrealloc(monitors, &allocated_monitors,
4777 sizeof *monitors);
4778 }
4779 monitors[n_monitors++] = m;
4780 }
4781
4782 rule_collection_init(&rules);
4783 for (i = 0; i < n_monitors; i++) {
4784 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4785 }
4786
4787 ofpmp_init(&replies, oh);
4788 ofmonitor_compose_refresh_updates(&rules, &replies);
4789 ovs_mutex_unlock(&ofproto_mutex);
4790
4791 rule_collection_destroy(&rules);
4792
4793 ofconn_send_replies(ofconn, &replies);
4794 free(monitors);
4795
4796 return 0;
4797
4798 error:
4799 ovs_mutex_unlock(&ofproto_mutex);
4800
4801 for (i = 0; i < n_monitors; i++) {
4802 ofmonitor_destroy(monitors[i]);
4803 }
4804 free(monitors);
4805 return error;
4806 }
4807
4808 static enum ofperr
4809 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4810 OVS_EXCLUDED(ofproto_mutex)
4811 {
4812 struct ofmonitor *m;
4813 enum ofperr error;
4814 uint32_t id;
4815
4816 id = ofputil_decode_flow_monitor_cancel(oh);
4817
4818 ovs_mutex_lock(&ofproto_mutex);
4819 m = ofmonitor_lookup(ofconn, id);
4820 if (m) {
4821 ofmonitor_destroy(m);
4822 error = 0;
4823 } else {
4824 error = OFPERR_NXBRC_FM_BAD_ID;
4825 }
4826 ovs_mutex_unlock(&ofproto_mutex);
4827
4828 return error;
4829 }
4830
4831 /* Meters implementation.
4832 *
4833 * Meter table entry, indexed by the OpenFlow meter_id.
4834 * These are always dynamically allocated to allocate enough space for
4835 * the bands.
4836 * 'created' is used to compute the duration for meter stats.
4837 * 'list rules' is needed so that we can delete the dependent rules when the
4838 * meter table entry is deleted.
4839 * 'provider_meter_id' is for the provider's private use.
4840 */
4841 struct meter {
4842 long long int created; /* Time created. */
4843 struct list rules; /* List of "struct rule_dpif"s. */
4844 ofproto_meter_id provider_meter_id;
4845 uint16_t flags; /* Meter flags. */
4846 uint16_t n_bands; /* Number of meter bands. */
4847 struct ofputil_meter_band *bands;
4848 };
4849
4850 /*
4851 * This is used in instruction validation at flow set-up time,
4852 * as flows may not use non-existing meters.
4853 * Return value of UINT32_MAX signifies an invalid meter.
4854 */
4855 static uint32_t
4856 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
4857 {
4858 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4859 const struct meter *meter = ofproto->meters[of_meter_id];
4860 if (meter) {
4861 return meter->provider_meter_id.uint32;
4862 }
4863 }
4864 return UINT32_MAX;
4865 }
4866
4867 static void
4868 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4869 {
4870 free(meter->bands);
4871
4872 meter->flags = config->flags;
4873 meter->n_bands = config->n_bands;
4874 meter->bands = xmemdup(config->bands,
4875 config->n_bands * sizeof *meter->bands);
4876 }
4877
4878 static struct meter *
4879 meter_create(const struct ofputil_meter_config *config,
4880 ofproto_meter_id provider_meter_id)
4881 {
4882 struct meter *meter;
4883
4884 meter = xzalloc(sizeof *meter);
4885 meter->provider_meter_id = provider_meter_id;
4886 meter->created = time_msec();
4887 list_init(&meter->rules);
4888
4889 meter_update(meter, config);
4890
4891 return meter;
4892 }
4893
4894 static void
4895 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4896 OVS_REQUIRES(ofproto_mutex)
4897 {
4898 uint32_t mid;
4899 for (mid = first; mid <= last; ++mid) {
4900 struct meter *meter = ofproto->meters[mid];
4901 if (meter) {
4902 ofproto->meters[mid] = NULL;
4903 ofproto->ofproto_class->meter_del(ofproto,
4904 meter->provider_meter_id);
4905 free(meter->bands);
4906 free(meter);
4907 }
4908 }
4909 }
4910
4911 static enum ofperr
4912 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4913 {
4914 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4915 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4916 enum ofperr error;
4917
4918 if (*meterp) {
4919 return OFPERR_OFPMMFC_METER_EXISTS;
4920 }
4921
4922 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4923 &mm->meter);
4924 if (!error) {
4925 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4926 *meterp = meter_create(&mm->meter, provider_meter_id);
4927 }
4928 return error;
4929 }
4930
4931 static enum ofperr
4932 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4933 {
4934 struct meter *meter = ofproto->meters[mm->meter.meter_id];
4935 enum ofperr error;
4936 uint32_t provider_meter_id;
4937
4938 if (!meter) {
4939 return OFPERR_OFPMMFC_UNKNOWN_METER;
4940 }
4941
4942 provider_meter_id = meter->provider_meter_id.uint32;
4943 error = ofproto->ofproto_class->meter_set(ofproto,
4944 &meter->provider_meter_id,
4945 &mm->meter);
4946 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
4947 if (!error) {
4948 meter_update(meter, &mm->meter);
4949 }
4950 return error;
4951 }
4952
4953 static enum ofperr
4954 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
4955 struct ofputil_meter_mod *mm)
4956 OVS_EXCLUDED(ofproto_mutex)
4957 {
4958 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4959 uint32_t meter_id = mm->meter.meter_id;
4960 struct rule_collection rules;
4961 enum ofperr error = 0;
4962 uint32_t first, last;
4963
4964 if (meter_id == OFPM13_ALL) {
4965 first = 1;
4966 last = ofproto->meter_features.max_meters;
4967 } else {
4968 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4969 return 0;
4970 }
4971 first = last = meter_id;
4972 }
4973
4974 /* First delete the rules that use this meter. If any of those rules are
4975 * currently being modified, postpone the whole operation until later. */
4976 rule_collection_init(&rules);
4977 ovs_mutex_lock(&ofproto_mutex);
4978 for (meter_id = first; meter_id <= last; ++meter_id) {
4979 struct meter *meter = ofproto->meters[meter_id];
4980 if (meter && !list_is_empty(&meter->rules)) {
4981 struct rule *rule;
4982
4983 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
4984 if (rule->pending) {
4985 error = OFPROTO_POSTPONE;
4986 goto exit;
4987 }
4988 rule_collection_add(&rules, rule);
4989 }
4990 }
4991 }
4992 if (rules.n > 0) {
4993 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
4994 }
4995
4996 /* Delete the meters. */
4997 meter_delete(ofproto, first, last);
4998
4999 exit:
5000 ovs_mutex_unlock(&ofproto_mutex);
5001 rule_collection_destroy(&rules);
5002
5003 return error;
5004 }
5005
5006 static enum ofperr
5007 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5008 {
5009 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5010 struct ofputil_meter_mod mm;
5011 uint64_t bands_stub[256 / 8];
5012 struct ofpbuf bands;
5013 uint32_t meter_id;
5014 enum ofperr error;
5015
5016 error = reject_slave_controller(ofconn);
5017 if (error) {
5018 return error;
5019 }
5020
5021 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5022
5023 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5024 if (error) {
5025 goto exit_free_bands;
5026 }
5027
5028 meter_id = mm.meter.meter_id;
5029
5030 if (mm.command != OFPMC13_DELETE) {
5031 /* Fails also when meters are not implemented by the provider. */
5032 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5033 error = OFPERR_OFPMMFC_INVALID_METER;
5034 goto exit_free_bands;
5035 } else if (meter_id > ofproto->meter_features.max_meters) {
5036 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5037 goto exit_free_bands;
5038 }
5039 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5040 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5041 goto exit_free_bands;
5042 }
5043 }
5044
5045 switch (mm.command) {
5046 case OFPMC13_ADD:
5047 error = handle_add_meter(ofproto, &mm);
5048 break;
5049
5050 case OFPMC13_MODIFY:
5051 error = handle_modify_meter(ofproto, &mm);
5052 break;
5053
5054 case OFPMC13_DELETE:
5055 error = handle_delete_meter(ofconn, oh, &mm);
5056 break;
5057
5058 default:
5059 error = OFPERR_OFPMMFC_BAD_COMMAND;
5060 break;
5061 }
5062
5063 exit_free_bands:
5064 ofpbuf_uninit(&bands);
5065 return error;
5066 }
5067
5068 static enum ofperr
5069 handle_meter_features_request(struct ofconn *ofconn,
5070 const struct ofp_header *request)
5071 {
5072 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5073 struct ofputil_meter_features features;
5074 struct ofpbuf *b;
5075
5076 if (ofproto->ofproto_class->meter_get_features) {
5077 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5078 } else {
5079 memset(&features, 0, sizeof features);
5080 }
5081 b = ofputil_encode_meter_features_reply(&features, request);
5082
5083 ofconn_send_reply(ofconn, b);
5084 return 0;
5085 }
5086
5087 static enum ofperr
5088 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5089 enum ofptype type)
5090 {
5091 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5092 struct list replies;
5093 uint64_t bands_stub[256 / 8];
5094 struct ofpbuf bands;
5095 uint32_t meter_id, first, last;
5096
5097 ofputil_decode_meter_request(request, &meter_id);
5098
5099 if (meter_id == OFPM13_ALL) {
5100 first = 1;
5101 last = ofproto->meter_features.max_meters;
5102 } else {
5103 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5104 !ofproto->meters[meter_id]) {
5105 return OFPERR_OFPMMFC_UNKNOWN_METER;
5106 }
5107 first = last = meter_id;
5108 }
5109
5110 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5111 ofpmp_init(&replies, request);
5112
5113 for (meter_id = first; meter_id <= last; ++meter_id) {
5114 struct meter *meter = ofproto->meters[meter_id];
5115 if (!meter) {
5116 continue; /* Skip non-existing meters. */
5117 }
5118 if (type == OFPTYPE_METER_STATS_REQUEST) {
5119 struct ofputil_meter_stats stats;
5120
5121 stats.meter_id = meter_id;
5122
5123 /* Provider sets the packet and byte counts, we do the rest. */
5124 stats.flow_count = list_size(&meter->rules);
5125 calc_duration(meter->created, time_msec(),
5126 &stats.duration_sec, &stats.duration_nsec);
5127 stats.n_bands = meter->n_bands;
5128 ofpbuf_clear(&bands);
5129 stats.bands
5130 = ofpbuf_put_uninit(&bands,
5131 meter->n_bands * sizeof *stats.bands);
5132
5133 if (!ofproto->ofproto_class->meter_get(ofproto,
5134 meter->provider_meter_id,
5135 &stats)) {
5136 ofputil_append_meter_stats(&replies, &stats);
5137 }
5138 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5139 struct ofputil_meter_config config;
5140
5141 config.meter_id = meter_id;
5142 config.flags = meter->flags;
5143 config.n_bands = meter->n_bands;
5144 config.bands = meter->bands;
5145 ofputil_append_meter_config(&replies, &config);
5146 }
5147 }
5148
5149 ofconn_send_replies(ofconn, &replies);
5150 ofpbuf_uninit(&bands);
5151 return 0;
5152 }
5153
5154 bool
5155 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5156 struct ofgroup **group)
5157 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5158 {
5159 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5160 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5161 hash_int(group_id, 0), &ofproto->groups) {
5162 if ((*group)->group_id == group_id) {
5163 ovs_rwlock_rdlock(&(*group)->rwlock);
5164 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5165 return true;
5166 }
5167 }
5168 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5169 return false;
5170 }
5171
5172 void
5173 ofproto_group_release(struct ofgroup *group)
5174 OVS_RELEASES(group->rwlock)
5175 {
5176 ovs_rwlock_unlock(&group->rwlock);
5177 }
5178
5179 static bool
5180 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5181 struct ofgroup **group)
5182 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5183 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5184 {
5185 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5186 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5187 hash_int(group_id, 0), &ofproto->groups) {
5188 if ((*group)->group_id == group_id) {
5189 ovs_rwlock_wrlock(&(*group)->rwlock);
5190 return true;
5191 }
5192 }
5193 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5194 return false;
5195 }
5196
5197 static bool
5198 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5199 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5200 {
5201 struct ofgroup *grp;
5202
5203 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5204 hash_int(group_id, 0), &ofproto->groups) {
5205 if (grp->group_id == group_id) {
5206 return true;
5207 }
5208 }
5209 return false;
5210 }
5211
5212 static void
5213 append_group_stats(struct ofgroup *group, struct list *replies)
5214 OVS_REQ_RDLOCK(group->rwlock)
5215 {
5216 struct ofputil_group_stats ogs;
5217 struct ofproto *ofproto = group->ofproto;
5218 long long int now = time_msec();
5219 int error;
5220
5221 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5222
5223 error = (ofproto->ofproto_class->group_get_stats
5224 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5225 : EOPNOTSUPP);
5226 if (error) {
5227 ogs.ref_count = UINT32_MAX;
5228 ogs.packet_count = UINT64_MAX;
5229 ogs.byte_count = UINT64_MAX;
5230 ogs.n_buckets = group->n_buckets;
5231 memset(ogs.bucket_stats, 0xff,
5232 ogs.n_buckets * sizeof *ogs.bucket_stats);
5233 }
5234
5235 ogs.group_id = group->group_id;
5236 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5237
5238 ofputil_append_group_stats(replies, &ogs);
5239
5240 free(ogs.bucket_stats);
5241 }
5242
5243 static enum ofperr
5244 handle_group_stats_request(struct ofconn *ofconn,
5245 const struct ofp_header *request)
5246 {
5247 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5248 struct list replies;
5249 enum ofperr error;
5250 struct ofgroup *group;
5251 uint32_t group_id;
5252
5253 error = ofputil_decode_group_stats_request(request, &group_id);
5254 if (error) {
5255 return error;
5256 }
5257
5258 ofpmp_init(&replies, request);
5259
5260 if (group_id == OFPG_ALL) {
5261 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5262 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5263 ovs_rwlock_rdlock(&group->rwlock);
5264 append_group_stats(group, &replies);
5265 ovs_rwlock_unlock(&group->rwlock);
5266 }
5267 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5268 } else {
5269 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5270 append_group_stats(group, &replies);
5271 ofproto_group_release(group);
5272 }
5273 }
5274
5275 ofconn_send_replies(ofconn, &replies);
5276
5277 return 0;
5278 }
5279
5280 static enum ofperr
5281 handle_group_desc_stats_request(struct ofconn *ofconn,
5282 const struct ofp_header *request)
5283 {
5284 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5285 struct list replies;
5286 struct ofputil_group_desc gds;
5287 struct ofgroup *group;
5288
5289 ofpmp_init(&replies, request);
5290
5291 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5292 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5293 gds.group_id = group->group_id;
5294 gds.type = group->type;
5295 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5296 }
5297 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5298
5299 ofconn_send_replies(ofconn, &replies);
5300
5301 return 0;
5302 }
5303
5304 static enum ofperr
5305 handle_group_features_stats_request(struct ofconn *ofconn,
5306 const struct ofp_header *request)
5307 {
5308 struct ofproto *p = ofconn_get_ofproto(ofconn);
5309 struct ofpbuf *msg;
5310
5311 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5312 if (msg) {
5313 ofconn_send_reply(ofconn, msg);
5314 }
5315
5316 return 0;
5317 }
5318
5319 /* Implements OFPGC11_ADD
5320 * in which no matching flow already exists in the flow table.
5321 *
5322 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5323 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5324 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5325 * initiated now but may be retried later.
5326 *
5327 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5328 * ownership remains with the caller.
5329 *
5330 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5331 * if any. */
5332 static enum ofperr
5333 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5334 {
5335 struct ofgroup *ofgroup;
5336 enum ofperr error;
5337
5338 if (gm->group_id > OFPG_MAX) {
5339 return OFPERR_OFPGMFC_INVALID_GROUP;
5340 }
5341 if (gm->type > OFPGT11_FF) {
5342 return OFPERR_OFPGMFC_BAD_TYPE;
5343 }
5344
5345 /* Allocate new group and initialize it. */
5346 ofgroup = ofproto->ofproto_class->group_alloc();
5347 if (!ofgroup) {
5348 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5349 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5350 }
5351
5352 ovs_rwlock_init(&ofgroup->rwlock);
5353 ofgroup->ofproto = ofproto;
5354 ofgroup->group_id = gm->group_id;
5355 ofgroup->type = gm->type;
5356 ofgroup->created = ofgroup->modified = time_msec();
5357
5358 list_move(&ofgroup->buckets, &gm->buckets);
5359 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5360
5361 /* Construct called BEFORE any locks are held. */
5362 error = ofproto->ofproto_class->group_construct(ofgroup);
5363 if (error) {
5364 goto free_out;
5365 }
5366
5367 /* We wrlock as late as possible to minimize the time we jam any other
5368 * threads: No visible state changes before acquiring the lock. */
5369 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5370
5371 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5372 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5373 goto unlock_out;
5374 }
5375
5376 if (ofproto_group_exists(ofproto, gm->group_id)) {
5377 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5378 goto unlock_out;
5379 }
5380
5381 if (!error) {
5382 /* Insert new group. */
5383 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5384 hash_int(ofgroup->group_id, 0));
5385 ofproto->n_groups[ofgroup->type]++;
5386
5387 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5388 return error;
5389 }
5390
5391 unlock_out:
5392 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5393 ofproto->ofproto_class->group_destruct(ofgroup);
5394 free_out:
5395 ofputil_bucket_list_destroy(&ofgroup->buckets);
5396 ofproto->ofproto_class->group_dealloc(ofgroup);
5397
5398 return error;
5399 }
5400
5401 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5402 * failure.
5403 *
5404 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5405 * if any. */
5406 static enum ofperr
5407 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5408 {
5409 struct ofgroup *ofgroup;
5410 struct ofgroup *victim;
5411 enum ofperr error;
5412
5413 if (gm->group_id > OFPG_MAX) {
5414 return OFPERR_OFPGMFC_INVALID_GROUP;
5415 }
5416
5417 if (gm->type > OFPGT11_FF) {
5418 return OFPERR_OFPGMFC_BAD_TYPE;
5419 }
5420
5421 victim = ofproto->ofproto_class->group_alloc();
5422 if (!victim) {
5423 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5424 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5425 }
5426
5427 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5428 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5429 goto free_out;
5430 }
5431 /* Both group's and its container's write locks held now.
5432 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5433 if (ofgroup->type != gm->type
5434 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5435 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5436 goto unlock_out;
5437 }
5438
5439 *victim = *ofgroup;
5440 list_move(&victim->buckets, &ofgroup->buckets);
5441
5442 ofgroup->type = gm->type;
5443 list_move(&ofgroup->buckets, &gm->buckets);
5444 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5445
5446 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5447 if (!error) {
5448 ofputil_bucket_list_destroy(&victim->buckets);
5449 ofproto->n_groups[victim->type]--;
5450 ofproto->n_groups[ofgroup->type]++;
5451 ofgroup->modified = time_msec();
5452 } else {
5453 ofputil_bucket_list_destroy(&ofgroup->buckets);
5454
5455 *ofgroup = *victim;
5456 list_move(&ofgroup->buckets, &victim->buckets);
5457 }
5458
5459 unlock_out:
5460 ovs_rwlock_unlock(&ofgroup->rwlock);
5461 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5462 free_out:
5463 ofproto->ofproto_class->group_dealloc(victim);
5464 return error;
5465 }
5466
5467 static void
5468 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5469 OVS_RELEASES(ofproto->groups_rwlock)
5470 {
5471 /* Must wait until existing readers are done,
5472 * while holding the container's write lock at the same time. */
5473 ovs_rwlock_wrlock(&ofgroup->rwlock);
5474 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5475 /* No-one can find this group any more. */
5476 ofproto->n_groups[ofgroup->type]--;
5477 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5478
5479 ofproto->ofproto_class->group_destruct(ofgroup);
5480 ofputil_bucket_list_destroy(&ofgroup->buckets);
5481 ovs_rwlock_unlock(&ofgroup->rwlock);
5482 ovs_rwlock_destroy(&ofgroup->rwlock);
5483 ofproto->ofproto_class->group_dealloc(ofgroup);
5484 }
5485
5486 /* Implements OFPGC_DELETE. */
5487 static void
5488 delete_group(struct ofproto *ofproto, uint32_t group_id)
5489 {
5490 struct ofgroup *ofgroup;
5491
5492 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5493 if (group_id == OFPG_ALL) {
5494 for (;;) {
5495 struct hmap_node *node = hmap_first(&ofproto->groups);
5496 if (!node) {
5497 break;
5498 }
5499 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5500 delete_group__(ofproto, ofgroup);
5501 /* Lock for each node separately, so that we will not jam the
5502 * other threads for too long time. */
5503 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5504 }
5505 } else {
5506 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5507 hash_int(group_id, 0), &ofproto->groups) {
5508 if (ofgroup->group_id == group_id) {
5509 delete_group__(ofproto, ofgroup);
5510 return;
5511 }
5512 }
5513 }
5514 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5515 }
5516
5517 static enum ofperr
5518 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5519 {
5520 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5521 struct ofputil_group_mod gm;
5522 enum ofperr error;
5523
5524 error = reject_slave_controller(ofconn);
5525 if (error) {
5526 return error;
5527 }
5528
5529 error = ofputil_decode_group_mod(oh, &gm);
5530 if (error) {
5531 return error;
5532 }
5533
5534 switch (gm.command) {
5535 case OFPGC11_ADD:
5536 return add_group(ofproto, &gm);
5537
5538 case OFPGC11_MODIFY:
5539 return modify_group(ofproto, &gm);
5540
5541 case OFPGC11_DELETE:
5542 delete_group(ofproto, gm.group_id);
5543 return 0;
5544
5545 default:
5546 if (gm.command > OFPGC11_DELETE) {
5547 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5548 ofproto->name, gm.command);
5549 }
5550 return OFPERR_OFPGMFC_BAD_COMMAND;
5551 }
5552 }
5553
5554 static enum ofperr
5555 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5556 {
5557 struct ofputil_table_mod tm;
5558 enum ofperr error;
5559
5560 error = reject_slave_controller(ofconn);
5561 if (error) {
5562 return error;
5563 }
5564
5565 error = ofputil_decode_table_mod(oh, &tm);
5566 if (error) {
5567 return error;
5568 }
5569
5570 /* XXX Actual table mod support is not implemented yet. */
5571 return 0;
5572 }
5573
5574 static enum ofperr
5575 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5576 OVS_EXCLUDED(ofproto_mutex)
5577 {
5578 const struct ofp_header *oh = msg->data;
5579 enum ofptype type;
5580 enum ofperr error;
5581
5582 error = ofptype_decode(&type, oh);
5583 if (error) {
5584 return error;
5585 }
5586
5587 switch (type) {
5588 /* OpenFlow requests. */
5589 case OFPTYPE_ECHO_REQUEST:
5590 return handle_echo_request(ofconn, oh);
5591
5592 case OFPTYPE_FEATURES_REQUEST:
5593 return handle_features_request(ofconn, oh);
5594
5595 case OFPTYPE_GET_CONFIG_REQUEST:
5596 return handle_get_config_request(ofconn, oh);
5597
5598 case OFPTYPE_SET_CONFIG:
5599 return handle_set_config(ofconn, oh);
5600
5601 case OFPTYPE_PACKET_OUT:
5602 return handle_packet_out(ofconn, oh);
5603
5604 case OFPTYPE_PORT_MOD:
5605 return handle_port_mod(ofconn, oh);
5606
5607 case OFPTYPE_FLOW_MOD:
5608 return handle_flow_mod(ofconn, oh);
5609
5610 case OFPTYPE_GROUP_MOD:
5611 return handle_group_mod(ofconn, oh);
5612
5613 case OFPTYPE_TABLE_MOD:
5614 return handle_table_mod(ofconn, oh);
5615
5616 case OFPTYPE_METER_MOD:
5617 return handle_meter_mod(ofconn, oh);
5618
5619 case OFPTYPE_BARRIER_REQUEST:
5620 return handle_barrier_request(ofconn, oh);
5621
5622 case OFPTYPE_ROLE_REQUEST:
5623 return handle_role_request(ofconn, oh);
5624
5625 /* OpenFlow replies. */
5626 case OFPTYPE_ECHO_REPLY:
5627 return 0;
5628
5629 /* Nicira extension requests. */
5630 case OFPTYPE_FLOW_MOD_TABLE_ID:
5631 return handle_nxt_flow_mod_table_id(ofconn, oh);
5632
5633 case OFPTYPE_SET_FLOW_FORMAT:
5634 return handle_nxt_set_flow_format(ofconn, oh);
5635
5636 case OFPTYPE_SET_PACKET_IN_FORMAT:
5637 return handle_nxt_set_packet_in_format(ofconn, oh);
5638
5639 case OFPTYPE_SET_CONTROLLER_ID:
5640 return handle_nxt_set_controller_id(ofconn, oh);
5641
5642 case OFPTYPE_FLOW_AGE:
5643 /* Nothing to do. */
5644 return 0;
5645
5646 case OFPTYPE_FLOW_MONITOR_CANCEL:
5647 return handle_flow_monitor_cancel(ofconn, oh);
5648
5649 case OFPTYPE_SET_ASYNC_CONFIG:
5650 return handle_nxt_set_async_config(ofconn, oh);
5651
5652 case OFPTYPE_GET_ASYNC_REQUEST:
5653 return handle_nxt_get_async_request(ofconn, oh);
5654
5655 /* Statistics requests. */
5656 case OFPTYPE_DESC_STATS_REQUEST:
5657 return handle_desc_stats_request(ofconn, oh);
5658
5659 case OFPTYPE_FLOW_STATS_REQUEST:
5660 return handle_flow_stats_request(ofconn, oh);
5661
5662 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5663 return handle_aggregate_stats_request(ofconn, oh);
5664
5665 case OFPTYPE_TABLE_STATS_REQUEST:
5666 return handle_table_stats_request(ofconn, oh);
5667
5668 case OFPTYPE_PORT_STATS_REQUEST:
5669 return handle_port_stats_request(ofconn, oh);
5670
5671 case OFPTYPE_QUEUE_STATS_REQUEST:
5672 return handle_queue_stats_request(ofconn, oh);
5673
5674 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5675 return handle_port_desc_stats_request(ofconn, oh);
5676
5677 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5678 return handle_flow_monitor_request(ofconn, oh);
5679
5680 case OFPTYPE_METER_STATS_REQUEST:
5681 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5682 return handle_meter_request(ofconn, oh, type);
5683
5684 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5685 return handle_meter_features_request(ofconn, oh);
5686
5687 case OFPTYPE_GROUP_STATS_REQUEST:
5688 return handle_group_stats_request(ofconn, oh);
5689
5690 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5691 return handle_group_desc_stats_request(ofconn, oh);
5692
5693 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5694 return handle_group_features_stats_request(ofconn, oh);
5695
5696 /* FIXME: Change the following once they are implemented: */
5697 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5698 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5699 return OFPERR_OFPBRC_BAD_TYPE;
5700
5701 case OFPTYPE_HELLO:
5702 case OFPTYPE_ERROR:
5703 case OFPTYPE_FEATURES_REPLY:
5704 case OFPTYPE_GET_CONFIG_REPLY:
5705 case OFPTYPE_PACKET_IN:
5706 case OFPTYPE_FLOW_REMOVED:
5707 case OFPTYPE_PORT_STATUS:
5708 case OFPTYPE_BARRIER_REPLY:
5709 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5710 case OFPTYPE_DESC_STATS_REPLY:
5711 case OFPTYPE_FLOW_STATS_REPLY:
5712 case OFPTYPE_QUEUE_STATS_REPLY:
5713 case OFPTYPE_PORT_STATS_REPLY:
5714 case OFPTYPE_TABLE_STATS_REPLY:
5715 case OFPTYPE_AGGREGATE_STATS_REPLY:
5716 case OFPTYPE_PORT_DESC_STATS_REPLY:
5717 case OFPTYPE_ROLE_REPLY:
5718 case OFPTYPE_FLOW_MONITOR_PAUSED:
5719 case OFPTYPE_FLOW_MONITOR_RESUMED:
5720 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5721 case OFPTYPE_GET_ASYNC_REPLY:
5722 case OFPTYPE_GROUP_STATS_REPLY:
5723 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5724 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5725 case OFPTYPE_METER_STATS_REPLY:
5726 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5727 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5728 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5729 default:
5730 return OFPERR_OFPBRC_BAD_TYPE;
5731 }
5732 }
5733
5734 static bool
5735 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5736 OVS_EXCLUDED(ofproto_mutex)
5737 {
5738 int error = handle_openflow__(ofconn, ofp_msg);
5739 if (error && error != OFPROTO_POSTPONE) {
5740 ofconn_send_error(ofconn, ofp_msg->data, error);
5741 }
5742 COVERAGE_INC(ofproto_recv_openflow);
5743 return error != OFPROTO_POSTPONE;
5744 }
5745 \f
5746 /* Asynchronous operations. */
5747
5748 /* Creates and returns a new ofopgroup that is not associated with any
5749 * OpenFlow connection.
5750 *
5751 * The caller should add operations to the returned group with
5752 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5753 static struct ofopgroup *
5754 ofopgroup_create_unattached(struct ofproto *ofproto)
5755 OVS_REQUIRES(ofproto_mutex)
5756 {
5757 struct ofopgroup *group = xzalloc(sizeof *group);
5758 group->ofproto = ofproto;
5759 list_init(&group->ofproto_node);
5760 list_init(&group->ops);
5761 list_init(&group->ofconn_node);
5762 return group;
5763 }
5764
5765 /* Creates and returns a new ofopgroup for 'ofproto'.
5766 *
5767 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5768 * connection. The 'request' and 'buffer_id' arguments are ignored.
5769 *
5770 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5771 * If the ofopgroup eventually fails, then the error reply will include
5772 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5773 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5774 *
5775 * The caller should add operations to the returned group with
5776 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5777 static struct ofopgroup *
5778 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
5779 const struct ofp_header *request, uint32_t buffer_id)
5780 OVS_REQUIRES(ofproto_mutex)
5781 {
5782 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
5783 if (ofconn) {
5784 size_t request_len = ntohs(request->length);
5785
5786 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
5787
5788 ofconn_add_opgroup(ofconn, &group->ofconn_node);
5789 group->ofconn = ofconn;
5790 group->request = xmemdup(request, MIN(request_len, 64));
5791 group->buffer_id = buffer_id;
5792 }
5793 return group;
5794 }
5795
5796 /* Submits 'group' for processing.
5797 *
5798 * If 'group' contains no operations (e.g. none were ever added, or all of the
5799 * ones that were added completed synchronously), then it is destroyed
5800 * immediately. Otherwise it is added to the ofproto's list of pending
5801 * groups. */
5802 static void
5803 ofopgroup_submit(struct ofopgroup *group)
5804 OVS_REQUIRES(ofproto_mutex)
5805 {
5806 if (!group->n_running) {
5807 ofopgroup_complete(group);
5808 } else {
5809 list_push_back(&group->ofproto->pending, &group->ofproto_node);
5810 group->ofproto->n_pending++;
5811 }
5812 }
5813
5814 static void
5815 ofopgroup_complete(struct ofopgroup *group)
5816 OVS_REQUIRES(ofproto_mutex)
5817 {
5818 struct ofproto *ofproto = group->ofproto;
5819
5820 struct ofconn *abbrev_ofconn;
5821 ovs_be32 abbrev_xid;
5822
5823 struct ofoperation *op, *next_op;
5824 int error;
5825
5826 ovs_assert(!group->n_running);
5827
5828 error = 0;
5829 LIST_FOR_EACH (op, group_node, &group->ops) {
5830 if (op->error) {
5831 error = op->error;
5832 break;
5833 }
5834 }
5835
5836 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
5837 LIST_FOR_EACH (op, group_node, &group->ops) {
5838 if (op->type != OFOPERATION_DELETE) {
5839 struct ofpbuf *packet;
5840 ofp_port_t in_port;
5841
5842 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
5843 &packet, &in_port);
5844 if (packet) {
5845 struct rule_execute *re;
5846
5847 ovs_assert(!error);
5848
5849 ofproto_rule_ref(op->rule);
5850
5851 re = xmalloc(sizeof *re);
5852 re->rule = op->rule;
5853 re->in_port = in_port;
5854 re->packet = packet;
5855
5856 if (!guarded_list_push_back(&ofproto->rule_executes,
5857 &re->list_node, 1024)) {
5858 ofproto_rule_unref(op->rule);
5859 ofpbuf_delete(re->packet);
5860 free(re);
5861 }
5862 }
5863 break;
5864 }
5865 }
5866 }
5867
5868 if (!error && !list_is_empty(&group->ofconn_node)) {
5869 abbrev_ofconn = group->ofconn;
5870 abbrev_xid = group->request->xid;
5871 } else {
5872 abbrev_ofconn = NULL;
5873 abbrev_xid = htonl(0);
5874 }
5875 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
5876 struct rule *rule = op->rule;
5877
5878 /* We generally want to report the change to active OpenFlow flow
5879 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
5880
5881 - The operation failed.
5882
5883 - The affected rule is not visible to controllers.
5884
5885 - The operation's only effect was to update rule->modified. */
5886 if (!(op->error
5887 || ofproto_rule_is_hidden(rule)
5888 || (op->type == OFOPERATION_MODIFY
5889 && op->actions
5890 && rule->flow_cookie == op->flow_cookie))) {
5891 /* Check that we can just cast from ofoperation_type to
5892 * nx_flow_update_event. */
5893 enum nx_flow_update_event event_type;
5894
5895 switch (op->type) {
5896 case OFOPERATION_ADD:
5897 case OFOPERATION_REPLACE:
5898 event_type = NXFME_ADDED;
5899 break;
5900
5901 case OFOPERATION_DELETE:
5902 event_type = NXFME_DELETED;
5903 break;
5904
5905 case OFOPERATION_MODIFY:
5906 event_type = NXFME_MODIFIED;
5907 break;
5908
5909 default:
5910 NOT_REACHED();
5911 }
5912
5913 ofmonitor_report(ofproto->connmgr, rule, event_type,
5914 op->reason, abbrev_ofconn, abbrev_xid);
5915 }
5916
5917 rule->pending = NULL;
5918
5919 switch (op->type) {
5920 case OFOPERATION_ADD:
5921 if (!op->error) {
5922 uint16_t vid_mask;
5923
5924 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
5925 if (vid_mask == VLAN_VID_MASK) {
5926 if (ofproto->vlan_bitmap) {
5927 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
5928 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
5929 bitmap_set1(ofproto->vlan_bitmap, vid);
5930 ofproto->vlans_changed = true;
5931 }
5932 } else {
5933 ofproto->vlans_changed = true;
5934 }
5935 }
5936 } else {
5937 oftable_remove_rule(rule);
5938 ofproto_rule_unref(rule);
5939 }
5940 break;
5941
5942 case OFOPERATION_DELETE:
5943 ovs_assert(!op->error);
5944 ofproto_rule_unref(rule);
5945 op->rule = NULL;
5946 break;
5947
5948 case OFOPERATION_MODIFY:
5949 case OFOPERATION_REPLACE:
5950 if (!op->error) {
5951 long long int now = time_msec();
5952
5953 rule->modified = now;
5954 if (op->type == OFOPERATION_REPLACE) {
5955 rule->created = rule->used = now;
5956 }
5957 } else {
5958 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
5959 ovs_mutex_lock(&rule->mutex);
5960 rule->idle_timeout = op->idle_timeout;
5961 rule->hard_timeout = op->hard_timeout;
5962 ovs_mutex_unlock(&rule->mutex);
5963 if (op->actions) {
5964 struct rule_actions *old_actions;
5965
5966 ovs_mutex_lock(&rule->mutex);
5967 old_actions = rule->actions;
5968 rule->actions = op->actions;
5969 ovs_mutex_unlock(&rule->mutex);
5970
5971 op->actions = NULL;
5972 rule_actions_unref(old_actions);
5973 }
5974 rule->flags = op->flags;
5975 }
5976 break;
5977
5978 default:
5979 NOT_REACHED();
5980 }
5981
5982 ofoperation_destroy(op);
5983 }
5984
5985 ofmonitor_flush(ofproto->connmgr);
5986
5987 if (!list_is_empty(&group->ofproto_node)) {
5988 ovs_assert(ofproto->n_pending > 0);
5989 ofproto->n_pending--;
5990 list_remove(&group->ofproto_node);
5991 }
5992 if (!list_is_empty(&group->ofconn_node)) {
5993 list_remove(&group->ofconn_node);
5994 if (error) {
5995 ofconn_send_error(group->ofconn, group->request, error);
5996 }
5997 connmgr_retry(ofproto->connmgr);
5998 }
5999 free(group->request);
6000 free(group);
6001 }
6002
6003 /* Initiates a new operation on 'rule', of the specified 'type', within
6004 * 'group'. Prior to calling, 'rule' must not have any pending operation.
6005 *
6006 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
6007 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6008 *
6009 * Returns the newly created ofoperation (which is also available as
6010 * rule->pending). */
6011 static struct ofoperation *
6012 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6013 enum ofoperation_type type,
6014 enum ofp_flow_removed_reason reason)
6015 OVS_REQUIRES(ofproto_mutex)
6016 {
6017 struct ofproto *ofproto = group->ofproto;
6018 struct ofoperation *op;
6019
6020 ovs_assert(!rule->pending);
6021
6022 op = rule->pending = xzalloc(sizeof *op);
6023 op->group = group;
6024 list_push_back(&group->ops, &op->group_node);
6025 op->rule = rule;
6026 op->type = type;
6027 op->reason = reason;
6028 op->flow_cookie = rule->flow_cookie;
6029 ovs_mutex_lock(&rule->mutex);
6030 op->idle_timeout = rule->idle_timeout;
6031 op->hard_timeout = rule->hard_timeout;
6032 ovs_mutex_unlock(&rule->mutex);
6033 op->flags = rule->flags;
6034
6035 group->n_running++;
6036
6037 if (type == OFOPERATION_DELETE) {
6038 hmap_insert(&ofproto->deletions, &op->hmap_node,
6039 cls_rule_hash(&rule->cr, rule->table_id));
6040 }
6041
6042 return op;
6043 }
6044
6045 static void
6046 ofoperation_destroy(struct ofoperation *op)
6047 OVS_REQUIRES(ofproto_mutex)
6048 {
6049 struct ofopgroup *group = op->group;
6050
6051 if (op->rule) {
6052 op->rule->pending = NULL;
6053 }
6054 if (op->type == OFOPERATION_DELETE) {
6055 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6056 }
6057 list_remove(&op->group_node);
6058 rule_actions_unref(op->actions);
6059 free(op);
6060 }
6061
6062 /* Indicates that 'op' completed with status 'error', which is either 0 to
6063 * indicate success or an OpenFlow error code on failure.
6064 *
6065 * If 'error' is 0, indicating success, the operation will be committed
6066 * permanently to the flow table.
6067 *
6068 * If 'error' is nonzero, then generally the operation will be rolled back:
6069 *
6070 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6071 * restores the original rule. The caller must have uninitialized any
6072 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6073 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6074 * and 7 for the new rule, calling its ->rule_dealloc() function.
6075 *
6076 * - If 'op' is a "modify flow" operation, ofproto restores the original
6077 * actions.
6078 *
6079 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6080 * allowed to fail. It must always succeed.
6081 *
6082 * Please see the large comment in ofproto/ofproto-provider.h titled
6083 * "Asynchronous Operation Support" for more information. */
6084 void
6085 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6086 {
6087 struct ofopgroup *group = op->group;
6088
6089 ovs_assert(group->n_running > 0);
6090 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6091
6092 op->error = error;
6093 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6094 /* This function can be called from ->rule_construct(), in which case
6095 * ofproto_mutex is held, or it can be called from ->run(), in which
6096 * case ofproto_mutex is not held. But only in the latter case can we
6097 * arrive here, so we can safely take ofproto_mutex now. */
6098 ovs_mutex_lock(&ofproto_mutex);
6099 ovs_assert(op->rule->pending == op);
6100 ofopgroup_complete(group);
6101 ovs_mutex_unlock(&ofproto_mutex);
6102 }
6103 }
6104 \f
6105 static uint64_t
6106 pick_datapath_id(const struct ofproto *ofproto)
6107 {
6108 const struct ofport *port;
6109
6110 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6111 if (port) {
6112 uint8_t ea[ETH_ADDR_LEN];
6113 int error;
6114
6115 error = netdev_get_etheraddr(port->netdev, ea);
6116 if (!error) {
6117 return eth_addr_to_uint64(ea);
6118 }
6119 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6120 ofproto->name, netdev_get_name(port->netdev),
6121 ovs_strerror(error));
6122 }
6123 return ofproto->fallback_dpid;
6124 }
6125
6126 static uint64_t
6127 pick_fallback_dpid(void)
6128 {
6129 uint8_t ea[ETH_ADDR_LEN];
6130 eth_addr_nicira_random(ea);
6131 return eth_addr_to_uint64(ea);
6132 }
6133 \f
6134 /* Table overflow policy. */
6135
6136 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6137 * to NULL if the table is not configured to evict rules or if the table
6138 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6139 * or with no timeouts are not evictable.) */
6140 static bool
6141 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6142 OVS_REQUIRES(ofproto_mutex)
6143 {
6144 struct eviction_group *evg;
6145
6146 *rulep = NULL;
6147 if (!table->eviction_fields) {
6148 return false;
6149 }
6150
6151 /* In the common case, the outer and inner loops here will each be entered
6152 * exactly once:
6153 *
6154 * - The inner loop normally "return"s in its first iteration. If the
6155 * eviction group has any evictable rules, then it always returns in
6156 * some iteration.
6157 *
6158 * - The outer loop only iterates more than once if the largest eviction
6159 * group has no evictable rules.
6160 *
6161 * - The outer loop can exit only if table's 'max_flows' is all filled up
6162 * by unevictable rules. */
6163 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6164 struct rule *rule;
6165
6166 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6167 *rulep = rule;
6168 return true;
6169 }
6170 }
6171
6172 return false;
6173 }
6174
6175 /* Searches 'ofproto' for tables that have more flows than their configured
6176 * maximum and that have flow eviction enabled, and evicts as many flows as
6177 * necessary and currently feasible from them.
6178 *
6179 * This triggers only when an OpenFlow table has N flows in it and then the
6180 * client configures a maximum number of flows less than N. */
6181 static void
6182 ofproto_evict(struct ofproto *ofproto)
6183 {
6184 struct oftable *table;
6185
6186 ovs_mutex_lock(&ofproto_mutex);
6187 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6188 evict_rules_from_table(ofproto, table, 0);
6189 }
6190 ovs_mutex_unlock(&ofproto_mutex);
6191 }
6192 \f
6193 /* Eviction groups. */
6194
6195 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6196 * rules. The priority contains low-order random bits to ensure that eviction
6197 * groups with the same number of rules are prioritized randomly. */
6198 static uint32_t
6199 eviction_group_priority(size_t n_rules)
6200 {
6201 uint16_t size = MIN(UINT16_MAX, n_rules);
6202 return (size << 16) | random_uint16();
6203 }
6204
6205 /* Updates 'evg', an eviction_group within 'table', following a change that
6206 * adds or removes rules in 'evg'. */
6207 static void
6208 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6209 OVS_REQUIRES(ofproto_mutex)
6210 {
6211 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6212 eviction_group_priority(heap_count(&evg->rules)));
6213 }
6214
6215 /* Destroys 'evg', an eviction_group within 'table':
6216 *
6217 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6218 * rules themselves, just removes them from the eviction group.)
6219 *
6220 * - Removes 'evg' from 'table'.
6221 *
6222 * - Frees 'evg'. */
6223 static void
6224 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6225 OVS_REQUIRES(ofproto_mutex)
6226 {
6227 while (!heap_is_empty(&evg->rules)) {
6228 struct rule *rule;
6229
6230 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6231 rule->eviction_group = NULL;
6232 }
6233 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6234 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6235 heap_destroy(&evg->rules);
6236 free(evg);
6237 }
6238
6239 /* Removes 'rule' from its eviction group, if any. */
6240 static void
6241 eviction_group_remove_rule(struct rule *rule)
6242 OVS_REQUIRES(ofproto_mutex)
6243 {
6244 if (rule->eviction_group) {
6245 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6246 struct eviction_group *evg = rule->eviction_group;
6247
6248 rule->eviction_group = NULL;
6249 heap_remove(&evg->rules, &rule->evg_node);
6250 if (heap_is_empty(&evg->rules)) {
6251 eviction_group_destroy(table, evg);
6252 } else {
6253 eviction_group_resized(table, evg);
6254 }
6255 }
6256 }
6257
6258 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6259 * returns the hash value. */
6260 static uint32_t
6261 eviction_group_hash_rule(struct rule *rule)
6262 OVS_REQUIRES(ofproto_mutex)
6263 {
6264 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6265 const struct mf_subfield *sf;
6266 struct flow flow;
6267 uint32_t hash;
6268
6269 hash = table->eviction_group_id_basis;
6270 miniflow_expand(&rule->cr.match.flow, &flow);
6271 for (sf = table->eviction_fields;
6272 sf < &table->eviction_fields[table->n_eviction_fields];
6273 sf++)
6274 {
6275 if (mf_are_prereqs_ok(sf->field, &flow)) {
6276 union mf_value value;
6277
6278 mf_get_value(sf->field, &flow, &value);
6279 if (sf->ofs) {
6280 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6281 }
6282 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6283 unsigned int start = sf->ofs + sf->n_bits;
6284 bitwise_zero(&value, sf->field->n_bytes, start,
6285 sf->field->n_bytes * 8 - start);
6286 }
6287 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6288 } else {
6289 hash = hash_int(hash, 0);
6290 }
6291 }
6292
6293 return hash;
6294 }
6295
6296 /* Returns an eviction group within 'table' with the given 'id', creating one
6297 * if necessary. */
6298 static struct eviction_group *
6299 eviction_group_find(struct oftable *table, uint32_t id)
6300 OVS_REQUIRES(ofproto_mutex)
6301 {
6302 struct eviction_group *evg;
6303
6304 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6305 return evg;
6306 }
6307
6308 evg = xmalloc(sizeof *evg);
6309 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6310 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6311 eviction_group_priority(0));
6312 heap_init(&evg->rules);
6313
6314 return evg;
6315 }
6316
6317 /* Returns an eviction priority for 'rule'. The return value should be
6318 * interpreted so that higher priorities make a rule more attractive candidates
6319 * for eviction. */
6320 static uint32_t
6321 rule_eviction_priority(struct rule *rule)
6322 OVS_REQUIRES(ofproto_mutex)
6323 {
6324 long long int hard_expiration;
6325 long long int idle_expiration;
6326 long long int expiration;
6327 uint32_t expiration_offset;
6328
6329 /* Calculate time of expiration. */
6330 ovs_mutex_lock(&rule->mutex);
6331 hard_expiration = (rule->hard_timeout
6332 ? rule->modified + rule->hard_timeout * 1000
6333 : LLONG_MAX);
6334 idle_expiration = (rule->idle_timeout
6335 ? rule->used + rule->idle_timeout * 1000
6336 : LLONG_MAX);
6337 expiration = MIN(hard_expiration, idle_expiration);
6338 ovs_mutex_unlock(&rule->mutex);
6339 if (expiration == LLONG_MAX) {
6340 return 0;
6341 }
6342
6343 /* Calculate the time of expiration as a number of (approximate) seconds
6344 * after program startup.
6345 *
6346 * This should work OK for program runs that last UINT32_MAX seconds or
6347 * less. Therefore, please restart OVS at least once every 136 years. */
6348 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6349
6350 /* Invert the expiration offset because we're using a max-heap. */
6351 return UINT32_MAX - expiration_offset;
6352 }
6353
6354 /* Adds 'rule' to an appropriate eviction group for its oftable's
6355 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6356 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6357 * own).
6358 *
6359 * The caller must ensure that 'rule' is not already in an eviction group. */
6360 static void
6361 eviction_group_add_rule(struct rule *rule)
6362 OVS_REQUIRES(ofproto_mutex)
6363 {
6364 struct ofproto *ofproto = rule->ofproto;
6365 struct oftable *table = &ofproto->tables[rule->table_id];
6366 bool has_timeout;
6367
6368 ovs_mutex_lock(&rule->mutex);
6369 has_timeout = rule->hard_timeout || rule->idle_timeout;
6370 ovs_mutex_unlock(&rule->mutex);
6371
6372 if (table->eviction_fields && has_timeout) {
6373 struct eviction_group *evg;
6374
6375 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6376
6377 rule->eviction_group = evg;
6378 heap_insert(&evg->rules, &rule->evg_node,
6379 rule_eviction_priority(rule));
6380 eviction_group_resized(table, evg);
6381 }
6382 }
6383 \f
6384 /* oftables. */
6385
6386 /* Initializes 'table'. */
6387 static void
6388 oftable_init(struct oftable *table)
6389 {
6390 memset(table, 0, sizeof *table);
6391 classifier_init(&table->cls);
6392 table->max_flows = UINT_MAX;
6393 }
6394
6395 /* Destroys 'table', including its classifier and eviction groups.
6396 *
6397 * The caller is responsible for freeing 'table' itself. */
6398 static void
6399 oftable_destroy(struct oftable *table)
6400 {
6401 ovs_rwlock_rdlock(&table->cls.rwlock);
6402 ovs_assert(classifier_is_empty(&table->cls));
6403 ovs_rwlock_unlock(&table->cls.rwlock);
6404 oftable_disable_eviction(table);
6405 classifier_destroy(&table->cls);
6406 free(table->name);
6407 }
6408
6409 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6410 * string, then 'table' will use its default name.
6411 *
6412 * This only affects the name exposed for a table exposed through the OpenFlow
6413 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6414 static void
6415 oftable_set_name(struct oftable *table, const char *name)
6416 {
6417 if (name && name[0]) {
6418 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6419 if (!table->name || strncmp(name, table->name, len)) {
6420 free(table->name);
6421 table->name = xmemdup0(name, len);
6422 }
6423 } else {
6424 free(table->name);
6425 table->name = NULL;
6426 }
6427 }
6428
6429 /* oftables support a choice of two policies when adding a rule would cause the
6430 * number of flows in the table to exceed the configured maximum number: either
6431 * they can refuse to add the new flow or they can evict some existing flow.
6432 * This function configures the former policy on 'table'. */
6433 static void
6434 oftable_disable_eviction(struct oftable *table)
6435 OVS_REQUIRES(ofproto_mutex)
6436 {
6437 if (table->eviction_fields) {
6438 struct eviction_group *evg, *next;
6439
6440 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6441 &table->eviction_groups_by_id) {
6442 eviction_group_destroy(table, evg);
6443 }
6444 hmap_destroy(&table->eviction_groups_by_id);
6445 heap_destroy(&table->eviction_groups_by_size);
6446
6447 free(table->eviction_fields);
6448 table->eviction_fields = NULL;
6449 table->n_eviction_fields = 0;
6450 }
6451 }
6452
6453 /* oftables support a choice of two policies when adding a rule would cause the
6454 * number of flows in the table to exceed the configured maximum number: either
6455 * they can refuse to add the new flow or they can evict some existing flow.
6456 * This function configures the latter policy on 'table', with fairness based
6457 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6458 * 'n_fields' as 0 disables fairness.) */
6459 static void
6460 oftable_enable_eviction(struct oftable *table,
6461 const struct mf_subfield *fields, size_t n_fields)
6462 OVS_REQUIRES(ofproto_mutex)
6463 {
6464 struct cls_cursor cursor;
6465 struct rule *rule;
6466
6467 if (table->eviction_fields
6468 && n_fields == table->n_eviction_fields
6469 && (!n_fields
6470 || !memcmp(fields, table->eviction_fields,
6471 n_fields * sizeof *fields))) {
6472 /* No change. */
6473 return;
6474 }
6475
6476 oftable_disable_eviction(table);
6477
6478 table->n_eviction_fields = n_fields;
6479 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6480
6481 table->eviction_group_id_basis = random_uint32();
6482 hmap_init(&table->eviction_groups_by_id);
6483 heap_init(&table->eviction_groups_by_size);
6484
6485 ovs_rwlock_rdlock(&table->cls.rwlock);
6486 cls_cursor_init(&cursor, &table->cls, NULL);
6487 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6488 eviction_group_add_rule(rule);
6489 }
6490 ovs_rwlock_unlock(&table->cls.rwlock);
6491 }
6492
6493 /* Removes 'rule' from the oftable that contains it. */
6494 static void
6495 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6496 OVS_REQUIRES(ofproto_mutex)
6497 {
6498 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6499
6500 ovs_rwlock_wrlock(&cls->rwlock);
6501 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6502 ovs_rwlock_unlock(&cls->rwlock);
6503
6504 cookies_remove(ofproto, rule);
6505
6506 eviction_group_remove_rule(rule);
6507 if (!list_is_empty(&rule->expirable)) {
6508 list_remove(&rule->expirable);
6509 }
6510 if (!list_is_empty(&rule->meter_list_node)) {
6511 list_remove(&rule->meter_list_node);
6512 list_init(&rule->meter_list_node);
6513 }
6514 }
6515
6516 static void
6517 oftable_remove_rule(struct rule *rule)
6518 OVS_REQUIRES(ofproto_mutex)
6519 {
6520 oftable_remove_rule__(rule->ofproto, rule);
6521 }
6522
6523 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6524 * the same cls_rule. */
6525 static void
6526 oftable_insert_rule(struct rule *rule)
6527 OVS_REQUIRES(ofproto_mutex)
6528 {
6529 struct ofproto *ofproto = rule->ofproto;
6530 struct oftable *table = &ofproto->tables[rule->table_id];
6531 bool may_expire;
6532
6533 ovs_mutex_lock(&rule->mutex);
6534 may_expire = rule->hard_timeout || rule->idle_timeout;
6535 ovs_mutex_unlock(&rule->mutex);
6536
6537 if (may_expire) {
6538 list_insert(&ofproto->expirable, &rule->expirable);
6539 }
6540
6541 cookies_insert(ofproto, rule);
6542
6543 if (rule->actions->provider_meter_id != UINT32_MAX) {
6544 uint32_t meter_id = ofpacts_get_meter(rule->actions->ofpacts,
6545 rule->actions->ofpacts_len);
6546 struct meter *meter = ofproto->meters[meter_id];
6547 list_insert(&meter->rules, &rule->meter_list_node);
6548 }
6549 ovs_rwlock_wrlock(&table->cls.rwlock);
6550 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6551 ovs_rwlock_unlock(&table->cls.rwlock);
6552 eviction_group_add_rule(rule);
6553 }
6554 \f
6555 /* unixctl commands. */
6556
6557 struct ofproto *
6558 ofproto_lookup(const char *name)
6559 {
6560 struct ofproto *ofproto;
6561
6562 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6563 &all_ofprotos) {
6564 if (!strcmp(ofproto->name, name)) {
6565 return ofproto;
6566 }
6567 }
6568 return NULL;
6569 }
6570
6571 static void
6572 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6573 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6574 {
6575 struct ofproto *ofproto;
6576 struct ds results;
6577
6578 ds_init(&results);
6579 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6580 ds_put_format(&results, "%s\n", ofproto->name);
6581 }
6582 unixctl_command_reply(conn, ds_cstr(&results));
6583 ds_destroy(&results);
6584 }
6585
6586 static void
6587 ofproto_unixctl_init(void)
6588 {
6589 static bool registered;
6590 if (registered) {
6591 return;
6592 }
6593 registered = true;
6594
6595 unixctl_command_register("ofproto/list", "", 0, 0,
6596 ofproto_unixctl_list, NULL);
6597 }
6598 \f
6599 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6600 *
6601 * This is deprecated. It is only for compatibility with broken device drivers
6602 * in old versions of Linux that do not properly support VLANs when VLAN
6603 * devices are not used. When broken device drivers are no longer in
6604 * widespread use, we will delete these interfaces. */
6605
6606 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6607 * (exactly) by an OpenFlow rule in 'ofproto'. */
6608 void
6609 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6610 {
6611 const struct oftable *oftable;
6612
6613 free(ofproto->vlan_bitmap);
6614 ofproto->vlan_bitmap = bitmap_allocate(4096);
6615 ofproto->vlans_changed = false;
6616
6617 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6618 const struct cls_table *table;
6619
6620 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6621 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
6622 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6623 const struct cls_rule *rule;
6624
6625 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6626 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6627 bitmap_set1(vlan_bitmap, vid);
6628 bitmap_set1(ofproto->vlan_bitmap, vid);
6629 }
6630 }
6631 }
6632 ovs_rwlock_unlock(&oftable->cls.rwlock);
6633 }
6634 }
6635
6636 /* Returns true if new VLANs have come into use by the flow table since the
6637 * last call to ofproto_get_vlan_usage().
6638 *
6639 * We don't track when old VLANs stop being used. */
6640 bool
6641 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6642 {
6643 return ofproto->vlans_changed;
6644 }
6645
6646 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6647 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6648 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6649 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6650 * then the VLAN device is un-enslaved. */
6651 int
6652 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6653 ofp_port_t realdev_ofp_port, int vid)
6654 {
6655 struct ofport *ofport;
6656 int error;
6657
6658 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6659
6660 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6661 if (!ofport) {
6662 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6663 ofproto->name, vlandev_ofp_port);
6664 return EINVAL;
6665 }
6666
6667 if (!ofproto->ofproto_class->set_realdev) {
6668 if (!vlandev_ofp_port) {
6669 return 0;
6670 }
6671 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6672 return EOPNOTSUPP;
6673 }
6674
6675 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6676 if (error) {
6677 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6678 ofproto->name, vlandev_ofp_port,
6679 netdev_get_name(ofport->netdev), ovs_strerror(error));
6680 }
6681 return error;
6682 }